Mathematical

Economics
Application of Fractional
Calculus
Edited by
Vasily E. Tarasov
Printed Edition of the Special Issue Published in Mathematics

www.mdpi.com/journal/mathematics
Mathematical Economics
Mathematical Economics
Application of Fractional Calculus

Special Issue Editor

Vasily E. Tarasov

MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin
Special Issue Editor
Vasily E. Tarasov
Skobeltsyn Institute of Nuclear Physics,
Lomonosov Moscow State University
Information Technologies and Applied Mathematics Faculty,
Moscow Aviation Institute (National Research University)
Russia

Editorial Office
MDPI
St. Alban-Anlage 66
4052 Basel, Switzerland

This is a reprint of articles from the Special Issue published online in the open access journal
Mathematics (ISSN 2227-7390) (available at: https://www.mdpi.com/journal/mathematics/special
issues/Mathematical Economics).

For citation purposes, cite each article independently as indicated on the article page online and as
indicated below:

LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. Journal Name Year, Article Number,
Page Range.

ISBN 978-3-03936-118-2 (Pbk)

ISBN 978-3-03936-119-9 (PDF)


c 2020 by the authors. Articles in this book are Open Access and distributed under the Creative
Commons Attribution (CC BY) license, which allows users to download, copy and build upon
published articles, as long as the author and publisher are properly credited, which ensures maximum
dissemination and a wider impact of our publications.
The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons
license CC BY-NC-ND.
Contents

About the Special Issue Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Vasily E. Tarasov
Mathematical Economics: Application of Fractional Calculus
Reprinted from: Mathematics 2020, 8, 660, doi:10.3390/math8050660 . . . . . . . . . . . . . . . . . 1

Vasily E. Tarasov
On History of Mathematical Economics: Application of Fractional Calculus
Reprinted from: Mathematics 2019, 7, 509, doi:10.3390/math7060509 . . . . . . . . . . . . . . . . . 5

Francesco Mainardi
On the Advent of Fractional Calculus in Econophysics via Continuous-Time Random Walk
Reprinted from: Mathematics 2020, 8, 641, doi:10.3390/math8040641 . . . . . . . . . . . . . . . . . 33

Vasily E. Tarasov
Rules for Fractional-Dynamic Generalizations: Difficulties of Constructing Fractional
Dynamic Models
Reprinted from: Mathematics 2019, 7, 554, doi:10.3390/math7060554 . . . . . . . . . . . . . . . . . 43

Anatoly N. Kochubei and Yuri Kondratiev

Growth Equation of the General Fractional Calculus
Reprinted from: Mathematics 2019, 7, 615, doi:10.3390/math7070615 . . . . . . . . . . . . . . . . . 93

Jean-Philippe Aguilar, Jan Korbel and Yuri Luchko

Applications of the Fractional Diffusion Equation to Option Pricing and Risk Calculations
Reprinted from: Mathematics 2019, 7, 796, doi:10.3390/math7090796 . . . . . . . . . . . . . . . . . 101

Jonathan Blackledge, Derek Kearney, Marc Lamphiere, Raja Rani and Paddy Walsh
Econophysics and Fractional Calculus: Einstein’s Evolution Equation, the Fractal Market
Hypothesis, Trend Analysis and Future Price Prediction
Reprinted from: Mathematics 2019, 7, 1057, doi:10.3390/math7111057 . . . . . . . . . . . . . . . . 125

Tomas Skovranek
The Mittag-Leffler Fitting of the Phillips Curve
Reprinted from: Mathematics 2019, 7, 589, doi:10.3390/math7070589 . . . . . . . . . . . . . . . . . 183

Yingkang Xie, Zhen Wang and Bo Meng

Stability and Bifurcation of a Delayed Time-Fractional Order Business Cycle Model
with a General Liquidity Preference Function and Investment Function
Reprinted from: Mathematics 2019, 7, 846, doi:10.3390/math7090846 . . . . . . . . . . . . . . . . . 195

José A. Tenreiro Machado, Maria Eugénia Mata and António M. Lopes

Fractional Dynamics and Pseudo-Phase Space of Country Economic Processes
Reprinted from: Mathematics 2020, 8, 81, doi:10.3390/math8010081 . . . . . . . . . . . . . . . . . . 205

Inés Tejado, Emiliano Pérez and Duarte Valério

Fractional Derivatives for Economic Growth Modelling of the Group of Twenty:
Application to Prediction
Reprinted from: Mathematics 2020, 8, 50, doi:10.3390/math8010050 . . . . . . . . . . . . . . . . . . 223

v
Hao Ming, JinRong Wang and Michal Fečkan
The Application of Fractional Calculus in Chinese Economic Growth Models
Reprinted from: Mathematics 2019, 7, 665, doi:10.3390/math7080665 . . . . . . . . . . . . . . . . . 245

Ertuğrul Karaçuha, Vasil Tabatadze, Kamil Karaçuha, Nisa Özge Önal and Esra Ergün
Deep Assessment Methodology Using Fractional Calculus on Mathematical Modeling and
Prediction of Gross Domestic Product per Capita of Countries
Reprinted from: Mathematics 2020, 8, 633, doi:10.3390/math8040633 . . . . . . . . . . . . . . . . . 251

vi
About the Special Issue Editor
Vasily E. Tarasov is a leading researcher with the Skobeltsyn Institute of Nuclear Physics,
Lomonosov Moscow State University, and Professor with the Information Technologies and Applied
Mathematics Faculty, Moscow Aviation Institute (National Research University). His fields of
research include several topics of applied mathematics, theoretical physics, mathematical economics.
He has published over 200 papers in international peer-reviewed scientific journals, including
over 50 papers in mathematical economics. He has published 12 scientific books, including
“Fractional Dynamics: Application of Fractional Calculus to Dynamics of Particles, Fields and Media”
(Springer, 2010) and “Quantum Mechanics of Non-Hamiltonian and Dissipative Systems” (Elsevier,
2008). He was an Editor of “Handbook of Fractional Calculus with Applications. Volumes 4
and 5” (De Gruyter, 2019). Vasily E. Tarasov is a member of the Editorial Boards of 12 journals,
including Fractional Calculus and Applied Analysis (De Gruyter), The European Physical Journal Plus
(Springer), Communications in Nonlinear Science and Numerical Simulations (Elsevier), Entropy (MDPI),
Computational and Applied Mathematics (Springer), International Journal of Applied and Computational
Mathematics (Springer), Mathematics (MDPI), Journal of Applied Nonlinear Dynamics (LH Scientific
Publishing LLC), Fractional Differential Calculus (Ele-Math), Fractal and Fractional (MDPI), and others.
At present, his h-index is 35 in Web of Science and Scopus with an h-index of 45 on Google
Scholar Citations. Web of Science ResearcherID: D-6851-2012; Scopus Author ID: 7202004582;
ORCID: 0000-0002-4718-6274; Google Scholar Citations ID: sfFN5B4AAAAJ; IstinaResearcher ID
(IRID): 394056.

vii
 mathematics
Editorial
Mathematical Economics: Application of
Fractional Calculus
Vasily E. Tarasov 1,2
1 Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, 119991 Moscow, Russia;
[email protected]; Tel.: +7-495-939-5989
2 Faculty “Information Technologies and Applied Mathematics”, Moscow Aviation Institute (National
Research University), 125993 Moscow, Russia

Received: 22 April 2020; Accepted: 24 April 2020; Published: 27 April 2020 

Mathematical economics is a theoretical and applied science in which economic objects, processes,
and phenomena are described by using mathematically formalized language. In this science, models are
formulated on the basis of mathematical formalizations of economic concepts and notions. An important
purpose of mathematical economics is the formulation of notions and concepts in form, which will
be mathematically adequate and self-consistent, and then, on their basis, to construct models of
processes and phenomena of economy. The standard mathematical language, which is actively used
in mathematical modeling of economy, is the calculus of derivatives and integrals of integer orders,
the differential and difference equations. These operators and equations allowed economists to
formulate models in mathematical form, and, on this basis, to describe a wide range of processes and
phenomena in economy. It is known that the integer-order derivatives of functions are determined
by the properties of these functions in an infinitely small neighborhood of the point, in which the
derivatives are considered. As a result, economic models, which are based on differential equations of
integer orders, cannot describe processes with memory and non-locality. As a result, this mathematical
language cannot take into account important aspects of economic processes and phenomena.
Fractional calculus is a branch of mathematics that studies the properties of differential and
integral operators that are characterized by real or complex orders. The methods of fractional calculus
are powerful tools for describing the processes and systems with memory and nonlocality. There are
various types of fractional integral and differential operators that are proposed by Riemann, Liouville,
Grunwald, Letnikov, Sonine, Marchaud, Weyl, Riesz, Hadamard, Kober, Erdelyi, Caputo and other
mathematicians. The fractional derivatives have a set of nonstandard properties such as a violation
of the standard product and chain rules. The violation of the standard form of the product rule is a
main characteristic property of derivatives of non-integer orders that allows us to describe complex
properties of processes and systems.
Recently, fractional integro-differential equations are actively used to describe a wide class
of economical processes with power-law memory and spatial nonlocality. Generalizations of
basic economic concepts and notions of the economic processes with memory were proposed.
New mathematical models with continuous time are proposed to describe the economic dynamics
with a long memory.
The purpose of this Special Issue is to create a collection of articles reflecting the latest mathematical
and conceptual developments in mathematical economics with memory and non-locality, based on
applications of modern fractional calculus.
The proposed collection of works can be conditionally divided into three parts: historical,
mathematical and applied.
This collection opens with two review articles, [1], by Vasily E. Tarasov, and [2], by Francesco
Mainardi, purpose of which is a brief description of the history of the application of fractional calculus
in economics and finance.

Mathematics 2020, 8, 660; doi:10.3390/math8050660 1 www.mdpi.com/journal/mathematics

Mathematics 2020, 8, 660

The collection continues with a review work, [3], by Vasily E. Tarasov, the purpose of which is a
description of the problems and difficulties arising in the construction of fractional-dynamic analogs
of standard models by using fractional calculus. In article [4], by Anatoly N. Kochubei and Yuri
Kondratiev, the authors proposed correct mathematical statements for growth models with memory in
general cases, for application in mathematical economics of processes with memory and distributed
lag. In article [5], by Jean-Philippe Aguilar, Jan Korbel and Yuri Luchko, applications of the fractional
diffusion equation to option pricing and risk calculations are described. In work [6], by Jonathan
Blackledge, Derek Kearney, Marc Lamphiere, Raja Rani and Paddy Walsh, authors discuss a range
of results that are connected to Einstein’s evolution equation, focusing on the Lévy distribution.
In article [7], by Tomas Skovranek, a mathematical model, which is based on the one-parameter
Mittag-Leffler function, is proposed to describe the relation between the unemployment rate and the
inflation rate, also known as the Phillips curve. In article [8], by Yingkang Xie, Zhen Wang and Bo
Meng, it is considered a fractional generalization of business cycle model with memory and time delay.
Further, this collection continues with works in which fractional calculus is applied to describe
economy of different countries. In article [9], by José A. Tenreiro Machado, Maria Eugénia Mata and
António M. Lopes, the fractional calculus and concept of pseudo-phase space are used for modeling the
dynamics of world economies and forecasting a country’s gross domestic product. In work [10], by Inés
Tejado, Emiliano Pérez and Duarte Valério, the fractional calculus is applied to study the economic
growth of the countries in the Group of Twenty (G20). In article [11], by Hao Ming, JinRong Wang and
Michal Fečkan, the application of fractional calculus to economic growth models of Chinese economy
is proposed. In work [12], by Ertuğrul Karaçuha, Vasil Tabatadze, Kamil Karaçuha, Nisa Özge Önal
and Esra Ergün, the fractional calculus approach and the time series modeling are applied to describe
the Gross Domestic Product (GDP) per capita for nine countries (Brazil, China, India, Italy, Japan, UK,
USA, Spain and Turkey) and the European Union.

Funding: This research received no external funding.

Conflicts of Interest: The author declares no conflict of interest.

References
1. Tarasov, V.E. On History of Mathematical Economics: Application of Fractional Calculus. Mathematics 2019,
7, 509. [CrossRef]
2. Mainardi, F. On the Advent of Fractional Calculus in Econophysics via Continuous-Time Random Walk.
Mathematics 2020, 8, 641. [CrossRef]
3. Tarasov, V.E. Rules for Fractional-Dynamic Generalizations: Difficulties of Constructing Fractional Dynamic
Models. Mathematics 2019, 7, 554. [CrossRef]
4. Kochubei, A.N.; Kondratiev, Y. Growth Equation of the General Fractional Calculus. Mathematics 2019, 7, 615.
[CrossRef]
5. Aguilar, J.-P.; Korbel, J.; Luchko, Y. Applications of the Fractional Diffusion Equation to Option Pricing and
Risk Calculations. Mathematics 2019, 7, 796. [CrossRef]
6. Blackledge, J.; Kearney, D.; Lamphiere, M.; Rani, R.; Walsh, P. Econophysics and Fractional Calculus:
Einstein’s Evolution Equation, the Fractal Market Hypothesis, Trend Analysis and Future Price Prediction.
Mathematics 2019, 7, 1057. [CrossRef]
7. Skovranek, T. The Mittag-Leffler Fitting of the Phillips Curve. Mathematics 2019, 7, 589. [CrossRef]
8. Xie, Y.; Wang, Z.; Meng, B. Stability and Bifurcation of a Delayed Time-Fractional Order Business Cycle
Model with a General Liquidity Preference Function and Investment Function. Mathematics 2019, 7, 846.
[CrossRef]
9. Tenreiro Machado, J.A.; Mata, M.E.; Lopes, A.M. Fractional Dynamics and Pseudo-Phase Space of Country
Economic Processes. Mathematics 2020, 8, 81. [CrossRef]
10. Tejado, I.; Pérez, E.; Valério, D. Fractional Derivatives for Economic Growth Modelling of the Group of
Twenty: Application to Prediction. Mathematics 2020, 8, 50. [CrossRef]

2
Mathematics 2020, 8, 660

11. Ming, H.; Wang, J.; Fečkan, M. The Application of Fractional Calculus in Chinese Economic Growth Models.
Mathematics 2019, 7, 665. [CrossRef]
12. Karaçuha, E.; Tabatadze, V.; Karaçuha, K.; Önal, N.Ö.; Ergün, E. Deep Assessment Methodology Using
Fractional Calculus on Mathematical Modeling and Prediction of Gross Domestic Product per Capita of
Countries. Mathematics 2020, 8, 633. [CrossRef]

© 2020 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

3
 mathematics
Review
On History of Mathematical Economics: Application
of Fractional Calculus
Vasily E. Tarasov
Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow 119991, Russia;
[email protected]; Tel.: +7-495-939-5989

Received: 15 May 2019; Accepted: 31 May 2019; Published: 4 June 2019 

Abstract: Modern economics was born in the Marginal revolution and the Keynesian revolution.
These revolutions led to the emergence of fundamental concepts and methods in economic theory,
which allow the use of differential and integral calculus to describe economic phenomena, effects, and
processes. At the present moment the new revolution, which can be called “Memory revolution”, is
actually taking place in modern economics. This revolution is intended to “cure amnesia” of modern
economic theory, which is caused by the use of differential and integral operators of integer orders.
In economics, the description of economic processes should take into account that the behavior of
economic agents may depend on the history of previous changes in economy. The main mathematical
tool designed to “cure amnesia” in economics is fractional calculus that is a theory of integrals,
derivatives, sums, and differences of non-integer orders. This paper contains a brief review of the
history of applications of fractional calculus in modern mathematical economics and economic theory.
The first stage of the Memory Revolution in economics is associated with the works published in 1966
and 1980 by Clive W. J. Granger, who received the Nobel Memorial Prize in Economic Sciences in
2003. We divide the history of the application of fractional calculus in economics into the following
five stages of development (approaches): ARFIMA; fractional Brownian motion; econophysics;
deterministic chaos; mathematical economics. The modern stage (mathematical economics) of the
Memory revolution is intended to include in the modern economic theory new economic concepts
and notions that allow us to take into account the presence of memory in economic processes. The
current stage actually absorbs the Granger approach based on ARFIMA models that used only the
Granger–Joyeux–Hosking fractional differencing and integrating, which really are the well-known
Grunwald–Letnikov fractional differences. The modern stage can also absorb other approaches by
formulation of new economic notions, concepts, effects, phenomena, and principles. Some comments
on possible future directions for development of the fractional mathematical economics are proposed.

Keywords: mathematical economics; economic theory; fractional calculus; fractional dynamics; long
memory; non-locality

1. Introduction: General Remarks about Mathematical Economics

Mathematical economics is a theoretical and applied science, whose purpose is a mathematically
formalized description of economic objects, processes, and phenomena. Most of the economic theories
are presented in terms of economic models. In mathematical economics, the properties of these models
are studied based on formalizations of economic concepts and notions. In mathematical economics,
theorems on the existence of extreme values of certain parameters are proved, properties of equilibrium
states and equilibrium growth trajectories are studied, etc. This creates the impression that the proof of
the existence of a solution (optimal or equilibrium) and its calculation is the main aim of mathematical
economics. In reality, the most important purpose is to formulate economic notions and concepts in
mathematical form, which will be mathematically adequate and self-consistent, and then, on their basis

Mathematics 2019, 7, 509; doi:10.3390/math7060509 5 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 509

to construct mathematical models of economic processes and phenomena. Moreover, it is not enough
to prove the existence of a solution and find it in an analytic or numerical form, but it is necessary to
give an economic interpretation of these obtained mathematical results.
We can say that modern mathematical economics began in the 19th century with the use of
differential (and integral) calculus to describe and explain economic behavior. The emergence of
modern economic theory occurred almost simultaneously with the appearance of new economic
concepts, which were actively used in various economic models. “Marginal revolution” and “Keynesian
revolution” in economics led to the introduction of the new fundamental concepts into economic theory,
which allow the use of mathematical tools to describe economic phenomena and processes. The most
important mathematical tools that have become actively used in mathematical modeling of economic
processes are the theory of derivatives and integrals of integer orders, the theory of differential and
difference equations. These mathematical tools allowed economists to build economic models in a
mathematical form and on their basis to describe a wide range of economic processes and phenomena.
However, these tools have a number of shortcomings that lead to the incompleteness of descriptions
of economic processes. It is known that the integer-order derivatives of functions are determined
by the properties of these functions in an infinitely small neighborhood of the point, in which the
derivatives are considered. As a result, differential equations with derivatives of integer orders, which
are used in economic models, cannot describe processes with memory and non-locality. In fact, such
equations describe only economic processes, in which all economic agents have complete amnesia and
interact only with the nearest neighbors. Obviously, this assumption about the lack of memory among
economic agents is a strong restriction for economic models. As a result, these models have drawbacks,
since they cannot take into account important aspects of economic processes and phenomena.

2. A Short History of Fractional Mathematical Economics

“Marginal revolution” and “Keynesian revolution” introduced fundamental economic concepts,
including the concepts of “marginal value”, “economic multiplier”, “economic accelerator”, “elasticity”
and many others. These revolutions led to the use of mathematical tools based on the derivatives and
integrals of integer orders, and the differential and difference equations. As a result, the economic
models with continuous and discrete time began to be mathematically described by differential
equations with derivatives of integer orders or difference equations of integer orders.
It can be said that at the present moment new revolutionary changes are actually taking place
in modern economics. These changes can be called a revolution of memory and non-locality. It is
becoming increasingly obvious in economics that when describing the behavior of economic agents,
we must take into account that their behavior may depend on the history of previous changes in the
economy. In economic theory, we need new economic concepts and notions that allow us to take into
account the presence of memory in economic agents. New economic models and methods are needed,
which take into account that economic agents may remember the changes of economic indicators
and factors in the past, and that this affects the behavior of agents and their decision making. To
describe this behavior we cannot use the standard mathematical apparatus of differential (or difference)
equations of integer orders. In fact, these equations describe only such economic processes, in which
agents actually have an amnesia. In other words, economic models, which use only derivatives
of integer orders, can be applied when economic agents forget the history of changes of economic
indicators and factors during an infinitesimally small period of time. At the moment it is becoming
clear that this restriction holds back the development of economic theory and mathematical economics.
In modern mathematics, derivatives and integrals of arbitrary order are well known [1–5]. The
derivative (or integral), order of which is a real or complex number and not just an integer, is
called fractional derivative and integral. Fractional calculus as a theory of such operators has a long
history [6–15]. There are different types of fractional integral and differential operators [1–5]. For
fractional differential and integral operators, many standard properties are violated, including such
properties as the standard product (Leibniz) rule, the standard chain rule, the semi-group property

6
Mathematics 2019, 7, 509

for orders of derivatives, the semi-group property for dynamic maps [16–21]. We can state that
the violation of the standard form of the Leibniz rule is a characteristic property of derivatives of
non-integer orders [16]. The most important application of fractional derivatives and integrals of
non-integer order is fading memory and spatial non-locality.
The new revolution (“Memory revolution”) is intended to include in the modern economic
theory and mathematical economics different processes with long memory and non-locality. The main
mathematical tool designed to “cure amnesia” in economics is the theory of derivatives and integrals
of non-integer order (fractional calculus), fractional differential and difference equations [1–5]. This
revolution has led to the emergence of a new branch of mathematical economics, which can be called
“fractional mathematical economics.”
Fractional mathematical economics is a theory of fractional dynamic models of economic processes,
phenomena and effects. In this framework of mathematical economics, the fractional calculus methods
are being developed for application to problems of economics and finance. The field of fractional
mathematical economics is the application of fractional calculus to solve problems in economics (and
finance) and for the development of fractional calculus for such applications. Fractional mathematical
economics can be considered as a branch of applied mathematics that deals with economic problems.
However, this point of view is obviously a narrowing of the field of research, goals and objectives of
this area. An important part of fractional mathematical economics is the use of fractional calculus to
formulate new economic concepts, notions, effects and phenomena. This is especially important due
to the fact that the fractional mathematical economics is now only being formed as an independent
science. Moreover, the development of the fractional calculus itself and its generalizations will largely
be determined precisely by such goals and objectives in economics, physics and other sciences.
This “Memory revolution” in the economics, or rather the first stage of this revolution, can be
associated with the works, which were published in 1966 and 1980 by Clive W. J. Granger [22–26], who
received the Nobel Memorial Prize in Economic Sciences in 2003 [27].
The history of the application of fractional calculus in economics can be divided into the following
stages of development (approaches): ARFIMA; fractional Brownian motion; econophysics; deterministic
chaos; mathematical economics. The appearance of a new stage obviously does not mean the cessation
of the development of the previous stage, just as the appearance of quantum theory did not stop the
development of classical mechanics.
Further in Sections 2.1–2.5, we briefly describe these stages of development, and then in Section 3
we outline possible ways for the further development of fractional mathematical economics.

2.1. ARFIMA Stage (Approach)

ARFIMA Stage (Approach): This stage is characterized by models with discrete time and
application of the Grunwald–Letnikov fractional differences.
More than fifty years ago, Clive W. J. Granger (see preprint [22], paper [23], the collection of
the works [24,25]) was the first to point out long-term dependencies in economic data. The articles
demonstrated that spectral densities derived from the economic time series have a similar shape. This
fact allows us to say that the effect of long memory in the economic processes was found by Granger.
Note that, he received the Nobel Memorial Prize in Economics in 2003 “for methods of analyzing
economic time series with general trends (cointegration)” [27].
Then, Granger and Joyeux [26], and Hosking [28] proposed the fractional generalization of
ARIMA(p,d,q) models (the ARFIMA (p,d,q) models) that improved the statistical methods for
researching of processes with memory. As the main mathematical tool for describing memory,
fractional differencing and integrating (for example, see books [29–34] and reviews [35–38]) were
proposed for discrete time case. The suggested generalization of the ARIMA(p,d,q) model is realized
by considering non-integer (positive and negative) order d instead of positive integer values of d. The
Granger–Joyeux–Hosking (GJH) operators were proposed and used without relationship with the
fractional calculus. As was proved in [39,40], these GJH operators are actually the Grunwald–Letnikov

7
Mathematics 2019, 7, 509

fractional differences (GLF-difference), which have been suggested more than a hundred and fifty years
ago and are used in the modern fractional calculus [1,3]. We emphasize that in the continuous limit
these GLF-differences give the GLF-derivatives that coincide with the Marchaud fractional derivatives
(see Theorem 4.2 and Theorem 4.4 of [1]).
Among economists, the approach proposed by Gravers (and based on the discrete operators
proposed by them) is the most common and is used without an explicit connection with the
development of fractional calculus. It is obvious that the restriction of mathematical tools only
to the Grunwald–Letnikov fractional differences significantly reduces the possibilities for studying
processes with memory and non-locality. The use of fractional calculus in economic models will
significantly expand the scope and allows us to obtain new results.

2.2. Fractional Brownian Motion (Mathematical Finance) Stage (Approach)

Fractional Brownian Motion Stage (Approach): This stage is characterized by financial models
and the application of stochastic calculus methods and stochastic differential equations.
Andrey N. Kolmogorov, who is one of the founders of modern probability theory, was the first
who considered in 1940 [41] the continuous Gaussian processes with stationary increments and with
the self-similarity property A.N. Kolmogorov called such Gaussian processes “Wiener Spirals”. Its
modern name is the fractional Brownian motion that can be considered as a continuous self-similar
zero-mean Gaussian process and with stationary increments.
Starting with the article by L.C.G. Rogers [42], various authors began to consider the use of
fractional Brownian motion to describe different financial processes. The fractional Brownian motion
is not a semi-martingale and the stochastic integral with respect to it is not well-defined in the classical
Ito’s sense. Therefore, this approach is connected with the development of fractional stochastic
calculus [43–45]. For example, in the paper [43] a stochastic integration calculus for the fractional
Brownian motion based on the Wick product was suggested.
At the present time, this stage (approach), which can be called as a fractional mathematical finance,
is connected with the development of fractional stochastic calculus, the theory fractional stochastic
differential equations and their application in finance. The fractional mathematical finance is a field
of applied mathematics, concerned with mathematical modeling of financial markets by using the
fractional stochastic differential equations.
As a special case of fractional mathematical finance, we can note the fractional generalization of
the Black–Scholes pricing model. In 1973, Fischer Black and Myron Scholes [46] derived the famous
theoretical valuation formula for options. In 1997, the Royal Swedish Academy of Sciences has decided
to award the Bank of Sweden Prize in Economic Sciences in Memory of Alfred Nobel [47] to Myron
S. Scholes, for the so-called Black–Scholes model published in 1973: “Robert C. Merton and Myron
S. Scholes have, in collaboration with the late Fischer Black, developed a pioneering formula for the
valuation of stock options.” [47].)
For the first time a fractional generalization of the Black–Scholes equation was proposed in [48] by
Walter Wyss in 2000. Wyss [48] considered the pricing of option derivatives by using the time-fractional
Black–Scholes equation and derived a closed form solution for European vanilla options. The
Black–Scholes equation is generalized by replacing the first derivative in time by a fractional derivative
in time of the order α ∈ (0, 1). The solution of this fractional Black–Scholes equation is considered.
However, in the Wyss paper, there are no financial reasons to explain why a time-fractional derivative
should be used.
The works of Cartea and Meyer-Brandis [49] and Cartea [50] proposed a stock price model
that uses information about the waiting time between trades. In this model the arrival of trades is
driven by a counting process, in which the waiting-time between trades processes is described by the
Mittag–Leffler survival function (see also [51]). In the paper [50], Cartea proposed that the value of
derivatives satisfies the fractional Black–Scholes equation that contains the Caputo fractional derivative

8
Mathematics 2019, 7, 509

with respect to time. It should be noted that, in general, the presence of a waiting time and a delay
time does not mean the presence of memory in the process.
In the framework of the fractional Brownian motion Stage, a lot of papers [50–71] and books [72,73]
were written on the description of financial processes with memory and non-locality.
As a rule, in fractional mathematical finance, fractional dynamic models are created without
establishing links with economic theory and without formulating new economic or financial concepts,
taking only observable market prices as input data. In the fractional mathematical finance, the main
requirement is the mathematical consistency and the compatibility with economic theory is not the
key point.

2.3. Econophysics Stage (Approach)

Econophysics Stage (Approach): This stage is characterized by financial models and the
application of physical methods and equations.
Twenty years ago, a new branch of the econophysics, which is connected with the application of
fractional calculus, has appeared. In fact, this branch, which can be called fractional econophysics, was
born in 2000 and it can be primarily associated with the works of Francesco Mainardi, Rudolf Gorenflo,
Enrico Scalas, Marco Raberto [74–76] on the continuous-time finance.
In fractional econophysics, the fractional diffusion models [74–76] are used in finance, where price
jumps replace the particle jumps in the physical diffusion model. The corresponding stochastic models
are called continuous time random walks (CTRWs), which are random walks that also incorporate a
random waiting time between jumps. In finance, the waiting times measure delay between transactions.
These two random variables (price change and waiting time) are used to describe the long-time behavior
in financial markets. The diffusion (hydrodynamic) limit, which is used in physics, is considered for
continuous time random walks [74–76]. It was shown that the probability density function for the
limit process obeys a fractional diffusion equation [74–76].
After the pioneering works [74–76] that laid the foundation for the new direction of econophysics
(fractional econophysics), various papers were written on the application of fractional dynamics
methods and physical models to describe processes in finance and economics (for example, see [77–84]).
The history and achievements of the econophysics stage in the first five years are described by Enrico
Scalas in the article [85] in 2016.
The fractional econophysics, as a branch of econophysics, can be defined as a new direction of
research applying methods developed in physical sciences, to describe processes in economics and
finance, basically those including power-law memory and spatial non-locality. The mathematical tool
of this branch of econophysics is the fractional calculus. For example, application to the study of
continuous time finance by using methods and results of fractional kinetics and anomalous diffusion.
Another example, which is not related to finance, is the time-dependent fractional dynamics with
memory in quantum and economic physics [86].
In this stage, the fractional calculus was applied mainly to financial processes. In the papers
on fractional econophysics, generalization of basic economic concepts and principles for economic
processes with memory (and non-locality) are not suggested.
Unfortunately, economists do not always understand the analogies with the methods and concepts
of modern physics, which restricts the possibilities for economists to use this approach. As a result, it
holds back and limits both the development and application of the fractional econophysics approach
to describing economic processes with memory and non-locality.
It can be said that the time has come for economists and econophysicists to work together on the
formulation of economic analogs of physical concepts and methods used in fractional econophysics,
and linking them with existing concepts and methods of economic theory. For the development of a
fractional mathematical economics, a translation should be made from the language of physics into the
language of economics.

9
Mathematics 2019, 7, 509

2.4. Deterministic Chaos Stage (Approach)

Deterministic Chaos Stage (Approach): This stage is characterized by financial (and economic)
models and application of methods of nonlinear dynamics. Strictly speaking, this approach should be
attributed to the econophysics stage/approach.
Nonlinear dynamics models are useful to explain irregular and chaotic behavior of complex
economic and financial processes. The complex behaviors of nonlinear economic processes restrict the
use of analytical methods to study nonlinear economic models.
In 2008, for the first time, Wei-Ching Chen proposed in [87] a fractional generalization of a financial
model with deterministic chaos. Chen [87] studied the fractional-dynamic behaviors and describes
fixed points, periodic motions, chaotic motions, and identified period doubling and intermittency
routes to chaos in the financial process that is described by a system of three equations with the Caputo
fractional derivatives. He demonstrates by numerical simulations that chaos exists when orders of
derivatives are less than three and that the lowest order at which chaos exists was 2.35. The work [88]
studied the chaos control method of such a kind of system by feedback control, respectively.
In the framework of the deterministic chaos stage, many papers [89–99] have been devoted to
the description of financial processes with memory. In some papers [100–105], economic models
were considered.
We should note that the various stages/approaches of development of fractional mathematical
economics did not develop in complete isolation from each other. For example, for the fractional Chen
model of dynamic chaos in the economy, Tomáš Škovránek, Igor Podlubny, Ivo Petráš [106] applied
the concept of the state space (the configuration space, the phase space) that arose in physics more
than a hundred years ago. As state variables authors consider the gross domestic product, inflation,
and unemployment rate. The dynamics of the modeled economy in time, which is represented by
the values of these three variables, is described as a trajectory in state-space. The system of three
fractional order differential equations is used to describe dynamics of the economy by fitting the
available economic data. Then José A. Tenreiro Machado, Maria E. Mata, Antonio M. Lopes suggested
the development of the state space concept in the papers [107–109]. The economic growth is described
by using the multidimensional scaling (MDS) method for visualizing information in data. The state
space is used to represent the sequence of points (the fractional state space portrait, FSSP, and pseudo
phase plane, PPP) corresponding to the states over time.

2.5. Mathematical Economics Stage (Approach)

Mathematical Economics Stage (Approach): This stage is characterized by macroeconomic
and microeconomic models with continuous time and generalization of basic economic concepts
and notions.
The fractional calculus approach has been used to describe the concept of memory itself for
economic processes in [39,40,110–117], and to define basic concepts of economic processes with memory
and non-locality in works [118–139].
From a subjective point of view, this stage began with a proposal of generalizations of the
basic economic concepts and notions at the beginning of 2016, when the concept of elasticity for
economic processes with memory was proposed [132–135]. Then in 2016, the concepts of the marginal
values with memory [118–120], the concept of accelerator and multiplier with memory [123,130]
and others were suggested [132–136,138]. These concepts are used in fractional generalizations of
some standard economic models [140–167] with the continuous time, which were proposed in 2016
and subsequent years [168–189]. These dynamic models describe fractional dynamics of economic
processes with memory.
The fractional calculus approach has been used to define basic concepts of economic processes with
memory in works [118–139], and to describe dynamics of economic processes with memory [168–189]
in the framework of the continuous time models.

10
Mathematics 2019, 7, 509

It should be noted that formal replacements of derivatives of integer order by fractional derivatives
in standard differential equations, which describe economic processes, and solutions of the obtained
fractional differential equations were considered in papers published before 2016. However, these
papers were purely mathematical works, in which generalizations of economic concepts and notions
were not proposed. In these works, fractional differential equations have not been derived, since a
formal replacement of integer-order derivatives by fractional derivatives cannot be recognized as
a derivation of the equations. Formulations of economic conclusions and interpretations from the
obtained solutions are not usually suggested in these papers. Examples of incorrectness and errors
in such generalizations are given in the work [189]. In the paper [189], we formulate five principles
of the fractional-dynamic generalization of standard dynamic models and then we illustrate these
principles by examples from fractional mathematical economics. We can state that in the works with
formal fractional generalizations of standard economic equations the Principles of Derivability and
Interpretability [189] were neglected. Let us give a brief formulation of the Principles of Derivability
and Interpretability.
Derivability Principle: It is not enough to generalize the differential equations describing the
dynamic model. It is necessary to generalize the whole scheme of obtaining (all steps of derivation)
these equations from the basic principles, concepts and assumptions. In this sequential derivation
of the equations we should take into account the non-standard characteristic properties of fractional
derivatives and integrals. If necessary, generalizations of the notions, concepts and methods, which
are used in this derivation, should also be obtained.
Interpretability Principle: The subject (physical, economic) interpretation of the mathematical
results, including solutions and their properties, should be obtained. Differences, and first of all
qualitative differences, from the results based on the standard model should be described.
The most important purpose of the modern stage of development of fractional mathematical
economics is the inclusion of memory and non-locality into the economic theory, into the basic
economic concepts and methods. The economics should be extended and generalized such that it
takes into account the memory and non-locality. Fractional generalizations of standard economic
models should be constructed only on this conceptual basis. The most important purpose of studying
such generalizations is the search and formulation of qualitatively new effects and phenomena caused
by memory and non-locality in the behavior of economic processes. In this case, these results in
mathematical economics, which are based on fractional calculus, can be further used in computer
simulations of real economic processes and in econometric studies.
Let us list some generalizations of economic concepts and fractional generalizations of economic
models that have already been proposed in recent years. Using the fractional calculus approach to
describe the processes with memory and non-locality, the generalizations of some basic economic
notions were proposed in the works [118–139]. The list of these new notions and concepts primarily
include the following:

• The marginal value of non-integer order [118–122,190] with memory and non-locality;
• The economic multiplier with memory [123,124];
• The economic accelerator with memory [123,124];
• The exact discretization of economic accelarators and multiplier [125–128] based on exact fractional
differences [129];
• The accelerator with memory and crisis periodic sharp bursts [130,131];
• The duality of the multiplier with memory and the accelerator with memory [123,124];
• The elasticity of fractional order [132–135] for processes with memory and non-locality;
• The measures of risk aversion with non-locality [136] and with memory [137];
• The warranted (technological) rate of growth with memory [112,170,174–176,189];
• The non-local methods of deterministic factor analysis for [138,139];

And some other.

11
Mathematics 2019, 7, 509

The use of these notions and concepts makes it possible for us to generalize some classical
economic models, including those proposed by the following well-known economists:

• Henry Roy F. Harrod [140–142] and Evsey D. Domar, [143,144];

• John M. Keynes [145–148];
• Wassily W. Leontief [149,150];
• Alban W.H. Phillips [151,152];
• Roy G.D. Allen [153–156];
• Robert M. Solow [157,158] and Trevor W. Swan [159];
• Nicholas Kaldor [160–162];

And other scientists.

Valentina V. Tarasova and the author built various economic models with power-law memory,
which are generalizations of the classical models. For example, the following economic models have
been proposed.

• The natural growth model [168,169];

• The growth model with constant pace [170,171];
• The Harrod–Domar model [172,173] and [112,174–176];
• The Keynes model [177–179];
• The dynamic Leontief (intersectoral) model [86,180–182];
• The dynamics of fixed assets (or capital stock) [170,171];
• The logistic growth model with memory [183,189];
• The model of logistic growth with memory and crises [183];
• The time-dependent dynamic intersectoral model with memory [86,182];
• The Phillips model with memory and lag [185];
• The Harrod–Domar growth model with memory and distributed lag [186];
• The dynamic Keynesian model with memory and lag [187];
• The model of productivity with fatigue and memory [188];
• The Solow–Swan model [189];
• The Kaldor-type model of business cycles (the Van der Pol model) [189];

And some other economic models.

Let us also note works, in which fractional dynamic generalizations of economic models were
proposed without introducing new economic notions and concepts.

(1) Michele Caputo proposed some fractional dynamic model of economy [191–200]:

• In the standard relaxation equation, which describes the relaxation economy to equilibrium,
the memory has been introduced in the reactivity of investment to the interest rate.
• The continuous-time IS–LM model with memory [192];
• The tax version of the Fisher model with memory for stock prices and inflation rates [199]
that can be used to predict nominal and real interest rate behavior with memory.
(2) Mathematical description of some fractional generalization of economic models was proposed by
the Kabardino–Balkarian group: Adam M. Nakhushev [201,202], Khamidbi Kh. Kalazhokov [203],
Zarema A. Nakhusheva [204].
(3) Mathematical description of some fractional generalization of economic models were proposed
by the Kamchatka group: Viktoriya V. Samuta, Viktoriya A. Strelova, and Roman I. Parovik [205],
Yana E. Shpilko, Anastasiya E. Solomko., Roman I. Parovik [206] Danil M. Makarov [207].
(4) Shiou-Yen Chu and Christopher Shane proposed the hybrid Phillips curve model with memory
to describe the dynamic process of inflation with memory in the work [208].

12
Mathematics 2019, 7, 509

(5) Rituparna Pakhira, Uttam Ghosh and Susmita Sarkar derived [209–213] some inventory models
with memory.
(6) Computer simulation for modeling the national economies in the framework of the fractional
generalizations of the Gross domestic product (GDP) model was proposed by Inés Tejado, Duarte
Valério, Nuno Valério, Emiliano Pérez [214–220] in 2014–2019, and by Dahui Luo, JinRong Wang,
Michal Feckan in 2018 in the paper [221].
(7) In addition, we may note the works with economic models that were proposed in [100–105] that
are related to the deterministic chaos stage.

Let us note that the problems and difficulties arising in the construction of fractional-dynamic
analogs of standard economic models by using the fractional calculus are described in [189] with details.
Some of proposed models can be considered as econophysics approach, which are based on
fractional generalization of the standard damped harmonic oscillator equation, where the memory has
been introduced in the frictional term by using fractional derivative instead of first-orderr derivative.
New principles, effects and phenomena have been suggested for fractional economic dynamics
with memory and non-locality (for example, see [174–176,189,222]). Qualitatively new effects due to
the presence of memory in the economic process are described in the works [174–176,189,222].
In my opinion, this stage of the development of fractional mathematical economics actually
includes (absorbs) approaches based on the ARFIMA model using only the Granger–Joyeux–Hosking
fractional differencing and integrating, which in really are the well-known Grunwald–Letnikov
fractional differences [39]. This opinion is based on the obvious fact that the new stage allows the
AFRIMA approach to go beyond the restrictions of the Grunwald–Letnikov operators, and use different
types of fractional finite differences and fractional derivatives of non-integer orders.
Moreover, this stage can include (absorbs) approaches based on the fractional econophysics and
deterministic chaos. For the econophysics approach, new opportunities are opening up on the way to
formulating economic analogues of physical concepts and notions that will be more understandable
to economists. This will significantly simplify the implementation of the concepts and methods of
fractional econophysics in economic theory and application.
The most important element in the construction of the fractional mathematical economics as a new
theory is the emergence and the formation of new notions, concepts, effects, phenomena, principles
and methods, which are specific only to this theory. This gives rise to a new scientific direction (the
fractional mathematical economics), since there is something of their own that others do not have.
We have now entered the stage of forming a new direction in mathematical economics and
economic theory, when concepts and methods are not borrowed from other sciences and areas, but
their own are created.

3. New Future Stages and Approaches

There is a natural question. What stages and approaches will appear and develop in the future?
In this section we propose some assumptions about the future use of fractional calculus in economics
and discuss the direction of development of fractional mathematical economics.

3.1. Self-Organization in Fractional Economic Dynamics

Self-organization processes play an important role in both the natural and social sciences. In
the description of self-organization in economic (and social) processes, it is necessary to abandon
the assumption that all economic agents suffer from amnesia. They should be considered as agents
with memory that interact with each other. We can consider self-organization with memory [222] in
economics and social sciences. Therefore, the fractional mathematical economics and economic theory
can be developed by considering the generalization of different economic models of self-organization
that are described in the books [223–225].

13
Mathematics 2019, 7, 509

3.2. Distributed Lag Fractional Calculus

The continuously distributed lag has been considered in economics starting with the works of
Michal A. Kalecki [163] and Alban W.H. Phillips [151,152]. The continuous uniform distribution of
delay time is considered by M.A. Kalecki in 1935 [163], (see also Section 8.4 of [154], (pp. 251–254)) for
dynamic models of business cycles. The continuously distributed lag with the exponential distribution
of delay time is considered by A.W.H. Phillips [151,152] in 1954. In the Phillips growth models,
generalization of the economic concepts of accelerator and multipliers were proposed by taking into
account the distributed lag. The operators with continuously distributed lag were considered by Roy
G.D. Allen [153,154], (pp. 23–29), in 1956. Currently, economic models with delay are actively used to
describe the processes in the economy.
The time delay is caused by finite speeds of processes and therefore it cannot be considered as
processes with long memory (for some details about concept of memory, see Section 3.12 of this paper
and [29–35,112,113,116,117]). For example, in physics the propagation of the electromagnetic field with
finite speed in a vacuum does not mean the presence of memory in this process. In economics and
electrodynamics, processes with time delay (lag) are not referred to as processes with memory and
effects of time delay are not interpreted as a memory.
Note that the operators with exponentially distributed lag, which were defined in the works of
Caputo and Fabrizio [226,227], cannot be interpreted as fractional derivatives of non-integer orders and
cannot describe processes with memory. Note that exponential distribution is the continuous analogue
of the geometric distribution that has the key property of being memoryless. The Caputo–Fabrizio
operators are integer-order derivatives with the exponentially distributed delay time [228]. The
fractional generalizations of the Caputo–Fabrizio operator are proposed in [185–187,228].
The distributed lag fractional calculus was proposed by the author and Svetlana S. Tarasova
in [228]. To take into account the memory and lag in economic and physical models, the fractional
differential and integral operators with continuously distributed lag (time delay) were proposed in [228].
The distributed lag fractional operators are compositions of fractional differentiation or integration and
continuously distributed translation (shift). The kernels of these operators are the Laplace convolution
of probability density function and the kernels of fractional derivatives or integrals. The random
variable is the delay time that is distributed by probability law (distribution) on positive semiaxis.
Examples of economic application of the lag-distributed fractional operators have been suggested in
the works [185–187], where the economic concepts of accelerator and multipliers with distributed lag
and memory were proposed.

3.3. Distributed Order Fractional Calculus

In general, the order parameter α, which can be interpreted as the parameter of memory
fading [112,113], can be distributed on the interval [α1 , α2 ], where the distribution is described by a
weight function ρ(α). The functions (α) describes distribution of the parameter of the memory fading
on a set of economic agents. This is important for the economics, since various types of economic
agents may have different parameters of memory fading. In this case, we should consider the fractional
operators, which depends on the weight function ρ(α) and the interval [α1 , α2 ].
The concept of the integrals and derivatives with distributed orders was first proposed by Michele
Caputo in [229] in 1995, and then these operators are applied and developed in different works (for
example see [230–234]).
Let us note the following three cases.

(1) The simplest distribution of the order of fractional derivatives and integrals is the continuous
uniform distribution. The fractional operators with uniform distribution were proposed in [112,
113,129] and were called as the Nakhushev operators. Adam M. Nakhushev [235,236] proposed
the continual fractional derivatives and integrals in 1998. The fractional operators, which
are inversed to the continual fractional derivatives and integrals, were suggested by Arsen V.

14
Mathematics 2019, 7, 509

Pskhu [237,238]. In papers [112,113,129], we proved that the fractional integrals and derivatives of
the uniform distributed order could be expressed (up to a numerical factor) through the continual
fractional integrals and derivatives, which have been suggested by A. M. Nakhushev [235,236].
The proposed fractional integral and derivatives of uniform distributed order have been called in
our paper [129] as the Nakhushev fractional integrals and derivatives. The corresponding inverse
operators, which contains the two-parameter Mittag–Leffler functions in the kernel, were called
as the Pskhu fractional integrals and derivatives [129].
(2) In the papers [112,113], we proposed the concept of “weak” memory and the distributed order
fractional operators with the truncated normal distribution of the order. The truncated normal
distribution with integer mean and small variance can be used to describe economic processes
with memory, which is distributed around the classical case.
(3) As a special case of the general fractional operators, which were proposed by Anatoly N. Kochubei,
the fractional derivatives and integrals of distributed order are investigated in the works [239,240].

Fractional differential equations of distributed orders are actively used to describe physical
processes. However, at the present time, equations with distributed order operators have not yet
been used to describe economic processes. We hope that new interesting effects in economics can be
described by using order-distributed fractional operators.

3.4. Generalized Fractional Calculus in Economics

Generalized fractional calculus was proposed by Virginia Kiryakova and described in detail in
the book [2] in 1994. The brief history of the generalized fractional calculus is given in the paper [10].
Operators of generalized fractional calculus [2,10] can be use to describe complex processes with
memory and non-locality in real economy. In the application of the generalized fractional operators, an
important question arises about the correct economic (and physical) interpretation of these operators.
It is important to emphasize that not all fractional operators can describe the processes with memory.
It is important to clearly understand what type of phenomena can be described by a given operator.
For example, among these types of phenomena, in addition to memory, we can specify the time delay
(lag) and the scaling. Let us give a few examples to clarify this problem.

Example 1. The Abel-type fractional integral (and differential) operator with Kummer function in the
kernel, which is described in the classic book [1] (see equation 37.1 in [1], (p. 731)) can be interpreted
as the Riemann–Liouville fractional integral (and derivatives) with gamma distribution of delay
time [187,228]. Some Prabhakar fractional operators with the three-parameter Mittag–Leffler functions
in the kernel can also be interpreted as a Laplace convolution of the Riemann–Liouville (or Caputo)
fractional operators with continuously distributed lag (time delay) [186,228].

Example 2. We can state that the Kober fractional integration of non-integer order [1,2,4] can be
interpreted as an expected value of a random variable up to a constant factor [241] (see also Section
9 in [228]). In this interpretation, the random variable describes dilation (scaling), which has the
gamma distribution. The Erdelyi–Kober fractional integration also has a probabilistic interpretation.
Fractional differential operators of Kober and the Erdelyi–Kober type have analogous probabilistic
interpretation, i.e., these operators cannot describe the memory. These operators describe integer-order
operator with continuously distributed dilation (scaling). The fractional generalizations of the Kober
and Erdelyi–Kober operators, which can be used to describe memory and distributed dilation (scaling),
were proposed in [228].

Example 3. The Riesz fractional integro-differentiation (See Section 2.10 of [4]) cannot be used to
describe memory since this operator violates the causality principle, if it is written in the standard
form. For economic and physical processes with memory, the causality can be described by the

15
Mathematics 2019, 7, 509

Kramers–Kronig relations [116]. The Riesz fractional integro-differentiation can be used to describe
power-law non-locality and power-law spatial dispersion.

Therefore, an important part of the application of generalized fractional calculus is a clear

understanding of what types of processes and phenomena can describe fractional operators of
non-integer order.

3.5. General Fractional Calculus

The concept of general fractional calculus was suggested by Anatoly N. Kochubei [239,240] by
using the differential-convolution operators. The works [239,240] describe the conditions under which
the general operator has a right inverse (a kind of a fractional integral) and produce, as a kind of a
fractional derivative, equations of evolution type. A solution of the relaxation equations with the
Kochubei general fractional derivative with respect to the time variable is described  [239]. In the
works [239,240] the Cauchy problem (A) is considered for the relaxation equation D(k) X (t) = λX(t),
where λ < 0. This Cauchy problem has a solution X(λ, t), which is continuous on R+ , infinitely
differentiable and completely monotone on R+ , if the Kochubei conditions (*) are satisfied.
In the economics, various growth models are used to describe real processes in economy. Therefore,
it
 is very
 important to describe conditions, for which the Cauchy problem (A) for the growth equation
D(k) X (t) = λX(t), where λ > 0, has a solution X(λ, t).
The growth equation is considered in [242] for the special case of a distributed order derivative,
where it was proved that a smooth solution exists and is monotone increasing. In addition, the
solution of the growth equation has been proposed for the case of fractional differential operators with
distributed lag in [185–187,228]. The existence of a solution in the growth case has been also proved in
2018 by Chung-Sik Sin [243] for nonlinear equation with a generalized derivative like the Kochubei
fractional derivative. The growth equation for physics is discussed in [244].
Solving the problem in the general case will allow us to accurately describe the conditions on
the operator kernels (the memory functions), under which equations for models of economic growth
with memory have solutions. A paper dedicated to solving this mathematical problem was written by
Anatoly N. Kochubei and Yuri Kondratiev [245] in 2019 for the Special Issue “Mathematical Economics:
Application of Fractional Calculus” of Mathematics. The application of these mathematical results in
economics and their economic interpretation is an open question at the moment.
To understand the warranted (technological) growth rate of the economy, it is important to obtain
the asymptotic behavior for solutions of the general growth equation.
In the application of the general fractional operators, it is also important to have correct economic
and physical interpretations that will connect the types of operator kernels with the types of phenomena.
For example, it is obvious that the kernels of general fractional operators satisfying the normalization
condition will describe distributed delays in time (lag), and not memory.

3.6. Partial Differential Equations in Economics

Usually the mathematical formulation of macroeconomic models is reduced to systems of difference
equations or ordinary differential equations that describe the dynamics of a relatively small number
of macroeconomic aggregates. However, it is known that, in macroeconomics, partial differential
equations (PDEs) naturally arise, and they are used in macroeconomics [246,247]. Accounting for
non-locality (for example, a power type) in the state space leads us to the necessity of using fractional
partial differential equations.

3.7. Fractional Variational Calculus in Economics

In mathematical economics, theorems on the existence of extreme values of certain parameters
are proved, the properties of equilibrium points and equilibrium growth trajectories are actively

16
Mathematics 2019, 7, 509

studied. The existence of optimal solutions for fractional differential equations should be considered
for economic processes with memory and non-locality.
Methods of the fractional calculus of variations are actively developing [248,249]. However, at
the present time, none of the variational problems, which are well known in economics, has been
generalized to the case of processes with memory using fractional calculus.
In the variation approach, there are some problems that restrict the possibilities of its application.
One of the problems associated with the property of integration in parts, which actually turns the
left-second fractional derivative into a right-sided derivative. As a result, we will obtain equations in
which, in addition to being dependent on the past, there is a dependence on the future, that is, the
principle of causality is violated.
We assume that this problem cannot be solved within the framework of using the principle of
stationarity of the holonomic functional (action). It is necessary to use non-holonomic functionals. We
can also consider non-holonomic constraints with fractional derivatives of non-integer orders [250].
We can also consider variations of non-integer order [251] and fractional variational derivatives [252].
Another problem is the mathematical interpretation and the economic (and physical)
interpretations of extreme values. The non-holonomic constraints and variations of fractional orders
should also have a correct economic (and physical) interpretation.
However, we emphasize that for the economics, finding the optimality and stability of the solution
is very important.

3.8. Fractional Differential Games in Economics

Models of differential games in which derivatives of non-integer orders are used and, thereby,
the power-fading memory is taken into account were proposed in the works of Arkadiy A. Chikriy
(Arkadii Chikrii), Ivan I. Matychyn and Alexander G. Chentsov [253–257] (see also [258–260]), which
are clearly not related to economy. Note that the construction of models of economic behavior, using
differential games with power memory, instead of games with full memory, currently remains an open
question. The construction of such models requires further research on economic behavior within the
framework of game theory. The basis for such constructions can be the methods and results described
in [253–260].

3.9. Economic Data and Fractional Calculus in Economic Modelling

We note the importance of using fractional calculus in computer simulations and modeling of real
economic data, including data related to both macroeconomics and microeconomics.
The first works that can be attributed to the mathematical economics stage/approach are works
published in 2014–2016 by Inés Tejado, Duarte Valério, Nuno Valério, [214–216]. An application
of fractional calculus for modeling the national economies in the framework of the fractional
generalizations of the Gross domestic product (GDP) model, which are described by the fractional
differential equations are used, were considered by Inés Tejado, Duarte Valério, Nuno Valério, Emiliano
Pérez [214–220] in 2014–2019, and by Dahui Luo, JinRong Wang, Michal Feckan in 2018 in the
paper [221]. The fractional differential equation used in the fractional GDP model was obtained by
replacing first-order derivatives with fractional derivatives. Therefore, this model requires theoretical
justification and consistent derivation.
To describe the real economic processes in the framework of fractional dynamic models, it is
necessary to combine theoretical constructions and computer modeling.

3.10. Big Data

It is obvious that Big Data that describes behavior of peoples and other economic agents should
contain information that can be considered as “Traces of People’s Memory”. It would be strange
if these Big Data neglected memory, since people have memory if they don’t suffer from amnesia.
We can assume that economic modeling in the era of Big Data will describe the memory effects in

17
Mathematics 2019, 7, 509

microeconomics and macroeconomics. The Big Data will give us a possibility to take into account the
effects of memory and non-locality in those economic and financial processes in which they were not
even suspected.

3.11. Fractional Econometrics

Economic theory is a branch of economics that employs mathematical models and abstractions of
economic processes and objects to rationalize, explain and predict economic phenomena.
One of the main goals of economic theory and mathematical economics is to explain the processes
and phenomena in economy and make predictions. To achieve this goal, within the framework of
economic theory and mathematical economics, new notions, concepts, tools and methods should
be developed for describing and interpreting economic processes with memory and non-locality.
Obviously, it is impossible to explain an economy with memory without having adequate concepts.
Economic theory and mathematical economics are branches of economics in which the creation of
concepts and theoretical (primarily mathematical) models of phenomena and their comparison with
reality is used as the main method of understanding economy. Economic theory is a separate way
of studying economy, although its content is naturally formed taking into account the observations
of economy. The methodology of economic theory consists in constructing main economic concepts;
in formulating (in mathematical language) the principles and laws of economics connecting these
concepts; in explaining observable economic phenomena and effects by using the formulated concepts
and laws; in predicting new phenomena that may be discovered.
Mathematical economics, when viewed in a narrow sense as a branch of mathematics, is reduced
only to the study of the properties of economic models at the mathematical level of rigor. In this
approach, mathematical economics is often denied in the choice of economic concepts, interpretation
and comparison of models with economic reality. For fractional mathematical economics, to describe
processes with memory this view is erroneous. Obviously, it is impossible to explain economic processes
with memory without having adequate concepts, since many standard concepts are not applicable.
Now the economics is undergoing a new revolution, “Memory revolution”. A mathematic
tool in the revolution is a new mathematics (fractional calculus), which was not previously used in
mathematical economics. As a result, the present stage is a stage of the formation of new concepts and
methods. In this stage, the mathematical economics cannot be isolated from the process of formation
of new concepts and methods. Non-standard properties of fractional operators should be reflected in
new economic concepts that take into account memory effects. Economic theory and mathematical
economics can explain and predict processes with memory in real economy only if they create a solid
foundation of new adequate economic concepts and principles.
To explain and predict the processes and phenomena with memory in economy, we must have
a good instrument for conducting observations and their adequate description. This instrument is
econometrics, or rather fractional econometrics. Econometrics is a link that connects economic theory
and mathematical economics with the phenomena and processes in real economy.
Econometrics mainly based on statistics for formulating and testing models and hypotheses
about economic processes or estimating parameters for them. Theoretical econometrics considers
the statistical properties of assessments and tests, while applied econometrics deals with the use
of econometric methods for evaluating economic models. Theoretical econometrics develops tools
and methods, and also studies the properties of econometric methods. Applied econometrics uses
theoretical econometrics and economic data to evaluate economic theories, develop econometric
models, analyzing economic dynamics, and forecasting.
Fractional econometrics is based on statistics of long-memory processes [29–32]. Currently,
fractional econometrics methods are actually related to ARFIMA models and the well-known
Grunwald–Letnikov fractional differences in the form of the Granger–Joyeux–Hosking fractional
differencing and integrating.

18
Mathematics 2019, 7, 509

Fractional econometrics can reach new opportunities in the development of new econometric
methods and their use in describing economic reality by applying methods of modern fractional calculus,
various types of fractional finite differences, differential and integral operators of non-integer order.
As a result, we can state that the main goals of fractional economics such as explain the economy
and to make predictions for processes with memory, and correctly describe economic events, data,
processes, and to give adequate predictions, we should have fractional econometrics.
One of the main goals of fractional mathematical economics and economic theory is to explain the
processes and phenomena with memory in economy and make predictions. However, to explain the
economic processes with memory, it is necessary to understand what memory is and how to describe it.
Note that a clear understanding of the memory does not even exist within the framework of fractional
calculus approach.

3.12. Development Concept of Memory

Further development of the concept of memory for economic processes in the framework of
fractional calculus is of great importance. It should be emphasized that not all types of fractional
derivatives and integrals of non-integer order can describe processes with memory. The concept of
memory itself for economic processes is discussed in [29–35], and [39,40,110–117]. One of the possible
criteria of memory proposed in the work [116]. Let us give some comments about memory concept.
From mathematical point of view, economic models may be classified as stochastic or deterministic
and as discrete or continuous. For simplification, let us consider the deterministic approach with
continuous time. In this case, memory can be defined as a property of the process, when there exists at
least one endogenous variable Y(t), and an associated exogenous (or endogenous) variable X(t), such
that the variable Y(t) at the time t > t0 depends on the history of the change of X(τ) on the interval
τ ∈ (t0 , t). However, not all such dependencies are due to the memory effect.
To describe processes with memory, this dependence of one variable on another should satisfy the
causality principle. For economic and physical processes with memory, the causality can be described
by the Kramers–Kronig relations [116].
An important property of memory is described by the principle of memory fading that was
proposed by Ludwig Boltzmann in 1874 and 1876. Then, it was significantly developed by Vito Volterra
in 1928 and 1930. The principle of memory fading states that the increasing of the time interval leads
to a decrease in the corresponding contribution to the variable Y(t).
Note that in physics the concept of fading memory assumes a set of stronger restrictions on
memory. For example, it is often assumed that the memory is described by functions, which tends to
zero monotonically with increasing the time variable. In this form, the principle of fading memory
assumes that it is less probable to expect strengthening of the memory with respect to the more distant
events. However, in some economic processes, it should be taken into account that the economic agents
may remember sharp and significant changes of the exogenous variable X(τ), despite the fact that
these changes were a more distant past compared to weaker changes in the near past. For this reason,
in economics we can use memory functions that are not monotonic decrease.
For a simple case, we can consider the dependence in the form
 t
Y (t) = M(t, τ)X(τ)dτ,
t0

where the kernel M(t, τ) of this integral operator is called the memory function (or the linear response
function). Obviously, the derivative of the integer orders of some variable can be considered as an
associated variable X(τ). We also can consider integer-order derivatives of Y(t) as endogenous variables.
It is obvious that not every kernels M(t, τ) can be used to describe the memory in the economic
processes. Possible restrictions on the memory function are discussed in paper [112,113].

19
Mathematics 2019, 7, 509

In this paper [112,113], we describe some general restrictions that can be imposed on the structure
and properties of memory. In addition, to the causality principle [116], these restrictions include the
following three principles:

• The principle of fading memory;

• The principle of memory homogeneity on time (the principle of non-aging memory);
• The principle of memory reversibility (the principle of memory recovery).

Mathematically, the principle of non-aging memory means that the memory function has a
property M(t, τ) = M(t − τ). In this case, the integral operator can be described by the convolution
Y(t) = (M ∗ X)(t), [112,113].
The principle of memory reversibility is connected with the principle of duality of accelerator
with memory and multiplier with memory, which is proposed in [123,124]. In general, fractional
calculus, which was proposed in [239,240] and based on the use of differential-convolution operators,
the principle of memory reversibility means that the general operators should have a right inverse
(a kind of a fractional integral).
Note that the Kober and Erdelyi–Kober fractional operators, which are interpreted in [112,113] as
operators that describe memory with generalized power-law fading, really are integer-order operators
with continuously distributed scaling or dilation (see [241] and Section 9 in [228]), and therefore, these
operators cannot describe the memory.
Note that time delay, which is sometimes interpreted as a complete (perfect, ideal) memory [112,
113], cannot describe memory. In economics and electrodynamics, processes with time delay (lag)
are not referred to as processes with memory and time delay is not interpreted as a memory. The
interpretation of the time delay, which is usually called a lag in economics, as some kind of memory
seems to be incorrect for the following reasons.
From economic and physical points of view, the time delay is caused by finite speeds of processes,
i.e., the change of one variable does not lead to instant changes of another variable. Therefore, the time
delay cannot be considered as memory in processes. This fact is well-known in physics as the retarded
potential of an electromagnetic field, when a change in the electromagnetic field at the observation
point is delayed with respect to the change in the sources of the field located at another point. The
processes of propagation of the electromagnetic field in a vacuum are not interpreted in physics as
presence of memory in these processes.
From a mathematical point of view, the kernels of integral operators for distributed time delay
(lag) and fading memory are distinguished by the fact that the normalization condition holds for the
time delay case. Note that the probability distribution functions as kernels, which are usually called
the weighting function in economics, are actively used for macroeconomic models with distributed
delay time. Equivalent differential equations of integer orders in economics are usually used instead
of equations with integro-differential operators, in which the weighting function in the kernels. It is
known that under certain conditions, equations with continuously distributed lag are equivalent to
differential equations with standard derivatives of integer orders. Mathematically, this means that
processes with time delay can be described by equations containing only a finite number of derivatives
of integer orders. The integer-order derivatives of functions are determined by the properties of these
functions in small neighborhood of the considered point. As a result, differential equations of integer
orders cannot describe a memory. To describe processes with fading memory and distributed time
delay, we should use the distributed lag fractional calculus [228], (see also [185–187]).
As a result, within the framework of fractional calculus, it is necessary to distinguish between
fractional operators that describe distributed time delay and distributed scaling from operators
describing memory, and the combination of memory with these phenomena. However, there are open
questions about what types of memory we can describe by using fractional calculus (for example,
see [116,117]), and in what directions the concept of memory for economic processes will develop.

20
Mathematics 2019, 7, 509

4. Conclusions
In this brief historical description, an attempt was made to draw a sketch picture of the development
of fractional calculus applications in economics, the birth of a new direction in mathematical economics,
a new revolution in economic theory. Due to brevity and schematics, this picture obviously cannot
reflect the fullness and complexity of the development of a fractional mathematical economics. As a
result, it was possible that some directions and approaches, results and works close in the described
history were missed. One can hope that the written short history will be perceived with understanding
and will be supplemented in the future with new works on the history of the use of fractional calculus
in the economics.
We can hope that the further development of the use of fractional calculus to describe economic
phenomena and processes will take an important place with modern mathematical economics and
economic theory. Generally speaking, it is strange to neglect memory in the economics, since the most
important actors are people with memory.

Funding: This research received no external funding.

Conflicts of Interest: The author declares no conflict of interest.

References
1. Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives Theory and Applications; Gordon
and Breach: New York, NY, USA, 1993; 1006p, ISBN 9782881248641.
2. Kiryakova, V. Generalized Fractional Calculus and Applications; Longman and John Wiley: New York, NY, USA,
1994; 360p, ISBN 9780582219779.
3. Podlubny, I. Fractional Differential Equations; Academic Press: San Diego, CA, USA, 1998; 340p.
4. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier:
Amsterdam, The Netherlands, 2006; 540p.
5. Diethelm, K. The Analysis of Fractional Differential Equations: An Application-Oriented Exposition Using Differential
Operators of Caputo Type; Springer: Berlin, Germany, 2010; 247p.
6. Letnikov, A.V. On the historical development of the theory of differentiation with arbitrary index. Math.
Collect. 1868, 3, 85–112. (In Russian)
7. Ross, B. A brief history and exposition of the fundamental theory of fractional calculus. In Fractional Calculus
and Its Applications. Proceedings of the International Conference Held at the University of New Haven, June 1974;
Series: Lecture Notes in Mathematics; Springer: Berlin/Heidelberg, Germany, 1975; Volume 457, pp. 1–36.
[CrossRef]
8. Ross, B. The development of fractional calculus 1695–1900. Hist. Math. 1977, 4, 75–89. [CrossRef]
9. Ross, B. Fractional Calculus. Math. Mag. 1977, 50, 115–122. [CrossRef]
10. Kiryakova, V. A brief story about the operators of the generalized fractional calculus. Fract. Calc. Appl. Anal.
2008, 11, 203–220.
11. Tenreiro Machado, J.; Kiryakova, V.; Mainardi, F. Recent history of fractional calculus. Commun. Nonlinear
Sci. Numer. Simul. 2011, 16, 1140–1153. [CrossRef]
12. Tenreiro Machado, J.A.; Galhano, A.M.; Trujillo, J.J. Science metrics on fractional calculus development since
1966. Fract. Calc. Appl. Anal. 2013, 16, 479–500. [CrossRef]
13. Tenreiro Machado, J.A.; Galhano, A.M.; Trujillo, J.J. On development of fractional calculus during the last
fifty years. Scientometrics 2014, 98, 577–582. [CrossRef]
14. Valerio, D.; Machado, J.; Kiryakova, V. Some pioneers of the applications of fractional calculus. Fract. Calc.
Appl. Anal. 2014, 17, 552–578. [CrossRef]
15. Tenreiro Machado, J.A.; Kiryakova, V. The chronicles of fractional calculus. Fract. Calc. Appl. Anal. 2017, 20,
307–336. [CrossRef]
16. Tarasov, V.E. No violation of the Leibniz rule. No fractional derivative. Commun. Nonlinear Sci. Numer. Simul.
2013, 18, 2945–2948. [CrossRef]
17. Ortigueira, M.D.; Tenreiro Machado, J.A. What is a fractional derivative? J. Comput. Phys. 2015, 293, 4–13.
[CrossRef]

21
Mathematics 2019, 7, 509

18. Tarasov, V.E. On chain rule for fractional derivatives. Commun. Nonlinear Sci. Numer. Simul. 2016, 30, 1–4.
[CrossRef]
19. Tarasov, V.E. Leibniz rule and fractional derivatives of power functions. J. Comput. Nonlinear Dyn. 2016, 11,
31014. [CrossRef]
20. Tarasov, V.E. No nonlocality. No fractional derivative. Commun. Nonlinear Sci. Numer. Simul. 2018, 62,
157–163. [CrossRef]
21. Sayevand, K.; Tenreiro Machado, J.; Baleanu, D. A new glance on the Leibniz rule for fractional derivatives.
Commun. Nonlinear Sci. Numer. Simul. 2018, 62, 244–249. [CrossRef]
22. Granger, C.W.J. The Typical Spectral Shape of an Economic Variable; Technical Report No. 11; Department of
Statistics, Stanford University: Stanford, CA, USA, 1964; 21p, Available online: https://statistics.stanford.edu/
sites/default/files/PAR%20NSF%2011.pdf (accessed on 29 May 2019).
23. Granger, C.W.J. The typical spectral shape of an economic variable. Econometrica 1966, 34, 150–161. [CrossRef]
24. Granger, C.W.J. Essays in Econometrics: Collected Papers of Clive W.J. Granger. Volume I: Spectral Analysis,
Seasonality, Nonlinearity, Methodology, and Forecasting; Ghysels, E., Swanson, N.R., Watson, M.W., Eds.;
Cambridge University Press: Cambridge, NY, USA, 2001; 523p.
25. Granger, C.W.J. Essays in Econometrics: Collected Papers of Clive W.J. Granger. Volume II: Causality, Integration
and Cointegration, and Long Memory; Ghysels, E., Swanson, N.R., Watson, M.W., Eds.; Cambridge University
Press: Cambridge, NY, USA, 2001; 398p.
26. Granger, C.W.J.; Joyeux, R. An introduction to long memory time series models and fractional differencing. J.
Time Ser. Anal. 1980, 1, 15–39. [CrossRef]
27. NobelPrize.org. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 2003; Press Release;
Nobel Media AB: Stockholm, Sweden, 2019; Available online: https://www.nobelprize.org/prizes/economic-
sciences/2003/summary/ (accessed on 29 May 2019).
28. Hosking, J.R.M. Fractional differencing. Biometrika 1981, 68, 165–176. [CrossRef]
29. Beran, J. Statistics for Long-Memory Processes; Capman and Hall: New York, NY, USA, 1994; 315p, ISBN
0-412-04901-5.
30. Palma, W. Long-Memory Time Series: Theory and Methods; Wiley-InterScience: Hoboken, NJ, USA, 2007; 304p,
ISBN 978-0-470-11402-5.
31. Beran, J.; Feng, Y.; Ghosh, S.; Kulik, R. Long-Memory Processes: Probabilistic Properties and Statistical Methods;
Springer: Berlin/Heidelberg, Germany; New York, NY, USA, 2013; 884p. [CrossRef]
32. Robinson, P.M. (Ed.) Time Series with Long Memory; Series: Advanced Texts in Econometrics; Oxford
University Press: Oxford, UK, 2003; 392p, ISBN 978-0199257300.
33. Teyssiere, G.; Kirman, A.P. (Eds.) Long Memory in Economics; Springer: Berlin/Heidelberg, Germany,
2007; 390p.
34. Tschernig, R. Wechselkurse, Unsicherheit und Long Memory; Physica-Verlag: Heidelberg, Germany, 1994; 232p,
(In German). ISBN 978-3-7908-0753-0. [CrossRef]
35. Baillie, R.N. Long memory processes and fractional integration in econometrics. J. Econom. 1996, 73, 5–59.
[CrossRef]
36. Parke, W.R. What is fractional integration? Rev. Econ. Stat. 1999, 81, 632–638. [CrossRef]
37. Banerjee, A.; Urga, G. Modelling structural breaks, long memory and stock market volatility: An overview. J.
Econom. 2005, 129, 1–34. [CrossRef]
38. Gil-Alana, L.A.; Hualde, J. Fractional Integration and Cointegration: An Overview and an Empirical
Application. In Palgrave Handbook of Econometrics. Volume 2: Applied Econometrics; Mills, T.C., Patterson, K.,
Eds.; Springer: Berlin, Germany, 2009; pp. 434–469.
39. Tarasov, V.E.; Tarasova, V.V. Long and short memory in economics: Fractional-order difference and
differentiation. IRA-Int. J. Manag. Soc. Sci. 2016, 5, 327–334. [CrossRef]
40. Tarasova, V.V.; Tarasov, V.E. Comments on the article «Long and short memory in economics: Fractional-order
difference and differentiation». Probl. Mod. Sci. Educ. 2017, 31, 26–28. [CrossRef]
41. Kolmogoroff, A.N. Wiener Spirals and some other interesting curves in Gilbert’s room. (Wienersche Spiralen
und einige andere interessante Kurven im Hilbertschen Raum, cr (Doklady). Acad. Sci. Urss 1940, 26,
115–118. (In German)
42. Rogers, L.C.G. Arbitrage with fractional Brownian motion. Math. Financ. 1997, 7, 95–105. Available online:
http://www.skokholm.co.uk/wp-content/uploads/2013/02/fbm.pdf (accessed on 29 May 2019). [CrossRef]

22
Mathematics 2019, 7, 509

43. Duncan, T.E.; Hu, Y.; Pasik-Duncan, B. Stochastic calculus for fractional Brownian motion. I. Theory. Siam J.
Control Optim. 2000, 38, 582–612. [CrossRef]
44. Tudor, C.; Viens, F. Statistical aspects of the fractional stochastic calculus. Ann. Stat. 2007, 35, 1183–1212.
Available online: https://projecteuclid.org/download/pdfview_1/euclid.aos/1185304003 (accessed on 29 May
2019). [CrossRef]
45. Mishura, Y.S. Stochastic Calculus for Fractional Brownian Motion and Related Processes; Lecture Notes in
Mathematics; Springer: Berlin/Heidelberg, Germany, 2008; 393p, ISBN 978-3540758723. [CrossRef]
46. Black, F.; Scholes, M.S. The pricing of options and corporate liabilities. J. Polit. Econ. 1973, 81, 637–654.
Available online: https://www.cs.princeton.edu/courses/archive/fall09/cos323/papers/black_scholes73.pdf
(accessed on 29 May 2019). [CrossRef]
47. NobelPrize.org. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1997; Press Release;
Nobel Media AB: Stockholm, Sweden, 2019; Available online: https://www.nobelprize.org/prizes/economic-
sciences/1997/press-release (accessed on 29 May 2019).
48. Wyss, W. The fractional Black–Scholes equation. Fract. Calc. Appl. Anal. 2000, 3, 51–61.
49. Cartea, A.; Meyer-Brandis, T. How duration between trades of underlying securities affects option prices.
Rev. Financ. 2010, 14, 749–785. [CrossRef]
50. Cartea, A. Derivatives pricing with marked point processes using tick-by-tick data. Quant. Financ. 2013, 13,
111–123. [CrossRef]
51. Akrami, M.H.; Erjaee, G.H. Examples of analytical solutions by means of Mittag-Leffler function of fractional
Black-Scholes option pricing equation. Fract. Calc. Appl. Anal. 2015, 18, 38–47. [CrossRef]
52. Muniandy, S.V.; Lim, S.C. Modeling of locally self-similar processes using multifractional Brownian motion
of Riemann-Liouville type. Phys. Rev. E 2001, 63, 7. [CrossRef] [PubMed]
53. Muniandy, S.V.; Lim, S.C.; Murugan, R. Inhomogeneous scaling behaviors in Malaysian foreign currency
exchange rates. Phys. A Stat. Mech. Appl. 2001, 301, 407–428. [CrossRef]
54. Kluppelberg, C.; Kuhn, C. Fractional Brownian motion as a weak limit of Poisson shot noise processes—with
applications to finance. Stoch. Process. Appl. 2004, 113, 333–351. [CrossRef]
55. Thao, T.H. An approximate approach to fractional analysis for finance. Nonlinear Anal. Real World Appl. 2006,
7, 124–132. [CrossRef]
56. Cartea, A.; Del-Castillo-Negrete, D. Fractional diffusion models of option prices in markets with jumps. Phys.
A Stat. Mech. Appl. 2007, 374, 749–763. [CrossRef]
57. Li, M.; Lim, S.C.; Hu, B.J.; Feng, H. Towards describing multi-fractality of traffic using local Hurst function.
In Proceedings of the 7th International Conference on Computational Science (ICCS 2007: Computational
Science), Beijing, China, 27–30 May 2007; Lecture Notes in Computer Science. Springer: Berlin/Heidelberg,
Germany, 2007; Volume 4488, pp. 1012–1020. [CrossRef]
58. Blackledge, J. Application of the fractal market hypothesis for modelling macroeconomic time series. Isast
Trans. Electron. Signal. Process 2008, 2, 89–110. [CrossRef]
59. Mendes, R.V. A fractional calculus interpretation of the fractional volatility model. Nonlinear Dyn. 2009, 55,
395–399. [CrossRef]
60. Kuroda, K.; Murai, J. Long memory in finance and fractional Brownian motion. Prog. Theor. Phys. Suppl.
2009, 179, 26–37. [CrossRef]
61. Marom, O.; Momoniat, E. A comparison of numerical solutions of fractional diffusion models in finance.
Nonlinear Anal. Real World Appl. 2009, 10, 3435–3442. [CrossRef]
62. Blackledge, J. Application of the fractional diffusion equation for predicting market behavior. Int. J. Appl.
Math. 2010, 40, 130–158. [CrossRef]
63. Tenreiro Machado, J.; Duarte, F.B.; Duarte, G.M. Fractional dynamics in financial indices. Int. J. Bifurc. Chaos
2012, 22, 12. [CrossRef]
64. Jiang, Y.; Wang, X.; Wang, Y. On a stochastic heat equation with first order fractional noises and applications
to finance. J. Math. Anal. Appl. 2012, 396, 656–669. [CrossRef]
65. Dung, N.T. Fractional stochastic differential equations with applications to finance. J. Math. Anal. Appl. 2013,
397, 334–348. [CrossRef]
66. Zeng, C.; Chen, Y.Q.; Yang, Q. Almost sure and moment stability properties of fractional order Black-Scholes
model. Fract. Calc. Appl. Anal. 2013, 16, 317–331. [CrossRef]

23
Mathematics 2019, 7, 509

67. Rostek, S.; Schobel, R. A note on the use of fractional Brownian motion for financial modeling. Econ. Model.
2013, 30, 30–35. [CrossRef]
68. Kerss, A.; Leonenko, N.; Sikorskii, A. Fractional Skellam processes with applications to finance. Fract. Calc.
Appl. Anal. 2014, 17, 532–551. [CrossRef]
69. Kleinerta, H.; Korbel, J. Option pricing beyond Black-Scholes based on double-fractional diffusion. Phys. A
Stat. Mech. Its Appl. 2016, 449, 200–214. [CrossRef]
70. Korbel, J.; Luchko, Y. Modeling of financial processes with a space-time fractional diffusion equation of
varying order. Fract. Calc. Appl. Anal. 2016, 19, 1414–1433. [CrossRef]
71. Aguilar, J.P.; Coste, C.G. Non-Gaussian analytic option pricing: A closed formula for the Levy-stable model.
arXiv 2016, arXiv:1609.00987. Available online: https://ssrn.com/abstract=2828673 (accessed on 29 May 2019).
[CrossRef]
72. Rostek, S. Option Pricing in Fractional Brownian Markets; Springer: Berlin/Heidelberg, Germany, 2009.
[CrossRef]
73. Fallahgoul, H.A.; Focardi, S.M.; Fabozzi, F.J. Fractional Calculus and Fractional Processes with Applications to
Financial Economics, Theory and Application; Academic Press: London, UK, 2016; 118p, ISBN 9780128042489.
74. Scalas, E.; Gorenflo, R.; Mainardi, F. Fractional calculus and continuous-time finance. Phys. A Stat. Mech. Its
Appl. 2000, 284, 376–384. [CrossRef]
75. Mainardi, F.; Raberto, M.; Gorenflo, R.; Scalas, E. Fractional calculus and continuous-time finance II: The
waiting-time distribution. Phys. A Stat. Mech. Appl. 2000, 287, 468–481. [CrossRef]
76. Gorenflo, R.; Mainardi, F.; Raberto, M.; Scalas, E. Fractional diffusion in finance: Basic theory. In Proceedings
of the MDEF2000—Workshop “Modelli Dinamici in Economia e Finanza”, Urbino, Italy, 28–30 September 2000;
Available online: http://www.mdef.it/fileadmin/user_upload/mdef/meetings/MDEF2000/MainardiMDEF.pdf
(accessed on 29 May 2019).
77. Gorenflo, R.; Mainardi, F.; Scalas, E.; Raberto, M. Fractional calculus and continuous-time finance III: The
diffusion limit. In Mathematical Finance; Kohlmann, A., Tang, S., Eds.; Birkhäuser: Basel, Switzerland, 2001;
pp. 171–180. [CrossRef]
78. Raberto, M.; Scalas, E.; Mainardi, F. Waiting-times and returns in high-frequency financial data: An empirical
study. Phys. A Stat. Mech. Its Appl. 2002, 314, 749–755. [CrossRef]
79. Scalas, E.; Gorenflo, R.; Luckock, H.; Mainardi, F.; Mantelli, M.; Raberto, M. Anomalous waiting times in
high-frequency financial data. Quant. Financ. 2004, 4, 695–702. [CrossRef]
80. Scalas, E. The application of continuous-time random walks in finance and economics. Phys. A Stat. Mech.
Its Appl. 2006, 362, 225–239. [CrossRef]
81. Meerschaert, M.M.; Scalas, E. Coupled continuous time random walks in finance. Phys. A Stat. Mech. Its
Appl. 2006, 370, 114–118. [CrossRef]
82. Laskin, N. Fractional market dynamics. Phys. A Stat. Mech. Its Appl. 2000, 287, 482–492. [CrossRef]
83. West, B.J.; Picozzi, S. Fractional Langevin model of memory in financial time series. Phys. Rev. E 2002, 65,
37106. [CrossRef] [PubMed]
84. Picozzi, S.; West, B.J. Fractional Langevin model of memory in financial markets. Phys. Rev. E 2002, 66, 12.
[CrossRef] [PubMed]
85. Scalas, E. Five years of continuous-time random walks in econophysics. In The Complex Networks of Economic
Interactions; Lecture Notes in Economics and Mathematical Systems; Springer: Berlin, Germany, 2006; Volume
567, pp. 3–16. [CrossRef]
86. Tarasov, V.E.; Tarasova, V.V. Time-dependent fractional dynamics with memory in quantum and economic
physics. Ann. Phys. 2017, 383, 579–599. [CrossRef]
87. Chen, W.C. Nonlinear dynamics and chaos in a fractional-order financial system. Chaos Solitons Fract. 2008,
36, 1305–1314. [CrossRef]
88. Chen, W.C. Dynamics and control of a financial system with time-delayed feedbacks. Chaos Solitons Fract.
2008, 37, 1198–1207. [CrossRef]
89. Abd-Elouahab, M.S.; Hamri, N.-E.; Wang, J. Chaos control of a fractional-order financial system. Math. Probl.
Eng. 2010, 2010, 18. [CrossRef]
90. Chen, L.; Chai, Y.; Wu, R. Control and synchronization of fractional-order financial system based on linear
control. Discret. Dyn. Nat. Soc. 2011, 2011, 21. [CrossRef]

24
Mathematics 2019, 7, 509

91. Wang, Z.; Huang, X.; Shi, G. Analysis of nonlinear dynamics and chaos in a fractional order financial system
with time delay. Comput. Math. Appl. 2011, 62, 1531–1539. [CrossRef]
92. Pan, I.; Korre, A.; Das, S.; Durucan, S. Chaos suppression in a fractional order financial system using
intelligent regrouping PSO based fractional fuzzy control policy in the presence of fractional Gaussian noise.
Nonlinear Dyn. 2012, 70, 2445–2461. [CrossRef]
93. Ma, S.; Xu, Y.; Yue, W. Numerical solutions of a variable order fractional financial system. J. Appl. Math.
2012, 2012, 14. [CrossRef]
94. Danca, M.F.; Garrappa, R.; Tang, W.K.S.; Chen, G. Sustaining stable dynamics of a fractional-order chaotic
financial system by parameter switching. Comput. Math. Appl. 2013, 66, 702–716. [CrossRef]
95. Xu, Y.; He, Z. Synchronization of variable-order fractional financial system via active control method. Cent.
Eur. J. Phys. 2013, 11, 824–835. [CrossRef]
96. Yue, Y.; He, L.; Liu, G. Modeling and application of a new nonlinear fractional financial model. J. Appl. Math.
2013, 2013, 9. [CrossRef]
97. Jajarmi, A.; Hajipour, M.; Baleanu, D. New aspects of the adaptive synchronization and hyperchaos
suppression of a financial model. Chaos Solitons Fract. 2017, 99, 285–296. [CrossRef]
98. Hajipour, A.; Hajipour, M.; Baleanu, D. On the adaptive sliding mode controller for a hyperchaotic
fractional-order financial system. Phys. A Stat. Mech. Its Appl. 2018, 497, 139–153. [CrossRef]
99. Tacha, O.I.; Munoz-Pacheco, J.M.; Zambrano-Serrano, E.; Stouboulos, I.N.; Pham, V.-T. Determining the
chaotic behavior in a fractional-order finance system with negative parameters. Nonlinear Dyn. 2018, 94,
1303–1317. [CrossRef]
100. Dadras, S.; Momeni, H.R. Control of a fractional-order economical system via sliding mode. Phys. A Stat.
Mech. Its Appl. 2010, 389, 2434–2442. [CrossRef]
101. Wang, Z.; Huang, X.; Shen, H. Control of an uncertain fractional order economic system via adaptive sliding
mode. Neurocomputing 2012, 83, 83–88. [CrossRef]
102. Baskonus, H.M.; Mekkaoui, T.; Hammouch, Z.; Bulut, H. Active control of a chaotic fractional order economic
system. Entropy 2015, 17, 5771–5783. [CrossRef]
103. He, L.; Yi, L.; Tang, P. Numerical scheme and dynamic analysis for variable-order fractional van der Pol
model of nonlinear economic cycle. Adv. Differ. Equ. 2016, 2016, 195. [CrossRef]
104. David, S.A.; Tenreiro Machado, J.A.; Quintino, D.D.; Balthazar, J.M. Partial chaos suppression in a fractional
order macroeconomic model. Math. Comput. Simul. 2016, 122, 55–68. [CrossRef]
105. David, S.A.; Fischer, C.; Tenreiro Machado, J.A. Fractional electronic circuit simulation of a nonlinear
macroeconomic model. AEU Int. J. Electron. Commun. 2018, 84, 210–220. [CrossRef]
106. Skovranek, T.; Podlubny, I.; Petras, I. Modeling of the national economies in state-space: A fractional calculus
approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]
107. Tenreiro Machado, J.A.; Mata, M.E. A fractional perspective to the bond graph modelling of world economies.
Nonlinear Dyn. 2015, 80, 1839–1852. [CrossRef]
108. Tenreiro Machado, J.A.; Mata, M.E. Pseudo phase plane and fractional calculus modeling of western global
economic downturn. Commun. Nonlinear Sci. Numer. Simul. 2015, 22, 396–406. [CrossRef]
109. Tenreiro Machado, J.A.; Mata, M.E.; Lopes, A.M. Fractional state space analysis of economic systems. Entropy
2015, 17, 5402–5421. [CrossRef]
110. Tarasova, V.V.; Tarasov, V.E. Criteria hereditarity of economic process and memory effect. Young Sci. 2016, 14,
396–399. (In Russian). Available online: https://elibrary.ru/item.asp?id=26389117 (accessed on 29 May 2019).
111. Tarasova, V.V.; Tarasov, V.E. Economic indicators: Ambiguity and memory effects. Econ. Manag. Law 2016,
3, 3–5. (In Russian). Available online: https://istina.msu.ru/media/publications/article/925/f43/26341086/
EUP2016.pdf (accessed on 29 May 2019).
112. Tarasova, V.V.; Tarasov, V.E. Concept of dynamic memory in economics. Commun. Nonlinear Sci. Numer.
Simul. 2018, 55, 127–145. [CrossRef]
113. Tarasova, V.V.; Tarasov, V.E. Notion of dynamic memory in economic theory. J. Econ. Entrep. 2017, 6, 868–880.
(In Russian). Available online: https://elibrary.ru/item.asp?id=29679697 (accessed on 29 May 2019).
114. Tarasova, V.V.; Tarasov, V.E. Long-term and short-term memory in discrete economic models. Compet. A
Glob. World Econ. Sci. Technol. 2017, 7, 155–160. (In Russian). Available online: https://elibrary.ru/item.asp?
id=29901818 (accessed on 29 May 2019).

25
Mathematics 2019, 7, 509

115. Tarasova, V.V.; Tarasov, V.E. Comments to the article “Long-term and short-term memory in discrete economic
models”. Probl. Sci. 2017, 8, 12–13. Available online: https://elibrary.ru/item.asp?id=29862856 (accessed on
29 May 2019).
116. Tarasov, V.E.; Tarasova, V.V. Criterion of existence of power-law memory for economic processes. Entropy
2018, 20, 414. [CrossRef]
117. Tarasov, V.E. Generalized memory: Fractional calculus approach. Fractal Fract. 2018, 2, 23. [CrossRef]
118. Tarasova, V.V.; Tarasov, V.E. Marginal utility for economic processes with memory. Alm. Mod. Sci. Educ. 2016,
7, 108–113. (In Russian). Available online: http://www.gramota.net/materials/1/2016/7/28.html (accessed on
29 May 2019).
119. Tarasova, V.V.; Tarasov, V.E. Economic indicator that generalizes average and marginal values. J. Econ. Entrep.
2016, 11, 817–823. (In Russian). Available online: https://elibrary.ru/item.asp?id=27340824 (accessed on 29
May 2019).
120. Tarasova, V.V.; Tarasov, V.E. Marginal values of non-integer order in economic analysis. Azimuth Sci. Res.
Econ. Manag. 2016, 3, 197–201. (In Russian). Available online: https://elibrary.ru/item.asp?id=27198798
(accessed on 29 May 2019).
121. Tarasova, V.V.; Tarasov, V.E. Economic interpretation of fractional derivatives. Prog. Fract. Differ. Appl. 2017,
3, 1–7. [CrossRef]
122. Tarasova, V.V.; Tarasov, V.E. Microeconomic meaning of derivatives of non-integer order. Sci. Educ. Today 2017,
8, 32–39. (In Russian). Available online: https://elibrary.ru/item.asp?id=29870013 (accessed on 29 May 2019).
123. Tarasova, V.V.; Tarasov, V.E. A generalization of concepts of accelerator and multiplier to take into account
memory effects in macroeconomics. J. Econ. Entrep. 2016, 10, 1121–1129. (In Russian). Available online:
https://elibrary.ru/item.asp?id=27211611 (accessed on 29 May 2019).
124. Tarasov, V.E.; Tarasova, V.V. Accelerator and multiplier for macroeconomic processes with memory. IRA Int.
J. Manag. Soc. Sci. 2017, 9, 86–125. [CrossRef]
125. Tarasova, V.V.; Tarasov, V.E. Exact discretization of economic accelerator and multiplier with memory. Fractal
Fract. 2017, 1, 6. [CrossRef]
126. Tarasova, V.V.; Tarasov, V.E. Exact discretization of economic accelerators and multipliers with memory. J.
Econ. Entrep. 2017, 7, 1063–1069. (In Russian). Available online: https://elibrary.ru/item.asp?id=29774025
(accessed on 29 May 2019).
127. Tarasova, V.V.; Tarasov, V.E. Accelerators in macroeconomics: A comparison of discrete and continuous
approaches. Sci. J. 2017, 8, 4–14. (In Russian). Available online: https://elibrary.ru/item.asp?id=30561974
(accessed on 29 May 2019). [CrossRef]
128. Tarasova, V.V.; Tarasov, V.E. Accelerators in macroeconomics: Comparison of discrete and continuous
approaches. Am. J. Econ. Bus. Adm. 2017, 9, 47–55. [CrossRef]
129. Tarasov, V.E. Exact discretization by Fourier transforms. Commun. Nonlinear Sci. Numer. Simul. 2016, 37,
31–61. [CrossRef]
130. Tarasova, V.V.; Tarasov, V.E. Economic accelerator with memory: Discrete time approach. Probl. Mod. Sci.
Educ. 2016, 36, 37–42. [CrossRef]
131. Tarasova, V.V.; Tarasov, V.E. Discrete accelerator with memory in macroeconomics. Economics 2017,
8, 32–40. (In Russian). Available online: https://elibrary.ru/item.asp?id=29944864 or https://economic-
theory.com/blizhajshij-nomer/matematicheskie-i-instrumentalnye/385-discrete-accelerato.html (accessed on
29 May 2019).
132. Tarasova, V.V.; Tarasov, V.E. Elasticity for economic processes with memory: Fractional differential calculus
approach. Fract. Differ. Calc. 2016, 6, 219–232. [CrossRef]
133. Tarasova, V.V.; Tarasov, V.E. Price elasticity of demand with memory. Econ. Sociol. Law 2016, 4, 98–106. (In
Russian). Available online: https://elibrary.ru/item.asp?id=26093967& (accessed on 29 May 2019).
134. Tarasova, V.V.; Tarasov, V.E. Elasticity of OTC cash turnover of currency market of Russian Federation. Actual
Probl. Humanit. Nat. Sci. 2016, 7, 207–215. (In Russian). Available online: https://publikacia.net/archive/
uploads/pages/2016_7_1/49.pdf (accessed on 29 May 2019).
135. Tarasova, V.V.; Tarasov, V.E. On applicability of point elasticity of demand on price to exchange trading on US
dollar. Sci. Perspect. 2016, 6, 6–11. (In Russian). Available online: https://elibrary.ru/item.asp?id=26344263 or
http://naupers.ru/6archive.html or https://docplayer.ru/55299569-O-primenimosti-tochechnoy-elastichnosti-
sprosa-po-cene-dlya-birzhevyh-torgov-po-dollaru-ssha.html (accessed on 29 May 2019).

26
Mathematics 2019, 7, 509

136. Tarasova, V.V.; Tarasov, V.E. Non-local measures of risk aversion in the economic process. Econ. Theory
Pract. 2016, 44, 54–58. (In Russian). Available online: https://elibrary.ru/item.asp?id=27541974 or http:
//econ.kubsu.ru/images/4_2016.pdf (accessed on 29 May 2019).
137. Tarasova, V.V.; Tarasov, V.E. Risk aversion for investors with memory: Hereditary generalizations of
Arrow-Pratt measure. Financ. J. 2017, 2, 46–63. (In Russian). Available online: https://elibrary.ru/item.asp?
id=28966889 or http://www.nifi.ru/images/FILES/Journal/Archive/2017/2/statii_2/fm_2017_2_05 (accessed on
29 May 2019).
138. Tarasova, V.V.; Tarasov, V.E. Deterministic factor analysis: Methods of integro-differentiation of non-integral
order. Actual Probl. Econ. Law 2016, 10, 77–87. (In Russian) [CrossRef]
139. Tarasova, V.V.; Tarasov, V.E. Fractional deterministic factor analysis of economic processes with memory
and nonlocality. In Chaotic, Fractional, and Complex Dynamics: New Insights and Perspectives; Edelman, M.,
Macau, E., Sanjuan, M.A.F., Eds.; Springer International Publishing: Berlin, Germany, 2018; pp. 173–189.
ISBN 978-3-319-68109-2. [CrossRef]
140. Harrod, R. An Essay in dynamic theory. Econ. J. 1939, 49, 14–33. [CrossRef]
141. Harrod, R. Towards a Dynamic Economics; Macmillan: London, UK, 1948.
142. Harrod, R. Economic Dynamics; Macmillan: London, UK, 1973; ISBN 978-0-333-14247-9.
143. Domar, E.D. Capital expansion, rate of growth and employment. Econometrica 1946, 14, 137–147. [CrossRef]
144. Domar, E.D. Expansion and employment. Am. Econ. Rev. 1947, 37, 34–55.
145. Keynes, J.M. The General Theory of Employment, Interest and Money; Macmillan: London, UK, 1936.
146. Keynes, J.M. The General Theory of Employment, Interest and Money: With the Economic Consequences of the Peace,
Series: Classics of World Literature; Wordsworth Editions ed; Ware: Hertfordshire, UK, 2017; 576p, ISBN
978-1840227475.
147. Keynes, J.M. The General Theory of Employment, Interest and Money: With the Economic Consequences of the Peace;
Palgrave Macmillan: London, UK, 2018; 438p. [CrossRef]
148. Keynes, J.M.; Johnson, E.; Moggridge, D.; Robinson, A. The Collected Writings of John Maynard Keynes, Reprint
Edition ed; Cambridge University Press: Cambridge, UK, 2013; 15126p, ISBN 978-1107677722.
149. Leontief, W.W. The Structure of the American Economy 1919–1939: An Empirical Application of Equilibrium
Analysis, 2nd ed.; Oxford University Press: New York, NY, USA, 1951; 282p.
150. Leontief, W.W. Input–Output Economics, 2nd ed.; Oxford University Press: New York, NY, USA, 1986.
151. Phillips, A.W. Stabilisation policy in a closed economy. Econ. J. 1954, 64, 290–323. [CrossRef]
152. Phillips, A.W. A. W. H. Phillips Collected Works in Contemporary Perspective; Leeson, R., Ed.; Cambridge
University Press: Cambridge, UK, 2000; 515p, ISBN 9780521571357.
153. Allen, R.G.D. Mathematical Economics; Macmillan: London, UK, 1956; 812p.
154. Allen, R.G.D. Mathematical Economics, 2nd ed.; Macmillan: London, UK, 1960; 812p, ISBN 978-1-349-81547-0.
[CrossRef]
155. Allen, R.G.D. Macro-Economic Theory. A Mathematical Treatment; Palgrave Macmillan: London, UK, 1967;
420p, ISBN 978-1-349-81541-8. [CrossRef]
156. Allen, R.G.D. Macro-Economic Theory. A Mathematical Treatment; Macmillan: London, UK, 1968; 420p, ISBN
978-033304112.
157. Solow, R.M. A contribution to the theory of economic growth. Q. J. Econ. 1956, 70, 65–94. Available online:
http://piketty.pse.ens.fr/files/Solow1956.pdf (accessed on 29 May 2019). [CrossRef]
158. Solow, R.M. Neoclassical growth theory. Handb. Macroecon. 1999, 1, 637–667. [CrossRef]
159. Swan, T.W. Economic growth and capital accumulation. Econ. Rec. 1956, 32, 334–361. [CrossRef]
160. Kaldor, N. A model of the trade cycle. Econ. J. 1940, 50, 78–92. [CrossRef]
161. Kaldor, N. Capital accumulation and economic growth. In The Theory of Capital; Hague, D.C., Ed.; Palgrave
Macmillan: London, UK, 1961; pp. 177–222. Available online: http://www.masterhdfs.org/masterHDFS/wp-
content/uploads/2014/05/Kaldor_1961.pdf (accessed on 29 May 2019).
162. Chang, W.W.; Smyth, D.J. The existence and persistence of cycles in a nonlinear model: Kaldor’s 1940 model
re-examined. Rev. Econ. Stud. 1971, 38, 37–44. [CrossRef]
163. Kalecki, M. A macrodynamic theory of business cycles. Econometrica 1935, 3, 327–344. Available online:
https://archive.org/details/Kalecki1935.AMacrodynamicTheoryOfBusinessCycles/page/n1 (accessed on 29
May 2019). [CrossRef]

27
Mathematics 2019, 7, 509

164. Lancaster, K. Mathematical Economics; The Macmillan Company: New York, NY, USA, 1968; ISBN
978-0023221316. (Revised edition, Dover Publications: Mineola, NY, USA, 2011; 448p, ISBN 978-0486653914).
165. Romer, D. Advanced Macroeconomics, 4th ed.; McGrow-Hill: New York, NY, USA, 2012; 716p, ISBN
978-0-07-351137-5.
166. Nicholson, W.; Snyder, C. Microeconomic Theory. Basic Principles and Extensions, 11th ed.; South-Western,
Cengage Learning: Masson, OH, USA, 2012; 758p, ISBN 978-111-1-52553-8.
167. Varian, H. Microeconomic Analysis, 3rd ed.; W.W. Norton: New York, NY, USA, 1992; 548p, ISBN 0-393-95735-7.
168. Tarasova, V.V.; Tarasov, V.E. Fractional dynamics of natural growth and memory effect in economics. Eur.
Res. 2016, 12, 30–37. [CrossRef]
169. Tarasova, V.V.; Tarasov, V.E. Economic model of natural growth with dynamic memory. Actual Probl.
Humanit. Nat. Sci. 2017, 4, 51–58. (In Russian). Available online: https://elibrary.ru/item.asp?id=28855511 or
https://publikacia.net/archive/uploads/pages/2017_4_2/15.pdf (accessed on 29 May 2019).
170. Tarasova, V.V.; Tarasov, V.E. Model of economic growth with constant rate and dynamic memory. Econ.
Sociol. Law 2017, 8, 18–24. (In Russian). Available online: https://elibrary.ru/item.asp?id=29938971 (accessed
on 29 May 2019).
171. Tarasova, V.V.; Tarasov, V.E. Economic growth model with constant pace and dynamic memory. Probl. Mod.
Sci. Educ. 2017, 2, 40–45. [CrossRef]
172. Tarasova, V.V.; Tarasov, V.E. Hereditary generalization of Harrod-Domar model and memory effects. J. Econ.
Entrep. 2016, 10, 72–78. (In Russian). Available online: https://elibrary.ru/item.asp?id=27220550 (accessed on
29 May 2019).
173. Tarasova, V.V.; Tarasov, V.E. Memory effects in hereditary Harrod-Domar model. Probl. Mod. Sci. Educ. 2016,
32, 38–44. (In Russian). Available online: https://elibrary.ru/item.asp?id=27253247 (accessed on 29 May 2019).
[CrossRef]
174. Tarasova, V.V.; Tarasov, V.E. Macroeconomic models with dynamic memory. J. Econ. Entrep. 2017, 3, 26–35.
(In Russian). Available online: https://elibrary.ru/item.asp?id=29078471 (accessed on 29 May 2019).
175. Tarasova, V.V.; Tarasov, V.E. Macroeconomic models with dynamic memory: Fractional calculus approach.
Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
176. Tarasov, V.E. Economic models with power-law memory. In Handbook of Fractional Calculus with Applications.
Volume 8: Applications in Engineering, Life and Social Sciences, Part B; Baleanu, D., Lopes, A.M., Tenreiro
Machado, J.A., Eds.; De Gruyter: Berlin, Germany; Boston, MA, USA, 2019; pp. 1–32. ISBN 978-3-11-057092.
[CrossRef]
177. Tarasova, V.V.; Tarasov, V.E. Keynesian model of economic growth with memory. Econ. Manag. Probl.
Solut. 2016, 10, 21–29. (In Russian). Available online: https://elibrary.ru/item.asp?id=27542599 (accessed on
29 May 2019).
178. Tarasova, V.V.; Tarasov, V.E. Memory effects in hereditary Keynes model. Probl. Mod. Sci. Educ. 2016,
38, 56–61. (In Russian). Available online: https://cyberleninka.ru/article/n/effekty-pamyati-v-ereditarnoy-
modeli-keynsa (accessed on 29 May 2019). [CrossRef]
179. Tarasova, V.V.; Tarasov, V.E. Influence of memory effects on world economics and business. Azimuth Sci.
Res. Econ. Manag. 2016, 5, 369–372. (In Russian). Available online: https://cyberleninka.ru/article/n/vliyanie-
effektov-pamyati-na-mirovuyu-ekonomiku-i-biznes (accessed on 29 May 2019).
180. Tarasova, V.V.; Tarasov, V.E. Dynamic intersectoral models with memory that generalize Leontief model.
J. Econ. Entrep. 2017, 2, 913–924. (In Russian). Available online: https://elibrary.ru/item.asp?id=28791089
(accessed on 29 May 2019).
181. Tarasova, V.V.; Tarasov, V.E. Dynamic intersectoral models with power-law memory. Commun. Nonlinear Sci.
Numer. Simul. 2018, 54, 100–117. [CrossRef]
182. Tarasova, V.V.; Tarasov, V.E. Chronological exponent for processes with memory and dynamic
intersectoral economic models. Sci. Educ. Today 2017, 4, 29–39. (In Russian). Available
online: https://cyberleninka.ru/article/n/hronologicheskaya-eksponenta-dlya-protsessov-s-pamyatyu-i-
dinamicheskie-mezhotraslevye-modeli-ekonomiki (accessed on 29 May 2019).
183. Tarasova, V.V.; Tarasov, V.E. Logistic map with memory from economic model. Chaos Solitons Fract. 2017, 95,
84–91. [CrossRef]
184. Tarasov, V.E.; Tarasova, V.V. Logistic equation with continuously distributed lag and application in economics.
Nonlinear Dyn. 2019. Submitted.

28
Mathematics 2019, 7, 509

185. Tarasov, V.E.; Tarasova, V.V. Phillips model with exponentially distributed lag and power-law memory.
Comput. Appl. Math. 2019, 38, 13. [CrossRef]
186. Tarasov, V.E.; Tarasova, V.V. Harrod-Domar growth model with memory and distributed lag. Axioms 2019, 8,
9. [CrossRef]
187. Tarasov, V.E.; Tarasova, V.V. Dynamic Keynesian model of economic growth with memory and lag. Mathematics
2019, 7, 178. [CrossRef]
188. Tarasova, V.V.; Tarasov, V.E. Productivity with fatigue and long memory: Fractional calculus approach. Int. J.
Appl. Comput. Math. 2019, 5, 16. [CrossRef]
189. Tarsasov, V.E. Rules for fractional-dynamic generalizations: Difficulties of constructing fractional dynamic
models. Mathematics 2019. (Special Issue “Mathematical Economics: Application of Fractional Calculus”.
Submitted).
190. Rehman, H.U.; Darus, M.; Salah, J. A note on Caputo’s derivative operator interpretation in economy. J. Appl.
Math. 2018, 2018, 7. [CrossRef]
191. Caputo, M. The memory function of the reactivity of investments caused by step of the prime rate. Econ.
Politica 2005, 22, 365–379. Available online: https://www.rivisteweb.it/doi/10.1428/21175 (accessed on 29 May
2019). [CrossRef]
192. Caputo, M. Planning output dynamics with memory. Econ. Politica 2009, 26, 79–97. Available online:
https://www.rivisteweb.it/doi/10.1428/29093 (accessed on 29 May 2019). [CrossRef]
193. Caputo, M. The memory response of populations and markets to extreme events. Econ. Politica 2012, 29,
261–282. Available online: https://www.rivisteweb.it/doi/10.1428/37632 (accessed on 29 May 2019). [CrossRef]
194. Caputo, M. The memory response of the populations to extreme events. In Atti dell’Accademia delle Scienze di
Ferrara; Volume 88, Anno Accademico 188, 2010–2011; Accademia delle Scienze di Ferrara: Ferrara, Italy,
2012; pp. 225–240. 433p, ISBN 88-7400-041-3.
195. Caputo, M. Memory oscillator for modeling socio-economic oscillatory responses to extreme events. Econ.
Politica 2013, 30, 221–239. Available online: https://www.rivisteweb.it/doi/10.1428/74241 (accessed on 29 May
2019). [CrossRef]
196. Caputo, M. The convergence of economic developments. In Stud. Nonlinear Dyn. Econom. 2012, 16, 23.
[CrossRef]
197. Caputo, M. The evolution and homogeneity of EU economies (with an econometric approach). Meccanica
2014, 49, 2237–2246. [CrossRef]
198. Caputo, M. The role of memory in modeling social and economic cycles of extreme events. Chapter 11
in A Handbook of Alternative Theories of Public Economics; Forte, F., Mudambi, R., Navarra, P.M.,
Eds.; Edward Elgar Publishing: Cheltenham, UK, 2014; pp. 245–259. 576p, ISBN 9781781004715.
Available online: https://books.google.ru/books?hl=ru&lr=&id=bekyAwAAQBAJ&oi=fnd&pg=PA245&ots=
4pzWJ8won7&sig=BKAM1zBH1dezS9CD78TOKK8wK8o&redir_esc=y#v=onepage&q&f=false (accessed
on 29 May 2019).
199. Caputo, M.; Kolari, J. An analytical model of the Fisher equation with memory functions. Altern. Perspect.
Financ. Account. 2001, 1, 1–14.
200. Caputo, M.; Di Giorgio, G. Monetary Policy, Memory and Output Dynamics. In Monetary Policy and
Institutions; Di Giorgio, G., Neri, F., Eds.; LUISS University Press: Rome, Italy, 2006; pp. 165–176.
201. Nakhushev, A.M. Loaded equations of mathematical economics. Reports of the Adyghe (Circassian).
Int. Acad. Sci. 2010, 12, 91–97. Available online: https://elibrary.ru/item.asp?id=16950586 (accessed on
29 May 2019).
202. Nakhushev, A.M. Section 4.6. Loaded equations of mathematical economics. In Loaded equations and their
application; Nakhushev, A.M., Ed.; Nauka: Moscow, Russia, 2012; pp. 145–153. 232p.
203. Kalazhokov, K.K. Some classes of dynamic models of productive and economic systems with “memory”.
News Kabard. -Balkar. Sci. Cent. Russ. Acad. Sci. 2012, 4, 26–34.
204. Nakhusheva, Z.A. On a one-sector macroeconomic model of long-term prognosis. News Kabard. -Balkar. Sci.
Cent. Russ. Acad. Sci. 2012, 4, 35–38.
205. Samuta, V.V.; Strelova, V.A.; Parovik, R.I. Nonlocal Solow model of neoclassical economic growth. Bull.
Krasec. Phys. Math. Sci. 2012, 2, 37–41. Available online: http://mi.mathnet.ru/eng/vkam72 (accessed on 29
May 2019). [CrossRef]

29
Mathematics 2019, 7, 509

206. Shpilko, Y.E.; Solomko, A.A.; Parovik, R.I. Parametrization Samuelson equation model for Evans fixing,
equilibrium price of the same product market. Bull. Krasec. Phys. Math. Sci. 2012, 2, 33–36. Available online:
http://mi.mathnet.ru/eng/vkam71 (accessed on 29 May 2019). [CrossRef]
207. Makarov, D.V. On a dynamic hereditarity system that simulates the economic cycle. Bull. Krasec. Phys.
Math. Sci. 2016, 2, 55–61. Available online: http://mi.mathnet.ru/eng/vkam132 (accessed on 29 May 2019).
[CrossRef]
208. Chu, S.Y.; Shane, C. Using the hybrid Phillips curve with memory to forecast US inflation. Stud. Nonlinear
Dyn. Econom. 2017, 21, 20160088. [CrossRef]
209. Pakhira, R.; Ghosh, U.; Sarkar, S. Study of memory effects in an inventory model using fractional calculus.
Appl. Math. Sci. 2018, 12, 797–824. Available online: http://www.m-hikari.com/ams/ams-2018/ams-17-20-
2018/p/ghoshAMS17-20-2018.pdf (accessed on 29 May 2019). [CrossRef]
210. Pakhira, R.; Ghosh, U.; Sarkar, S. Application of memory effects in an inventory model with linear
demand and no shortage. Int. J. Res. Advent Technol. 2018, 6, 1853–1871. Available online: http:
//www.ijrat.org/downloads/Vol-6/august-2018/paperID-68201804.pdf (accessed on 29 May 2019).
211. Pakhira, R.; Ghosh, U.; Sarkar, S. Study of memory effect in an inventory model with linear demand and
salvage value. Int. J. Appl. Eng. Res. 2018, 13, 14741–14751. Available online: https://www.ripublication.com/
ijaer18/ijaerv13n20_40.pdf (accessed on 29 May 2019).
212. Pakhira, R.; Ghosh, U.; Sarkar, S. Study of memory effect in an inventory model with linear demand and
shortage. Int. J. Math. Sci. Comput. 2019, 5, 54–70. [CrossRef]
213. Pakhira, R.; Ghosh, U.; Sarkar, S. Study of memory effect in an inventory model with quadratic type demand
rate and salvage value. Appl. Math. Sci. 2019, 13, 209–223. Available online: http://www.m-hikari.com/ams/
ams-2019/ams-5-8-2019/p/pakhiraAMS5-8-2019.pdf (accessed on 29 May 2019). [CrossRef]
214. Tejado, I.; Valerio, D.; Valerio, N. Fractional calculus in economic growth modeling. The Portuguese case.
In Proceedings of the Conference: 2014 International Conference on Fractional Differentiation and its
Applications (FDA’14), Catania, Italy, 23–25 June 2014; ISBN 978-1-4799-2591-9. [CrossRef]
215. Tejado, I.; Valerio, D.; Valerio, N. Fractional calculus in economic growth modelling. The Spanish case.
In CONTROLO’2014—Proceedings of the 11th Portuguese Conference on Automatic Control; Moreira, A.P.,
Matos, A., Veiga, G., Eds.; Lecture Notes in Electrical Engineering; Springer International Publishing: Cham,
Switzerland, 2015; Volume 321, pp. 449–458.
216. Tejado, I.; Valerio, D.; Perez, E.; Valerio, N. Fractional calculus in economic growth modelling: The economies
of France and Italy. In Proceedings of the International Conference on Fractional Differentiation and
its Applications, Novi Sad, Serbia, 18–20 July 2016; Spasic, D.T., Grahovac, N., Zigic, M., Rapaic, M.,
Atanackovic, T.M., Eds.; University of Novi Sad: Novi Sad, Serbia, 2016; pp. 113–123.
217. Tejado, I.; Valerio, D.; Perez, E.; Valerio, N. Fractional calculus in economic growth modelling: The Spanish
and Portuguese cases. Int. J. Dyn. Control 2017, 5, 208–222. [CrossRef]
218. Tejado, I.; Perez, E.; Valerio, D. Economic growth in the European Union modelled with fractional derivatives:
First results. Bull. Pol. Acad. Sci. Tech. Sci. 2018, 66, 455–465. [CrossRef]
219. Tejado, I.; Perez, E.; Valerio, D. Fractional calculus in economic growth modelling of the group of seven. Ssrn
Electron. J. 2018, 22, 139–157. [CrossRef]
220. Tejado, I.; Perez, E.; Valerio, D. Fractional calculus in economic growth modelling of the group of seven.
Fract. Calc. Appl. Anal. 2019, 22, 139–157. [CrossRef]
221. Luo, D.; Wang, J.R.; Feckan, M. Applying fractional calculus to analyze economic growth modelling. J. Appl.
Math. Stat. Inform. 2018, 14, 25–36. [CrossRef]
222. Tarasov, V.E. Self-organization with memory. Commun. Nonlinear Sci. Numer. Simul. 2019, 72, 240–271.
[CrossRef]
223. Zhang, W.-B. Synergetic Economics. Time and Change in Nonlinear Economics; Springer: Berlin/Heidelberg,
Germany, 1991; 246p, ISBN 978-3-642-75911-6.
224. Koblo, R.B. The Visible Hand: Synergetic Microfoundation of Macroeconomic Dynamics; Springer:
Berlin/Heidelberg, Germany, 1991; 131p, ISBN 978-3-540-54595-8.
225. Stanley, H.E.; Amaral, L.A.N.; Buldyrev, S.V.; Gopikrishnan, P.; Plerou, V.; Salinger, M.A. Self-organized
complexity in economics and finance. Proc. Natl. Acad. Sci. United States Am. 2002, 99 (Suppl. S1), 2561–2565.
[CrossRef] [PubMed]

30
Mathematics 2019, 7, 509

226. Caputo, M.; Fabrizio, M. A new definition of fractional derivative without singular kernel. Prog. Fract. Differ.
Appl. 2015, 1, 73–85. [CrossRef]
227. Caputo, M.; Fabrizio, M. Applications of new time and spatial fractional derivatives with exponential kernels.
Prog. Fract. Differ. Appl. 2016, 2, 1–11. [CrossRef]
228. Tarasov, V.E.; Tarasova, S.S. Fractional and integer derivatives with continuously distributed lag. Commun.
Nonlinear Sci. Numer. Simul. 2019, 70, 125–169. [CrossRef]
229. Caputo, M. Mean fractional-order-derivatives differential equations and filters. Ann. Dell’universita Di
Ferrara 1995, 41, 73–84. [CrossRef]
230. Bagley, R.L.; Torvik, P.J. On the existence of the order domain and the solution of distributed order
equations—Part I. Int. J. Appl. Math. 2000, 2, 865–882.
231. Bagley, R.L.; Torvik, P.J. On the existence of the order domain and the solution of distributed order
equations—Part II. Int. J. Appl. Math. 2000, 2, 965–987.
232. Caputo, M. Distributed order differential equations modelling dielectric induction and diffusion. Fract. Calc.
Appl. Anal. 2001, 4, 421–442.
233. Lorenzo, C.F.; Hartley, T.T. Variable order and distributed order fractional operators. Nonlinear Dyn. 2002, 29,
57–98. [CrossRef]
234. Jiao, Z.; Chen, Y.Q.; Podlubny, I. Distributed-Order Dynamic Systems: Stability, Simulation, Applications and
Perspectives; Springer: London, UK, 2012; 90p.
235. Nakhushev, A.M. On the positivity of continuous and discrete differentiation and integration operators that
are very important in fractional calculus and in the theory of equations of mixed type. Differ. Equ. 1998, 34,
103–112. Available online: http://mi.mathnet.ru/eng/de9635 (accessed on 29 May 2019).
236. Nakhushev, A.M. Fractional Calculus and its Application; Fizmatlit: Moscow, Russia, 2003; 272p, ISBN
5-9221-0440-3. (In Russian)
237. Pskhu, A.V. On the theory of the continual integro-differentiation operator. Differ. Equ. 2004, 40, 128–136.
[CrossRef]
238. Pskhu, A.V. Partial Differential Equations of Fractional Order; Nauka: Moscow, Russia, 2005; 199p, ISBN
5-02-033721-8. (In Russian)
239. Kochubei, A.N. General fractional calculus, evolution equations, and renewal processes. Integral Equ. Oper.
Theory 2011, 71, 583–600. [CrossRef]
240. Kochubei, A.N. General fractional calculus. Chapter 5. In Handbook of Fractional Calculus with Applications;
Kochubei, A., Luchko, Y., Tenreiro Machado, J.A., Eds.; De Gruyter: Berlin, Germany; Boston, MA, USA,
2019; pp. 111–126. [CrossRef]
241. Tarasov, V.E.; Tarasova, S.S. Probabilistic interpretation of Kober fractional integral of non-integer order. Prog.
Fract. Differ. Appl. 2019, 5, 1–5. [CrossRef]
242. Kochubei, A.N. Distributed order calculus and equations of ultraslow diffusion. J. Math. Anal. Appl. 2008,
340, 252–281. [CrossRef]
243. Sin, C.-S. Well-posedness of general Caputo-type fractional differential equations. Fract. Calc. Appl. Anal.
2018, 21, 819–832. [CrossRef]
244. Kochubei, A.N.; Kondratiev, Y. Fractional kinetic hierarchies and intermittency. Kinetic and Related Models.
Am. Inst. Math. Sci. 2017, 10, 725–740. [CrossRef]
245. Kochubei, A.N.; Kondratiev, Y. Growth equation of the general fractional calculus. Mathematics 2019. (Special
Issue “Mathematical Economics: Application of Fractional Calculus” edited by V.E. Tarasov. Submitted).
246. Achdou, Y.; Buera, F.J.; Lasry, J.-M.; Lions, P.-L.; Moll, B. Partial differential equation models in
macroeconomics. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2014, 372, 201303977. [CrossRef]
247. Burger, M.; Caffarelli, L.; Markowich, P.A. Partial differential equation models in the socio-economic
sciences. Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2014, 372, 20130406. Available online:
https://royalsocietypublishing.org/doi/pdf/10.1098/rsta.2013.0406 (accessed on 29 May 2019). [CrossRef]
248. Malinowska, A.B.; Odzijewicz, T.; Torres, D.F.M. Advanced Methods in the Fractional Calculus of Variations;
Springer: Heidelberg, Germany; New York, NY, USA, 2015; 135p.
249. Almeida, R.; Tavares, D.; Torres, D.F.M. The Variable-Order Fractional Calculus of Variations; Springer:
Heidelberg, Germany; New York, NY, USA, 2019; 124p.
250. Tarasov, V.E.; Zaslavsky, G.M. Nonholonomic constraints with fractional derivatives. J. Phys. A 2006, 39,
9797–9815. [CrossRef]

31
Mathematics 2019, 7, 509

251. Tarasov, V.E. Fractional Stability. ArXiv 2007, arXiv:0711.2117. Available online: https://arxiv.org/pdf/0711.
2117.pdf (accessed on 29 May 2019).
252. Tarasov, V.E. Fractional-order variational derivative. Int. J. Appl. Math. 2014, 27, 491–518. [CrossRef]
253. Chikrii, A.A. Game dynamic problems for systems with fractional derivatives. In Pareto Optimality, Game
Theory and Equilibria; Chinchuluun, A., Pardalos, P.M., Migdalas, A., Pitsoulis, L., Eds.; Springer: New York,
NY, USA, 2008; Volume 17, pp. 349–386.
254. Chikrii, A.A.; Matychyn, I.I. Game problems for fractional-order linear systems. Proc. Steklov Inst. Math.
2010, 268, 54–70. (In Russian). Available online: http://mi.mathnet.ru/eng/timm419 (accessed on 29 May
2019). [CrossRef]
255. Chikriy, A.A.; Chentsov, A.G.; Matychyn, I.I. Differential games of the fractional order with separated
dynamics. J. Autom. Inf. Sci. 2009, 41, 17–27. [CrossRef]
256. Chikrii, A.A.; Matychyn, I.I. Riemann–Liouville, Caputo, and sequential fractional derivatives in differential
games. In Advances in Dynamic Games; Breton, M., Szajowski, K., Eds.; Springer Science Business Media:
Berlin, Germany, 2011; pp. 61–81.
257. Chikrii, A.A.; Matychyn, I.I. On linear conflict-controlled processes with fractional derivatives. Work.
Inst. Math. Mech. Ural Branch Russ. Acad. Sci. 2011, 17, 256–270. (In Russian). Available online:
http://mi.mathnet.ru/eng/timm711 (accessed on 29 May 2019).
258. Matychyn, I.I.; Chikriy, A.A.; Onyshchenko, V.V. Conflict-controlled processes involving fractional differential
equations with impulses. Math. Balk. 2012, 26, 159–168. [CrossRef]
259. Matychyn, I.I.; Onyshchenko, V.V. Differential games of fractional order with impulse effect. J. Autom. Inf.
Sci. 2015, 47, 43–53. [CrossRef]
260. Zhukovskiy, V.J.; Chikriy, A.A. On discrete conflict-controlled processes described by Grunvald-Letnikov
fractional systems. J. Autom. Inf. Sci. 2015, 47, 24–34. [CrossRef]

© 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

32
 mathematics
Review
On the Advent of Fractional Calculus in Econophysics
via Continuous-Time Random Walk
Francesco Mainardi
Department of Physics and Astronomy, University of Bologna, & The National Institute of Nuclear Physics
(INFN), Via Irnerio 46, I-40126 Bologna, Italy; [email protected]

Received: 30 March 2020; Accepted: 18 April 2020; Published: 21 April 2020 

Abstract: In this survey article, at first, the author describes how he was involved in the late 1990s
on Econophysics, considered in those times an emerging science. Inside a group of colleagues the
methods of the Fractional Calculus were developed to deal with the continuous-time random walks
adopted to model the tick-by-tick dynamics of financial markets Then, the analytical results of this
approach are presented pointing out the relevance of the Mittag-Leffler function. The consistence of
the theoretical analysis is validated with fitting the survival probability for certain futures (BUND and
BTP) traded in 1997 at LIFFE, London. Most of the theoretical and numerical results (including figures)
reported in this paper were presented by the author at the first Nikkei symposium on Econophysics,
held in Tokyo on November 2000 under the title “Empirical Science of Financial Fluctuations” on
behalf of his colleagues and published by Springer. The author acknowledges Springer for the license
permission of re-using this material.

Keywords: econophysics; continuous-time random walk (CTRW); fractional calculus; Mittag–Leffler

functions; Laplace transform; Fourier transform

1. Introduction
As we read from Wikipedia [1], “Econophysics is a heterodox interdisciplinary research field,
applying theories and methods originally developed by physicists in order to solve problems in
economics, usually those including uncertainty or stochastic processes and nonlinear dynamics.
Some of its application to the study of financial markets has also been termed statistical finance
referring to its roots in statistical physics”. Here we stress that the problems dealt with in Econophysics
are essentially devoted to statistical finance. The first book on Econophysics was by R. N. Mantegna and
H. E. Stanley in 2000 [2], followed by a number of books including that by Bouchaud and Potters [3].
The importance of random walks in finance has been known since the seminal work of Bachelier [4]
which was completed at the end of the XIXth century, more than a hundred years ago. The ideas of
Bachelier were further carried out and improved by many scholars see, for example, Mandelbrot [5],
Cootner [6], Samuelson [7], Black and Scholes [8], Merton [9], Mantegna and Stanley [2], Bouchaud and
Potters [3].
The term “Econophysics” was coined by H. Eugene Stanley (Boston University) to describe a
number of papers written by physicists (including his Ph.D. and Post-Doc students) in the problems
of stock and other markets. The inaugural meeting on Econophysics was organized in July 1997
in Budapest by János Kertész and Imre Kondor but the book of proceedings edited by them to be
published with the impressive title Econophysics, an Emerging Science was surprisingly cancelled by the
publisher Kluwer during the checking of galley proofs.
The author of this paper was alerted on this conference in Budapest from a national newspaper
so he was able to get an invitation to present his current research on Lévy stable distributions
with his former student Ph.D. Paolo Paradisi and his colleague, the late Professor Rudolf Gorenflo.

Mathematics 2020, 8, 641; doi:10.3390/math8040641 33 www.mdpi.com/journal/mathematics

Mathematics 2020, 8, 641

Indeed, the author was impressed by the derivation of Lévy distributions as fundamental solutions of
a space fractional diffusion equation so generalizing the well known Gaussian distribution and the
Brownian motion. However, because of the sad end of the book of Budapest proceedings, this paper
was not published, but later submitted as an E-print in arXiv, see [10].
Then, after the author’s seminar stable distributions held in Rome, at the University “La Sapienza”,
he was approached by Dr. Enrico Scalas of the University of Alessandria (Italy) on the possibility of
collaborating with him and his student Marco Raberto and Prof. Rudolf Gorenflo on applications of
methods of Fractional Calculus in Econophyscs. As a matter of fact, Enrico Scalas was the director
of our group (in Italian called as “Il Regista”, taking inspiration from directing movies) being expert
of finance. So a collaboration started inside our group after an introduction paper by Scalas et al. in
Physica A, see Reference [11], followed by a series of papers, see Mainardi et al. [12], Gorenflo et al. [13],
Raberto et al. [14]. In these papers, the authors have argued that “the continuous-time random walk
(CTRW) model, formerly introduced in Statistical Mechanics by Montroll and Weiss [15], can provide a
phenomenological description of tick-by-tick dynamics in financial markets” and they have discussed
some applications concerning “high frequency exchanges of bond futures”. A particular mention
is given to the paper by Mainardi et al. [16], being presented by the author under invitation of the
organizer H. Takayasu at an international conference on Econophysics held in Tokyo, November 2000.
Later, other papers were published by our group, see References [17,18] to summarize our
approach to CTRW via fractional master equations. Furthermore, Scalas published papers on
the relevance of Fractional Calculus in dealing with CTRW in Econophysics, see, for example,
Reference [19] and chapters in the book authored by Baleanu et al., see Reference [20,21].
The purpose of this paper is to survey our phenomenological theory of tick-by-tick dynamics in
financial markets, based on the continuous-time random walk (CTRW) model, by pointing out the
relevance of Fractional Calculus and of the Mittag-Leffler function, a special function almost unknown
in those times. Nowadays the Mittag-Leffler function has hundreds citations and a treatise on the
functions of the Mittag-Leffler type has been published in 2014 by Gorenflo et al. [22], of which a
second enlarged edition will hopefully appear in 2020.
The body of our paper is essentially based on the above references and on the conferences and
seminars that anybody of our group could give by invitation in several Institutions spread in the world.
In particular, most of the the theoretical and numerical results reported in this paper were presented
by the author at the first Nikkei symposium on Econophysics, held in Tokyo on November 2000 under
the title “Empirical Science of Financial Fluctuations” on behalf of his colleagues. published in the
Springer volume edited by Takayasu [16].
Our survey-article is divided as follows: Section 2 is devoted to revisit the theoretical framework
of the CTRW model. We provide the most appropriate form for the general master equation,
which is expected to govern the evolution of the probability density for non-local and non-Markovian
processes. In Section 3 we propose a master equation containing a time derivative of fractional order
to characterize non-Markovian processes with long memory. In this respect, we outline the central role
played by the Mittag-Leffler function which exhibits an algebraic tail consistent with such processes.
Section 4 is devoted to explain how the CTRW model can be used in describing the financial time
series of the log-prices of an asset, for which the time interval between two consecutive transactions
varies stochastically. In particular we test the theoretical predictions on the survival-time probability
against empirical market data. The empirical analysis concerns high-frequency prices time series of
German and Italian bond futures. Finally, in Section 5, we draw the main conclusions. For the sake
of convenience, the Appendix A introduces in a simple way the correct notion of time derivative of
fractional order.

34
Mathematics 2020, 8, 641

2. The CTRW Model in Statistical Physics

We recall that the CTRW model leads to the general problem of computing the probability density
function (pdf) p( x, t) (x ∈ IR , t ∈ IR + ) of finding, at position x at time t, a particle (the walker)
which performs instantaneous random jumps ξ i = x (ti ) − x(ti−1 ) at random instants ti (i ∈ IN),
where 0 = t0 < t1 < t2 < . . . . We denote by τi = ti − t1−1 the (so-called) waiting times. As usual, it is
assumed that the particle is located at x0 = 0 for t0 = 0 , which means p( x, 0) = δ( x) . We denote by
ϕ(ξ, τ ) the joint probability density for jumps and waiting times.
The CTRW is generally defined through the requirement that the ξ i and τi are independent
identically distributed (i.i.d.) random variables with pd f ’s independent of each other, so that we have
∞  +∞
the factorization ϕ(ξ, τ ) = w(ξ ) ψ(τ ) , which implies w(ξ ) = 0 ϕ(ξ, τ ) dτ , ψ(τ ) = −∞ ϕ(ξ, τ ) dξ .
The marginal probability densities w and ψ are called jump pdf and waiting-time pdf, respectively.
We now provide further details on the densities w(ξ ) , ϕ(τ ) in order to derive their relation with
the pdf p( x, t) .
The jump pdf w(ξ ) represents the pd f for transition of the walker from a point x to a point x + ξ , so it
is also called the transition pd f . The waiting-time pd f represents the pd f that a step is taken at the instant
ti−1 + τ after the previous one that happened at the instant ti−1 , so it is also called the pausing-time pd f .
Therefore, the probability that τ ≤ ti − ti−1 < τ + dτ is equal to ψ(τ ) dτ . The probability that a given
waiting interval is greater or equal to τ will be denoted by Ψ(τ ) , which is defined in terms of ψ(τ ) by
 ∞  τ
d
Ψ(τ ) = ψ(t ) dt = 1 − ψ(t ) dt , ψ(τ ) = − Ψ(τ ) . (1)
τ 0 dτ

τ
We note that 0 ψ(t ) dt represents the probability that at least one jump is taken at some instant
in the interval [0, τ ), hence Ψ(τ ) is the probability that the walker is sitting in x at least during the
time interval of duration τ after a jump. Recalling that t0 = 0 , we also note that Ψ(t) represents the so
called survival probability, namely the probability of finding the walker at the initial position x0 = 0
until time instant t .
Now, only based upon the previous probabilistic arguments, we can derive the evolution equation
for the pdf p( x, t) , that we shall call the master equation of the CTRW. In fact, we are led to write
 t  +∞

p( x, t) = δ( x ) Ψ(t) + ψ(t − t ) w( x − x  ) p( x  , t ) dx  dt , (2)
0 −∞

where we recognize the role of the survival probability Ψ(t) and of the pd f ’s ψ(t) , w( x ) . The first term
in the RHS of Equation (2) expresses the persistence (whose strength decreases with increasing time)
of the initial position x = 0. The second term (a spatio-temporal convolution) gives the contribution
to p( x, t) from the walker sitting in point x  ∈ IR at instant t < t jumping to point x just at instant t ,
after stopping (or waiting) time t − t . Furthermore, as a check for the correctness of Equation (2) we
 +∞
can easily verify that p( x, t) ≥ 0 for all t ≥ 0 and x ∈ IR , and −∞ p( x, t) dx = 1 for all t ≥ 0 .
Originally the master equation was derived by Montroll and Weiss in 1965, see [15], recurring to
the tools of the Fourier-Laplace transforms. These authors showed that the Fourier-Laplace transform
of p( x, t) satisfies a characteristic equation, now called the Montroll-Weiss equation, which reads

  (s)
p(κ, s) = Ψ
1
, with  (s) = 1 − ψ̃(s) .
Ψ (3)
1−w (s)
 (κ ) ψ s

Here, we have adopted the following standard notation for the generic Fourier and Laplace transforms:
 +∞  ∞
F { f ( x ); κ } = f(κ ) = e iκx f ( x ) dx , L { g(t); s} = g(s) = e −st g(t) dt ,
−∞ 0

35
Mathematics 2020, 8, 641

where f ( x ) (x ∈ IR) and g(t) (t ∈ IR + ) are sufficiently well-behaved functions of their arguments. It is
straightforward to verify the equivalence between the Equations (2) and (3) by recalling the well-known
properties of the Fourier and Laplace transforms with respect to the space and time convolution.
Hereafter, we present an alternative form to Equation (2), formerly proposed by Mainardi et al. [12],
which involves the first time derivative of p( x, t) (along with an additional auxiliary function), so that
the resulting equation can be interpreted as an evolution equation of Fokker-Planck-Kolmogorov type.
To this purpose, we re-write Equation (3) as

 (s) s 
Φ  (κ ) − 1] p(κ, s) ,
p(κ, s) − 1 = [w (4)

where
  
 (s) = 1 − ψ(s) = Ψ(s) = Ψ(s) .
Φ (5)

s ψ(s) 
ψ(s)  (s)
1 − sΨ
Then our master equation reads
 t  +∞
∂
Φ(t − t ) p( x, t ) dt = − p( x, t) + w( x − x  ) p( x  , t) dx  , (6)
0 ∂t −∞

where the “auxiliary” function Φ(t) , being defined through its Laplace transform in Equation (5),
t
is such that Ψ(t) = 0 Φ(t − t ) ψ(t ) dt . We remind the reader that Equation (6), combined with the
initial condition p( x, 0) = δ( x ) , is equivalent to Equation (4), and then its solution represents the
Green function or the fundamental solution of the Cauchy problem for Equation (6).
From Equation (6) we recognize the role of Φ(t) as a “memory function”. As a consequence,
the CTRW turns out to be in general a non-Markovian process. However, the process is “memoryless”,
namely “Markovian” if (and only if) the above memory function degenerates into a delta function
(multiplied by a certain positive constant) so that Ψ(t) and ψ(t) may only differ by a multiplying
positive constant. By appropriate choice of the unit of time we assume Φ  (s) = 1 , so Φ(t) = δ(t) , t ≥ 0 .
In this case we derive
1
ψ  (s) =
(s) = Ψ , so ψ(t) = Ψ(t) = e −t , t ≥ 0 . (7)
1+s

Then Equation (6) reduces to

 +∞
∂
p( x, t) = − p( x, t) + w( x − x  ) p( x  , t) dx  , p( x, 0) = δ( x ) . (8)
∂t −∞

This is up to a change of the unit of time (which means multiplication of the RHS by a
positive constant), the most general master equation for a Markovian CTRW; it is usually called the
Kolmogorov-Feller equation.

3. The Time-Fractional Master Equation with “Long-Memory”

Let us now consider “long-memory” processes, namely non-Markovian processes characterized
by a memory function Φ(t) exhibiting a power-law time decay. To this purpose a natural choice is

t− β
Φ(t) = , t ≥ 0, 0 < β < 1. (9)
Γ (1 − β )

Thus, Φ(t) is a weakly singular function that, in the limiting case β = 1 , reduces to Φ(t) = δ(t) ,
according to the formal representation of the Dirac generalized function, δ(t) = t−1 /Γ(0) , t ≥ 0 .

36
Mathematics 2020, 8, 641

As a consequence of the choice (9), we see that (in this peculiar non-Markovian situation) our
master equation (6) contains a time fractional derivative. In fact, by inserting into Equation (4) the
 (s) = 1/s1− β , we get
Laplace transform of Φ(t) , Φ

sβ   (κ ) − 1] p(κ, s) ,
p(κ, s) − s β−1 = [w 0 < β < 1, (10)

so that the resulting Equation (6) can be written as

 +∞
∂β
p( x, t) = − p( x, t) + w( x − x  ) p( x  , t) dx  , p( x, 0) = δ( x ) , (11)
∂t β −∞

where ∂ β /∂t β is the pseudo-differential operator explicitly defined in the Appendix, that we call the
Caputo fractional derivative of order β . Thus Equation (11) can be considered as the time-fractional
generalization of Equation (8) and consequently can be called the time-fractional Kolmogorov-Feller equation.
Our choice for Φ(t) implies peculiar forms for the functions Ψ(t) and ψ(t) that generalize the
exponential behaviour (7) of the Markovian case. In fact, working in the Laplace domain we get from
(5) and (9)
β −1 1
 (s) = s
Ψ , ψ (s) = , 0 < β < 1, (12)
1 + sβ 1 + sβ
from which by inversion we obtain for t ≥ 0

d
Ψ(t) = Eβ (−t β ) , ψ(t) = − E (−t β ) , 0 < β < 1, (13)
dt β
where Eβ denotes an entire transcendental function, known as the Mittag-Leffler function of order β ,
defined in the complex plane by the power series
∞
zn
Eβ ( z ) : = ∑ Γ ( β n + 1)
, β > 0, z ∈ C. (14)
n =0

For detailed information on the Mittag-Leffler-type functions and their Laplace transforms the
reader may consult e.g., the books [22,23] and the articles [24,25].
Hereafter, we find it convenient to summarize the features of the functions Ψ(t) and ψ(t) most
relevant for our purposes. We begin to quote their series expansions and asymptotic representations:
⎧ ∞
⎪
⎪ t βn
⎨= ∑ (−1)n Γ( β n + 1) , t≥ 0
Ψ(t) n =0 (15)
⎪
⎩ ∼ sin ( βπ ) Γ( β) ,
⎪
t → ∞,
π tβ
and ⎧ ∞
⎪
⎪ 1 t βn
⎨=
t 1− β
∑ (−1)n Γ( β n + β) , t≥ 0
ψ(t) n =0 (16)
⎪
⎩ ∼ sin ( βπ ) Γ( β + 1) ,
⎪
t → ∞.
π t β +1
In the limit for β → 1 we recover the exponential functions of the Markovian case. We note
that for 0 < β < 1 both functions ψ(t), Ψ(t), even if losing their exponential decay by exhibiting
power-law tails for large times, keep the “completely monotonic” character. Complete monotonicity of
the functions ψ(t), Ψ(t), t > 0, means:

dn dn
(−1)n Ψ(t) ≥ 0 , (−1)n ψ(t) ≥ 0 , n = 0, 1, 2, . . . (17)
dtn dtn

37
Mathematics 2020, 8, 641

or equivalently, their representability as (real) Laplace transforms of non-negative functions. It may

be instructive to note that for sufficiently small times Ψ(t) exhibits a behaviour similar to that of a
stretched exponential; in fact we have

tβ
Eβ (−t β ) 1 − exp{−t β /Γ(1 + β)} , 0 ≤ t
1. (18)
Γ ( β + 1)

4. The CTRW Model in Statistical Finance

The price dynamics in financial markets can be mapped onto a random walk whose properties
are studied in continuous, rather than discrete, time, see, for example [9]. As a matter of fact, there are
various ways in which to embed a random walk in continuous time. Here, we shall base our approach
on the CTRW discussed in Section 2, in which time intervals between successive steps are i.i.d.
random variables.
Let S(t) denote the price of an asset or the value of an index at time t. In finance, returns
rather than prices are considered. For this reason, in the following we shall take into account
the variable x (t) = log S(t), that is the logarithm of the price. Indeed, for a small price variation
ΔS = S(ti ) − S(ti−1 ), the return r = ΔS/S(ti−1 ) and the logarithmic return rlog = log[S(ti )/S(ti−1 )]
virtually coincide. The statistical physicist will recognize in x the position of a random walker jumping
in one dimension. Thus, in the following, we shall use the language and the notations of Section 2.
In financial markets, prices are fixed when demand and offer meet and a transaction occurs.
In this case, we say that a trade takes place. As a consequence, not only prices but also waiting times
between two consecutive transactions can be modelled as random variables. In agreement with the
assumptions of Section 2, we consider the returns ξ i = x (ti ) − x (ti−1 ) as i.i.d random variables with
pdf w(ξ ) and the waiting times τi = ti − ti−1 as i.i.d. random variables with pdf ψ(τ ) . In real processes
of financial markets this independence hypothesis may not strictly hold for their duration or not be
verified at all. Therefore, it may be considered with caution.
In the following, we limit ourselves to investigate the consistency of the long-memory process
analyzed in Section 3 with respect to the empirical data concerning exchanges of certain financial
derivatives. We have considered the waiting time distributions of certain futures traded at LIFFE in 1997
and estimated the corresponding empirical survival probabilities. LIFFE stands for London International
Financial Futures (and Options) Exchange. It is a London-based derivative market; for further information,
see http://www.liffe.com. Futures are derivative contracts in which a party agrees to sell and the other
party to buy a fixed amount of an underlying asset at a given price and at a future delivery date.
As underlying assets, we have chosen German and Italian Government bonds, called BUND
and BTP respectively, for both of which the delivery dates are June and September 1997. BUND
and BTP (Buoni del Tesoro Poliennali) are respectively the German and Italian word for BOND
(middle and long term Government bonds with fixed interest rate). Usually, for a future with a
certain maturity, transactions begin 4 or 5 months before the delivery date. At the beginning, there
are few trades a day, but closer to the delivery there may be more than 1000 transactions a day. For
each maturity, the total number of transactions is greater than 160,000. Hence, these types of financial
instruments are particularly interesting for the analysis of the waiting times distributions between
consecutive transactions.
In Figures 1 and 2 we plot Ψ(τ ) for the four cases (June and September delivery dates for BUND
and BTP). The circles refer to market data and represent the probability of a waiting time greater than
the abscissa τ. We have determined about 500–600 values of Ψ(τ ) for τ in the interval between 1 s and
50,000 s, neglecting the intervals of market closure. The solid line is a two-parameter fit obtained by
using the Mittag-Leffler type function

Ψ(τ ) = Eβ −(γτ ) β , (19)

38
Mathematics 2020, 8, 641

where β is the index of the Mittag-Leffler function and γ is a time-scale factor, depending on the time
unit. The dash-dotted line is the stretched exponential function exp{−(γτ ) β )/Γ(1 + β)} , see the RHS
of Equation (18), whereas the dashed line is the power law function (γτ )− β /Γ(1 − β), see the RHS of
the second Equation in (15), noting that Γ( β) sin( βπ )/π = 1/Γ(1 − β) . The Mittag-Leffler function
well interpolates between these two limiting behaviours—the stretched exponential for small times,
and the power law for large ones.
As regards the BUND futures we can summarize as follows. For the June delivery date we get
an index β = 0.96 , and a scale factor γ = 1/12 , whereas, for the September delivery date, we have
β = 0.95 , and γ = 1/12 . The fits in the plots of Figure 1 have a reduced chi square χ2 0.3. As regards
the BTP futures we can summarize as follows. For both the June and September delivery dates we get
the same index β = 0.96 , and the same scale factor γ = 1/13 . The fits in the plots of Figure 2 have a
reduced chi square χ2 0.2.
To the possible objection that, in all four cases here treated, β does not differ significantly from 1
and so the process still could be Markovian, we answer that then we would have Ψ(τ ) = exp(−γτ )
and the graph of Ψ(τ ) would look completely different for sufficiently long times.

0 0
10 10

−1 BUND June ’97 −1 BUND Sept. ’97

10 10

−2 −2
10 10
Ψ(τ)

Ψ(τ)

−3 −3
10 10

−4 −4
10 10

−5 −5
10 10

−6 −6
10 0 1 2 3 4 5
10 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10 10
τ {s} τ {s}

Figure 1. Survival probability for BUND futures with delivery date: June 1997 (left) September 1997
(right). The Mittag-Leffler function (solid line) is compared with the stretched exponential (dash-dotted
line) and the power (dashed line) functions: {β = 0.96 , γ = 1/12} (left); {β = 0.95 , γ = 1/12)} (right).
0 0
10 10

−1 BTP June ’97 −1 BTP Sept. ’97

10 10

−2 −2
10 10
Ψ(τ)

Ψ(τ)

−3 −3
10 10

−4 −4
10 10

−5 −5
10 10

−6 −6
10 0 1 2 3 4 5
10 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10 10
τ {s} τ {s}

Figure 2. Survival probability for BTP futures with delivery date: June 1997 (left) September 1997
(right). The Mittag-Leffler function (solid line) is compared with the stretched exponential (dash-dotted
line) and the power (dashed line) functions: {β = 0.96 , γ = 1/13} (left); {β = 0.96 , γ = 1/13)} (right).

39
Mathematics 2020, 8, 641

5. Conclusions
In this paper, we have reviewed our phenomenological theory of tick-by-tick dynamics in financial
markets, based on the continuous-time random walk (CTRW) model. The theory can take into account
the possibility of the non-Markovian character of financial time series by means of a generalized master
equation with a time fractional derivative. We have presented predictions of the behaviour of the
waiting-time probability density by introducing a special function of Mittag-Leffler type whose decay
interpolates from a stretched exponential at small times to a power-law for long times. This function
has been successfully applied in the empirical analysis of high-frequency prices time series of German
and Italian bond futures. We may note the common behaviour of the survival probabilities found from
the trading of the above assets. This might be corroborated or not in other cases.

Funding: This research received no external funding.

Acknowledgments: The author is grateful to V. Tarasov for having invited him to survey the former approach to
Econophysics using the methods of Fractional Calculus that the author has shared with the colleagues R. Gorenflo,
M. Raberto and E. Scalas. The research activity of the author has been carried out in the framework of the activities
of the National Group of Mathematical Physics (GNFM, INdAM).
Conflicts of Interest: The author declares no conflict of interest.

Appendix A. The Caputo Fractional Derivative

For the sake of convenience of the reader here we present an introduction to the Caputo fractional
derivative starting from its representation in the Laplace domain and pointing out its difference with
respect to the standard Riemann-Liouville fractional derivative. So doing, we avoid the subtleties lying
in the inversion of fractional integrals. If f (t) is a (sufficiently well-behaved) function with Laplace
∞
transform L { f (t); s} = f(s) = 0 e −st f (t) dt , we have
 
dβ
L f ( t ); s = s β f(s) − s β−1 f (0+ ) , 0 < β < 1, (A1)
dt β

if we define  t
dβ 1 d f (τ ) dτ
f (t) := . (A2)
dt β Γ (1 − β ) 0 dτ (t − τ ) β
We can also write
 
dβ 1 d t dτ
f (t) = [ f (τ ) − f (0+ )] , (A3)
dt β Γ(1 − β) dt 0 (t − τ ) β
 
dβ 1 d t f (τ ) t− β
f (t) = dτ − f (0+ ) . (A4)
dt β Γ(1 − β) dt 0 (t − τ ) β Γ (1 − β )
The reader should observe that first term in the R.H.S. of (A4) provides the most usual Riemann-Liouville
fractional derivative, see e.g., [26].For more details on the Caputo fractional derivative we refer to
Refs. [24,27].

References
1. Wikipedia. Econophysics. 2020. Available online: https://en.wikipedia.org/wiki/Econophysics (accessed on
17 April 2020).
2. Mantegna, R.N.; Stanley, H.E. An Introduction to Econophysics; Cambridge University Press: Cambridge,
UK, 2000.
3. Bouchaud, J.-P.; Potters, M. Theory of Financial Risk: From Statistical Physics to Risk Management; Cambridge
Univ. Press: Cambridge, UK, 2000.

40
Mathematics 2020, 8, 641

4. Bachelier, L.J.B. Théorie de la Speculation. Ann. École Norm. Supérieure 1900, 17, 21–86; Reprinted by Editions
Jaques: Gabay, Paris, 1995. (English Translation with Editor’s Notes in The Random Character of Stock
Market Prices, pp. 17–78). [CrossRef]
5. Mandelbrot, B.B. The variation of certain speculative prices. J. Bus. 1963, 36, 394–419. (Reprinted in The
Random Character of Stock Market Prices, pp. 307–332). [CrossRef]
6. Cootner, P.H. (Ed.) The Random Character of Stock Market Prices; MIT Press: Cambridge, MA, USA, 1964.
7. Samuelson, P.A. Rational theory of warrant pricing. Ind. Manag. Rev. 1965, 6, 13–31.
8. Black, F.; Scholes, M. The pricing of options and corporate liabilities. J. Polit. Econ. 1973, 81, 637–659. [CrossRef]
9. Merton, R.C. Continuous Time Finance; Blackwell: Cambridge, MA, USA, 1990.
10. Mainardi, F.; Paradisi, P.; Gorenflo, R. Probability distributions generated by fractional diffusion equations.
In Proceedings of the Workshop on Econophysics, Budapest, Hungary, 21–27 July 1997; Invited Lecture; Department
of Physics: Bologna, Italy, January 1998; pp. ii+39.
11. Scalas, E.; Gorenflo, R.; Mainardi, F. Fractional calculus and continuous-time finance. Physica A
2000, 284, 376–384. [CrossRef]
12. Mainardi, F.; Raberto, M.; Gorenflo, R.; Scalas, E. Fractional calculus and continuous-time finance II:
The waiting-time distribution. Physica A 2000, 287, 468–481. [CrossRef]
13. Gorenflo, R.; Mainardi, F.; Scalas, E.; Raberto, R. Fractional calculus and continuous-time finance III:
The diffusion limit. In Mathematical Finance; Kohlmann, M., Tang, S., Eds.; Birkhäuser Verlag: Basel,
Switzerland; Boston, MA, USA; Berlin, Germany, 2001; pp. 171–180.
14. Raberto, M.; Scalas, E.; Mainardi, F. Waiting-times and returns in high-frequency financial data: An empirical
study. Physica A 2001, 314, 751–757. [CrossRef]
15. Montroll, E.W.; Weiss, G.H. Random walks on lattices, II. J. Math. Phys. 1965, 6, 167–181. [CrossRef]
16. Mainardi, F.; Raberto, M.; Scalas, E.; Gorenflo, R. Survival probability of LIFFE bond futures via the Mittag-Leffler
function. In Empirical Science of Financial Fluctuations; Takayasu, H., Ed.; Springer: Tokyo, Japan, 2002; pp. 195–206.
17. Gorenflo, R.; Mainardi, F. Fractional diffusion processes: Probability distributions and continuous-time
random walk. In Processes with Long-Range Correlations; Rangarajan, G., Ding, M., Eds.; Springer: Berlin,
Germany, 2003; pp. 148–166.
18. Scalas, E.; Gorenflo, R.; Mainardi, F. Uncoupled continuous-time random walks: Solution and limiting
behavior of the master equation. Phys. Rev. E 2004, 69, 011107. [CrossRef] [PubMed]
19. Scalas, E. The application of continuous-time random walks in finance and economics. Physica A
2006, 362, 225–239. [CrossRef]
20. Scalas, E. Continuous-Time Random Walks and Fractional Diffusion Models. In Fractional Calculus: Models
and Numerical Methods, 2nd ed.; Baleanu, D., Diethelm, K., Scalas, E., Trujillo, J.J., Eds.; World Scientific:
Singapore, 2017; Chapter 6, pp. 291–320; (1st Edition, 2012).
21. Scalas, E. Applications of Continuous-Time Random Walks. In Fractional Calculus: Models and Numerical
Methods, 2nd ed.; Baleanu, D., Diethelm, K., Scalas, E., Trujillo, J.J., Eds.; World Scientific: Singapore, 2017;
Chapter 7, pp. 321–344; (1st Edition, 2012).
22. Gorenflo, R.; Kilbas, A.A.; Mainardi, F.; Rogosin, S.V. Mittag-Leffler Functions. Related Topics and Applications;
Springer Monographs in Mathematics; Springer: Berlin, Germany, 2014; p. XIV-443, ISBN 978-3-662-43929-6.
23. Erdélyi, A.; Magnus, W.; Oberhettinger, F.; Tricomi, F.G. Higher Transcendental Functions; McGraw-Hill:
New York, NY, USA, 1955; Volume 3, pp. 206–227.
24. Gorenflo, R.; Mainardi, F. Fractional calculus, integral and differential equations of fractional order. In Fractals
and Fractional Calculus in Continuum Mechanics; CISM Lecture Notes; Carpinteri, A., Mainardi, F., Eds.;
Springer: Wien, Austria; New York, NY, USA, 1997; Volume 378, pp. 223–276.
25. Mainardi, F.; Gorenflo, R. On Mittag-Leffler type functions in fractional evolution processes. J. Comput.
Appl. Math. 2000, 118, 283–299. [CrossRef]
26. Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives, Theory and Applications; (Translated
from the Russian (Nauka i Tekhnika, Minsk, 1987)); Gordon and Breach: Amsterdam, The Netherlands, 1993.
27. Podlubny, I. Fractional Differential Equations; Academic Press: San Diego, CA, USA, 1999.

c 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

41
 mathematics
Review
Rules for Fractional-Dynamic Generalizations:
Difficulties of Constructing Fractional
Dynamic Models
Vasily E. Tarasov 1,2
1 Skobeltsyn Institute of Nuclear Physics, Lomonosov Moscow State University, Moscow 119991, Russia;
[email protected]; Tel.: +7-495-939-5989
2 Faculty “Information Technologies and Applied Mathematics”, Moscow Aviation Institute
(National Research University), Moscow 125993, Russia

Received: 9 May 2019; Accepted: 4 June 2019; Published: 18 June 2019 

Abstract: This article is a review of problems and difficulties arising in the construction of
fractional-dynamic analogs of standard models by using fractional calculus. These fractional
generalizations allow us to take into account the effects of memory and non-locality, distributed lag,
and scaling. We formulate rules (principles) for constructing fractional generalizations of standard
models, which were described by differential equations of integer order. Important requirements
to building fractional generalization of dynamical models (the rules for “fractional-dynamic
generalizers”) are represented as the derivability principle, the multiplicity principle, the solvability
and correspondence principles, and the interpretability principle. The characteristic properties of
fractional derivatives of non-integer order are the violation of standard rules and properties that
are fulfilled for derivatives of integer order. These non-standard mathematical properties allow
us to describe non-standard processes and phenomena associated with non-locality and memory.
However, these non-standard properties lead to restrictions in the sequential and self-consistent
construction of fractional generalizations of standard models. In this article, we give examples of
problems arising due to the non-standard properties of fractional derivatives in construction of
fractional generalizations of standard dynamic models in economics.

Keywords: fractional calculus; fractional dynamics; fractional generalization; long memory;

non-locality; mathematical economics; economic theory

MSC: 26A33 Fractional derivatives and integrals; 91B02 Fundamental topics (basic mathematics,
applicable to economics in general)

1. Introduction
In mathematics, in addition to derivatives and integrals of integer order, fractional differentiation
and integration of non-integer orders (for example, see the comprehensive encyclopedic-type
monograph [1], the unsurpassed monograph on generalized fractional calculus [2], the very important
and remarkable books on fractional calculus and fractional differential equations [3–5]). These operators
have been known for several centuries (for example, see comments to Chapters in [1], the first description
of the history of fractional calculus (FC), written 150 years ago [6], brief history of FC [7–9], and the
first review of history of generalized fractional calculus [10]). The recent history of fractional calculus
is described in [11], the chronicles and science metrics of recent development of FC [12–14], and some
pioneers in applications of FC [15]. The fractional differential equations are a powerful tool to describe
processes with long memory and spatial non-locality. Recently, the fractional calculus and fractional
differential equations have become actively used to describe various phenomena in natural and social

Mathematics 2019, 7, 554; doi:10.3390/math7060554 43 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 554

sciences. The most important results in this area are collected in the eight-volume encyclopedic
handbook on fractional calculus and its applications [16].
At the present time, in some works, fractional differential equations of dynamic models, which are
intended to describe physical and economic processes, are proposed without carefully deducing
them from some physical and economic assumptions, interpretations and generalizations of concepts.
The fractional differential equations are obtained by simply replacing the integer derivatives with
fractional derivatives of non-integer order in the equations of standard model. Moreover, it is usually
not discussed how such fractional equations can be obtained and justified. After obtaining the solutions
of fractional differential equations, which can be presented in an analytical or approximate form,
the physical/economic interpretation and analysis of these solutions is not carried out. This way of
obtaining fractional generalizations of standard dynamic models can be called a formal generalization,
which is a mathematical exercise, and it cannot be considered as mathematical models of the natural
and social processes.
In our opinion, the goals of fractional generalizations of models in natural and social sciences
cannot be reduced only to a mathematical consideration of fractional differential equations and its
solutions. In case of this reduction, the connection with the physics and economics is lost, and it leads
to the fact that the results of such generalizations cannot be used directly in these areas of science.
The mathematical analysis of fractional differential equations and its solutions should be a bridge,
connecting the initial economic or physical assumptions and concepts on the one side, and economic
or physical interpretations, effects and conclusions on the other side. All this leads to the need to
formulate rules and principles that are important for the development of applications of fractional
calculus in natural and social sciences.
Let us formulate basic rules (the principles of fractional-dynamic generalizer) for constructing
fractional generalizations of standard dynamic models, i.e., models that are described by differential
equations of integer orders.

(1) Derivability Principle: It is not enough to generalize the differential equations describing the dynamic
model. It is necessary to generalize the whole scheme of obtaining (all steps of derivation) these equations
from the basic principles, concepts and assumptions. In this sequential derivation of the equations we
should take into account the non-standard characteristic properties of fractional derivatives and integrals.
If necessary, generalizations of the notions, concepts and methods, which are used in this derivation,
should also be obtained. The derivability principle states that we should realize a correct fractional
generalization of the derivation of the model equations. It is necessary to generalize not only
and not so much the differential equation of the model itself, but a generalization of all steps
of deriving the standard (non-fractional) equations of the model. In the general case, this will
not be an equation that is obtained by simply replacing the integer derivatives with fractional
derivatives of non-integer order. Often, the consistent construction of a fractional-dynamic
model is associated with the need to introduce new concepts and notions that generalize the
concepts and notions of standard models. Note that fractional generalizations of basic concepts
are not so much a part of this particular model, but in fact are the common basis of different
models, and basis of all fractional dynamics (fractional mathematical economics), and not just the
model. An important part of this derivation is the need to take into account the non-standard
characteristic properties of fractional derivatives and integrals [17–22]. These properties include
(a) violation of the standard chain rule (for example, see [3], pp. 97–98, [5], pp. 35–36, [19] and
Section 2.1); (b) violation of the standard semi-group property for orders of derivatives (see [1], pp.
46–47, [5], p. 30, and Section 2.2); (c) violation of the standard product (Leibniz) rule (for example,
see [1], pp. 280–284, [3], pp. 91–97, [5], pp. 33, 59, [17,20,22] and Section 2.3); (d) violation of
the standard semi-group property for dynamic maps (see the explanations and references in
Section 2.4). These properties narrow the field for maneuver and make it difficult to obtain
fractional generalizations. These non-standard properties are obstacles that must be overcome
to build correct fractional dynamic models. At the same time, these non-standard properties

44
Mathematics 2019, 7, 554

allow us to get correct fractional dynamic models to describe non-standard effects, processes and
phenomena. Schematically, this principle is represented by Figure 5.
(2) Multiplicity Principle: For one standard model, there is a set of fractional dynamical generalizations, due
to the existence of various types of fractional operators and violation of s-equivalence for fractional differential
equations. In addition to existence a large number of different types of fractional derivatives and
integrals, the violation of the standard rules generate an additional uncertainty of fractional
generalizations. Fractional generalizations of solution-equivalent (s-equivalent) representations
of integer-order differential equations of standard models, as a rule, lead to different fractional
differential equations that have non-equivalent solutions. This situation is partially analogous
to the fact that quantization of equivalent classical models leads to nonequivalent quantum
theories. As a result, fractional generalizations of one standard model (which is represented
by s-equivalent differential equations of integer order) can lead to different fractional-dynamic
models that will predict different behaviors of a process and only some of them may be useful in a
given context. We can state that for one standard model, there is a family of fractional dynamical
generalizations, due to the existence of various types of fractional operators and violation of
s-equivalence for fractional differential equations. In this regard, it is important to investigate
and describe the properties of solutions of fractional dynamic equations, which are (qualitatively
and/or quantitatively) the same, and the properties of solutions that are (first of all, qualitatively)
different. Schematically, this principle is given by Figure 1.
(3) Solvability Principle: The properties of process types (such as long memory, spatial nonlocality,
distributed delay, distributed scaling) and the properties of the corresponding types of fractional operators
must be taken into account in the existence of solutions, and in obtaining correct analytical and numerical
solutions. The solvability principle states that the existence of solution, and the possibility
of obtaining an exact analytical solution or correct numerical solutions for some conditions.
Obviously, the existence conditions should allow us to obtain solutions for those cases and
properties that the described process has. In addition, we should take into account that different
types of fractional derivatives and integrals are known in fractional calculus [1,2,4]. Therefore, in
fractional dynamic generalization, it is important that type of fractional operators correspond
to the type of natural or social process. It should be noted that not all well-known fractional
operators can describe the long memory and spatial non-locality (see Section 2.5 of this paper).
For example, some fractional operators can be used to describe the distributed lag (time delay)
and the distributed scaling (dilation) and they are not suitable for memory and non-locality.
Additionally, we need to verify the existence condition for properties of solutions obtained.
For example, if we describe processes with long memory then derivation of numerical solution
must take into account not only local information, but the numerical scheme must contain memory
terms. Schematically, this principle is represented by Figure 4.
(4) Correspondence Principle: The limiting procedure, when orders of fractional derivatives tend to integer
values, applied to the equations of the fractional dynamic model and their solutions, should give the
standard model equations and their solutions. The correspondence principle means a possibility
of obtaining equations and solutions of standard model by using a limit procedure, when the
orders of the fractional derivatives tend to an integer values. The principle of correspondence
must be fulfilled both for the equation itself and for its solution. It should be noted if the order
of the derivative tends to the integer value, then the limit on the left and the limit on the right
can give different results in the general case. Schematically, this principle is depicted in Figure 2.
The Correspondence Principle can also be represented by the formal expression:

lim Frac − Eq[α] = Int − Eql[n], (1)

α→n

lim Frac − Sol[α] = Int − Sol[n], (2)

α→n

45
Mathematics 2019, 7, 554

where n ∈ N. It should be noted that the limit on the left and the limit on the right do not coincide
in the general case:
lim Frac − Sol[α]  lim Frac − Sol[α]. (3)
α→n− α→n+

(5) Interpretability Principle: The subject (physical, economic) interpretation of the mathematical results,
including solutions and its properties, should be obtained. Differences, and first of all qualitative differences,
from the results based on the standard model should be described. The subject interpretation of the
solutions should be obtained. The properties of solutions should be described in details with
their economic or physical meaning (interpretation). It is important to have an interpretability of
mathematical results. The differences between results, which were obtained for the proposed
generalization and the standard model, should be clearly indicated. An important purpose is to
find qualitative differences between the properties of solutions for the fractional dynamic model
and the properties of the solutions of the standard model. Schematically, this principle is given
by Figure 3.

Figure 1. The non-equivalence (multiplicity) diagram. This diagram is non-commutative. The following
notation is used in the diagram: Int-Eq[n] is a set of differential and/or integral equations of integer
orders that describe the standard dynamic model; Frac-Eq[α] is a set of fractional differential and/or
integral equations of non-integer orders that describe the fractional dynamic model. The S-equivalence
of some equations of standard models and non-equivalence of fractional generalizations of these
equations are considered in Section 5.

46
Mathematics 2019, 7, 554

Figure 2. The correspondence diagram. This diagram should be commutative. The following notation
is used in the diagram: Int-Eq[n] is a set of differential and/or integral equations of integer orders that
describe the standard dynamic model; Int-Sol[n] denotes solutions of differential and/or integral equations
of the standard dynamic model; Frac-Eq[α] is a set of fractional differential and/or integral equations of
non-integer orders that describe the fractional dynamic model; Frac-Sol[n] denotes solutions of fractional
differential and/or integral equations of the fractional dynamic model; Lim is a limit transition when the
non-integer orders α tend to integer values n from the left (α→ n−) or form the right (α→ n+).

Figure 3. The interpretability diagram. The following notation is used in the diagram: Int-Sol[n]
denotes solutions of differential and/or integral equations of the standard dynamic model; Frac-Sol[n]
denotes solutions of fractional differential and/or integral equations of the fractional dynamic model;
“Int-Math Results” and “Frac-Math Results” denote mathematical results (for example, asymptotic
behaviors) obtained from solutions of integer-order and fractional-order differential and/or integral
equations; “Comp” denotes a comparison of solutions, mathematical results and subject (economic,
physical) results based on the interpretation.

47
Mathematics 2019, 7, 554

Figure 4. The solvability diagram. This diagram should be commutative. The following notation
is used in the diagram: Frac-Eq[α] is a set of fractional differential and/or integral equations of
non-integer orders that describe the fractional dynamic model; Frac-Sol[n] denotes solutions of
fractional differential and/or integral equations of the fractional dynamic model; DER denoted a
derivation of exact analytical solutions or correct numerical solutions for the fractional differential
and/or integral equations; VER is a verification of the fulfillment of the condition of existence of
solutions for properties of solutions obtained.

The proposed five principles are designed primarily to eliminate errors that are usually made
when building fractional dynamic generalizations of standard models. The most important element
is the requirement that in fractional generalization of economic (or physical) model the “output” of
the research should be an economic (physical) conclusions (phenomena, effects) and new economic
(physical) effects that are a consequence of subject assumptions on the “input”. Here, mathematics
(fractional calculus) is the tool that mathematically strictly connects “economic/physical input” and
“economic/physical output”. If mathematical equations and solutions are not rigidly connected with
subject “input” and “output”, they will fly away into “airless space”. In this case, the results will turn
from economics and physics into formal manipulations, which may not even have mathematical value
from the point of view of pure mathematics (fractional calculus).
An important goal of fractional generalizations is to obtain qualitatively new effects and phenomena
in natural and social sciences. The results obtained in a science by using the new mathematical apparatus
(fractional calculus) should give qualitatively new results and predict new effects and phenomena for
this science. First of all, it is precisely such qualitatively new results are interesting in the first place.
In this paper, we illustrate these rules (principles) by using examples of fractional generalizations
of standard economic models.
In Section 2 of this paper, we describe the non-standard rules for fractional operators of non-integer
orders. The violation of the standard chain rule is described in Section 2.1. The violation of the standard
semi-group property for orders of derivatives is discussed in Section 2.2. We consider the violation of
the standard product (Leibniz) rule in Section 2.3. The violation of the standard semi-group property
for dynamic maps is described in Section 2.4. A correspondence between the types of fractional
operators of non-integre orders and the types of phenomena is discussed in Section 2.5.

48
Mathematics 2019, 7, 554

Figure 5. The derivability diagram. This diagram is non-commutative in general. The reason for
this noncommutativity is a violation of standard rules and properties for fractional derivatives of
non-integer order. The following notation is used in the diagram: Con[n] and Con[α] are concepts of
the standard dynamic model and their fractional generalization; Ass[n] and Ass[α] are assumptions
that are used in the standard dynamic model and in its fractional generalization; FG is fractional
generalization based on the fractional calculus; Int-Eq[n] is a set of differential and/or integral equations
of integer orders that describe the standard dynamic model; Frac-Eq[α] is a set of fractional differential
and/or integral equations of non-integer orders that describe the fractional dynamic model.

In Section 3, we consider an application of the Derivability Principle and we give examples of

the problem with the violation of the standard rules for fractional operators of non-integer orders.
In Section 3.1, to illustrate problems that are connected with the non-standard form of the chain rule,
we consider a fractional generalization of the Kaldor-type model of business cycles. In Section 3.2,
problem with violation of the standard semi-group rule for orders of derivatives is shown for the
fractional generalization of the Phillips model of the multiplier-accelerator. In Section 3.3, to illustrate
the problems arising from the non-standard form of the product (Leibniz) rule, we consider the
fractional generalization of the standard Solow–Swan model. In Section 3.4, the problem with the
violation of the standard semi-group property of dynamic map is described using the examples of
fractional generalization of the dynamic Leontief (intersectoral) model and logistic growth model.
In Section 3.4, the definitions of new economic concepts and notions are described.
In Section 4, the Solvability Principle and the Correspondence Principle are discussed and some
examples are suggested. In Section 4.1, we discuss the Solvability Principle by using the general
fractional calculus as an example. In Section 4.2, for illustration we consider the distributed lag
fractional calculus and growth-relaxation equations with gamma distributed delay time. In Section 4.3,
a simple example of the Correspondence Principle for the case, when the order of the derivative tends
toward integer values from the left and from the right, is considered. In Section 4.4, the Solvability
Principle is discussed by using example from numerical simulation of fractional differential equations.
In Section 5, we describe some problems (“Non-Equivalence” and “Unpredictability”) of fractional
generalizations that are associated with non-equivalent fractional equations, which are formal
generalization of equivalent differential equations of integer orders. In Section 5.1, we give definitions
of equivalence of equations by solutions (s-equivalence). In Section 5.2, we illustrate non-equivalence

49
Mathematics 2019, 7, 554

of fractional generalization for relaxation and growth differential equations. In Section 5.3, we describe
non-equivalence of fractional generalization of the fractional logistic equation that in economics
describes growth in a competitive environment with memory. In Section 5.4, we formulate that
fractional generalization of standard model can generate non-equivalent models.
In Section 6, we consider example of application of the Interpretability Principle by describing
some examples of new effects and phenomena in economics. In Section 6.1, we describe a simple
economic model with memory. Fractional differential equation, its solution and asymptotic behavior
are proposed. In Section 6.2, we give an interpretation of the mathematical results by using suggested
new concept of the warranted rate of growth with memory. In Section 6.3, we describe the interpretation
of mathematical results in the form of economic phenomena for economic growth and decline with
memory. In Section 6.4, we describe an interpretation of relaxation of economic processes with memory.
In Section 7, we give a short conclusion.

2. Non-Standard Properties of Fractional Derivatives

In this section we describe some properties (rules) of fractional derivatives causing problems
when constructing fractional generalizations of standard dynamic models.
The fractional derivatives of non-integer orders have a set of non-standard properties and rules
such as the violation of the standard product (Leibniz) and the standard chain rules, the violation of
semigroup rules for orders of the derivatives and the violation of semigroup rules for dynamical map.
The non-standard properties of fractional derivatives should be taken into account, when constructing
fractional generalization of dynamic models. These properties create problems in realization of the
derivability principle.

2.1. Violation of Standard Chain Rule

The standard chain rule for the first order derivative has the form:

D1t f ( g(t)) = (D1g f ( g)) g= g(t) D1t g(t), (4)

where D1t = d/dt is the derivative of first order. The standard chain rule for the derivative of integer
order n ∈ N can be written by the equation:


n 
n  ar
1 1 r
Dnt f ( g(t)) = n! g f ( g)) g= g(t)
(Dm Dt g ( t ) , (5)
ar ! r!
m=1 r=1

which is called the Faá di Bruno’s formula [23].

The standard chain rules shown in Equations (4) and (5) are not satisfied for fractional derivatives of
non-integer orders. Foe example, the chain rule for the Riemann–Liouville fractional derivative of the
order α > 0 (see equation (2.209) in section 2.7.3 of [3], pp. 97–98, [5], pp. 35–36, and [19]) has the form:

tα f ( g(t))
DαRL,0+ f ( g(t)) = Γ(1−α)
+

∞ k−α k  
k  a r (6)
Cαk Γ(k!k−α
t
+1)
Dmg f ( g) ar ! r! Dt g(t) ,
1 1 r
m=1 g= g(t) r=1
k =1

where t > 0, Dm
g and Drt
are derivatives of integer orders, extends over all combinations of

non-negative integer values of a1 , a2 , . . . , ak such that kr=1 rar = k and kr ar = m.
The chain rules for other type of fractional derivatives have a similar form. We see that standard
chain rules (4) and (5) do not satisfied for fractional derivatives of non-integer order.

50
Mathematics 2019, 7, 554

2.2. Violation of Semi-Group Rule for Orders of Derivatives

The standard semi-group rule for orders of integer-order derivatives has the form of the equality:

n+m
t X ( t ) = Dt
Dnt Dm X (t), (7)

which holds for n, m ∈ N, if the function X(t) is smooth or X(t) is a continuous function that has
continuous first n + m derivatives (for example, X(t) ∈ Cn+m (R)). It is well known that this property
may be broken for discontinuous functions X(t) and if the derivatives are not continuous.
For fractional derivatives, the standard semi-group rule (7) is not satisfied in the general case
(for example, see [1], pp. 46–47 and [5], p. 30). For example, the Caputo fractional derivatives of the
orders 0 < α, β < 1 satisfy the equality:
 β
  α+β  1
DαC,0+ DC,0+ X (t) = DC,0+ X (t) + X(1) (0)t1−α−β , (8)
Γ (2 − α − β)

β
where DαC,0+ and DC,0+ are the Caputo fractional derivative of the orders 0 < α, β < 1 is defined by the
equation:
   t
1
Dα0+ X (t) = (t − τ)−α X(1) (τ)dτ, (9)
Γ (1 − α) 0
where Γ(α) is the gamma function. Equality (8) means the violation of the semi-group property for
orders of derivatives, i.e., in general, we have the inequality:
 β
  α+β 
DαC,0+ DC,0+ X (t)  DC,0+ X (t), (10)

if the orders of these fractional derivatives are non-integer. In the order α of the Caputo fractional

derivative in (10) is non-integer and the order β = n ∈ N, then we have the equality DαC,0+ Dnt X (t) =
 
+n
DαC,0 +
X (t). If the order α = n ∈ N and β is non-integer, then the standard semi-group property is
 β
  n+β 
violated, i.e., the inequality Dnt DC,0+ X (t)  DC,0+ X (t) holds in general.

2.3. Violation of the Standard Product Rule

The standard product (Leibniz) rule for first-order derivative (for n = 1) has the form:
   
D1t ( f (t) g(t)) = D1x f(x) g(x) + f (x) D1x g(x) . (11)

The standard product rule for the derivative of integer order n ∈ N has the form:


n
n!   
Dnt ( f (t) g(t)) = Dn−k f (t) Dkt g(t) . (12)
(n − k)! k! t
k =0

The Leibniz rule for derivative of non-integer order α  1 cannot have the simple form:
   
Dαt ( f (t) g(t)) = Dαt f (t) g(t) + f (t) Dαt g(t) . (13)

A violation of relation in Equation (13) is a characteristic property of all derivatives of integer-orders

n ∈ N greater than one and for all types derivatives of the non-integer order α > 0 (for example,
see [1], pp. 280–284, [3], pp. 91–97, [5], p. 33, 59, and [17,20,22]). In [17], the following theorem has
been proved:

51
Mathematics 2019, 7, 554

Theorem 1 (“No violation of the Leibniz rule. No fractional derivative”). If a linear operator Dαt
satisfies the product rule in the form of Equation (13), then the operator Dαt is the differential operator of first
order, that can be represented in the form Dαt = a(t)D1t , where a(t) is function on R.

As a result, we can states that derivatives of non-integer orders α  1 cannot satisfy the standard
product rule of Equation (13). For example, the fractional generalization of the Leibniz rule for the
Riemann–Liouville derivatives has the form (see section 15 in [1], pp. 277–284), of the infinite series:


∞
Γ (α + 1)   
DαRL ( f (t) g(t)) = Dα−k f (t) Dkt g(t) , (14)
Γ (α − k + 1) Γ (k + 1) RL
k =0

where f (t) and g(t) are analytic functions on [a, b] (see theorem 15.1 in [1]), DαRL is the Riemann–Liouville
derivative; Dk is derivative of integer order k ∈ N. It should be noted that the sum of Equation (14) is
k−α = Dα−k of non-integer orders (k − α) for the values
infinite and it contains the fractional integrals IRL RL
k > [α] + 1.

2.4. Violation of the Standard Semi-Group Rule for Dynamic Maps

Let us consider linear ordinary differential equation equations of first order in the form:

dX(t)
= A X (t), (15)
dt

where X(t) is an unknown function (with values in a Banach space) and A is a constant linear bounded
operator acting in the space (or A is the linear operator having an everywhere dense domain of
definition D(A) in the Banach space). We can consider the Cauchy problem of finding a solution
of Equation (15) for 0 < t < ∞, satisfying the given initial condition X(0) = X0 ∈ D(A). A unique
solution of the Cauchy problem exists for the differential equation of first order (Equation (15)) with a
constant bounded operator A and it can be written (for example, see [24], pp. 119–157) in the form:

X (t) = U (t) X (0), (16)

where the operator U (t) is defined by the series:


∞ k
t
U (t) = exp(t A) = Ak , (17)
k!
k =0

which converges in the operator norm. The operator U (t) is called the dynamic map or the phase
flow [25].
A family of bounded linear operators U (t), depending on the parameter 0 < t < ∞, forms a
semi-group if the condition U (0) = I and the equality:

U (t1 ) U (t2 ) = U (t1 + t2 ) (18)

hold for all t1 , t2 where 0 < t1 , t2 < ∞. Equation (18) is the standard semi-group rule for dynamical map.
 
The set U (t), t > 0 is called one-parameter dynamical semi-group. In quantum theory the operator A
is called the infinitesimal generator of the quantum dynamical semi-group (see classical papers [26–29]).
The class of differential equations for which A is a generator for a semigroup of class (C0 ) coincides
with the class of differential equations for which the Cauchy problem is uniformly correct [24].
Daftardar-Gejji and Babakhani [30] (see also [31] and [4], p. 142) have studied the existence,
uniqueness, and stability of solutions for the fractional differential equations:
 
DαC,0+ X (t) = A X(t), (19)

52
Mathematics 2019, 7, 554

where DαC,0+ is the Caputo fractional derivative of the order 0 < α < 1, X(t). is the column vector and
A. is real square N x N matrix. They obtained the unique solution of Equation (19) in the form:

X(t) = Uα (t) X(0), (20)

where the operator Uα (t) is defined by the series:


∞
(tα )k
Uα (t) = Eα [tα A] = Ak . (21)
Γ(αk + 1)
k =0

Here, Eα [tα A] is the Mittag–Leffler function with matrix arguments [32].

For α = 1, we have E1 [z] = exp(z). Therefore, we have U1 (t) = U (t) = exp(t A).
The standard semi-group rule (Equation (18)) for dynamical maps Uα (t) does not hold for
non-integer values of α ∈ (0, 1), i.e., we have the inequality:

Uα (t1 ) Uα (t2 )  Uα (t1 + t2 ) (22)

that follows from the property of the Mittag–Leffler function (for example, see [33,34], and some
additional information in [35–37]) in the form:
  
Eα tα1 A Eα tα2 A  Eα (t1 + t2 )α A . (23)

As a result, the dynamical maps Uα (t) with α  N cannot form a semigroup.

The operator Uα (t) describes the dynamical map with power-law fading memory for non-integer
values of α. The violation of the standard semigroup rule for dynamical maps is a characteristic
 
property of dynamics with memory. We can only state that the set Uα (t), t > 0 of the dynamical map
with memory forms a dynamical groupoid [34,37] for on-integer values of α ∈ (0, 1).
It should be noted that the fractional differential Equation (19) describes the fractional
generalization of N-level open quantum system and the Leontief dynamic model of N-sectors
in economy, in which the power-law memory is taken into account (see Section 3.4.1).

2.5. What Effects Are Fractional Derivatives Described?

In fractional calculus, many different types of fractional derivatives and integrals are known [1–4].
In construction of a fractional generalization of a standard dynamic model, an important part of
the work is an adequate choice of the type of the fractional derivative or/and integral. First of all,
fractional operators must correspond to the type of process to be described. It is well known that
fractional derivatives and integrals are a powerful tool for describing processes with memory and
nonlocality. However, not all fractional operators can describe the effects of memory (or non-locality).
In application of the generalized and general fractional operators, an important question arises about
the correct subject interpretation of these operators (for example, see informational [38], physical [39],
and economic [40–42] interpretations). It is important to emphasize that not all fractional operators can
describe the processes with memory (for example, see [43–46]). It is important to clearly understand
what type of phenomena a given operator can describe. Let us give some examples for illustration.

2.5.1. First Example: Kober and Erdelyi–Kober Operators

The Kober fractional integration of non-integer order [1,2,4] can be interpreted as an expected
value of a random variable up to a constant factor (for example, see [43,45] and section 10 in [46]),
where the random variable describes scaling (dilation) with the gamma distribution. The Erdelyi–Kober
integral operator, the differential operators of Kober and Erdelyi–Kober type have analogous
interpretation [43,45,46]. As a result, these operators are integer-order operator with continuously
distributed scaling (dilation), and these operators cannot describe the memory. Note that the fractional

53
Mathematics 2019, 7, 554

generalizations of the Kober and Erdelyi–Kober operators, which can be used to describe memory and
distributed scaling (dilation) simultaneously, were proposed in [46].
The Kober fractional integral of the order α > 0 [4], p. 106, is defined as:

  
t−α−η t η
α
IK;0 +;η ϕ (t) = Γ (α) τ (t − τ)α−1 ϕ(τ)dτ, (24)
0

where α > 0 is the order of integration and η ∈ R. Using the variable x = τ/t, this operator can be
represent by the equation:

  
α Γ (η + α + 1) 1
IK;0 +;η ϕ (t) = Γ (η + 1)
fη+1;α (x)(Sx ϕ)(t)dx, (25)
0

where Sx is the operator [1], pp. 95–96 and [4], p. 11 such that (Sx ϕ)(t) = ϕ(x t) and fα;β (x) is the
probability density function (pdf) of the beta-distribution such that:

1
fα;β (x) = xα−1 (1 − x)β−1 (26)
B(α, β)

for x ∈ [0, 1] and fα;β (x) = 0 if x  [0, 1], where B(α, β) is the beta function. We see that the Kober
integral operator describes beta distributed scaling up to numerical factor. For details see [43,45] and
section 10 in [46].

2.5.2. Second Example: Causality Principle and Kramers–Kronig Relations

To describe processes with memory [47–49], the operators should satisfy the causality principle.
For natural and social processes with memory, the causality can be described by the Kramers–Kronig
relations [50]. The Riesz fractional operators (see section 2.10 of [4]) cannot be used to describe memory
since this operator violates the causality principle. The Riesz fractional operators can be used to
describe power-law non-locality and power-law spatial dispersion (for example, see [51,52]).
The principle of causality is represented in the form of the Kramers–Kronig relations (the Hilbert
transform pair) by using the Fourier transforms. Let us consider the Fourier transform M  (ω) of the
memory function M(t) . In general,  
M(ω) is the complex function M( ω) = M  
  1 (ω) + i M2 (ω), where the
real part M 1 (ω) = Re M (ω) and the imaginary part M  2 (ω) = Im M  (ω) are real-valued functions.
The Kramers–Kronig relations state that the real part and the imaginary parts of the memory function
are not independent, and the full function can be reconstructed given just one of its parts. Let us
assume that the function M  (ω) is analytic in the closed upper half-plane of frequency ω and vanishes
like 1/|ω| or faster as |ω| → ∞ . The Kramers–Kronig relations are given by:

+∞

 1 (ω) = 1 P.V.
M
1 
M2 (Ω)dΩ, (27)
π Ω−ω
−∞

+∞

 2 (ω) = − 1 P.V.
M
1 
M1 (Ω)dΩ, (28)
π Ω−ω
−∞

where P.V. denotes the Cauchy principal value. For details see [50].

2.5.3. Third Example: Abel-type operator with Kummer Function in Kernel

The Abel-type fractional integral (and differential) operator with Kummer function (or the three
parameter Mittag–Leffler functions) in the kernel (see the classic book [1] and equation (37.1) in [1],
p. 731) can be interpreted as the Riemann–Liouville fractional integral (and derivatives) with gamma
distribution of delay time [43,53,54].

54
Mathematics 2019, 7, 554

It is known that the Abel-type (AT) fractional integral operator with Kummer function in the
kernel (see equation (37.1) in [1], p. 731) is defined by the equation:

 α,β,γ,   t
1
IAT;0+ Y (t) = (t − τ)α−1 F1,1 (β; α; γ(t − τ))Y(τ) dτ, (29)
Γ (α) 0

and F1,1 (a; b; z) is the confluent hypergeometric Kummer function. Using equality F1,1 (a; c; z) =
Γ(c)Ea1,c (z) the memory kernel in Equation (29) can be expressed through the three parameter
γ
Mittag–Leffler functions Eα,β (z).
The fractional integration with the gamma distributed lag in the form:

      t  
Iλ,a;α
T;RL;0+
Y ( t ) = M λ,a
T
( τ ) ∗ I α
RL,0 + Y ( t ) = Mλ,a
T
α
(τ) IRL,0 + Y (t − τ) dτ, (30)
0
 
α
where * denotes the Laplace convolution, IRL,0 Y (t) is the Riemann–Liouville fractional integral [1,4],
+
Mλ,a
T
(τ) is the probability density function (weighting functions) of the gamma distribution:

λa τa−1
Mλ,a (τ) = exp(−λ τ) (31)
T Γ (a)

for τ > 0 and Mλ,a

T
(τ) = 0 for τ ≤ 0, where a > 0 is the shape parameter and λ > 0 is the rate parameter.
Equation (30) can be written thought the Laplace convolution of memory and weighting functions:
    
Iλ,a;α
T;C;0+
Y (t) = Mλ,a
T
∗ MαRL ∗ Y (t), (32)

where MαRL (t) = (t − τ)α−1 /Γ(α) is the kernel of the Riemann–Liouville fractional integral.
The associativity of the Laplace convolution allows us to represent operator in the form:

   t
Iλ,a;α
T;C;0+
Y ( t ) = Mλ,a;α
TRL
(t − τ)Y(τ) dτ, (33)
0
 
where Mλ,a;α
TRL
(t) = Mλ,a
T
∗ MαRL (t) is the memory-and-lag function of the form:

λa Γ ( a )
Mλ,a;α (t) = ta+α−1 F1,1 (a; a + α; −λt), (34)
TRL Γ (a + n − α)

where F1,1 (a; b; z) is the confluent hypergeometric Kummer function.

As a result, we obtain the relation:
  1  
+α,a,−λ, λ,a;α
IaAT;0 +
Y (t) = . IT;RL;0 +
Y (t). (35)
λa Γ (a)

This equation shows that the AT fractional integral can be expressed through the Riemann–Liouville
fractional integral with gamma distributed lag for wide range of parameters.

2.5.4. Fourth Example: Abel-type Operator with Kummer Function in Kernel

In application it is important to have conditions for the operator kernel, which make it possible
to assign this operator to one or another type of phenomena or processes. For example, it is obvious
that the kernels of general fractional convolution operators satisfying the normalization condition
will describe distributed delays in time (lag), and not memory (for example, see [44,46], and some
additional comments in [53–55]). It is well known in physics that the time delay is related to the finite
speed of the process and not to the memory. For example, the Caputo–Fabrizio operators, which were

55
Mathematics 2019, 7, 554

misinterpreted as fractional derivatives of non-integer orders, are integer-order derivatives with the
exponentially distributed delay time [43,44]. Therefore, these operators cannot be used to describe
processes with memory. Note that the fractional derivatives with exponentially distributed is suggested
in [43] and then applied in economics [53–55].

2.5.5. Fifth Example: Fractional operators with Uniform Distributed Order

The continual fractional derivatives and integrals were proposed by A.M. Nakhushev [56,57].
The fractional operators, which are inversed to the continual fractional integrals and derivatives,
have been proposed by A.V. Pskhu [58,59]. In papers [47,60], we proved that the fractional integrals
and derivatives of the uniform distributed order can be expressed (up to a numerical factor)
thought the continual fractional integrals and derivatives that were suggested by A.M. Nakhushev.
Therefore, the proposed fractional integral and derivatives of uniform distributed order we called
in our paper [60] as the Nakhushev fractional integrals and derivatives. The corresponding inverse
operators, which contain the two parameter Mittag–Leffler functions in the kernel, were called as the
Pskhu fractional integrals and derivatives [60].
For example, the Riemann–Liouville fractional integral of distributed order is defined as:
   α2
[α1 ,α2 ] α
IRL,0 +
X ( t ) = ρ(α) (IRL,0 + X )(t)dα, (36)
α1

where α2 > α1 ≥ 0, and the weight function ρ(α) satisfies the normalization condition:
 α2
ρ(α)dα = 1. (37)
α1

In Equation (36) the integration with respect to time and the integration with respect to order can be
permuted for a wide class of functions X(τ). As a result, Equation (36) is written in the form:
   t
[α1 ,α2 ] [α ,α ]
IRL,0 +
X (t) = Mρ(1α) 2 (t − τ)X(τ)dτ, (38)
0

[α ,α ]
where the kernel Mρ(1α) 2 (t − τ) is defined by the equation:

 α2
[α ,α ] ρ(α) 1
Mρ(1α) 2 (t − τ) = dα, (39)
α1 Γ(α) (t − τ)1−α

where α2 > α1 ≥ 0. In the simplest case, we can use the continuous uniform distribution (CUD) that is
defined by the expression:
 1
α2 −α1 f or α ∈ [α1 , α2 ]
ρ(α) = . (40)
0 f or α ∈ (−∞, α1 ) ∪ (α2 , ∞)

For the probability density function (Equation (40)), the memory function (Equation (39)) has the form:
 α2
[α ,α ] 1 tξ dξ
1 2
MCUD (t) = W (α1 , α2 , t) = . (41)
( α2 − α1 ) t α1 Γ (ξ)

As a result, the fractional integral of uniform distributed order is defined by the equation:
 β   t
[α,β] 1 ξ
IN,RL X(t) = IRL,0 +
X (t)dξ = W (α, β, t − τ) X(τ)dτ, (42)
β−α α 0

56
Mathematics 2019, 7, 554

where β > α > 0. The fractional integrals and derivatives of the uniform distributed order can be
expressed thought the continual fractional integrals and derivatives, which have been suggested by
A.M. Nakhushev [56,57].

2.5.6. Sixth Example: Left-Sided and Right-Sided Fractional Operators

The right-sided Riemann–Liouville, Liouville, and Caputo fractional derivatives [4] cannot
describe the memory processes. Using only the left-sided derivatives of bon-integer orders, we take
into account the history of changes of variable in the past, that is for τ < t. The right-sided operators
are defined by integration over τ > t, where t is the present moment of time. Using right-sided
operators actually means that the present state depends on the future states, and not on the past states
of the process.

2.5.7. Seventh Example: Fading Memory, Spetial Non-Locality, Time Delay (Lag), Scaling
Fractional calculus approach allows us to describe the spatial non-locality and fading memory of
power-law type, the openness of processes and systems, intrinsic dissipation, long-range interactions,
and some other type of phenomena. The most well-known phenomena in physics that can be described
by fractional differential equations, are the fractional relaxation-oscillation, fractional diffusion-wave,
fractional viscoelasticity, spatial and frequency dispersion of power type, nonexponential relaxation,
anomalous diffusion, and some others [61,62].
As a result, we can state that the following type of phenomena can be independent of each other:

• fading memory (forgetting) (for example, see [47–50] and references therein) and power-law
frequency dispersion;
• spatial non-locality (for example, see [63]) and power-law spatial dispersion (for example, see [64]);
• lag (time delay) (for example, see [43,53–55,65] and references therein); and
• scaling (dilation) (for example, see section 9 in [43] and references therein).

As a result, these phenomena are described by certain types of operator kernels. For other types
of processes and phenomena, we do not have mathematical conditions on the kernel of operators,
which allow us to uniquely identify one or another type of process. In this part of applied mathematics,
the fractional calculus requires its development. Mathematically strict conditions on the operator
kernels are necessary to initially distinguish between various types of processes and phenomena.
It should be emphasized that we must first clearly distinguish between the types of processes and
phenomena, but simply list various examples of their specific manifestations in the reality surrounding
us, described by the natural and social sciences. It is necessary to establish a clear correspondence
between the types of operator kernels and the types of phenomena.

3. Examples of Problems from Non-Standard Properties of Fractional Derivatives

In this section, we present examples illustrating the problems and difficulties of fractional
generalization of standard dynamic models, which arise from non-standard properties of fractional
derivatives. As an example of the problem with the non-standard form of the chain rule, we consider a
fractional generalization of the Kaldor-type model of business cycles. Problem with the violation of the
standard semi-group rule for orders of derivatives is shown for the fractional generalization of the
Phillips model of the multiplier-accelerator. To illustrate the problems arising from the non-standard
form of the product (Leibniz) rule, we consider the fractional generalization of the standard Solow–Swan
model. Problem with the violation of the standard semi-group property of dynamic map is described
on the examples of fractional generalization of the dynamic Leontief (intersectoral) model and logistic
growth model.

57
Mathematics 2019, 7, 554

3.1. Example of Problems with Chain Rule: Kaldor-Type Model of Business Cycles and Slutsky Equation
In this subsection, we demonstrate that the violation of the standard chain rule gives a restriction in
fractional generalization of dynamic models. For this purpose, we consider a fractional generalization
of the Kaldor-type model of business cycles and the economic model [66–68] based on the van der Pol
equation [69,70].
Economic models, which are based on the van der Pol equation, are considered as prototypes of
model for complex economic dynamics [69,70]. Nonlinear dynamic models are used to explain irregular
and chaotic behavior of complex economic and financial processes (for example, see the business cycle
theory [71,72], nonlinear economic dynamics and chaos [73,74], and stabilization [75]). Some models of
business cycles, which are based on the Kaldor nonlinear investment-savings functions [69–72] and the
Goodwin nonlinear accelerator-multiplier (for example, see the Goodwin’s paper [76], and [77–79]),
can be reduced to the van der Pol equation, which describes damped oscillations [69–72].

3.1.1. Standard Kaldor-Type Model of Business Cycles

In the framework of Keynesian approach to theory of national income, Nicholas Kaldor
formulated [66–68] the first nonlinear model of endogenous business cycles in 1940. Kaldor consider
the interactions between the investment I (Y) and the savings S(Y), where Y = Y(t) denotes national
income. Using the fact that the linear functions I (Y) and S(Y) cannot describe processes of business
cycle, Kaldor proposed nonlinear form for I (Y) and S(Y), which leads to oscillatory processes of
business cycles [69,70].
Let us derive the equation of the Kaldor model of business cycles by using approach proposed
by Chang and Smyth [68] (see also [70–72]). In the Kaldor model, instead of the standard accelerator
equation I (t) = vY(1) (t) the dependence of investments on the rate of change of national income is
considered in the form:
I (Y, K) − S(Y, K) = v Y(1) (t), (43)

which takes into account the savings, where K = K(t) denotes the capital stock, Y = Y(t) is the
national income, v is the accelerator coefficient and Y(1) (t) denotes its time derivatives of first
order. The parameter a = 1/v is an adjustment coefficient. In this model assumes that IK (Y, K) =
∂I (Y, K)/∂K < 0 and SK (Y, K) = ∂S(Y, K)/∂K > 0.
Differentiation of Equation (43) with respect to time and using the standard chain rule, we obtain:

v Y(2) (t) = (IY (Y, K) − SY (Y, K))Y(1) (t) + (IK (Y, K) − SK (Y, K))K(1) (t). (44)

In the paper [68] it is assumed that the actual change in the capital stock is determined by savings
decisions, such that:
S(Y, K) = K(1) (t), (45)

where K(1) (t) denotes the time derivatives of first order of the capital stock K(t). Substitution of
Equation (45) into Equation (44) gives:

v Y(2) (t) = (IY (Y, K) − SY (Y, K)) Y(1) (t) + (IK (Y, K) − SK (Y, K)) S(Y, K). (46)

In the paper [68], it is also assumed that the function I (Y, K) is linear in K(t) and savings is
independent of the capital stock, i.e., the function S(Y, K) = S(Y). In this case, the expression
(IK (Y, K) − SK (Y, K)) is independent of the capital stock K(t) and Equation (46) takes the form:

v Y(2) (t) = (IY − SY )(Y) Y(1) (t) + IK (Y)S(Y). (47)

58
Mathematics 2019, 7, 554

Using the variable y(t) = Y(t) − Y, where Y is the equilibrium value, Equation (47) can be rewritten [70]
in the form of the Lienard equation:

y(2) (t) + g( y(t)) y(1) (t) + f ( y(t)) = 0, (48)

which is used in mechanics to describe the dynamics of a spring-mass system.

Assuming symmetric shapes of the  investment
 and savings functions, the parabolic form of the
function of their difference, g( y) = μ y2 − 1 , and the linear form of f ( y) = y, we obtain the Van der
Pol equation:  
y(2) (t) + μ y2 (t) − 1 y(1) (t) + y(t) = 0. (49)

This equation is used in economic modeling of the business cycles in the framework of nonlinear
economic models with continuous-time. The Van der Pol Equation (49) can be written in the
two-dimensional form: ⎧
⎪
⎪ y(1) = x,
⎨
⎪
⎪ (50)
⎩ x ( 1 ) = μ 1 − y2 x − y ( t ) .

This form of the Van der Pol equation is used in computer simulation on the phase space.

3.1.2. Fractional Generalization of Kaldor-Type Model of Business Cycles

To generalize Equation (49) for the case of processes with memory, we cannot simply replace the
derivatives of integer order by fractional derivatives to get the fractional Van der Pol equation:
    β 
DαC,0+ X (t) + μ y2 (t) − 1 DC,0+ X (t) + y(t) = 0, (51)

where α > β > 0. The fractional generalization of the Van der Pol equation are considered in physics
(for example, see [80–82]) and in economics [83,84].
To correctly generalize the standard model, it is necessary to take into account the process of
obtaining Equations (49) and (50) from Equation (43). Note that the replacement of the derivatives of
the integer order in Equations (43) and (44) by fractional derivatives also does not allow obtaining the
fractional differential Equation (51). This is because, when deriving Equation (49) from Equations (43)
and (44), we must use the standard chain rules in the form:

D1t F(Y(t), K(t)) = FY (Y, K)Y(1) (t) + FK (Y, K) K(1) (t), (52)

where D1t = d/dt.

The chain rule for fractional derivative has more complicated form (see equation (2.209) in section
2.7.3 of [3,19]). As a result, we should restrict ourselves to the assumption of the presence of a memory
only for Equation (43). Let us assume that the excess of investment over saving, i.e., the difference
I (Y, K) − S(Y, K) is determined by changes in the growth rate of the national income in the past:

t
I (Y(t), K(t)) − S(Y(t), K(t)) = v(t − τ)Y(1) (τ)dτ, (53)
0

where the time variable is considered as dimensionless variable. For the case v(t − τ) = v δ(t − τ)
Equation (53) gives Equation (43) of the standard model.
The memory with one-parameter power-law fading is described [47,48,60] by the function:
vα
v(t − τ) = (t − τ)−α , (54)
Γ (1 − α)

59
Mathematics 2019, 7, 554

 
where Γ(α) is the gamma function and 0 < α ≤ 1, and DαC,0+ Y (t) is the Caputo fractional derivative:

     t
(n) 1
DαC,0+ Y (t) = IRL,0
n−α
+Y (t) = (t − τ)n−α−1 Y(n) (τ)dτ, (55)
Γ (n − α) 0

where n = [α] + 1 for α  N and n = α for α ∈ N, and the function Y(τ) has integer-order derivatives
Y( j) (τ), j = 1, . . . , (n − 1), that are absolutely continuous.
Equation (53) with kernel (54) can be rewritten through the Caputo fractional derivative:
 
I (Y(t), K(t)) − S(Y(t), K(t)) = vα DαC,0+ Y (t). (56)

Action of the first-order derivative D1t = d/dt. with respect to time on Equation (56) and using the
standard chain rule, we obtain:
 
vα D1t DαC,0+ Y (t) = (IY (Y, K) − SY (Y, K))Y(1) (t)+
(57)
(IK (Y, K) − SK (Y, K))K(1) (t)

Substituting Equation (45) into Equation (57) gives:

 
vα D1t Dα (1)
C,0+ Y (t) = (IY (Y, K ) − SY (Y, K ))Y (t)+ (58)
(IK (Y, K) − SK (Y, K))S(Y, K).

Using the assumptions that are proposed in the paper [68], Equation (58) takes the form:
 
vα D1t DαC,0+ Y (t) = (IY − SY )(Y) Y(1) (t) + IK (Y)S(Y), (59)

and:  
D1t DαC,0+ y (t) + g( y) y(1) (t) + f ( y) = 0. (60)
   
+1
Note that D1t DαC,0+ y (t)  DαC,0 +
y (t) since the standard semi-group rule for order of derivatives is
violated in general.
To obtain two-dimensional form of fractional differential Equation (60), we can use the
Riemann–Liouville fractional derivative that is defined by the equation:

     t
1 dn
DαRL,0+ Y (t) = Dnt IRL,0
n−α
+ Y (t) = (t − τ)n−α−1 Y(τ)dτ. (61)
Γ(n − α) dtn 0

Using Equation (61), we can get the equalities:

        
(1) (1)
D1t DαC,0+ Y (t) = Dt IRL,0+ Y
1 1−α
(t) = D1t IRL,0
1−α
+ Y (t) = DαRL,0+ Y(1) (t). (62)

This allows us to rewrite Equation (60) as:

 
DαRL,0+ y(1) (t) + g( y) y(1) (t) + f ( y) = 0. (63)

As a result, the Kaldor-type model of business cycles with power-law memory can be described
by the fractional Van der Pol Equation (63). Equation (63) can be written in the two-dimensional form:
⎧
⎪
⎪ D1t y = x, 
⎨
⎪
⎪ (64)
⎩ DαRL,0+ x = μ 1 − y2 x − y.

60
Mathematics 2019, 7, 554

This form of the fractional Van der Pol equation can be used in computer simulation of the
Kaldor-type model of business cycles with power-law memory by analogy with the papers in
physics [80–82], and in economics [83,84].

3.1.3. Fractional Generalization of Slutsky Equation

The Slutsky Equation (see classical paper [85], and its available copies [86–89]), which is used in
microeconomics [90–92], allows us to calculate the unobservable functions (compensated (Hicksian)
demand function) from observable functions such as the derivatives of the ordinary (Marshallian)
demand function with respect to price and income. The difficulties of the fractional generalization
of the standard Slutsky equation is connected with the using the chain rule in the derivation of this
equation in microeconomics.
Let us describe the derivation of the standard Slutsky equation. For simplification we will assume
that there are only two goods (x and  y). In microeconomics,
 two type demand function are used: the
compensated demand function, xc px , p y , U , and the ordinary (uncompensated) demand function,
 
x px , p y , I . The compensated (Hicksian) demand function describes the demand of a consumer over a
bundle of goods (x and y) that minimizes their expenditure while delivering a fixed level of utility.
The compensated demand functions are convenient from a mathematical point of view  since these

functions do not require income or wealth to be represented. In addition, the function xc px , p y , U is
linear in (x, y), which gives a simpler optimization problem. Unfortunately these functions
 are not
directly observable. The uncompensated (Marshallian) demand functions x px , p y , I are convenient
from an economic point  of view.
 However, this convenience is due to the fact that the uncompensated
demand function x px , p y , I describes demand given prices px , p y and income I that are easier to
observe directly in economics.
The compensated (Hicksian) demand function is defined by the equation
    
xc px , p y , U = x px , p y , E px , p y , U , (65)
 
where E px , p y , U is the expenditure function that gives the minimum wealth required to get to a given
utility
 level.
 Equation (65) is obtained by inserting that expenditure level into the demand function,
x px , p y , I . Note that the variables px , p y enter into the ordinary demand function in (65) in two places.
In 1915, Evgeny E. Slutsky proposed [85–89] an equation that allows us to calculate the compensated
(Hicksian) demand function from observable functions, namely, the derivative of the Marshallian
demand with respect to price and income.
To derive the Slutsky equation, we apply the partial differentiation of Equation (65) with respect
to px . This allows us to obtain the equation:
         
∂xc px , p y , U ∂ x px , p y , E px , p y , U ∂ x px , p y , E px , p y , U ∂E px , p y , U
= + , (66)
∂px ∂px ∂E ∂px

where we use the standard chain rule. Then we should change the notation and taking into account two
following economic effects. The first, we take into account the substitution effect that mathematically
is represented by the equality:
  ⎛    ⎞
∂xc px , p y , U ⎜⎜ ∂ x px , p y , E px , p y , U ⎟⎟
⎜
⎜ ⎟⎟
= ⎜⎜ ⎟⎟ , (67)
∂px ⎝ ∂px ⎠
U =const

that indicates movement along a single indifference curve (U = const). The second, we take into
account the income effect in the form:
    
∂ x px , p y , E px , p y , U ∂ x px , p y , I
= , (68)
∂E ∂I

61
Mathematics 2019, 7, 554

 
because changes in income or expenditures is the same thing in the function x px , p y , I . Then we can
use the Shephard’s lemma in the form:
 
∂E px , p y , U  
= xc px , p y , U . (69)
∂px

Substitution of Equations (67)–(69) into Equation (66) gives the Slutsky equation:
  ⎛    ⎞  
∂xc px , p y , U ⎜⎜ ∂ x px , p y , E px , p y , U ⎟⎟ ∂ x px , p y , I  
⎜⎜ ⎟⎟
= ⎜⎜ ⎟⎟ + xc px , p y , U , (70)
∂px ⎝ ∂px ⎠ ∂I
U =const
   
where we should see that xc px , p y , U (x∗ , y∗ ) = xc px , p y , I at the utility-maximizing point (x∗ , y∗ ).
In fractional generalization of the Slutsky equation, the violation of the standard chain rule leads
to the equation:
  pαx
     
Dαpx xc px , p y , U = Γ(1−α x p , p y , U + Dαpx x px , p y , E px , p y , U
) c x
+

∞
k ∂m x(p , p , I ) k−1

 r  ar U=const (71)
k! pk−α ∂ x ( p , p , U )
Cαk Γ(k−αx+1)
x y 1 1 c x y
∂Im ar ! r! ∂pr
,
k =1 m=1 r=0 x

which has a significant complication of the form in compared to the standard equation. In the fractional

Slutsky equation extends over all combinations of non-negative integer values of a1 , a2 , . . . , ak such
k
that r=1 rar = k and kr ar = m.
In addition, the fractional Slutsky equation does not make much sense from an economic point of
view, if we consider it as a description of the relationship of compensated (Hicksian) demand function
and ordinary (Marshallian) demand function. The standard equation describes the connection these
functions in full and this connection is local.
However, the Slutsky fractional equation is important from the other point of view. It is known
that the standard Slutsky equation can be represented in terms of elasticity. In this form the Slutsky
equation describes a connection of the compensated (Hicksian) price elasticity, the (uncompensated)
price elasticity, and the income elasticity of goods. The proposed fractional Slutsky equation describes
a connection of the fractional Hicksian elasticity of non-integer order [93–96] and the Marshallian
(uncompensated) price and income elasticities, which are special cases of the fractional elasticity [93–96]
for α ∈ N.
In this regard, we note that the fractional elasticity of a non-integral order can be represented as
an infinite sum of elasticities of a higher order, using an equation expressing a fractional derivative in
view of the infinite sum of the derivatives of integer orders (see lemma 15.3 in [1], p. 278).

3.2. Example of Problems with Semi-Group Rule for Orders of Derivatives: Phillips Model of
Multiplier-Accelerator
Let us consider a fractional generalization of the standard Phillips model of the
multiplier-accelerator to demonstrate the fact that the semi-group rule for orders of fractional derivatives
gives a restriction in the construction of such generalizations.
The Phillips model of the multiplier-accelerator has been proposed by Alban W.H. Phillips [97,98]
(see also [55,78,79,99]) in 1954 as a generalization of the Harrod–Domar macroeconomic growth model
with continuous time. The standard Phillips model is described by the ordinary differential equation
of second order in the form:

Y(2) (t) + a Y(1) (t) + bY(t) = λ1 λ2 A, (72)

where a = λ2 s + λ1 − λ1 λ2 v and b = λ1 λ2 s; Y(t) is the national income; 0 < s < 1 is the marginal
propensity to save; v is the investment coefficient; λ1 is the speed of response of output to changes in

62
Mathematics 2019, 7, 554

demand; λ2 is the speed of response of induced investment to changes in output. The autonomous
expenditure A(t) is assumed [78,79] to be constant (A(t) = A).
The formal generalization of Equation (72) by replacing the derivatives of integer orders by
fractional derivatives has the form:
 β
  
DC,0+ Y (t) + a DαC,0+ Y (t) + bY(t) = λ1 λ2 A, (73)

where β > α > 0 and DαC,0+ is the Caputo fractional derivative, for example. Such a generalization does
not take into account how the standard Phillips equation was obtained. It does not take into account
what assumptions are used in the basis and what economic concepts were applied for the derivation of
equation of standard model.
Let us briefly describe the process of obtaining the standard equation. The first assumption is
form of equation of the investment accelerator [78], p. 72. The value of the actual induced investment
I (t) at time t in response to changes in output Y(t) is given by:
 
I (1) (t) = −λ1 I (t) − v Y(1) (t) . (74)

The second assumption is the equation for the total demand Z(t) in the form:

Z(t) = C(t) + I (t) + A(t), (75)

where C(t) = cY(t) is the planned consumption, and we can use s = 1 − c, the marginal propensity
to save instead of the marginal propensity to consume c ∈ (0, 1). Then we have the equation:

Z(t) = cY(t) + I (t) + A(t). (76)

The third assumption is the multiplier equation [78], p. 73, in the form

Y(1) (t) = −λ2 (Y(t) − Z(t)). (77)

The equations of the standard model are Equations (74), (76)–(77). A differential equation for income
Y(t) is obtained by eliminating Z(t) and I(t) from the system of Equations (75)–(77). Substitution of
Equation (76) into Equation (77) allows us to obtain the expression for the induced investment in the form:

(1)
I (t) = λ−1
2 Y (t) + sY(t) − A(t). (78)

Substituting Equation (78) into Equation (76) under the assumption that the autonomous
expenditure A(t) = A is constant, we obtain Equation (72) of the standard Phillips model by the
first-order differentiation.
The type of Equations (74), (76), and (77), which are used in the derivation of the standard model
Equation (72), gives an impression that it is possible to propose a fractional generalization of the
standard model using a formal replacement of the derivatives of first order by fractional derivatives in
Equations (74) and (77). This gives the following system of equations:
⎧  α1     
⎪
⎪
⎪ DC,0+ I (t) = −λ1 I (t) − v DαC,0
2
+
Y (t) ,
⎪
⎨
⎪
⎪
⎪
⎪  αZ(t) =  cY(t) + I (t) + A(t), (79)
⎩ D 3 Y (t) = −λ2 (Y(t) − Z(t)),
C,0+

where the orders of fractional derivatives do not necessarily coincide, and 0 < α1 , α2 , α3 ≤ 1.
The last two equations of system (79) give an expression for the function I (t) in the form:
 α 
I (t) = λ−1
2 DC,0+ Y (t) + sY(t) − A(t).
3
(80)

63
Mathematics 2019, 7, 554

Substituting Equation (80) in the first equation of system (79) under the assumption that the autonomous
expenditure A(t) = A is constant, we obtain the equation:
 α1
 α   α     α 
DC,0 +
3
DC,0 +
Y (t) + λ2 s DC,0
1
+
Y (t) − λ1 λ2 v DαC,0
2
+
Y (t) + λ1 DC,0
3
+
Y (t)+
(81)
λ1 λ2 sY(t) = λ1 λ2 A.

For the case α1 = α2 = α3 = α, we have the equation:

    
DαC,0+ DαC,0+ Y (t) + a DαC,0+ Y (t) + bY(t) = λ1 λ2 A, (82)

where
a = λ2 s + λ1 − λ1 λ2 v ; b = λ1 λ2 s. (83)

As a result, we see that in general case the fractional generalization of the Phillips model can be
described by Equation (82) instead of Equation (73). We can also see that Equation (82) cannot contain
β = 2α as it used in Equation (73). In addition, the violation  of the
 standard semi-group rule for the
orders of derivatives led us to the fact that we have DαC,0+ DαC,0+ Y instead of D2α C,0+
Y.
It should be emphasized that the generalization given by equation system (79) is formal and
does not reflect the economic sense of the original Equations (74) and (77) of the standard model.
In Equations (74) and (77), the derivatives of the functions to the left of the equal sign in reality are part
of the operator of the exponential distributed lag [78], pp. 72–74.
The standard Phillips model of the multiplier-accelerator takes into account two continuously
distributed lags. The first lag characterize the output responding to demand with speed λ1 . The second
lag describes the induced investment responding to changes in output with speed λ2 . These economic
accelerator and multiplier can be described by the following operators.
The integer-order derivative with exponentially distributed lag can be defined [46] by the
first-order equation:
   t
 
Dλ,n
T,C
Y ( t ) = λ exp −λ (t − τ) Y(n) (τ)dτ, . (84)
−∞
where n ∈ N0 . For n = 0, we have:

   t
 
Dλ,0
T,C
Y ( t ) = λ exp −λ (t − τ) Y(τ)dτ. (85)
−∞

In reality, the first and third assumptions of the standard model, which are described by Equations
(74) and (77), should be written [78], pp. 25–27, in the form of the equations:
 λ ,1 
I(t) = v DT,C
1
Y (t), (86)

and:  
Y ( t ) = Dλ 2 ,0
T,C
Z (t). (87)

In standard macroeconomic models, the differential equations of exponentially distributed lag are
used in the form of Equations (74) and (77) instead of equations with integro-differential operators in
the form of Equations (86) and (87). Equations (74) and (77) are called the differential equations of the
exponential lag [78], p. 27. In economics, the use of differential equations of integer orders instead of the
integro-differential operators (86) and (87) is caused by the fact that there are considerable difficulties in
handling the integrals in Equations (86) and (87). It is seen that equations with continuously distributed
lag are equivalent to differential equations of integer orders under certain conditions. These differential
equations are easier to handle in comparison with equations that contain integro-differential operators
of the distributed lag.
As a result, to obtain a correct generalization of the standard Philips model, we should use the
fractional derivative with exponentially distributed lags [46,55] instead of the integer-order operators

64
Mathematics 2019, 7, 554

with exponentially distributed lags. For example, we can use the Caputo fractional derivative with
exponentially distributed lag:

   t
  
Dλ,α
T,C
Y ( t ) = λ exp −λ (t − τ) DαC,0 Y (τ)dτ, (88)
0
 
where λ > 0 is the rate parameter of exponential distribution and DαC,0 Y is the Caputo fractional
derivative of the order α > 0.
Another generalization method is to account for memory effects instead of the distributed lag
effect [55]. This generalization assumes to use fractional derivatives (without distributed lag) instead
of integer-order operators (86) and (87).
Self-consistent constructions of different fractional generalizations of the standard Phillips model
of the multiplier-accelerator were proposed in the work [55].
At the same time, Equation (73), which is a formal fractional generalization of the equation of
the standard Phillips model, does not have economic significance due to the violation of the principle
of derivability.

3.3. Example of Problems with Product Rule: Solow–Swan Model

In this subsection, we consider a fractional generalization of the standard Solow–Swan model
(see, classical papers [100–102], and books [103,104]) to demonstrate the fact that the violation of
the standard product (Leibniz) rule for fractional derivatives [17,20,22], which is main characteristic
property of these operators, gives a restriction in the construction of such generalizations.
The standard Solow–Swan model with continuous time is represented in the form of the single
nonlinear ordinary differential equation:

k(1) (t) = −(a + b)k(t) + p f (k(t)), (89)

which describes how an increase of capital stock leads to an increase of per capita production, when
the supply of labor changes as L(t) = L0 exp (at) at a constant rate a ∈ (−1, +1). Here k(t) = K(t)/L(t)
is the per capita capital; K(t) is capital expenditure; b ∈ (0, 1) is the capital retirement ratio; p ∈ (0, 1)
is the rate of accumulation. The function f (k(t)) describes the labor productivity, which is usually
considered in the form f (k(t)) = Akγ (t) with γ ∈ (0, 1).
The formal generalization, which is realized by replacing the first-order derivative by the fractional
derivative in Equation (89), has the form:
 
DαC,0+ k (t) = −(a + b)k(t) + p f (k(t)), (90)

where DαC,0+ is the Caputo fractional derivative, for example.

Unfortunately, the consistent construction of the fractional generalization of the standard model
Equation (89) cannot give a fractional differential equation in the form of Equation (90). In order
to prove this statement, we first briefly describe the consistent construction of the equation for the
standard Solow model.

3.3.1. Standard Solow Model with Continuous Time

The Solow model, which is also called the Solow–Swan model, is a dynamic single-sector model
of economic growth (see, Solow and Swan articles [100–102], and books [103,104]). In this model,
the economy is considered without structural subdivisions. The economy produces only universal
products, which can be consumed both in the non-production and production sectors. As a universal
product, one can consider a monetary value of the entire economy. Exports and imports are not taken
into account. This model describes the capital accumulation, labor or population growth, and increases

65
Mathematics 2019, 7, 554

in productivity, which is commonly called the technological progress. The Solow model can be used to
estimate the separate effects on economic growth of capital, labor and technological change.
The Solow model is a generalization of the Harrod–Domar model, which includes a productivity
growth as new effect. This relatively simple growth models was independently proposed by Robert M.
Solow and Trevor W. Swan in 1956 [100,101]. In 1987 Solow was awarded the Nobel Memorial Prize
in Economic Sciences for his contributions to the theory of economic growth [105]. Mathematically,
the Solow–Swan model is actually represented by one nonlinear ordinary differential equation
(Equation (89), which describes the evolution of the per capita stock of capital. Now it is a classical
nonlinear economic model that is actively used in economics [106–109].
In the Solow model, the state of the economy is given by the following five endogenous
state variables (defined within the model): Y(t) is the final product (production capacity), L(t) is
the labor input (available labor resources), K(t) describes the capital reserves (capital expenditure,
production assets), I (t) is the investment (investment rates), and C(t) is the amount of non-productive
consumption (instant consumption). All variables are functions of time t, which is assumed to be
continuous. In addition, the Solow model uses exogenous indicators (defined outside the model):
a ∈ (−1, +1) is the rate of increase in labor resources; b ∈ (0, 1) is the capital retirement ratio; p ∈ (0, 1)
is the rate of accumulation (the share of the final product used for investment). These exogenous
indicators are considered constant in time. The rate of accumulation is considered as a controlling
parameter. It is assumed that the production and labor resources are fully used in the production
of the final product. The final product at each moment in time is a function of the capital and labor:
Y = F(K(t), L(t)). This production function F (K, L) of the national economy is often specified to be a
function of the Cobb–Douglas type. It is assumed that Y = F(K, L) is a linearly homogeneous function
satisfying the constant scale, i.e.:
F(zK, zL) = zF (K, L). (91)

The final product is used for non-productive consumption and investment: Y(t) = C(t) + I (t).
The accumulation rate p ∈ (0, 1) is the fraction of the final product used for investment, i.e., I (t) = pY(t).
Therefore, we have the multiplier equation C(t) = (1 − p) Y(t).
If we assume that the increase in labor resources is proportional to the available labor resources,
then taking into account the growth rate of employed a ∈ (−1, +1), we can write the differential equation:

L(1) (t) = aL(t), (92)

where L(1) (t) = dL(t)/dt is the derivative of first order. Equation (92) with the initial condition
L(0) = L0 , has the solution L(t) = L0 exp (at), where L0 is the labor resources at the beginning of
observation at t = 0. The equation of labor resources can also be considered in the form of the logistic
equation (for example, see [106]).
Capital stock may change for two reasons: investment causes an increase in capital stock;
depreciation or disposal of capital causes a decrease in its reserves. If we assume that the retirement of
capital occurs with a constant retirement rate of b ∈ (0, 1), then the capital dynamics is described by
the equation K(1) (t) = I (t) − bK(t). Finally, taking into account I (t) = pY(t) and Y = F(K(t), L(t)),
we obtain:
K(1) (t) = pF(K(t), L(t)) − bK(t). (93)

To obtain the equation of the standard Solow model, the following relative variables are introduced.
The per capita capital (capital endowment) is defined as k(t) = K(t)/L(t). The labor productivity is:

y(t) = Y(t)/L(t) = F(K(t), L(t))/L(t) = F(K(t)/L(t), 1) = f (k), (94)

where we use the property (Equation (91)) of the linear homogeneity of the production function.
The dynamics of the output of the final product depends on the amount of the capital per employed
person, the per capita capital k(t) = K(t)/L(t).

66
Mathematics 2019, 7, 554

Substitution of K(t) = k(t) L(t) into Equation (93) gives:

(k(t) L(t))(1) = pF(k(t) L(t), L(t)) − bk(t) L(t). (95)

Using the standard product (Leibniz) rule:

(k(t) L(t))(1) = k(1) (t) L(t) + k(t) L(1) (t), (96)

and the property of the linearly homogeneity (Equation (91)), Equation (95) is rewritten in the form:

k(1) (t) L(t) + k(t) L(1) (t) = p f (k(t)) L(t) − bk(t) L(t). (97)

Using Equation (92) for the labor resources, we obtain:

k(1) (t) = −(a + b)k(t) + p f (k(t)). (98)

Equation (98) is the standard Solow–Swan model.

The behavior of the indicators of the standard Solow–Swan model is determined by the ordinary
differential equation (Equation (98)) of the first order and the dynamics of labor resources (Equation (92)).
The Cauchy problem, which consists of Equation (97) and an initial condition, has a unique solution.

3.3.2. Fractional Generalization of Solow Model

A fractional generalization of the labor resource Equation (92) and obtaining a solution to this
fractional differential equation is not difficult. If we take into account this consistent derivation of
Equation (98) of the standard model, we see that we cannot use the standard product (Leibniz) rule
for fractional derivative. Therefore, we cannot obtain a fractional generalization of the differential
Equation (98) for the per capita capital k(t) = K(t)/L(t) because of a violation of the standard Leibniz
rule for fractional derivatives of non-integer orders.
We emphasize that the violation of the standard product rule is a characteristic property of all
derivatives of non-integer order. Note that the implementation of the standard product rule for an
operator means that this operator is a differential operator of integer order [17], and such operators
cannot describe the effects of memory and nonlocality.
As a result, the fractional generalization of the standard Solow–Swan model, which will take
into account the power-law memory effects, should be represented as the system of the fractional
differential equation: ⎧  
⎪
⎪ DαC,0+ L (t) = aL(t),
⎨  
⎪
⎪ (99)
⎩ Dβ K (t) = pF(K(t), L(t)) − bK(t).
C,0+

The fractional dynamics of the per capita capital k(t) will be described as the ratio K(t)/L(t) of
solutions of these two fractional differential equations.
For production function of the national economy in the form the Cobb–Douglas function F (K, L) =
A Kγ (t)L1−γ (t), we have the system (99) in the form:
⎧  
⎪
⎪ DαC,0+ L (t) = aL(t),
⎨  
⎪
⎪ (100)
⎩ Dβ K (t) = pA Kγ (t)L1−γ (t) − bK(t).
C,0+

The fractional differential equation with n − 1 < α ≤ n, which describes the labor resources, has the
solution (theorem 5.15 of [4], p. 323) in the form:


n−1
L(t) = L(k) (0)tk Eα,k+1 [a tα ] (101)
k =0

67
Mathematics 2019, 7, 554

where L(k) (0) is integer-order derivatives of orders k ≥ 0 at t = 0, and Eα,k+1 [a tα ] is the two-parameter
Mittag–Leffler function [32]. In the case 0 < α ≤ 1 (n = 1) Equation (101) takes the form:

L(t) = L(0)Eα,1 [λ tα ]. (102)

For α = 1, we obtain the standard solution L(t) = L0 exp (at), where L0 = L(0).
Using Equation (102) that describes fractional dynamics of the labor resources, we can obtain the
nonlinear fractional differential equation for the capital expenditure K(t) in the form:
 β 
DC,0+ K (t) = pA Kγ (t)L0 (Eα,1 [λ tα ])1−γ − bK(t).
1−γ
(103)

In the case α = 1, this equation takes the form:

 β 
DC,0+ K (t) = pA Kγ (t)L0 ea(1−γ)t − bK(t).
1−γ
(104)

The question of the existence of solutions of nonlinear fractional differential Equations (103) and
(104) and computer modeling of capital expenditure dynamics remains open at the present time.
Note that the nonlinear fractional differential Equations (103) and (104) can be represented as
Volterra integral equations by using the results of the papers of Kilbas and Marzan [110,111]. In the space
Cr [0, T ] of continuously differentiable function the Cauchy problem for fractional differential equation:
 β 
DC,0+ K (t) = G(t, K(t)), (105)

where n − 1 < β ≤ n, is equivalent (see Theorem 3.24 of [4], pp.199–202, to the Volterra integral equation:


n−1 (m)  t
K (0) m 1
K (t) = t + (t − τ)β−1 G(τ, K(τ))dτ, (106)
m=0
m! Γ (α) 0

if the function G(t, K(t)) ∈ Cγ [0, T ] with 0 ≤ γ < 1 and γ ≤ β, the variable K(t) ∈ Cr [0, T ], where r = n
for integer values of β (β ∈ N) and r = n − 1 for non-integer values of β (β  N).
At the same time, Equation (90), which is a formal fractional generalization of the equation
(Equation (89)) of the standard model, does not have economic significance due to the violation of the
principle of derivability.

3.4. Example of Problem with Semi-Group Rule of Dynamic Map: Dynamic Leontief Model and Logistic
Growth Model
In this subsection, we consider fractional generalizations of the standard dynamic Leontief model
and logistic growth model to demonstrate that the violation of the standard semi-group rule of dynamic
map for fractional derivatives creates a restriction in the construction of such generalizations.

3.4.1. Dynamic Leontief (Intersectoral) Model

One of the famous multidimensional economic models is the dynamic intersectoral model that
was proposed Wassily W. Leontief [112,113] in 1951. The Royal Swedish Academy of Sciences has
awarded the 1973 year’s Prize in Economic Science in Memory of Alfred Nobel to W.W. Leontief
for “the development of the input-output method and for its application to important economic
problems” [114]. The Leontief dynamic model is an economic model of growth of gross national
product and national income [115,116].
The fractional generalization of the dynamic Leontief (intersectoral) model was proposed
in [117,118] in 2017 and in the works [119,120] for the case of time-dependent direct material costs and
the incremental capital intensity of production.

68
Mathematics 2019, 7, 554

Let us give the first example from the econophysics approach based on [117,118], and [119,120].
The fractional generalization of the equation for the dynamic Leontief (intersectoral) model [92,93] has
the form:  
Dαt0 + X (t) = H (t) X(t), (107)

where the vector X(t) = (Xk (t)) describes the gross product   (gross output) in monetary terms,
where k = 1, . . . , n are production sectors; the matrix A = aij describes the direct material costs;
 
the matrix B = bij describes the incremental capital intensity of production; the matrix E is the unit
diagonal matrix of n-th order; the matrix H is defined by the equation H = B−1 (E − A).
Equation (107) describes dynamics of the sectoral structure of the gross products in the closed
dynamic intersectoral model with power-law memory (for details, see [117,118], and [119,120]).
The solution of Equation (107) with constant operator H (t) = H = const has the form:

X ( t ) = Uα ( t ) X ( 0 ) , (108)

where the operator Uα (t) is defined through the Mittag–Leffler function with matrix arguments by the
equation Uα (t) = Eα [tα H]. Therefore, for the operator Uα (t), which describes the dynamic map with
power-law memory, we have the inequality:

Uα (t1 )Uα (t2 )  Uα (t1 + t2 ), (109)

which means the violation of the standard semi-group rule for non-integer values of α (for example,
0 < α < 1).
For the general case of the time-dependent matrix H (t), the solutions of Equation (110) are
given in [119,120]. To obtain these solutions, we proposed new concepts of the memory-ordered
exponential and memory-ordered product, which are a generalization to processes with memory
of such well-known concepts in quantum physics as time-ordered exponential (T-exponential) and
time-ordered product (T-product) [121,122].

3.4.2. Logistic Growth with Memory

The second example is taken from the economic model of logistic growth [104,123]. In economic
growth models, the competition effects are taken into account by assuming that price is a function of
the value of output. Model of natural growth in a competitive environment is often called a model of
logistic growth. The variables of this model are the function Y(t) that describes the value of output
at time t; the price P(t) is considered as a function of released product Y(t), i.e., P = P(Y(t)). It is
often assumed that this function is linear, i.e., P(Y(t)) = b–a Y(t), where b is the price, which is
independent of the output and the parameter a is the margin price. In addition, it is assumed that all
manufactured products are sold (the assumption of market unsaturation). The equation of this model
is the differential equation of the first order in the form:

dY(t) m
= (b − a Y(t)) Y(t), (110)
dt v
where v > 0 is the accelerator coefficient, 1/v is the marginal productivity of capital (rate of acceleration),
m is the norm of net investment (0 < m < 1) that describes the share of income, which is spent on the
net investment.
If a  0 and b  0, we can use the variable Z(t) and the parameter r, which are defined by the
equations Z(t) = ba Y(t), and r = mv . Then Equation (110) of the logistic growth model is represented in
the form:
dZ(t)
= r (1 − Z(t)) Z(t), (111)
dt

69
Mathematics 2019, 7, 554

The fractional generalization of the logistic growth model with power-law memory [123] gives
the fractional differential equation:
 
DαC,0+ Z (t) = r (1 − Z(t)) Z(t), (112)

where DαC;0+ is the Caputo fractional derivative. Equation (112) is the logistics fractional
differential equation.
The solution of nonlinear fractional differential equations is a difficult problem. Recently, Bruce J.
West has published the paper [124], where he proposed an analytical expression of the solution for the
fractional logistic equation with α ∈ (0, 1) in the form:
∞ 
 
Z(0) − 1
Z(t) = Eα [ k r ( α ) t α ] , (113)
Z(0)
k =0

where Eα (z) is the Mittag–Leffler function [4], p. 42. As it has been proved by I. Area, J. Losada,
J. Nieto in [125], the function (113) is not the solution to (11). The main reason is the violation the
semi-group property by the Mittag–Leffler function, i.e., we have (for example, see [33,34], and [35–37])
the inequality: 
Eα λ (t + s)α  Eα [λ tα ] Eα [λ sα ] (114)

for α ∈ (0, 1), and real constant λ. In [125] it has been proved that Equation (113), which is proposed
in [124], is not an exact solution of the fractional logistic Equation (112).
As a result, we see that the violation of the standard semi-group property for the dynamic map
us an important property of processes with memory that should be taken into account in dynamic
models. Neglect of this non-standard property of the dynamical map can lead to errors.

3.4.3. Principle of Optimality for Processes with Memory

The principle of optimality, which was originally proposed for dynamic programming by Belmann,
is very important for describing economic processes. The Bellman principle of optimality states that
any tail of an optimal trajectory is optimal too.
In considering optimal growth trajectories of economy, a concept known as the optimality principle
is very useful. Let us give the standard principle of optimality that describes processes without memory
(for example, see section 11.2 of [92]):

Principle of Optimality. Any optimal behavior has the property that whatever the initial state and corresponding
(initial) solution are, the subsequent solutions must constitute the optimal behavior with regards to the state
resulting from the initial solution.

Applied to economic growth theories, the optimality principle leads to the following conclusion.
If the trajectory is optimal, starts from point X(0) and passes through X(t) on the way to the end point
X(T ), then part of the trajectory from X(t) to X(T ) will be optimal with respect to the initial point X(t).
The implementation of the principle of optimality is based on the semi-group rule of dynamic
map. The violation of the standard semi-group rule of dynamic map for dynamics with memory leads
to violation of the standard principle of optimality.
Mathematically, the violation of the standard optimality principle is represented by the violation
of the semi-group rule of dynamic map.
Economically the reason for the violation of the standard principle of optimality is the cutting
off of part of the history of this process (that is, starting from the beginning of this process, but at a
different time point). In other words, if you put in place of the general director, whose age is 40 years
old, his same age 15 years, the company will develop differently.
As a result, we can formulate the following statement:

70
Mathematics 2019, 7, 554

Principle of Optimality in Processes with Memory. For economic processes with memory, any optimal
behavior has the property that whatever the initial state and corresponding (initial) solution are, the subsequent
solutions cannot constitute the optimal behavior with regards to the state resulting from the initial solution.
We can state that if there is no violation of the standard principle of optimality, then there is no memory in the
process “No Violation of Optimality Principle. No Memory”.

For economic growth models, the suggested optimality principle in processes with memory leads
to the following statement. If the trajectory is optimal, starts from point X(0) and passes through X(t)
on the way to the end point X(T ), then part of the trajectory from X(t) to X(T ) cannot be optimal with
respect to the initial point X(t) in general.
This principle actually means that the implementation of the standard optimality principle for
economic processes with memory in the general case is equivalent to the lack or absence of memory in
this process.

3.5. Generalizations of Economic Notions and Concepts

Derivability Principle states that it is not enough to get a fractional generalization of the differential
equations of economic model. It is necessary to generalize the whole scheme (all steps) of obtaining
these equations from the basic principles, concepts and assumptions that is used in economic theory
for standard model. In this sequential derivation of the equations, we should take into account the
non-standard characteristic properties of fractional derivatives and integrals. Another important
requirement of the derivability principle is the need to generalize economic the notions, concepts and
methods, which were used in the derivation of standard model.
It should be noted that formal replacements of derivatives of integer order by fractional derivatives
in standard differential equations, and then solutions of these fractional differential equations cannot
be considered as a correct and self-consistent fractional generalization of the standard dynamic models
in different sciences.
A very important part of the fractional generalization of dynamic models is the inclusion of
memory and non-locality into the economic theory and into the basic economic concepts and methods.
A fractional generalizations of basic economic concepts and notions are not so much a part of this
particular economic model, but in fact are the common basis of different models, and basis of fractional
mathematical economics, and not just an economic model.
The concept of memory for economics is considered in [47–50] and [126–131]. The fractional
dynamic models should be constructed on this conceptual basis. The most important task of studies of
such fractional generalizations is also the search for qualitatively new effects and phenomena caused
by memory and non-locality in the behavior of processes.
Let us give a list of some standard notions of economic theory, the generalization of which were
proposed to describe the processes with memory and non-locality in the last years.
The list of these new notions and concepts primarily include the following:

• the marginal value of non-integer order [132–134], (see also [40,41]) with memory and nonlocality;
• the multiplier with memory [60,135–139];
• the accelerator with memory [60,135–139] (see also [140,141]);
• the duality of the multiplier with memory and the accelerator with memory [60,135];
• the elasticity of fractional order [93–96];
• the measures of risk aversion with memory [142] and non-locality [143];
• the warranted (technological) rate of growth with memory [144–146]; and
• the non-local fractional deterministic factor analysis [147,148], and other.

The use of these notions and concepts makes it possible us to construct fractional generalizations of
some economic models. A brief description of the history of the use of fractional calculus in economics
is proposed as a separate article [149].

71
Mathematics 2019, 7, 554

4. Example of Application of the Solvability and Correspondence Principles

The Solvability Principle assumes the existence of solution, and the possibility of obtaining
an exact analytical solution or a correct numerical solution for some conditions. Obviously, these
conditions for the existence of solutions should allow us to describe the processes considered in natural
and social sciences.
The Correspondence Principle assumes that in the limit cases of integer orders the solution
(and equation) should exist and the expression of this solution (and equation) should give expression
of the standard solution. The principle of correspondence must be performed both for the fractional
differential equation itself and for its solution.

4.1. Solvability Principle: Example from General Fractional Calculus

A general concept of fractional calculus was proposed by Anatoly N. Kochubei [150] on the
basis of the differential-convolution operator. The general fractional calculus is described in the
works [150,151], where author describes the conditions under which the general operator has a right
inverse (a kind of a fractional integral) and produce, as a kind of fractional derivative, equations.
A solution of the relaxation equations with the Kochubei fractional derivative with respect to the time
variable is described. As a special case of the general fractional operators, the fractional derivatives
and integrals of distributed order are considered in [150,151].
In the works about the general fractional calculus [150,151] the Cauchy problem (A) is considered
for Equation (D(k) X)(t) = λX(t), where λ < 0 (see [151], p. 112). In Section 6 “Relaxation equations”,
Theorem 4 states that this Cauchy problem has a solution X(λ, t), which is continuous on R+ , infinitely
differentiable and completely monotone on R+ , if the Kochubei conditions (*) hold. The works [150,151]
consider only the case of relaxation, i.e., λ < 0. The case of growth (λ > 0) is not discussed.
In the economics, different growth models are actively studied. In the simplified form, these
growth models can be described by the ordinary differential equation D1t X(t) = λX(t), where λ > 0.
The fractional generalization of these models, in which the memory function k(t) is taken into account,
can be described by the Equation (D(k) X)(t) = λX(t) with λ > 0, i.e., “relaxation equations” is replaced
by “growth equations”.
It is known that for the Caputo fractional derivative, which is a special case of the Kochubei
fractional derivatives, the Cauchy problem (A) has a solution X(λ, t), for all real λ ∈ R, i.e., for λ < 0
and λ > 0 (see theorem 4.3 in [4], p. 231).
Therefore, the following questions, which are important for describing processes with memory in
economics, arise within the framework of general fractional calculus.

1. Is there a mathematical reason for using only the condition λ < 0 in general calculus, when the
Caputo fractional derivative there is no such restriction?
2. Could we tell something under what conditions on the memory function, which is described by
the kernel k(t) of the general fractional derivative, the solution exists for λ > 0?
3. Is it possible to specify a wider class of operators than the fractional Caputo derivative for the
existence of solutions of growth equations?
4. Do the conditions of existence of solutions for the general relaxation equation and the general
growth equation coincide?
5. What types of asymptotic behavior of solutions of general growth equations and type of growth
rates exist?

The growth equation is considered in [150] for the special case of a distributed order derivative
(see also [152]). In this paper it was proved that a smooth solution exists and is non-decreasing belongs
to C∞ (0, ∞). To understand the warranted (technological) growth rate of the economy, it is important
to know the asymptotic behavior of this solution. The description of the asymptotic behavior of such
solutions is an open question at the moment. This complicates the economic interpretation of solutions

72
Mathematics 2019, 7, 554

and, thus, prevents the implementation of the interpretability principle, when writing works on the
economics of processes with distributed memory fading parameter.
In addition, the solution of the growth equation has been proposed for the case of fractional
differential operators with distributed lag in [46,53,54]. This case will be briefly described in the
next subsection.
The existence of a solution in the growth case has been also considered by Chung-Sik Sin [153]
in 2018 for a much more general case of a nonlinear equation with a generalized derivative like the
Kochubei fractional derivative.
The solution of the Cauchy problem for general growth equation is an open question at the
present moment. The growth case of general fractional calculus was discussed by Kochubei and
Kondratiev [154] as a part of the intermittency property in fractional models of statistical mechanics.
Unfortunately, the results were not formulated separately for the general fractional growth equation
in [154].
Solving the existence problem in the general case will allow us to accurately describe the conditions
on the operator kernels (the memory functions), under which equations for models of economic growth
with memory have solutions. The asymptotic behavior of these solutions allows us to describe the
warranted (technological) growth rate in the economy, in which we take into account this type of
memory. An article dedicated to solving this mathematical problem was written by Anatoly N.
Kochubei and Yuri Kondratiev [155] in 2019 for Special Issue “Mathematical Economics: Application
of Fractional Calculus” of Mathematics. The application of these mathematical results in economics
and their economic interpretation is an open question at the moment.

4.2. Distributed Lag Fractional Calculus: Growth-Relaxation Equations

The fractional calculus with continuously distributed lag is proposed in [46]. In the
papers [46,53–55], we consider an application of this fractional calculus to describe economic growth
with power-law memory and distributed lag.
Let us consider the fractional integration with the gamma distributed lag that is defined by the
equation [46,53–55] in the form:

      t  
Iλ,a;α
T;RL;0+
Y ( t ) = M λ,a
T
( τ ) ∗ I α
RL,0+ Y ( t ) = Mλ,a
T
α
(τ) IRL,0 + Y (t − τ) dτ, (115)
0

where Mλ,a
T
(τ) is the probability density function of the gamma distribution:
⎧
⎪
⎪ λa τa−1
exp(−λ τ) if τ > 0
⎨
Mλ,a (τ) =⎪
⎪
Γ (a) (116)
T ⎩ 0 if τ ≤ 0

with the shape parameter a > 0 and the rate parameter λ > 0. If a = 0, Equation (116) describes the
exponential distribution. The function Mλ,a
T
(τ) describes the distribution of the delay time τ, which is
considered as a random variable.
In the papers [46,53,54], we prove that the Riemann–Liouville fractional integral with gamma
distribution of delay time can represented [40] by the equation:

   t
λa Γ(a)
Iλ,a;α
T;RL;0+
Y (t) = (t − τ)α+a−1 F1,1 (a; a + α; −λ(t − τ))Y(τ) dτ, (117)
Γ (a + α) 0

where F1,1 (a; c; z) is the confluent hypergeometric Kummer function, α > 0 is the order of integration
and the parameters a > 0, λ > 0 describe the shape and rate of the gamma distribution, respectively.
Note that the kernel of Equation (117) can be represented through the three parameter Mittag–Leffler
function instead of the confluent hypergeometric Kummer function [53,54].

73
Mathematics 2019, 7, 554

The fractional integral (Equation (117)) is the Abel-type fractional integral operator with Kummer
function in the kernel (see equation (37.1) in [1], p. 731, and [53]). This kernel can be considered as a
new memory function.
Note that the fractional integral (Equation (117)) with gamma distributed lag (Equation (116)) can
be represented as the series of the Riemann–Liouville fractional integrals:

  
∞
Γ (a + k )  
λ,a;α α+a+k
IT;RL;0 +
Y (t) = (−1)k λk+a IRL,0 +
Y (t). (118)
Γ (k + 1)
k =0

The Caputo fractional differential operator with gamma distributed lag can be expressed through
the Riemann–Liouville fractional integral operator in the form:
      
Dλ,a;α
T;C;0+
Y (t) = Mλ,a
T
(τ) ∗ DαC,0+ Y (t) = Iλ,a;n−α
T;RL;0+
Y (n) ( t ) , (119)

where n − 1 < α ≤ n, the parameters a > 0 and λ > 0 describe the shape and rate of the gamma
distribution of delay time, respectively.
Let us consider the growth-relaxation equation with the fractional operator (Equation (119)).
In the works [46,53,54], for the fractional differential equation:
 
λ,a;α
DT;C;0 +
Y (t) = ωY(t), (120)

we proposed has the solution:


n−1
α−j−1
Y (t) = Sα,a [ω λ−a , λ|t]Y( j) (0), (121)
j=0

γ

where n = [α] + 1, and Sα,δ [μ, λt] is the special function that is defined by the expression:

 
∞
tδ(k+1)−αk−γ−1
γ
Sα,δ [μ, λt] = − F1,1 (δ(k + 1); δ(k + 1) − αk − γ, −λt), (122)
k =0
μk+1 Γ(δ(k + 1) − αk − γ)

where F1,1 (a; b; z) is the confluent hypergeometric Kummer function.

To prove this statement we can use the Laplace transform method [46]. Using the Laplace
transform of Equation (120), we obtain:
⎛ ⎞
λa ⎜⎜⎜⎜ α 
n−1 ⎟⎟
⎟
a⎜⎜ ( )(
s LY s −) s Y 0 ⎟⎟⎟ = ω(LY)(s).
α−j−1 ( j)
( ) (123)
(s + λ) ⎝ j=0
⎠

Then we can write:


n−1
sα−j−1
(L Y)(s) = Y ( j) (0 ), (124)
j=0
sα − μ(s + λ)a

where μ = ω λ−a . Using equation (5.4.9) of [156] and [157] in the form:
  
sa 1
L−1 (s) = tc−a−1 F1,1 (c; c − a, −bt), (125)
(s + b)c Γ (c − a)

where Re(c − a) > 0, we obtain [46,53,54] the inverse Laplace transform:

⎛ ⎛ ⎞⎞
⎜⎜ −1 ⎜⎜ sγ ⎟⎟⎟⎟ 
⎜⎜L ⎜⎜ ⎟⎟⎟⎟(s) = Sγ [μ, λt]. (126)
⎝ ⎝ ⎠⎠ α,δ
sα − μ ( s + λ ) δ

74
Mathematics 2019, 7, 554

Then using equality (126), the solution of Equation (120) takes the form of Equation (121).
Note that we can use Equation (1.9.13) [4], p. 47 (see also [32,158,159]) in the form:
  
γ sαγ−β
L tβ−1 Eα,β (λtα ) (s) = , (127)
( sα − λ ) γ

where λ ∈ C Re(s) > 0, Re(β) > 0 and |s|α > |λ|, instead of Equation (126) to get the representation
of Equation (121) though three parameter Mittag–Leffler functions instead of the confluent
hypergeometric functions.
Equation (1.9.3) of [4], p. 45, in the form:

γ 1
E1,β (z) = F11 (γ, β; z), (128)
Γ (β)
γ
where Eα,β (λtα ) is the three parameter Mittag–Leffler function [31] allows us to represent of the
fractional differential Equation (120) with 0 < α < 1 in the form:


∞
Y0 (t) = − ω−k λak t(α+a)k Eak
1,(α+a)k+1
(−λt) Y(0). (129)
k =1

Note that the solution of homogeneous fractional differential equation that describes economic
growth with memory in absence of time delay (lag) is expressed thought the two parameter
Mittag–Leffler function, where the argument depends on ω tα [144–146], instead of the rate parameter
λ > 0 of gamma distribution.

4.3. Correspondence Principle: Order of Derivative Tends to Integer Value

Let us give an example to illustrate that if the order of the fractional derivative tends to the integer
value, then the limit on the left and the limit on the right can give different results in the general case.
The fractional differential equation:
 
DαC,0+ Y (t) = λ Y(t) (130)

has the following solution. If α ∈ (0, 1) the solution takes the form:

Y1 (t) = Y(0)Eα,1 [λ tα ]. (131)

For α ∈ (1, 2) Equation (130) has the solution:

Y2 (t) = Y(0) Eα,1 [λ tα ] + Y(1) (0) t Eα,2 [λ tα ], (132)

where Y(1) (0) is first-order derivative of Y(t) at t = 0.

Using the equalities 1.8.2, 1.8.18, 1.8.19 of the book [4] in the form:

E1,1 [z] = E1 [z] = exp(z), (133)

1
E1,2 [z] = E1 [z] = (exp(z) − 1), (134)
z
we obtain:
lim Y1 (t) = Y(0)E1,1 [λ t] = Y(0) exp(λ t), (135)
t→1−

1 (1)
lim Y2 (t) = Y(0) exp(λ t) + Y (0)(exp(λ t) − 1). (136)
t→1+ λ

75
Mathematics 2019, 7, 554

The correct solution of growth Equation (130) for the case α = 1 is Equation (135). We can see that
in the limit the solutions (135) and (136) of the growth equation (Equation (130)) coincide only if the
derivative Y(1) (0) is equal to zero.

4.4. Solvability Principle: Examples from Numerical Simulation

Computer simulation of processes with memory and non-locality should use such methods for the
numerical solution of fractional differential equations with derivatives of non-integer order, which take
into account non-locality and memory. Numerical approximation should not use only local information.
Numerical scheme should contain a term of the memory (or non-locality). Numerical methods that
neglect non-locality and memory are not reliable and often lead to incorrect results, since the non-local
nature of fractional differential operators with non-integer orders cannot be neglected. Examples of
such errors are given in the work of Roberto Garrappa [160].

5. “Non-Equivalence” and “Unpredictability” of Fractional Generalization

5.1. Equivalence of Equations by Solutions (s-Equivalence)

Differential equations will be called equivalent by solution (s-equivalent) if these equations have
the same solutions for a sufficiently wide class of functions and initial conditions [29,161,162]. Let us
give some detalizations of this notion.
Let us consider two differential equations:

E1 x, f (x), f (1) (x), . . . , f (n) (x) = 0, (137)

E2 x, u(x), u(1) (x), . . . , u(n) (x) = 0. (138)

Differential Equations (137) and (138) will be called equivalent by solution (s-equivalent) if there exists
a certain function g : u(x) = g( f (x)) such that the solution of Equation (138), which is expressed
through the function f (x), coincides with the solution of Equation (137).
For simplicity, we consider the first-order ordinary differential equation:

d f (x)
= E1 [ f (x), x, λ], and f (0) = C1 (x ∈ R+ ), (139)
dx

where λ denotes a set of parameters, and E1 [ f (x), x, λ] is such that Equation (139) has a unique solution
for x ≥ 0 or x ∈ R. The solution of Equation (139) will be denoted as:

f (x) = S1 (x, λ, C1 ) (140)

with the initial condition:

S1 (0, λ, C1 ) = C1 . (141)

Let us consider the second differential equation:

du(x)
= E2 [u(x), x, λ], and u(0) = C2 (x ∈ R+ ), (142)
dx

where E2 [u(x), x, λ] is such that Equation (142) has a unique solution for x ≥ 0 or x ∈ R. The solution of
Equation (142) will be denoted as:
u(x) = S2 (x, λ, C2 ) (143)

with the condition:

S2 (0, λ, C2 ) = C2 . (144)

Let us give the concept an equivalence of the solutions of these two differential equations.

76
Mathematics 2019, 7, 554

Definition 1. Equations (139) and (142) are called equivalent by solution (s-equivalent) if there exists a function
g : u(x) = g( f (x)) such that:
S2 (x, λ, C2 ) = g(S1 (x, λ, C1 )), (145)

and C2 = g(c1 ) for a sufficiently wide class of functions and initial conditions.

An s-equivalence can be considered as a map of differential equation into another differential

equation such that the solutions of these equations are also transformed by the same map. In the next
subsections we give simple examples to illustrate this concept.

5.2. Relaxation and Growth Differential Equations

Let us consider the “relaxation” differential equation:

d f (x)
= −λ f (x) (x ∈ R+ ) (146)
dx
with λ > 0. The solution of Equation (146) has the form:

f (x) = f (0) exp (−λ x). (147)

Let us consider also the “growth” differential equation:

du(x)
= λ u(x) (x ∈ R+ ) (148)
dx
with λ > 0. The solution has the form:

u(x) = u(0) exp (λ x). (149)

We can consider the function u(x) = g(u f x)) in the form:

1
u(x) = . (150)
f (x)

In this case, using Equations (146) and (150), we can get Equation (148):
 
du(x) d 1 f (1) ( x ) λ f (x) 1
= =− 2 = 2 =λ = λ u(x), (151)
dx dx f (x) f (x) f (x) f (x)

and the solution (149) can be obtained from (147) by the obvious transformations:

1 1 1
u(x) = = = u(0) = u(0)exp (λ x). (152)
f (x) f (0) exp (−λ x) exp (−λ x)

Let us consider the “relaxation” fractional differential equation:

 
DαC,0+ f (t) = −λ f (x) (x ∈ R+ ) (153)

with λ > 0, where DαC,0+ is the Caputo fractional derivative of order 0 < α < 1. The solution has
the form:
f (x) = f (0) Eα [−λ xα ], (154)

where Eα [tα A] is the Mittag–Leffler function [32]. Let us consider the “growth” differential equation:
 
DαC,0+ u (x) = λ u(x) (x ∈ R+ ) (155)

77
Mathematics 2019, 7, 554

with λ > 0. The solution has the form:

u(x) = u(0)Eα [λ xα ]. (156)

These fractional differential equations are not equivalent:

1 1 1
u(x) = = = u(0)  u(0)Eα [λxα]. (157)
f (x) f (0) Eα [−λ xα ] Eα [−λ xα ]

As a result, the solutions of fractional differential Equation (155) with λ > 0 and λ < 0 cannot be
considered as s-equivalent equations.
The fact of violation of s-equivalence is caused by the violation of the standard chain rule,
which is used in (151), for fractional derivatives of non-integer orders. As a result, the derivative
(DαC,0+ f −1 )(x) cannot be represented through (DαC,0+ f )(x) in the simple form. From the point of view
of solutions, this nonequivalence is caused by the properties of the Mittag–Leffler functions and the
violation of group (and semi-group) property of dynamic maps. This allows us to formulate the
following statement:

Principle “Violation of s-Equivalence by Fractional Generalization”. The s-equivalence property of

differential equations of integer order is violated by formal fractional generalization of these equations. As a
result, the equivalence of dynamic models is violated by the fractional dynamic generalization.

Another example is given in the next subsection.

5.3. Fractional Logistic Equation: Growth in Competitive Environment with Memory

The logistic differential equation can be derived from economic model of natural growth in a
competitive environment. This model is described in Section 3.4.2 of this paper. Differential equation
that describes logistic growth in competitive environment without memory (110) has the form:

dY(t) m
= (b − a Y(t))Y(t). (158)
dt v
Equation (158) is the logistic differential equation, i.e., the ordinary differential equation of first
order that describes the logistic growth. For a = 0, Equation (158) describes the natural growth in the
absence of competition. If a  0 and b  0, we can use the variable f (t) and the parameter μ that are
defined by the expressions:
a m
f (t) = Y(t) and μ = . (159)
b v
Then Equation (158) of the logistic growth is represented in the form:

d f (t)
= μ f (t)(1 − f (t)). (160)
dt
This is the standard logistics differential equation. The solution of this logistic equation has the form:

f (0) f (0) exp(μ t)

f (t) = = . (161)
f (0) + (1 − f (0)) exp(−μ t) 1 + f (0) (exp(μ t) − 1)

Using the variable:

1
u(t) = , (162)
f (t)

78
Mathematics 2019, 7, 554

Equation (160) can be represented as:

du(t)
= μ (1 − u(t)). (163)
dt
Equations (160) and (163) are integer-order differential equations that are s-equivalent.
Let us consider the fractional generalization of these equations, which are represented in the form:
 
DαC,0+ f (t) = μ f (t)(1 − f (t)), (164)
 
DαC,0+ u (x) = μ (1 − u(t)).) (165)

Equations (164) and (165) cannot be considered as s-equivalent equations. It is well known that
the analytical expression for solution of fractional differential Equation (164) for the function f (t) is
still unknown at the moment. The solution of equation of linear Equation (165) for the function u(t) is
given in the form:
u(t) = 1 + (u(0) − 1)Eα [−μ tα ], (166)

where μ > 0. Therefore, we obtain:

1 f (0)
f (t) = = . (167)
u(t) f (0) + (1 − f (0)) Eα [−μ tα ]

For α = 1, using that E1 [−μ t] = exp(−μ t), Equation (167) is the standard solution (161).
As a result, we can see that Equations (164) and (165) have different solutions and the fractional
generalization violates the s-equivalence of differential equations. The fractional generalizations of
equivalent models can give non-equivalent fractional dynamic models.

5.4. Fractional Generalization Generates Nonequivalent Models

In Sections 6.2 and 6.3, we prove that fractional generalizations of equivalent representations
of standard dynamic models, which are described by s-equivalent differential equations, as a rule,
lead to different fractional dynamic models that have non-equivalent solutions. This, in a sense,
is analogous to the situation in quantum theory when quantization of equivalent classical models leads
to nonequivalent quantum theories.
As a result, we can formulate the following statement:

Principle “Non-Equivalence of Equivalent”. Fractional generalizations of s-equivalent differential

equations of integer order are not equivalent in general.

This property of fractional generalization is caused by the violation of the standard rule (the chain
rule and other rules) for fractional derivatives of non-integer order. This non-equivalence of equations
in natural and social sciences generates uncertainty in the description of the processes. Note that
an additional unpredictability of fractional generalizations creates the presence of a large number
of different types of fractional derivatives and integrals. This fact of mathematical non-equivalence
allows us to formulate the following principle:

Principle “Unpredictability of Fractional Generalization”. A fractional generalization of one standard

model (which is represented by s-equivalent differential equations of integer order) can lead to different
fractional-dynamic models that will predict different behaviors of a process.

Due to this, the correct and self-consistent derivation of fractional differential equations and the
economic justification of existence of memory (or nonlocality) for one or another endogenous variable,
are of fundamental importance.

79
Mathematics 2019, 7, 554

As a result, the importance of the Derivability Principle and Interpretability Principle, which are
proposed at the beginning of this work, increases substantially.

6. Example of Application of the Interpretability Principle: Effects, Phenomena, and Principles

In this section, some effects and phenomena are considered only for illustration and explanation
of the interpretability principle. As an example, we give an economic interpretation of the solutions
of fractional differential equations, which describe the fractional generalization of the standard
Harrod–Domar growth model.

6.1. Economic Model with Memory: Equation, Solution, and Asymptotic Behavior
One of the simple models of economic growth was proposed by Roy Harrod [163] and Evsey
Domar [164,165] in 1946–1947. The fractional generalization of the standard Harrod–Domar growth
models was proposed in papers [166,167] in 2016 (see also [47,54,144,145]).
Let us consider the fractional generalization of the standard Harrod–Domar growth model, which
is described in Allen’s book [78], (pp. 64–65). The fractional generalization of this model was proposed
in section 3 of [54]. The fractional differential equation, which describes the Harrod–Domar model
with power-law memory, has the form:
 
DαC,0+ Y (t) = λ Y(t) + C(t), (168)

where the function Y(t) describes the national income; C(t) = −v−1 A(t) is the exogenous variable that is
independent of the national income Y(t); the function A(t) is the autonomous investment; the parameter
s ∈ (0, 1) is the marginal propensity to save; v > 0 is the investment coefficient indicating the power
of accelerator; B = v/s describes the capital intensity of the national
 income; λ = B−1 = s/v [115].
α ( )
In Equation (168), we use the Caputo fractional derivative DC,0+ Y t of the order 0 < α < 2. This order
of the Caputo fractional derivatives is interpreted as the memory fading parameter [47]. The absence of
memory corresponds to the positive integer values of α. For α = 1, Equation (168) gives the differential
equation of the first order that describes the standard Harrod–Domar model.
Equation (168) has the solution (Theorem 5.15 of [4], p. 323) in the form:


n−1  t 
Y (t) = Y(k) (0)tk Eα,k+1 [λ tα ] + (t − τ)α−1 Eα,α λ(t − τ)α C(τ)dτ, (169)
k =0 0

where Y(k) (0) is integer-order derivatives of the orders k ≥ 0 at t = 0, and Eα,β [z] is the two-parameter
Mittag–Leffler function [32].
In the case 0 < α ≤ 1 (n = 1) Equation (169) takes the form:
 t 
Y(t) = Y(0)Eα,1 [λ tα ] + (t − τ)α−1 Eα,α λ (t − τ)α C(τ)dτ. (170)
0

For 1 < α ≤ 2 (n = 2) Equation (169) gives:

t 
Y(t) = Y(0) Eα,1 [λ t ] + Y α (1)
(0) t Eα,2 [λ t ] + α
(t − τ)α−1 Eα,α λ (t − τ)α C(τ)dτ, (171)
0

where Y(1) (0) is first-order derivative of Y(t) at t = 0.

80
Mathematics 2019, 7, 554

The asymptotic behavior of solution (169) of the fractional differential Equation (168) with λ > 0
and C(t) = 0 at t → ∞ is described by the expression:
⎛ ⎞
   ⎟⎟
 λ−k/α ⎜⎜⎜⎜ Y(k) (0)λ− j k−αj
n−1 n−1 m
1 ⎟⎟
Y(t) = exp λ1/α t Y (k ) ( 0 ) + ⎜⎜ t +O ⎟⎟, (172)
α ⎝ Γ (k + 1 − α j) tα (m+1)−k ⎠
k =0 k =0 j=1

where n − 1 < α < n. For the non-integer values of the fading

 parameter
 0 < α < 2 the behavior of the
national income Y(t) is determined by the term with exp λ1/α t . The power-law terms tk−αj of (172)
do not determine the dominant behavior at t → ∞ .

6.2. Interpretation: Warranted Rate of Growth with Memory

An important economic concept of growth models is the technological growth rate [115], p. 49,
which is also called the Harrod’s warranted rate of growth [78], p. 67, of endogenous variables
(for example, national income). The technological (warranted) growth rate describes the growth rate in
the case of the constant structure of the economy and the absence of external influences. The constant
structure means that the parameters of the model do not change over time (for example, s, v are
constants). The absence of external influences means the absence of exogenous variables (C(t) = 0).
Mathematically, the technological growth rate is described by the asymptotic behavior of the solution
of homogeneous differential equations for the economic model.
In the standard Harrod–Domar model, the solution of Equation (168) with α = 1 and C(t) = 0
has the form Y(t) = Y(0) exp(λ t). Therefore, the technological growth rate of this model is described
by the value λ = s/v. The capital intensity of the national income B = λ−1 = v/s is the characteristic
time τ = B = λ−1 of growth without memory.
Using Equation (172) of the asymptotic behavior of the solution, we can formulate new economic
concept, which can be called the warranted (technological) rate of growth with memory [144–168].
This concept allows us to characterize the processes of economic growth with memory not only in the
fractional generalizations of the standard Harrod–Domar model, but also for a wide range of other
models described by fractional differential equations [144–146,168].
The warranted (technological) rate of growth with memory is defined by the equation:

λ(α) = λ1/α = B−1/α . (173)

Note that for parameter α = 1 this growth rate is equal to the standard warranted rate of growth
without memory, λe f f (1) = λ. It can be seen that the warranted (technological) rates of growth
(Equation (173)) with one-parametric memory do not coincide with the growth rates λ = B−1 of
standard models without memory (α = 1).
As a result, we can formulate [144–146,168] the following principles, which gives an economic
interpretation of obtained mathematical results:

Principle of Changing of the Warranted Growth Rate by Memory. The power-law memory with the
non-integer fading parameter 0 < α < 2 change of the warranted rate of growth with memory according to
the equation:
λ(α) = λ1/α , (174)

where λ = s/v is the warranted growth rate without memory (α = 1) for the same values of other parameters.

Using this concept and principle, we can give examples of the economic interpretation of solution
(169) of fractional differential Equation (168) that describes the fractional generalization of the standard
Harrod–Domar model.
Let us consider two phenomena that follow from the suggested principle of changing of the
warranted growth rate by memory for the case 0 < α < 1.

81
Mathematics 2019, 7, 554

Acceleration Phenomenon. The memory effect with α ∈ (0, 1) increases the warranted growth rate of
economic processes if growth rate of the processes without memory is λ > 1. In the case λ > 1, the memory effect
can increase the growth rate by many orders of magnitude.

Let us give a numerical example of acceleration phenomenon. For α = 0.1 and λ1 = 10 > 1,
the warranted rate of growth with memory (α = 0.1) is equal to λ1 (0.1) = 10000000000 instead of
λ1 (1) = λ1 = 10 for process without memory (α = 1), that is, the memory effect can increase the
growth rate by nine orders of magnitude.

Slowdown Phenomenon. The memory effect with α ∈ (0, 1) decreases the warranted growth rate of economic
processes if growth rate of the processes without memory is small 0 < λ < 1. In the case 0 < λ < 1, the memory
effect can decrease the growth rate by many orders of magnitude.

Let us give a numerical example of slowdown phenomenon. For α = 0.1 and λ2 = 0.1 < 1,
the warranted rate of growth with memory (α = 0.1) is equal to λ2 (0.1) = 0.00000000001 instead of
λ2 (1) = λ2 = 0.1 for the process without memory (α = 1), that is, the memory effect can decrease the
growth rate by nine orders of magnitude.
As a result, these examples demonstrate that the memory effect can significantly change the
warranted growth rate by many orders of magnitude for the case 0 < α < 1.
Note that the concept of the “warranted characteristic times” of processes with memory is proposed
in [168] for processes of growth (λ > 0), which can be called the amplification, and for processes of
relaxation (λ < 0).

6.3. Interpretation: Growth and Decline with Memory

Let us consider an economic interpretation of solution (169) of the fractional differential Equation
(168) with λ > 0 for the case of constant autonomous investment A(t) = const, i.e., C(t) = C = const.
This solution for n − 1 < α < n has the form:

1  n−1
Y (t) = C (Eα,1 [λ tα ] − 1) + Y(k) (0)tk Eα,k+1 [λ tα ]. (175)
λ
k =0

For 0 < α < 1 the behavior of solution (175) with λ > 0 at t → ∞ is described by the equation:
   
Y(t) = − λ1 C + α1 Y(0) + λ1 C exp λ1/α t −
⎛ ⎞
 ⎜⎜ m  ⎟⎟ (176)
λ− j
Y(0) + λ1 C ⎜⎜⎝ Γ(1−αj)
+ O α (
1
m + 1 )
⎟⎟.
⎠
j=1 t

For 1 < α < 2 the behavior of solution (175) with λ > 0 at t → ∞ has the form:
   
Y(t) = − λ1 C + α1 Y(0) + λ1 C + λ−1/α Y(1) (0) exp λ1/α t −

m    
Y (1) (0) t −j (177)
Y(0) + λ1 C + 1−α j Γ(λ1−αj) t−αj + O α(m+1 1)−1 .
j=1 t

For processes without memory (α = 1), we have:

 
1 1
Y (t) = − C + Y(0) + C exp(λt). (178)
λ λ

Using Equation (177), we can formulate conditions of growth and decline for economic processes
with memory that is described by solution (175) with λ > 0 and the memory fading parameter α ∈ (0, 2).

82
Mathematics 2019, 7, 554

For the case α ∈ (0, 1) and α = 1, the condition of growth with memory is represented by the inequality:

1
Y (0) + C > 0. (179)
λ

For the case α ∈ (0, 1) and α = 1, the condition of decline with memory is represented by the inequality:

1
Y (0) + C < 0. (180)
λ
We see that conditions (179) and (180) do not depend on the value of the memory fading parameter
α ∈ (0, 1). As a result, the conditions of growth and decline with memory have the same form as
for process without memory (α = 1). However, the growth rates of the processes with memory and
without memory may differ greatly.
Let us describe the conditions of growth and decline for the case 1 < α < 2.
For the case α ∈ (1, 2), the condition of growth with memory is given by the inequality:

1
Y (0) + C + λ−1/α Y(1) (0) > 0. (181)
λ

For the case α ∈ (1, 2), the condition of decline with memory is represented by the inequality:

1
Y (0) + C + λ−1/α Y(1) (0) < 0. (182)
λ
From inequalities (181) and (182), we see that these conditions of the growth and decline for
economy are determined not only by the initial conditions, but also by the memory fading parameter
α ∈ (1, 2), if Y(1) (0)  0.
Let us consider some special cases. If the condition (181) is satisfied, the effects of memory with
α ∈ (1, 2), can lead to faster growth, i.e., to increase of the warranted growth rate, if 0 < λ < 1, since
λ(α) > λ. The effects of memory with α ∈ (1, 2), can lead to a slowing of the decline, i.e., to decrease of
the warranted growth rate, if 0 < λ < 1, since λ(α) < λ.
Let us also note an important special case, when we have a decline for process without memory
and a growth for process with memory for the same other parameters [145,146,168].
For example, we can consider the case C(t) = −v−1 A < 0, λ > 0 and α ∈ (1, 2).

Phenomenon of Replacing Economic Decline with Growth. The process without memory shows decline,
while the process with memory (at the same other parameters) demonstrates a growth, if the inequalities:

1
Y (0) < |C| < Y(0) + λ−1/α Y(1) (0) (183)
λ

holds for C < 0, Y(1) (0) > 0, λ > 0 and 1 < α < 2.

Condition (183) means that decline is replaced by the growth, when the memory effect is taken into
account [145,146,168]. As a result, memory effects can change the decline by growth. In processes with
memory with α ∈ (1, 2), we can have a growth instead of decline, when the other process parameter is
unchanged. The decline of economic processes can be replaced by the growth, when the memory effect
is taken into account [145,146,168].
As an example, the mathematical results, which is represented by the solution of the fractional
differential equation allows us to give the following economic interpretation of these results for the
case λ ∈ (0, 1) and Y(1) (0) > 0.

Phenomenon of Amplification of Economy by Memory. For small values of warranted growth rates
λ ∈ (0, 1) and Y(1) (0) > 0, the effects of memory with α ∈ (1, 2) positively affect the economy, and lead to an
improvement in economic dynamics. In other words, for the case λ ∈ (0, 1) and Y(1) (0) > 0 effects of memory

83
Mathematics 2019, 7, 554

with α ∈ (1, 2) lead to positive results, such as a slowdown in the rate of decline, a replacement of the economic
decline by its growth, an increase in the rate of economic growth.

Finding qualitative differences in the behavior of the processes described by generalizations of

standard models is an important part of building fractional dynamic generalizations. The phenomenon
of replacing economic decline with growth demonstrates the qualitative difference (from an economic
point of view) of the behavior of the standard model from the fractional dynamic model.

6.4. Interpretation: Relaxation with Memory

The fractional differential equation with λ < 0 describes a relaxation process. As an example of
relaxation processes with memory, we can consider the dynamics of fixed assets (or capital stock),
where we take into account the memory effects [169,170]. Let us assume that the retirement of capital
occurs with a constant retirement rate of 0 < b < 1, where the parameter b > 0 can also be interpreted
as a coefficient of disposal of fixed assets. Let us assume that the investment is equal to I = const
monetary units. In the standard model the dynamics of fixed assets (or capital stock) without memory
and lag, the rate of change of the fixed assets is equal to the difference between investments and disposal
of fixed assets. Let us denote the fixed assets (or capital stock) at time t ≥ 0 by K(t). The fractional
generalization of this standard model, which is proposed in [169], describes the dynamics of the fixed
assets the fixed assets (the capital stock) with power-law memory by the fractional differential equation:
 
DαC,0+ K (t) = I − b K(t), (184)

where DαC,0+ is the Caputo derivative [4]. For α = 1, Equation (184) takes the form:

dK(t)
= I − b K (t), (185)
dt
which is equation of the standard dynamic model of fixed assets [104], p. 82, without memory and lag.
Equation (185) describes the relaxation to the equilibrium state K = I/b. The solution of fractional
differential Equation (184) can written in the form of Equations (169) and (175), where λ = −b < 0
and C = I = const. This solution describes a generalized relaxation processes since λ < 0. Note
that asymptotic behavior of the solutions cannot be represented in the form of Equations (172), (176),
and (177) since the asymptotic expressions for cases λ < 0 and λ > 0 have different forms. For economic
interpretation of solution (175) with λ < 0, we can use the fractional relaxation-oscillation phenomenon
that was proposed by Francesco Mainardi [171] in 1996. The detailed description of this phenomenon
is given in the works [172–174]. Some aspects of this interpretation are described in section 2.4 of [168],
where the concept of the “warranted” characteristic times of processes with memory (the “warranted”
relaxations times of processes with memory if λ < 0) has been proposed. Note that, in contrast to the
growth (amplification) processes (λ > 0), for relaxation processes (λ < 0) it is necessary to consider two
types of “warranted” characteristic times describing oscillations and damping (for details, see [168]).
Note that generalization of the fractional relaxation-oscillation phenomenon for distributed order
is described in [175–177].

7. Conclusions
In this paper we formulated some principles and rules that are important for constructing fractional
generalizations of standard dynamic models that are described by differential equations of integer
order. These rules emphasize the importance of taking into account the non-standard properties
of fractional derivatives of non-integer order. The violation of the standard form of the chain rule,
the semi-group rule for orders of derivatives, the product (Leibniz) rule, the semi-group property of
dynamic maps should be considered as the most important part of the mathematical tools designed to
describe non-locality and memory.

84
Mathematics 2019, 7, 554

The proposed principles of fractional generalization are illustrated by examples from economics.
We note that a brief review of the history of applications of fractional calculus in modern mathematical
economics and economic theory is proposed in [149].
We also think that these principles are general and can be applied to construct fractional
generalizations of standard models in mechanics, physics, biology, and other sciences. It is hoped
that various works will soon appear in which these principles will be illustrated with examples from
natural and social sciences.

Funding: This research received no external funding.

Conflicts of Interest: The author declares no conflict of interest.

References
1. Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives Theory and Applications; Gordon
and Breach: New York, NY, USA, 1993; 1006p, ISBN 9782881248641.
2. Kiryakova, V. Generalized Fractional Calculus and Applications; Longman and J. Wiley: New York, NY, USA,
1994; 360p, ISBN 9780582219779.
3. Podlubny, I. Fractional Differential Equations; Academic Press: San Diego, CA, USA, 1998; 340p.
4. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier:
Amsterdam, The Netherlands, 2006; 540p, ISBN 9780444518323.
5. Diethelm, K. The Analysis of Fractional Differential Equations: An Application-Oriented Exposition Using Differential
Operators of Caputo Type; Springer: Berlin, Germany, 2010; 247p. [CrossRef]
6. Letnikov, A.V. On the historical development of the theory of differentiation with arbitrary index. Sb. Math.
Mat. Sb. 1868, 3, 85–112. (In Russian)
7. Ross, B. A brief history and exposition of the fundamental theory of fractional calculus. In Fractional Calculus and
Its Applications, Proceedings of the International Conference Held at the University of New Haven, 15-16 June 1974; Series:
Lecture Notes in Mathematics; Springer: Berlin/Heidelberg, Germany, 1975; Volume 457, pp. 1–36. [CrossRef]
8. Ross, B. The development of fractional calculus 1695–1900. Hist. Math. 1977, 4, 75–89. [CrossRef]
9. Ross, B. Fractional Calculus. Math. Mag. 1977, 50, 115–122. [CrossRef]
10. Kiryakova, V. A brief story about the operators of the generalized fractional calculus. Fract. Calc. Appl. Anal.
2008, 11, 203–220.
11. Machado, J.T.; Kiryakova, V.; Mainardi, F. Recent history of fractional calculus. Commun. Nonlinear Sci.
Numer. Simul. 2011, 16, 1140–1153. [CrossRef]
12. Machado, J.A.T.; Galhano, A.M.; Trujillo, J.J. Science metrics on fractional calculus development since 1966.
Fract. Calc. Appl. Anal. 2013, 16, 479–500.
13. Machado, J.A.T.; Galhano, A.M.; Trujillo, J.J. On development of fractional calculus during the last fifty years.
Scientometrics 2014, 98, 577–582. [CrossRef]
14. Machado, J.T.; Kiryakova, V. The chronicles of fractional calculus. Fract. Calc. Appl. Anal. 2017, 20, 307–336.
[CrossRef]
15. Valerio, D.; Machado, J.T.; Kiryakova, V. Some pioneers of the applications of fractional calculus. Fract. Calc.
Appl. Anal. 2014, 17, 552–578. [CrossRef]
16. Machado, J.A.T. (Ed.) Handbook of Fractional Calculus with Applications; De Gruyter: Berlin, Germany; Boston, MA,
USA, 2019; Volumes 1–8, Available online: https://www.degruyter.com/view/serial/498377 (accessed on 9 May 2019).
17. Tarasov, V.E. No violation of the Leibniz rule. No fractional derivative. Commun. Nonlinear Sci. Numer. Simul.
2013, 18, 2945–2948. [CrossRef]
18. Ortigueira, M.D.; Machado, J.A.T. What is a fractional derivative? J. Comput. Phys. 2015, 293, 4–13. [CrossRef]
19. Tarasov, V.E. On chain rule for fractional derivatives. Commun. Nonlinear Sci. Numer. Simul. 2016, 30, 1–4.
[CrossRef]
20. Tarasov, V.E. Leibniz rule and fractional derivatives of power functions. J. Comput. Nonlinear Dyn. 2016, 11.
[CrossRef]
21. Tarasov, V.E. No nonlocality. No fractional derivative. Commun. Nonlinear Sci. Numer. Simul. 2018, 62, 157–163.
[CrossRef]

85
Mathematics 2019, 7, 554

22. Sayevand, K.; Tenreiro Machado, J.; Baleanu, D. A new glance on the Leibniz rule for fractional derivatives.
Commun. Nonlinear Sci. Numer. Simul. 2018, 62, 244–249. [CrossRef]
23. Hernandez Encinas, L.; Munoz Masque, J. A short proof of the generalized Faá di Bruno’s formula. Appl. Math.
Lett. 2003, 16, 975–979. [CrossRef]
24. Vilenkin, N.Y.; Gorin, E.A.; Kostyuchenko, A.G.; Krasnosel’skii, M.A.; Krein, S.G.; Maslov, V.P.; Mityagin, B.S.;
Petunin, Y.I.; Rutitskii, Y.B.; Sobolev, V.I.; et al. Functional Analysis; Vilenkin, N.Y., Gorin, E.A.,
Kostyuchenko, A.G., Krasnosel’skii, M.A., Krein, S.G., Eds.; Wolters-Noordhoff Publishing: Groningen,
The Netherlands, 1972; 380p, ISBN 90-01-90980-9.
25. Arnold, V.I. Ordinary Differential Equations; Springer: Berlin/Heidelberg, Germany; New York, NY, USA, 1992;
338p, ISBN 3-540-54813-0.
26. Gorini, V.; Kossakowski, A.; Sudarshan, E.C.G. Completely positive dynamical semigroups of N-level
systems. J. Math. Phys. 1976, 17, 821–825. [CrossRef]
27. Gorini, V.; Frigerio, A.; Verri, M.; Kossakowski, A.; Sudarshan, E.C.G. Properties of quantum Markovian
master equations. Rep. Math. Phys. 1978, 13, 149–173. [CrossRef]
28. Lindblad, G. On the generators of quantum dynamical semigroups. Commun. Math. Phys. 1979, 48, 119–130.
[CrossRef]
29. Tarasov, V.E. Quantum Mechanics of Non-Hamiltonian and Dissipative Systems; Elsevier Science: Amsterdam,
The Netherlands, 2008; 540p.
30. Daftardar-Gejji, V.; Babakhani, A. Analysis of a system of fractional differential equations. J. Math. Anal.
Appl. 2004, 293, 511–522. [CrossRef]
31. Daftardar-Gejji, V.; Jafari, H. Adomian decomposition: A tool for solving a system of fractional differential
equations. J. Math. Anal. Appl. 2005, 301, 508–518. [CrossRef]
32. Gorenflo, R.; Kilbas, A.A.; Mainardi, F.; Rogosin, S.V. Mittag-Leffler Functions, Related Topics and Applications;
Springer: Berlin, Germany, 2014; 443p.
33. Peng, J.; Li, K. A note on property of the Mittag-Leffler function. J. Math. Anal. Appl. 2010, 370, 635–638. [CrossRef]
34. Tarasov, V.E. Quantum dissipation from power-law memory. Ann. Phys. 2012, 327, 1719–1729. [CrossRef]
35. Elagan, S.K. On the invalidity of semigroup property for the Mittag-Leffler function with two parameters.
J. Egypt. Math. Soc. 2016, 24, 200–203. [CrossRef]
36. Sadeghi, A.; Cardoso, J.R. Some notes on properties of the matrix Mittag-Leffler function. Appl. Math.
Comput. 2018, 338, 733–738. [CrossRef]
37. Tarasov, V.E. Fractional Quantum Mechanics of Open Quantum Systems. Chapter 11. In Handbook of
Fractional Calculus with Applications; Applications in Physics, Part, B; Tarasov, V.E., Ed.; De Gruyter: Berlin,
Germany; Boston, MA, USA, 2019; Volume 5, pp. 257–277. [CrossRef]
38. Tarasov, V.E. Interpretation of fractional derivatives as reconstruction from sequence of integer derivatives.
Fundam. Inform. 2017, 151, 431–442. [CrossRef]
39. Podlubny, I. Geometrical and physical interpretation of fractional integration and fractional differentiation.
Fract. Calc. Appl. Anal. 2002, 5, 367–386.
40. Tarasova, V.V.; Tarasov, V.E. Economic interpretation of fractional derivatives. Prog. Fract. Differ. Appl. 2017, 3, 1–7.
[CrossRef]
41. Tarasova, V.V.; Tarasov, V.E. Microeconomic meaning of derivatives of non-integer order. Sci. Educ. Today
2017, 8, 32–39. (In Russian)
42. Rehman, H.U.; Darus, M.; Salah, J. A note on Caputo’s derivative operator interpretation in economy. J. Appl.
Math. 2018, 2018, 1260240. [CrossRef]
43. Tarasov, V.E.; Tarasova, S.S. Probabilistic interpretation of Kober fractional integral of non-integer order.
Prog. Fract. Differ. Appl. 2019, 5, 1–5. [CrossRef]
44. Tarasov, V.E. Caputo-Fabrizio operator in terms of integer derivatives: Memory or distributed lag? Comput.
Appl. Math. 2019, 38, 113. [CrossRef]
45. Mathai, A.M.; Haubold, H.J. Erdelyi-Kober fractional integral operators from a statistical perspective—I. Tbil.
Math. J. 2017, 10, 145–159. [CrossRef]
46. Tarasov, V.E.; Tarasova, S.S. Fractional and integer derivatives with continuously distributed lag.
Commun. Nonlinear Sci. Numer. Simul. 2019, 70, 125–169. [CrossRef]
47. Tarasova, V.V.; Tarasov, V.E. Concept of dynamic memory in economics. Commun. Nonlinear Sci. Numer.
Simul. 2018, 55, 127–145. [CrossRef]

86
Mathematics 2019, 7, 554

48. Tarasova, V.V.; Tarasov, V.E. Notion of dynamic memory in economic theory. J. Econ. Entrep. 2017, 6, 868–880.
(In Russian)
49. Tarasov, V.E. Generalized memory: Fractional calculus approach. Fractal Fract. 2018, 2, 23. [CrossRef]
50. Tarasov, V.E.; Tarasova, V.V. Criterion of existence of power-law memory for economic processes. Entropy
2018, 20, 414. [CrossRef]
51. Tarasov, V.E.; Trujillo, J.J. Fractional power-law spatial dispersion in electrodynamics. Ann. Phys. 2013, 334, 1–23.
[CrossRef]
52. Tarasov, V.E.; Aifantis, E.C. Non-standard extensions of gradient elasticity: Fractional non-locality, memory
and fractality. Commun. Nonlinear Sci. Numer. Simul. 2015, 22, 197–227. [CrossRef]
53. Tarasov, V.E.; Tarasova, V.V. Dynamic Keynesian model of economic growth with memory and lag. Mathematics
2019, 7, 178. [CrossRef]
54. Tarasov, V.E.; Tarasova, V.V. Harrod-Domar growth model with memory and distributed lag. Axioms 2019, 8, 9.
[CrossRef]
55. Tarasov, V.E.; Tarasova, V.V. Phillips model with exponentially distributed lag and power-law memory.
Comput. Appl. Math. 2019, 38. [CrossRef]
56. Nakhushev, A.M. On the positivity of continuous and discrete differentiation and integration operators that are
very important in fractional calculus and in the theory of equations of mixed type. Differ. Equ. 1998, 34, 103–112.
57. Nakhushev, A.M. Fractional Calculus and its Application; Fizmatlit: Moscow, Russia, 2003; 272p, ISBN
5-9221-0440-3. (In Russian)
58. Pskhu, A.V. On the theory of the continual integro-differentiation operator. Differ. Equ. 2004, 40, 128–136. [CrossRef]
59. Pskhu, A.V. Partial Differential Equations of Fractional Order; Nauka: Moscow, Russia, 2005; 199p,
ISBN 5-02-033721-8. (In Russian)
60. Tarasov, V.E.; Tarasova, V.V. Accelerator and multiplier for macroeconomic processes with memory. IRA-Int.
J. Manag. Soc. Sci. 2017, 9, 86–125. [CrossRef]
61. Tarasov, V.E. (Ed.) Handbook of Fractional Calculus with Applications. Vol. 4. Applications in Physics, Part A; De
Gruyter: Berlin, Germany; Boston, MA, USA, 2019; 306p, ISBN 978-3-11-057088-5.
62. Tarasov, V.E. (Ed.) Handbook of Fractional Calculus with Applications. Volumes 5. Applications in Physics, Part B;
De Gruyter: Berlin, Germany; Boston, MA, USA, 2019; 319p, ISBN 978-3-11-057089-2.
63. Tarasov, V.E. Fractional calculus and long-range interactions. In Handbook of Fractional Calculus with
Applications; Applications in Physics. Part A; Tarasov, V.E., Ed.; De Gruyter: Berlin, Germany; Boston, MA,
USA, 2019; Volume 4, pp. 75–102. [CrossRef]
64. Tarasov, V.E. Fractional electrodynamics with spatial dispersion. In Handbook of Fractional Calculus with
Applications; Applications in Physics. Part B; Tarasov, V.E., Ed.; De Gruyter: Berlin, Germany; Boston, MA,
USA, 2019; Volume 5, pp. 25–52. [CrossRef]
65. Tarasov, V.E.; Tarasova, V.V. Logistic equation with continuously distributed lag and application in economics.
Nonlinear Dyn. 2019. [CrossRef]
66. Kaldor, N. A model of the trade cycle. Econ. J. 1940, 50, 78–92. [CrossRef]
67. Kaldor, N. Capital accumulation and economic growth. In The Theory of Capital; Hague, D.C., Ed.; Palgrave
Macmillan: London, UK, 1961; Chapter 10; pp. 177–222. Available online: http://www.masterhdfs.org/
masterHDFS/wp-content/uploads/2014/05/Kaldor_1961.pdf (accessed on 9 May 2019). [CrossRef]
68. Chang, W.W.; Smyth, D.J. The existence and persistence of cycles in a nonlinear model: Kaldor’s 1940 model
re-examined. Rev. Econ. Stud. 1971, 38, 37–44. [CrossRef]
69. Chian, A.C.-L.; Rempel, E.L.; Rogers, C. Complex economic dynamics: Chaotic saddle, crisis and intermittency.
Chaos Solitons Fract. 2006, 29, 1194–1218. [CrossRef]
70. Chian, A.C.-L. Complex Systems Approach to Economic Dynamics; Springer: Berlin/Heidelberg, Germany;
New York, NY, USA, 2007; 101p, ISBN 978-3-540-39752-6.
71. Gabisch, G.; Lorenz, H.W. Business Cycle Theory: A Survey of Methods and Concepts; Springer: Berlin/Heidelberg,
Germany; New York, NY, USA, 1987; 229p. [CrossRef]
72. Gabisch, G.; Lorenz, H.W. Business Cycle Theory: A Survey of Methods and Concepts; Second, Revised, and
Enlarged Edition; Springer: Berlin/Heidelberg, Germany, 1989; 248p.
73. Lorenz, H.W. Nonlinear Dynamical Economics and Chaotic Motion; Springer: Berlin/Heidelberg, Germany, 1989;
248p. [CrossRef]

87
Mathematics 2019, 7, 554

74. Lorenz, H.W. Nonlinear Dynamical Economics and Chaotic Motion; Second, Revised and Enlarged Edition;
Springer: Berlin/Heidelberg, Germany; New York, NY, USA, 1993; 319p. [CrossRef]
75. Hori, H. Business Cycle Dynamics and Stabilization Policies. A Keynesian Approach; Springer: Singapore, 2017;
142p, ISSN 2197-8859. [CrossRef]
76. Goodwin, R.M. The nonlinear accelerator and the persistence of business cycles. Econometrica 1951, 19, 1–17.
[CrossRef]
77. Matsumoto, A.; Szidarovszky, F. Nonlinear multiplier-accelerator model with investment and consumption
delays. Struct. Chang. Econ. Dyn. 2015, 33, 1–9. [CrossRef]
78. Allen, R.G.D. Mathematical Economics, 2nd ed.; First Edition 1956; Macmillan: London, UK, 1960; 812p,
ISBN 978-1-349-81547-0. [CrossRef]
79. Allen, R.G.D. Mathematical Economics, 2nd ed.; Macmillan: London, UK, 1960; 812p, ISBN 978-1-349-81547-0.
80. Barвosa, R.S.; Macнado, J.A.T.; Vinagre, B.M.; Calder, A.J. Analysis of the Van der Pol oscillator containing
derivatives of fractional order. J. Vib. Control 2007, 13, 1291–1301. [CrossRef]
81. Machado, J.A.T.; Lopes, A.M. Chapter 1. Fractional Van der Pol oscillator. In Handbook of Fractional Calculus
with Applications; Applications in Physics, Part A; Tarasov, V.E., Ed.; De Gruyter: Berlin, Germany; Boston,
MA, USA, 2019; Volumes 4, ISBN 978-3-11-057088-5.
82. Chen, J.H.; Chen, W.C. Chaotic dynamics of the fractionally damped van der Pol equation. Chaos Solitons
Fract. 2008, 35, 188–198. [CrossRef]
83. Lei, H.; Li, Y.; Tang, P. Numerical scheme and dynamic analysis for variable-order fractional van der Pol
model of nonlinear economic cycle. Adv. Differ. Equ. 2016, 2016. [CrossRef]
84. Lin, Z.; Li, J.; Li, S. On a business cycle model with fractional derivative under narrow-band random
excitation. Chaos Solitons Fract. 2016, 87, 61–70. [CrossRef]
85. Slutsky, Е.Е. Sulla teoria del bilancio del consumatore. G. Degli Econ. 1915, 51, 1–26.
86. Slutsky, E.E. On the theory of the budget of the consumer. Giornale degli Economisti e Annali di Economia.
Nuova Serie. Vol. 71 (Anno 125). No. 2/3. Special Issue: Classics from the “Giornale Degli Economisti” In
English Translation (December 2012), pp. 173–200. Available online: https://www.jstor.org/stable/43828058
(accessed on 9 May 2019).
87. Slutsky, E.E. On the theory of the budget of the consumer. Bull. Mosc. Univ. Ser. 6 Econ. 2016, 4, 86–111.
(In Russian)
88. Slutsky, E.E. On the theory of the budget of the consumer. In Economic-Mathematical Methods. National Models:
Theoretical Problems of Consumption; Publishing House of the Academy of Sciences of the USSR: Moscow,
Russia, 1963; pp. 241–271. (In Russian)
89. Slutsky, E.E. Economic and Statistical Works: Favorites; Translation of P. N. Klyukin; Eksmo: Moscow, Russia,
2010; pp. 448–485. (In Russian)
90. Nicholson, W.; Snyder, C. Microeconomic Theory. Basic Principles and Extensions, 11th ed.; Cengage Learning:
Masson, OH, USA, 2012; 758p, ISBN 978-111-1-52553-8.
91. Varian, H. Microeconomic Analysis, 3rd ed.; W.W. Norton: New York, NY, USA, 1992; 548p, ISBN 0-393-95735-7.
92. Lancaster, K. Mathematical Economics; The Macmillan Company: New York, NY, USA, 1968; ISBN 978-0023221316.
(Revised edition. Dover Publications: Mineola, NY, USA, 2011; 448p. ISBN 978-0486653914).
93. Tarasova, V.V.; Tarasov, V.E. Elasticity for economic processes with memory: Fractional differential calculus
approach. Fract. Differ. Calc. 2016, 6, 219–232. [CrossRef]
94. Tarasova, V.V.; Tarasov, V.E. Price elasticity of demand with memory. Econ. Sociol. Law 2016, 4-1, 98–106.
(In Russian)
95. Tarasova, V.V.; Tarasov, V.E. Elasticity of OTC cash turnover of currency market of Russian Federation.
Actual Probl. Humanit. Nat. Sci. 2016, 7-1, 207–215. (In Russian)
96. Tarasova, V.V.; Tarasov, V.E. On applicability of point elasticity of demand on price to exchange trading on
US dollar. Sci. Perspect. 2016, 6, 6–11. (In Russian)
97. Phillips, A.W. Stabilisation policy in a closed economy. Econ. J. 1954, 64, 290–323. [CrossRef]
98. Phillips, A.W. A. W. H. Phillips Collected Works in Contemporary Perspective; Leeson, R., Ed.; Cambridge
University Press: Cambridge, UK, 2000; 515p, p. 1357. ISBN 9780521571357.
99. Allen, R.G.D. Macro-Economic Theory. A Mathematical Treatment; Macmillan: London, UK, 1968; 420p,
ISBN 978-033304112.
100. Solow, R.M. A contribution to the theory of economic growth. Q. J. Econ. 1956, 70, 65–94. [CrossRef]

88
Mathematics 2019, 7, 554

101. Swan, T.W. Economic growth and capital accumulation. Econ. Rec. 1956, 32, 334–361. [CrossRef]
102. Solow, R.M. Neoclassical growth theory. In Handbook of Macroeconomics; Part A; Elsevier Science B.V.:
Amsterdam, The Netherlands, 1999; Volume 1, Chapter 9; pp. 637–667. [CrossRef]
103. Romer, D. Advanced Macroeconomics, 4th ed.; Business and Economics; McGrow-Hill: New York, NY, USA,
2012; 716p, ISBN 978-0-07-351137-5.
104. Volgina, O.A.; Golodnaya, N.Y.; Odiako, N.N.; Shuman, G.I. Mathematical Modeling of Economic Processes and
Systems, 3rd ed.; Knorus: Moscow, Russia, 2014; 200p, ISBN 978-5-406-03252-7.
105. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1987. Press Release.
NobelPrize.org. Nobel Media AB 2019. Available online: https://www.nobelprize.org/prizes/economic-
sciences/1987/press-release/ (accessed on 9 May 2019).
106. Ferrara, M.; Guerrini, L. The neoclassical model of Solow and Swan with logistic population growth. In
Proceedings of the 2nd International Conference of IMBIC on Mathematical Sciences for Advancement of
Science and Technology (MSAST), Kolkata, India, 21–23 December 2017; pp. 119–127. Available online:
https://www.unirc.it/documentazione/materiale_didattico/600_2008_108_1691.pdf (accessed on 9 May 2019).
107. Capasso, V.; Engbers, R.; La Torre, D. On a spatial Solow model with technological diffusion and nonconcave
production function. Nonlinear Anal. Real World Appl. 2010, 11, 3858–3876. [CrossRef]
108. Fanti, L.; Manfredi, P. The Solow’s model with endogenous population: A neoclassical growth cycle model.
J. Econ. Dev. 2003, 28, 103–115.
109. Stamova, I.M.; Stamov, A.G. Impulsive control on the asymptotic stability of the solutions of a Solow model
with endogenous labor growth. J. Franklin Inst. 2012, 349, 2704–2716. [CrossRef]
110. Kilbas, A.A.; Marzan, S.A. Nonlinear differential equations with the Caputo fractional derivative in the space
of continuously differentiable functions. Differ. Equ. 2005, 41, 84–89. [CrossRef]
111. Kilbas, A.A.; Marzan, S.A. Cauchy problem for differential equation with Caputo derivative. Fract. Calc.
Appl. Anal. 2004, 7, 297–321.
112. Leontief, W.W. The Structure of the American Economy 1919–1939: An Empirical Application of Equilibrium
Analysis, 2nd ed.; Oxford University Press: New York, NY, USA, 1951; 282p.
113. Leontief, W.W. Input–Output Economics, 2nd ed.; Oxford University Press: New York, NY, USA, 1986.
114. The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel 1973. Press release.
NobelPrize.org. Nobel Media AB 2019. Available online: https://www.nobelprize.org/prizes/economic-
sciences/1973/press-release/ (accessed on 9 May 2019).
115. Granberg, A.G. Dynamic Models of the National Economy; Ekonomika: Moscow, Russia, 1985; 240p. (In Russian)
116. Pokrovskii, V.N. Econodynamics: The Theory of Social Production, 2nd ed.; Springer: Dordrecht, The Netherlands,
2012; 200p. [CrossRef]
117. Tarasova, V.V.; Tarasov, V.E. Dynamic intersectoral models with memory that generalize Leontief model.
J. Econ. Entrep. 2017, 2-1, 913–924. (In Russian)
118. Tarasova, V.V.; Tarasov, V.E. Dynamic intersectoral models with power-law memory. Commun. Nonlinear Sci.
Numer. Simul. 2018, 54, 100–117. [CrossRef]
119. Tarasova, V.V.; Tarasov, V.E. Chronological exponent for processes with memory and dynamic intersectoral
economic models. Sci. Educ. Today 2017, 4, 29–39. (In Russian)
120. Tarasov, V.E.; Tarasova, V.V. Time-dependent fractional dynamics with memory in quantum and economic
physics. Ann. Phys. 2017, 383, 579–599. [CrossRef]
121. Bogoliubov, N.N.; Shirkov, D.V. Quantum Fields; Benjamin/Cumming: New York, NY, USA, 1983; 388p.
122. Itzykson, C.; Zuber, J.B. Quantum Field Theory; McGraw-Hill: New York, NY, USA, 1980; 705p.
123. Tarasova, V.V.; Tarasov, V.E. Logistic map with memory from economic model. Chaos Solitons Fract. 2017, 95, 84–91.
[CrossRef]
124. West, B.J. Exact solution to fractional logistic equation. Phys. A Stat. Mech. Its Appl. 2015, 429, 103–108. [CrossRef]
125. Area, I.; Losada, J.; Nieto, J.J. A note on the fractional logistic equation. Phys. A Stat. Mech. Its Appl.
2016, 444, 182–187. [CrossRef]
126. Tarasova, V.V.; Tarasov, V.E. Criteria hereditarity of economic process and memory effect. Young Sci.
2016, 14, 396–399. (In Russian)
127. Tarasova, V.V.; Tarasov, V.E. Economic indicators: Ambiguity and memory effects. Econ. Manag. Law
2016, 3, 3–5. (In Russian)

89
Mathematics 2019, 7, 554

128. Tarasov, V.E.; Tarasova, V.V. Long and short memory in economics: Fractional-order difference and
differentiation. IRA-Int. J. Manag. Soc. Sci. 2016, 5, 327–334. [CrossRef]
129. Tarasova, V.V.; Tarasov, V.E. Comments on the article “Long and short memory in economics: Fractional-order
difference and differentiation”. Probl. Mod. Sci. Educ. 2017, 31, 26–28. [CrossRef]
130. Tarasova, V.V.; Tarasov, V.E. Long-term and short-term memory in discrete economic models. Compet. Glob.
World Econ. Sci. Technol. 2017, 7, 155–160. (In Russian)
131. Tarasova, V.V.; Tarasov, V.E. Comments to the article “Long-term and short-term memory in discrete economic
models”. Probl. Sci. 2017, 8, 12–13.
132. Tarasova, V.V.; Tarasov, V.E. Marginal utility for economic processes with memory. Alm. Mod. Sci. Educ.
2016, 7, 108–113. (In Russian)
133. Tarasova, V.V.; Tarasov, V.E. Economic indicator that generalizes average and marginal values. J. Econ. Entrep.
2016, 11-1, 817–823. (In Russian)
134. Tarasova, V.V.; Tarasov, V.E. Marginal values of non-integer order in economic analysis. Azimuth Sci. Res.
Econ. Manag. 2016, 3, 197–201. (In Russian)
135. Tarasova, V.V.; Tarasov, V.E. A generalization of concepts of accelerator and multiplier to take into account
memory effects in macroeconomics. J. Econ. Entrep. 2016, 10-3, 1121–1129. (In Russian)
136. Tarasova, V.V.; Tarasov, V.E. Economic accelerator with memory: Discrete time approach. Probl. Mod. Sci.
Educ. 2016, 36, 37–42. [CrossRef]
137. Tarasova, V.V.; Tarasov, V.E. Exact discretization of economic accelerator and multiplier with memory.
Fractal Fract. 2017, 1, 6. [CrossRef]
138. Tarasova, V.V.; Tarasov, V.E. Exact discretization of economic accelerators and multipliers with memory.
J. Econ. Entrep. 2017, 1, 6. [CrossRef]
139. Tarasova, V.V.; Tarasov, V.E. Discrete accelerator with memory in macroeconomics. Economics 2017, 8, 32–40.
(In Russian)
140. Tarasova, V.V.; Tarasov, V.E. Accelerators in macroeconomics: A comparison of discrete and continuous
approaches. Sci. J. 2017, 8, 4–14. (In Russian) [CrossRef]
141. Tarasova, V.V.; Tarasov, V.E. Accelerators in macroeconomics: Comparison of discrete and continuous
approaches. Am. J. Econ. Bus. Adm. 2017, 9, 47–55. [CrossRef]
142. Tarasova, V.V.; Tarasov, V.E. Risk aversion for investors with memory: Hereditary generalizations of
Arrow-Pratt measure. Financ. J. 2017, 2, 46–63. (In Russian)
143. Tarasova, V.V.; Tarasov, V.E. Non-local measures of risk aversion in the economic process. Econ. Theory Pract.
2016, 44, 54–58. (In Russian)
144. Tarasova, V.V.; Tarasov, V.E. Macroeconomic models with dynamic memory. J. Econ. Entrep. 2017, 3-2, 26–35.
(In Russian)
145. Tarasova, V.V.; Tarasov, V.E. Macroeconomic models with dynamic memory: Fractional calculus approach.
Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
146. Tarasov, V.E. Economic models with power-law memory. In Handbook of Fractional Calculus with Applications;
Applications in Engineering, Life and Social Sciences, Part B; Baleanu, D., Lopes, A.M., Eds.; De Gruyter:
Berlin, Germany; Boston, MA, USA, 2019; Volume 8, Chapter 1; pp. 1–32. ISBN 978-3-11-057092. [CrossRef]
147. Tarasova, V.V.; Tarasov, V.E. Deterministic factor analysis: Methods of integro-differentiation of non-integral
order. Actual Probl. Econ. Law 2016, 10, 77–87. (In Russian) [CrossRef]
148. Tarasova, V.V.; Tarasov, V.E. Fractional deterministic factor analysis of economic processes with memory
and nonlocality. In Chaotic, Fractional, and Complex Dynamics: New Insights and Perspectives; Chapter of the
Book Springer Book; Edelman, M., Macau, E.E.N., Sanjuan, M.A.F., Eds.; Springer International Publishing:
Berlin/Heidelberg, Germany, 2018; Chapter 9; pp. 173–189. [CrossRef]
149. Tarasov, V.E. On history of mathematical economics: Application of fractional calculus. Mathematics 2019, 7, 509.
[CrossRef]
150. Kochubei, A.N. General fractional calculus, evolution equations, and renewal processes. Integral Equ. Oper.
Theory 2011, 71, 583–600. [CrossRef]
151. Kochubei, A.N. General fractional calculus. In Handbook of Fractional Calculus with Applications; Basic Theory;
Kochubei, A., Luchko, Y., Eds.; De Gruyter: Berlin, Germany; Boston, MA, USA, 2019; Volume 1, Chapter 5;
pp. 111–126. [CrossRef]

90
Mathematics 2019, 7, 554

152. Kochubei, A.N. Distributed order calculus and equations of ultraslow diffusion. J. Math. Anal. Appl.
2008, 340, 252–281. [CrossRef]
153. Sin, C.-S. Well-posedness of general Caputo-type fractional differential equations. Fract. Calc. Appl. Anal.
2018, 21, 819–832. [CrossRef]
154. Kochubei, A.N.; Kondratiev, Y. Fractional kinetic hierarchies and intermittency. Kinetic and Related Models.
Am. Inst. Math. Sci. 2017, 10, 725–740. [CrossRef]
155. Kochubei, A.N.; Kondratiev, Y. Growth equation of the general fractional calculus. Mathematics 2019. Submitted.
156. Bateman, H. Tables of Integral Transforms; McGraw-Hill: New York, NY, USA, 1954; Volume I, 391p,
ISBN 07-019549-8.
157. WolframAlpha: Inverse Laplace Transform sˆa/(s+b)ˆc. Available online: https://www.wolframalpha.com/
input/?i=inverse+Laplace+transform+s%5Ea%2F(s%2Bb)%5Ec (accessed on 9 May 2019).
158. Prabhakar, T.R. A singular integral equation with a generalized Mittag-Leffler function in the kernel. Yokohama
Math. J. 1971, 19, 7–15.
159. Garra, R.; Garrappa, R. The Prabhakar or three parameter Mittag-Leffler function: Theory and application.
Commun. Nonlinear Sci. Numer. Simul. 2018, 56, 314–329. [CrossRef]
160. Garrappa, R. Neglecting nonlocality leads to unreliable numerical methods for fractional differential equations.
Commun. Nonlinear Sci. Numer. Simul. 2019, 70, 302–306. [CrossRef]
161. Hojman, S.; Zertuche, F. S-equivalence and symmetries of first-order differential systems. Il Nuovo Cim. B
1985, 88, 1–8. [CrossRef]
162. Filippov, V.M.; Savchin, V.M.; Shorokhov, S.G. Variational principles for nonpotential operators. J. Math. Sci.
1994, 68, 275–398. [CrossRef]
163. Harrod, R. An Essay in dynamic theory. Econ. J. 1939, 49, 14–33. [CrossRef]
164. Domar, E.D. Capital expansion, rate of growth and employment. Econometrica 1946, 14, 137–147. [CrossRef]
165. Domar, E.D. Expansion and employment. Am. Econ. Rev. 1947, 37, 34–55.
166. Tarasova, V.V.; Tarasov, V.E. Hereditary generalization of Harrod-Domar model and memory effects. J. Econ.
Entrep. 2016, 10-2, 72–78.
167. Tarasova, V.V.; Tarasov, V.E. Memory effects in hereditary Harrod-Domar model. Probl. Mod. Sci. Educ.
2016, 32, 38–44. (In Russian) [CrossRef]
168. Tarasov, V.E. Self-organization with memory. Commun. Nonlinear Sci. Numer. Simul. 2019, 72, 240–271. [CrossRef]
169. Tarasova, V.V.; Tarasov, V.E. Economic growth model with constant pace and dynamic memory. Probl. Mod.
Sci. Educ. 2017, 2, 40–45. [CrossRef]
170. Tarasova, V.V.; Tarasov, V.E. Model of economic growth with constant rate and dynamic memory. Econ. Sociol.
Law 2017, 8, 18–24. (In Russian)
171. Mainardi, F. Fractional relaxation-oscillation and fractional diffusion-wave phenomena. Chaos Solitons Fract.
1996, 7, 1461–1477. [CrossRef]
172. Mainardi, F. The fundamental solutions for the fractional diffusion-wave equation. Appl. Math. Lett. 1996, 9, 23–28.
[CrossRef]
173. Gorenflo, R.; Mainardi, F. Fractional calculus, integral and differential equations of fractional order. In Fractals
and Fractional Calculus in Continuum Mechanics; Carpinteri, A., Mainardi, F., Eds.; Springer: Wien, Austria,
1997; pp. 223–276.
174. Gorenflo, R.; Mainardi, F. Chapter 3. Fractional relaxation-oscillation phenomena. In Handbook of Fractional
Calculus with Applications; Applications in Physics, Part A; Tarasov, V.E., Ed.; De Gruyter: Berlin, Germany;
Boston, MA, USA, 2019; Volume 4, pp. 45–74. [CrossRef]
175. Gorenflo, R.; Mainardi, F. Fractional relaxation of distributed order. In Complexus Mundi: Emergent Patterns in
Nature; Novak, M., Ed.; World Scientific: Singapore, 2006; pp. 33–42.
176. Mainardi, F.; Mura, A.; Pagnini, G.; Gorenflo, R. Fractional relaxation and time-fractional diffusion of
distributed order. IFAC Proc. Vol. 2006, 39, 1–21. [CrossRef]
177. Mainardi, F.; Mura, A.; Gorenflo, R.; Stojanović, M. The two forms of fractional relaxation of distributed
order. J. Vib. Control 2007, 13, 1249–1268. [CrossRef]

© 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

91
 mathematics
Article
Growth Equation of the General Fractional Calculus
Anatoly N. Kochubei 1, * and Yuri Kondratiev 2
1 Institute of Mathematics, National Academy of Sciences of Ukraine, Tereshchenkivska 3, 01024 Kyiv, Ukraine
2 Department of Mathematics, University of Bielefeld, D-33615 Bielefeld, Germany
* Correspondence: [email protected]

Received: 2 May 2019; Accepted: 9 July 2019; Published: 11 July 2019 

Abstract: We consider the Cauchy problem pDpkq uqptq “ λuptq, up0q “ 1, where Dpkq is the general
convolutional derivative introduced in the paper (A. N. Kochubei, Integral Equations Oper. Theory 71
Þ Eα pλtα q, where 0 ă α ă 1, Eα
(2011), 583–600), λ ą 0. The solution is a generalization of the function t Ñ
is the Mittag–Leffler function. The asymptotics of this solution, as t Ñ 8, are studied.

Keywords: generalized fractional derivatives; growth equation; Mittag–Leffler function

1. Introduction
In several models of the dynamics of complex systems, the time evolution for observed quantities
has exponential asymptotics of two possible types. In the simplest cases, these asymptotics are related
with the solutions to the equations

u1 ptq “ zuptq, t ą 0; up0q “ 1,

where we will consider positive and negative z separately. For z ă 0 (the relaxation equation),
the solution decays to zero. In particular models such as, e.g., Glauber stochastic dynamics in the
continuum, this corresponds to an exponential convergence to an equilibrium; see [1]. The case z ą 0
may also appear in applications. We can mention the contact model in the continuum where for the
mortality below a critical value, the density of the population will grow exponentially fast [2,3], as
well as models of economic growth.
On the other hand, the observed behavior of specific physical and biological systems show an
emergence of other time asymptotics that may be far from exponential decay or growth. An attempt to
obtain other relaxation characteristics is related with a use of generalized time derivatives in dynamical
equations (see [4,5]). In this way, we may produce a wide spectrum of possible asymptotics to reflect a
demand coming from applications [6].
The general fractional calculus introduced in [7] is based on a version of the fractional derivative,
the differential-convolution operator

żt
d
pDpkq uqptq “ kpt ´ τqupτq dτ ´ kptqup0q,
dt
0

where k is a non-negative locally integrable function satisfying additional assumptions, under which
(A) the Cauchy problem
pDpkq uqptq “ ´λuptq, t ą 0; up0q “ 1, (1)

where λ ą 0, has a unique solution that is completely monotone;

Mathematics 2019, 7, 615; doi:10.3390/math7070615 93 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 615

(B) the Cauchy problem

pDpkq wqpt, xq “ Δwpt, xq, t ą 0, x P Rn ; wp0, xq “ w0 pxq,

is solvable (under appropriate conditions for w0 ) and possesses a fundamental solution, a kernel with
the property of a probability density.
A class of functions k, for which (A) and (B) hold, was found in [7] and is described below. The
simplest example is
t´α
kptq “ , t ą 0, (2)
Γp1 ´ αq
where 0 ă α ă 1, and for this case, Dpkq is the Caputo–Djrbashian fractional derivative Dpαq . Another
subclass is the one of distributed order derivatives; see [8] for the details.
Note that for the case where k has the form (2), the solution of (1) is uptq “ Eα p´λtα q, where Eα is
the Mittag–Leffler function; see [9], Lemma 2.23 (page 98). This solution has a slow decay at infinity
due to the asymptotic property of the Mittag–Leffler function; see [10]. Note that using particular
classes of fractional derivatives, we observe several specific asymptotics for the solution of Equation (1)
with λ ą 0. Some results in this direction were already obtained in [8,11,12]. A more detailed analysis
of this problem will be performed in a forthcoming paper.
In this paper, we consider the Cauchy problem with the opposite sign in the right-hand side,
that is
pDpkq uqptq “ λuptq, t ą 0; up0q “ 1; (3)

as before, λ ą 0. In the case of (2), we have uptq “ Eα pλtα q (see [9], Lemma 2.23 (page 98)), and due
to the well-known asymptotics of Eα [10], this is a function of exponential growth. The existence
and uniqueness of an absolutely continuous solution of (3) follows from the results of [13] dealing
with more general nonlinear equations. Here, we study the asymptotic behavior of the solution of (3).
Functions of this kind can be useful for fractional macroeconomic models with long dynamic memory;
see [14] and references therein. Let us explain this in a little greater detail.
In modern macroeconomics, the most important are so-called growth models, which in the
mathematical sense are reduced (for linear models) to the equation u1 ptq “ λuptq ` f ptq with λ ą 0.
In economics, an important role is played by processes with a distributed lag, starting with Phillips’
works [15] (see also [16]), and long memory, starting with Granger’s work [17] (see also [18]).
If we assume the presence of effects of distributed lag (time delay) or fading memory in economic
processes, then the fractional generalization of the linear classical growth models can be described
by the fractional differential equation D α uptq “ λuptq ` f ptq with λ ą 0, α ą 0. The fractional
generalizations of well-known economics models were first proposed for the Caputo–Djrbashian
fractional derivative D α . Solving the problem in a more general case will allow us to describe accurately
the conditions on the operator kernels (the memory functions), under which equations for models of
economic growth with memory have solutions.
In general fractional calculus, which was proposed in [7] (see also [4]), the case λ ą 0 is not
considered. The growth equation was considered in [8] for the special case of a distributed order
derivative, where it was proved that a smooth solution exists and is monotone increasing.
In this article, we propose correct mathematical statements for growth models with memory
in more general cases, for the general fractional derivative Dpkq with respect to the time variable.
Their application can be useful for mathematical economics for the description of processes with long
memory and distributed lag.
Note that the technique used below was developed initially in [12] for use in the study of
intermittency in fractional models of statistical mechanics.

94
Mathematics 2019, 7, 615

2. Preliminaries
Our conditions regarding the function k will be formulated in terms of its Laplace transform

ż8
Kppq “ e´ pt kptq dt. (4)
0

Denote Φppq “ pKppq.

We make the following assumptions leading to (A) and (B) (see [7]).
(*) The Laplace transform (4) exists for all positive numbers p. The function K belongs to the Stieltjes
class S , and
Kppq Ñ 8, as p Ñ 0; Kppq Ñ 0, as p Ñ 8; (5)

pKppq Ñ 0, as p Ñ 0; pKppq Ñ 8, as p Ñ 8. (6)

Recall that the Stieltjes class consists of the functions ψ admitting the integral representation

ż8
a 1
ψpzq “ `b` σpdtq,
z z`t
0

where a, b ě 0, σ is a Borel measure on r0, 8q, such that

ż8
p1 ` tq´1 σpdtq ă 8. (7)
0

For a detailed exposition of the theory of Stieltjes functions including properties of the measure σ,
see [19], and especially Chapters 2 and 6.
In particular, for the Stieltjes function K, the limit conditions (5) and (6) imply the representation

ż8
1
Kppq “ σpdtq. (8)
z`t
0

We can also write [7] that

ż8
kpsq “ e´ts σpdtq, 0 ă s ă 8.
0

The function Φ belongs to the class CBF of complete Bernstein functions, a subclass of the class
BF of Bernstein functions. Recall that a function f : p0, 8q Ñ R is called a Bernstein function if
f P C8 , f pzq ě 0 for all z ą 0, and

p´1qn´1 f pnq pzq ě 0 for all n ě 1, z ą 0,

so that the derivative of f is completely monotone. A function f belongs to CBF if it has an analytic
continuation to the cut complex plane Czp´8, 0s such that Im z ¨ Im f pzq ě 0 and there exists the real
limit
f p0`q “ lim f pzq.
p0,8qQzÑ0

Both the classes BF and CBF admit equivalent descriptions in terms of integral representations;
see [19].

95
Mathematics 2019, 7, 615

Below, we will need the following inequality for complete Bernstein functions (Proposition 2.4
in [20]), valid, in particular, for the function Φ. For any p outside the negative real semi-axis, we have
c d
1 ` cos ϕ 2
Φp|p|q ď |Φppq| ď Φp|p|q, ϕ “ arg p. (9)
2 1 ` cos ϕ

Solutions of the Cauchy problem (3) and a similar problem with the classical first order derivative
are connected by the subordination identity (see [7]; for the case of the Caputo–Djrbashian derivative,
see [21]), an integral transformation with the kernel Gps, tq constructed as follows.
Consider the function
gps, pq “ Kppqe´sΦp pq , s ą 0, p ą 0. (10)

It is proved [7] that g is a Laplace transform in the variable t of the required kernel Gps, tq, that is,

ż8
gps, pq “ e´ pt Gps, tq dt.
0

G is non-negative, and
ż8
Gps, tq ds “ 1 for each t.
0

3. Cauchy Problem for the Growth Equation

Let us consider the Cauchy problem (3). If uλ ptq is its solution, whose Laplace transform u
Ăλ ppq
exists for some p, then it follows from properties of the Laplace transform [22] that

uλ ppq “ Kppq.
uλ ppq ´ λĂ
ΦppqĂ

Hence,
Kppq
Ăλ ppq “
u , if Φppq ą λ. (11)
Φppq ´ λ
On the other hand, consider the function

ż8
Ept, λq “ eλs Gps, tq ds, t ą 0. (12)
0

The existence of the integral in (12) for almost all t ą 0 is, by the Fubini–Tonelli theorem, a
consequence of the absolute convergence of the repeated integral

ż8 ż8 ż8
Kppq
eλs ds e´ pt Gps, tq dt “ eλs gps, λq ds “ ,
Φppq ´ λ
0 0 0

where p ą 0 is such that Φppq ą λ.

The above calculation shows that Ept, λq “ uλ ptq, the solution of (3), and the identity (12) provides
an integral representation of this solution.
A more detailed analysis of its properties is based on the analytic properties of the Stieltjes function
K, or equivalently, of the complete Bernstein function Φ; in particular, we use the representation

ż8
p
Φppq “ σpdtq, (13)
p`t
0

96
Mathematics 2019, 7, 615

which follows from (7). The measure σ satisfies (8).

Since Φ is a Bernstein function, its derivative Φ1 is completely monotone. By our assumptions, Φ
is not a constant function, so that Φ1 is not the identical zero. It follows from Bernstein’s description
of completely monotone functions that Φ1 ppq ‰ 0 for any p ą 0 (see Remark 1.5 in [19]). Therefore,
Φ is strictly monotone, and for each z ą 0, there exists a unique p0 “ p0 pzq ą 0 such that Φpp0 q “ z.
The inequality Φppq ą z is equivalent to the inequality p ą p0 pzq. Since Φ, as a complete Bernstein
function, preserves the open upper and lower half-planes (in fact, this follows from (13)), we have
Φppq ‰ z for any nonreal p.
It is proved in [12] that the function p0 pzq, z ą 0 is strictly superadditive, that is

p0 px ` yq ą p0 pxq ` p0 pyq for any x, y ą 0.

Proposition 1. The solution uλ ptq of the Cauchy problem (3) admits a holomorphic continuation in the variable
t to a sector Σv “ reiθ : r ą 0, ´v ă θ ă v , 0 ă v ă π2 , and
 (

sup ˇe´ p0 t uλ ptqˇ ă 8, p0 “ p0 pλq.

ˇ ˇ
(14)
tPΣ v

Ăλ ppq is holomorphic in p on any sector

Proof. It follows from (11) and (13) that the Laplace transform u
p0 ` Σρ` π , 0 ă ρ ă π2 . In addition,
2

sup |pp ´ p0 qĂ
uλ ppq| ă 8. (15)
pP p0 `Σρ` π
2

Now the assertion is implied by (15) and the duality theorem for holomorphic continuations of a
function and its Laplace transform; see Theorem 2.6.1 in [23].

Now we are ready to formulate and prove our main result.

Theorem 1. Let the assumptions p˚q hold, and in addition,

ż8
ds
ă 8. (16)
sΦpsq
1

Then
λ
uλ ptq “ e p0 pλqt ` ope p0 pλqt q, t Ñ 8. (17)
Φ1 pp0 pλqqp0 pλq

Proof. The representation (11) can be written as

λ
ˆ ˙
1
Ăλ ppq “
u 1` .
p Φppq ´ λ

This implies the representation of uλ as uλ ptq “ 1 ` Bλ ptq, where Bλ has the Laplace transform

Ăλ ppq “ λ ¨
B
1
,
p Φppq ´ λ

for such p that Φppq ą λ.

Using the inequality (9), we find that |Φppq| ě ?1 Φp|p|q on any vertical line tp “ γ ` iτ, τ P Ru
2
where γ ą p0 . By our assumption (16), B
Ăλ is absolutely integrable on such a line. In addition, it follows

97
Mathematics 2019, 7, 615

Ăλ ppq Ñ 0, as p Ñ 8 in the half-plane Re p ą p0 . These properties make it possible (see

from (6) that B
Theorem 28.2 in [22]) to write the inversion formula

γ`
ż i8
λ dp
uλ ptq “ 1 ` e pt , γ ą p0 .
2πi ppΦppq ´ λq
γ ´i 8

Denote
r`
żi 8
λ dp
Vptq “ 1 ` e pt ,
2πi ppΦppq ´ λq
r ´i 8

where 0 ă r ă p0 . Then
ˇ 8 ˇ
ˇż ˇ
rt ˇ iτt dτ ˇ “ opert q,
|Vptq| ď 1 ` Ce ˇ t Ñ 8,
ˇ ˇ
e (18)
pr ` iτqpΦpr ` iτq ´ λq ˇ
´8
ˇ ˇ

by virtue of (9), (16), and the Riemann–Lebesgue theorem.

On the other hand, we may write
¨ ˛
λ ˚ dp
ż ż ż
‹ pt
uλ ptq ´ Vptq “ ` ` ‚e ,
2πi ppΦppq ´ λq
˝
Λ` Λ0 Λ´

where the contour Λ` consists of the vertical rays tRe p “ r, Im p ě Ru, tRe p “ γ, Im p ě Ru, and the
horizontal segment tr ď Re p ď γ, Im p “ Ru (R ą 0), Λ´ is a mirror reflection of Λ` with respect
to the real axis, Λ0 is the finite rectangle consisting of the vertical segments tRe p “ r, | Im p| ď Ru,
tRe p “ γ, | Im p| ď Ru, and the horizontal segments tr ď Re p ď γ, Im p “ ˘Ru.
We have
dp
ż
e pt “ 0,
ppΦppq ´ λq
Λ`

due to the Cauchy theorem, absolute integrability of the integrand on the vertical rays (see (16)) and
the estimate ˇ ˇ
ˇż ˇ
ˇ dp ˇ
ˇ e pt ˇ ď Ch´1 Ñ 0, h Ñ 8,
ˇ ˇ
ˇ ppΦppq ´ λq ˇ
ˇΠh ˇ

where Πh “ tr ď Re p ď γ, Im p “ hu, h ą R. In a similar way, we prove that

dp
ż
e pt “ 0.
ppΦppq ´ λq
Λ´

Due to the inequality Φ1 pp0 q ‰ 0, there exists a complex neighborhood W of the point λ “ Φpp0 q,
in which Φ has a single-valued holomorphic inverse function p “ ψpwq, so that Φpψpwqq “ w and
p0 “ ψpλq. In the above arguments, the numbers r, γ, R were arbitrary. Now we choose R and γ ´ r so

98
Mathematics 2019, 7, 615

small that that the curvilinear rectangle ΦpΛ0 q lies inside W. Making the change of variables p “ ψpwq
and using the Cauchy formula, we find that

λ dp λ 1 dw λ
ż ż
e pt “ e ψpwqt ¨ “ 1 e ψpλqt
2πi ppΦppq ´ λq 2πi Φ1 pψpwqqψpwq w ´ λ Φ pψpλqqψpλq
Λ0 ΦpΛ0 q
λ
“ e p0 pλqt .
Φ1 pp0 pλqqp0 pλq

Together with (18), this implies the required asymptotic relation (17).

Example 1. (1) In the case (2) of the Caputo–Djrbashian fractional derivative of order 0 ă α ă 1, we have
uλ ptq “ Eα pλtα q, Φppq “ pα , and the condition (16) is satisfied. Here the asymptotics (17) coincide with the
one given by the principal term of the asymptotic expansion of the Mittag–Leffler function. The above proof is
different from the classical proof of the latter (see [10]).
(2) Let us consider the case of a distributed order derivative with a weight function μ, that is,

ż1
Dpμq uptq “ pDpαq quptqμpαq dα.
0

Suppose that μ P C2 r0, 1s, μp1q ‰ 0. In this case [8],

ż1 ż1
s´α
kpsq “ μpαq dα, Φppq “ pα μpαq dα,
Γp1 ´ αq
0 0

and under the above assumptions,

μp1qp
` O p| log p|´2 ,
´ ¯
Φppq “ p Ñ 8.
log p

Condition (16) is satisfied, and our asymptotic result (17) is applicable in this case.

Author Contributions: Both authors contributed equally and significantly in writing this article. Both authors
read and approved the final manuscript.
Funding: The work of the first-named author was funded in part under the budget program of Ukraine No.
6541230 “Support to the development of priority research trends” and under the research work “Markov evolutions
in real and p-adic spaces” of the Dragomanov National Pedagogical University of Ukraine.
Acknowledgments: The authors are grateful to V. E. Tarasov for calling our attention to the problem studied in
this paper, as well as for his advice regarding its text.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Kondratiev, Y.; Kutoviy, O.; Minlos, R. On non-equilibrium stochastic dynamics for interacting particle
systems in continuum. J. Funct. Anal. 2008, 255, 200–227. [CrossRef]
2. Kondratiev, Y.; Kutoviy, O.; Pirogov, S. Correlation functions and invariant measures in continuous contact
model. Infin. Dimens. Anal. Quantum Probab. Relat. Top. 2008, 11, 231–258. [CrossRef]
3. Kondratiev, Y.; Skorokhod, A. On contact processes in continuum. Infin. Dimens. Anal. Quantum Probab.
Relat. Top. 2006, 9, 187–198. [CrossRef]
4. Handbook of Fractional Calculus with Applications. Volume 1: Basic Theory; Kochubei, A.N., Luchko, Y., Eds.;
De Gruyter: Berlin, Germany, 2019; 481p.

99
Mathematics 2019, 7, 615

5. Handbook of Fractional Calculus with Applications. Volume 2. Fractional Differential Equations; Kochubei, A.N.,
Luchko, Y., Eds.; De Gruyter: Berlin, Germany, 2019; 519p.
6. Handbook of Fractional Calculus with Applications; Tenreiro Machado, J.A., Ed.; De Gruyter: Berlin, Germany,
2019; Volumes 3–8.
7. Kochubei, A.N. General fractional calculus, evolution equations, and renewal processes. Integral Equ.
Oper. Theory 2011, 71, 583–600. [CrossRef]
8. Kochubei, A.N. Distributed order calculus and equations of ultraslow diffusion. J. Math. Anal. Appl. 2008,
340, 252–281. [CrossRef]
9. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier:
Amsterdam, The Netherlands, 2006.
10. Gorenflo, R.; Kilbas, A.A.; Mainardi, F.; Rogosin, S.V. Mittag-Leffler Functions, Related Topics and Applications;
Springer: Berlin, Germany, 2014.
11. Kochubei, A.N. Distributed order derivatives and relaxation patterns. J. Phys. A 2009, 42, 315203. [CrossRef]
12. Kochubei, A.N.; Kondratiev, Y. Fractional kinetic hierarchies and intermittency. Kinet. Relat. Models 2017, 10,
725–740. [CrossRef]
13. Sin, C.S. Well-posedness of general Caputo-type fractional differential equations. Frac. Calc. Appl. Anal. 2018,
21, 819–832. [CrossRef]
14. Tarasov, V.E.; Tarasova, V.V. Macroeconomic models with long dynamic memory: Fractional calculus
approach. Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
15. Phillips, A.W. Stabilisation policy in a closed economy. Econ. J. 1954, 64, 290–323. [CrossRef]
16. Allen, R.G.D. Macro-Economic Theory. A Mathematical Treatment; Macmillan: London, UK, 1968.
17. Granger, C.W.J. The typical spectral shape of an economic variable. Econometrica 1966, 34, 150–161. [CrossRef]
18. Baillie, R.N. Long memory processes and fractional integration in econometrics. J. Econom. 1996, 73, 5–59.
[CrossRef]
19. Schilling, R.L.; Song, R.; Vondrac̆ek, Z. Bernstein Functions. Theory and Applications; Walter de Gruyter: Berlin,
Germany, 2010.
20. Batty, C.; Chill, R.; Tomilov, Y. Fine scale of decay of operator semigroups. J. European Math. Soc. 2016, 18,
853–929. [CrossRef]
21. Bazhlekova, E. Subordination principle for fractional evolution equations. Frac. Calc. Appl. Anal. 2000, 3,
213–230.
22. Doetsch, G. Introduction to the Theory and Applications of the Laplace Transformation; Springer: Berlin, Germany,
1974.
23. Arendt, W.; Batty, C.J.K.; Hieber, M.; Neubrander, F. Vector-Valued Laplace Transforms and Cauchy Problems;
Birkhäuser: Basel, Switzerland, 2011.

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

100
 mathematics
Article
Applications of the Fractional Diffusion Equation to
Option Pricing and Risk Calculations
Jean-Philippe Aguilar 1 , Jan Korbel 2,3,4 and Yuri Luchko 5, *
1 BRED Banque Populaire, Modeling Department, 18 quai de la Râpée, 75012 Paris, France
2 Section for the Science of Complex Systems, Center for Medical Statistics, Informatics, and Intelligent
Systems (CeMSIIS), Medical University of Vienna, Spitalgasse 23, 1090 Vienna, Austria
3 Complexity Science Hub Vienna, Josefstädterstrasse 39, 1080 Vienna, Austria
4 Faculty of Nuclear Sciences and Physical Engineering, Czech Technical University, 11519 Prague,
Czech Republic
5 Beuth Technical University of Applied Sciences Berlin, Luxemburger Str. 10, 13353 Berlin, Germany
* Correspondence: [email protected]

Received: 15 August 2019; Accepted: 26 August 2019; Published: 1 September 2019 

Abstract: In this article, we first provide a survey of the exponential option pricing models
and show that in the framework of the risk-neutral approach, they are governed by the
space-fractional diffusion equation. Then, we introduce a more general class of models based
on the space-time-fractional diffusion equation and recall some recent results in this field concerning
the European option pricing and the risk-neutral parameter. We proceed with an extension of
these results to the class of exotic options. In particular, we show that the call and put prices can
be expressed in the form of simple power series in terms of the log-forward moneyness and the
risk-neutral parameter. Finally, we provide the closed-form formulas for the first and second order
risk sensitivities and study the dependencies of the portfolio hedging and profit-and-loss calculations
upon the model parameters.

Keywords: fractional diffusion equation; fundamental solution; option pricing; risk sensitivities;
portfolio hedging

1. Introduction
Fractional Calculus (FC) is nearly as old as conventional calculus. Many prominent
mathematicians including Leibniz, Fourier, Laplace, Liouville, Riemann, Weyl, and Riesz suggested
their own definitions of the fractional integrals and derivatives and studied their properties.
Whereas the mathematical theory of FC was nearly completed a long time ago (see, e.g., [1] or the recent
reference books [2,3]), only a few applications of FC outside mathematics were known until recently.
During the last three decades, the situation changed completely, and currently, the majority of FC
publications is devoted to modeling of a broad class of systems and processes using either the FC
operators or the so-called fractional ordinary or partial differential equations. In particular, the FC
models were successfully employed in physics [4,5], control theory [6], as well as in engineering, life,
and social sciences [7,8], to mention only a few of the many application areas.
The two probably most prominent and broadly-recognized FC applications are in linear
viscoelasticity [9] and for describing anomalous transport processes [10]. In both cases, the FC models
in the form of the fractional ODEs (linear viscoelasticity) and the fractional PDEs (anomalous transport
processes), respectively, cannot be derived from “first principles”. Instead, they are introduced as
interpolations between several models formulated in terms of the ODEs or PDEs, respectively. In the
case of linear viscoelasticity, the basic FC model interpolates between the Hooke model (elasticity, ODE

Mathematics 2019, 7, 796; doi:10.3390/math7090796 101 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 796

of the zero order) and the Newton model (viscosity, ODE of the first order). The FC model for the slow
anomalous diffusion interpolates between a stationary state (time-independent Poisson equation) and
the diffusion equation (first order equation with respect to time), whereas the fast diffusion is modeled
by the fractional diffusion-wave PDE that interpolates between the diffusion equation (first order
equation with respect to time) and the wave equation (second order equation with respect to time).
Of course, this kind of model needs an additional justification, say, in the form of a better fitting of
certain datasets or better predictions compared to ones delivered by the conventional models. One of
the advantages of the FC models is that they have some additional parameters (orders of the fractional
derivatives) that can be suitably chosen for a set of data at hand. To do this, an inverse problem for the
determination of the optimal parameters from the data has to be solved (see, e.g., [11] for an excellent
survey of the basic inverse problems for the fractional differential equations).
It is worth mentioning that the anomalous diffusion models have a clear stochastic interpretation
and can be formulated in terms of the continuous time random walk processes. The models in
the form of the time- and/or space-fractional differential equations follow from these stochastic
models for a special choice of the jump probability density functions with infinite first or/and second
moments [10,12,13]. In [14], the fundamental solution for a time-fractional diffusion equation with the
Caputo fractional derivative of order α ∈ (0, 1) and the spatial Laplace operator was shown to be a
probability density function evolving in time. In [15], a space-time fractional diffusion equation with
the spatial Riesz–Feller derivative of order α ∈ (0, 2] and skewness θ (|θ | ≤ min {α, 2 − α}) and the
time-fractional Caputo derivative of order β ∈ (0, 2] was investigated in detail. In particular, several
subordination formulas for the fundamental solutions of this equation with different values of α and β
and an extension of their probabilistic interpretation to the ranges {0 < α ≤ 2} ∩ {0 < β ≤ 1} and
{1 < β ≤ α ≤ 2} were derived in [15].
A close connection of the fractional diffusion equations with the stochastic processes (fractional
Brownian motion, Lévy flights, etc.) made them very promising models for different financial
applications. In particular, they were already employed in the hot problems of finance (for recent
reviews, see, e.g., [16,17]), particularly in financial markets [18–21], macroeconomics [22], mathematical
economics [23], and for describing the concept of memory in economics [24] or economic growth
[25]. One of the first applications of FC in finance was through the fractional Brownian motion,
which enables incorporating long-range auto-correlations, typically observed in finance [26,27],
volatility modeling [28], and option pricing [29]. FC has been specifically used in many option pricing
models [30–33], also in connection with the jump processes [34] or in pricing of more complicated
types of options, as American options [31], double barrier options [35], or currency options [36].
These models have been also investigated by numerical methods [37,38], and some applications to
implied volatility have been also discussed [39].
In this paper, we focus on FC applications to option pricing, which is one of the most important
tasks of financial mathematics. It is an important tool for market participants who want to hedge
their positions and to estimate the value and risks or their portfolios. The first option pricing
model was introduced by Black and Scholes [40] in 1973 based on the Gaussian assumptions for
the variation of the stock prices’ returns. After that, many generalizations and adaptations of
this model were derived in order to capture the behavior of financial markets more realistically
(let us mention, among others, models based on stochastic volatility [41], regime switching [42],
or jumps [43]). Particularly interesting are the approaches based on replacement of the underlying
stochastic process by the fractional Brownian motion or Lévy flights that leads to models driven by
the fractional diffusion equations. One of the first models of this kind explicitly designed for the
purpose of option pricing was introduced by Carr and Wu [44] by replacing the conventional Gaussian
noise by a maximally-skewed Lévy-stable process (in other words, by replacing the underlying
diffusion equation by a space-fractional diffusion equation). This model is far more realistic than the
Black-Scholes model because it incorporates the heavy-tailed distributions and thus allows reproducing
complex, but observable behavior in the distribution of prices (such as large drops) and in long-term

102
Mathematics 2019, 7, 796

volatility patterns. In this paper, we focus on a generalization of this model based on the space-time
fractional diffusion equation [15] originally introduced in [45] and further investigated in [46–49]. The
main advantage of this model is its interpretation: the model parameters play the role of the volatility
(general risk level) and the spatial and temporal risk redistributions, and therefore, this model captures
the complex market behaviors more accurately compared to the conventional models. In the first part
of this paper, we recall the main features and particular cases of the fractional diffusion model, as well
as some recent results on European option pricing. Then, we apply this mathematical framework to
more exotic types of options, i.e., binary options. The second part of the paper is devoted to the risk
sensitives and Profits and Losses (P&L) calculations, which are very important for market practitioners.
The paper is organized as follows. In Section 2, we provide an overview of some recent results on
fractional diffusion models and discuss their applications to financial modeling. Section 3 is devoted to
the pricing formulas for the European and binary options driven by the space-time fractional diffusion
equation. In Section 4, we apply our results to the risk sensitivities and portfolio management. The last
section is dedicated to the conclusions.

2. The Fractional Diffusion Model and Option Pricing

We start with some standard financial definitions. A call (resp. put) option with a strike K > 0 and
a maturity T > 0 is a financial instrument that gives the holder the right to buy (resp. to sell) a given
quantity S (an asset, an index, etc.) at an agreed price P at time t = T. We will denote the option payoff
by P (S( T ), K ).
In the case:
P (S, K ) = max{S − K, 0} := [S − K ]+ (1)

one speaks of an European call option. If the option has the payoff (1), but can be exercised at any time
t ∈ [0, T ], one speaks of an American call option. European and American options are often referred
to as the vanilla options, while the payoff of the so-called exotic options is determined by some more
sophisticated algorithms. Say, the payoff of the binary or the digital call options is given by the formula:

P (S, K ) = H (S − K ) (cash or nothing),
(2)
P (S, K ) = S × H (S − K ) (asset or nothing),

where H denotes the Heaviside step function. Let us also mention that there are even more complicated
options such as Asian or barrier options, which are path-dependent instruments. This means that the
payoff P (S( T ), K ) of these options depends on all values of the asset price S(t), t ≤ T, and not just
on S( T ). In this paper, the focus is mainly on the European and binary options.
Evidently, in order to calculate the options’ prices, the dynamics of the underlying asset
has to be described and modeled. A common way to do this is interpretation of S(t) as a
stochastic process, so that the payoff function P (S( T ), K ) becomes a random variable with a certain
probability distribution. In the next subsection, we discuss a particular class of stochastic models
for S(t), which are very popular among the market practitioners.

2.1. Exponential Market Models

In a wide class of option pricing models, it is assumed that the underlying asset price is
described by a stochastic process {S(t)}t≥0 on a filtered probability space (Ω, F , {Ft }t≥0 , P), whose
instantaneous variations can be written in local form as follows:

dS(t)
= rS(t) dt + σ dX (t), t ∈ [0, T ], (3)
S(t)

where r denotes the continuous risk-free interest rate and σ stands for the market volatility. For financial
application, the common choices for X (t) are:

103
Mathematics 2019, 7, 796

• Gaussian process: X (t) = W (t) (standard Brownian motion). In this case, the stock price S(t)
is said to follow a geometric Brownian motion. This is the basic hypothesis for the Black-Scholes
model [40].
• Lévy process: X (t) = Lα,β (t) (standardized Lévy-stable process with stability α ∈ [0, 2] and
skewness or asymmetry β ∈ [−1, 1]). In this case, one says that S(t) follows an exponential
Lévy-stable process (see the details in [43]). Exponential Lévy models generalize the Gaussian
framework (because L2,β (t) = W (t) for any β ∈ [−1, 1]), while allowing additional realistic
features such as the presence of price jumps with non-zero probability. Their relevance in financial
modeling has been known since the works of Mandelbrot and Fama [50,51]. For most financial
applications, only the so-called Lévy-Pareto distributions, i.e., the values α ∈ (1, 2], are relevant
(historically, Mandelbrot calibrated α = 1.7 for the cotton market).

It is important to note that the stochastic differential Equation (3) is specified under the risk-neutral
measure Q (also known as the martingale measure), that is the measure under which the discounted
market price is a martingale (for more information regarding the risk-neutral option pricing, see [52]
or any monograph on financial mathematics):

S(t) = e−rτ EQ [ST |Ft ], (4)

where τ := T − t. The martingale measure is given by the Esscher transform [53] of the physical
measure P:
dQ|Ft e X (t)
= P X (t) = e X (t)+μt , (5)
dP|Ft E [e ]
where the parameter μ is defined as the value of the negative cumulant-generating function evaluated
at the point 1: 
μ := − log EP e X (1) . (6)

In particular, if the process X (t) admits a probability distribution (or density) g( x, t), then μ reads:

+∞

μ = − log e x g( x, 1) dx (7)
−∞

and therefore, the existence of a risk-neutral measure is linked to the existence of the two-sided Laplace
transform of the density. As we will see later, for exponential Lévy models, the two-sided Laplace
transform diverges with the only exception in the case when the asymmetry parameter is maximal,
that is when the probability distribution has exponential decay on the positive real semi-axis and
polynomial decay (heavy tail) on the negative real semi-axis.
With the above notations, the solution to (3) is the exponential process:

S(t) = S(0)e(r+μ)t + σX (t) (8)

and the price of an option with strike K, maturity T, and payoff P is equal to the present value of its
expected payoff:
C (S, K, r, μ, τ ) = e−rτ EQ [P (S( T ), K )] . (9)

The expectation (9) can be computed by integrating all possible realizations of the terminal payoff
over the martingale measure: if X (t) admits a density g( x, t), then (9) reads:

+∞

C (S, K, r, μ, τ ) = e−rτ P (Se(r+μ)τ + x , K ) g( x, τ ) dx. (10)
−∞

104
Mathematics 2019, 7, 796

2.2. Generalizing Exponential Market Models: The Fractional Diffusion Model

2.2.1. Setup of the Model

By the space-time fractional diffusion model, we mean the following Cauchy problem:
⎧ 
⎪ ∗ γ θ α
⎪ 0 Dt + μα,θ,γ D x g( x, t) = 0,
⎪
⎪
x ∈ R, t > 0,
⎪
⎪
⎪
⎨ g(±∞, t) = 0,
(11)
⎪
⎪ g( x, 0) = f 0 ( x ),
⎪
⎪
⎪
⎪ ∂g
⎪
⎩ ( x, 0) = f 1 ( x ),
∂t

where the parameters α and γ are restricted as follows: α ∈ (0, 2], γ ∈ (0, α]. The asymmetry parameter
γ
θ belongs to the so-called Feller-Takayasu diamond |θ | ≤ min {α, 2 − α}. 0∗ Dt denotes the Caputo fractional
derivative, which is defined as:
 t
∗ ν 1 f ν (τ )
t0 D t f ( t ) = dτ, (12)
Γ( ν − ν) t0 (t − τ )ν+1− ν

and θ D xα denotes the Riesz–Feller fractional derivative, which is usually defined via its Fourier image:

F [θ D xν f ( x )](k) = −θ ψν (k) F [ f ( x )](k) = −μ|k|ν ei(signk)θπ/2 F [ f ( x )](k) . (13)

By definition, both fractional derivatives become ordinary derivative operators if the orders of
the derivatives are natural numbers. The fundamental solution or the Green function of (11), that is the
solution corresponding to the initial values f 0 ( x ) = δ( x ) and f 1 ( x ) = 0, was derived in [15] in form of
a Mellin-Barnes integral (for x > 0):
      t1
c1+i∞ t1 t1
1 1 α Γ 1 − α Γ (1 − t1 )
Γ x
gα,θ,γ ( x, t) =       dt1 . (14)
2πi αx Γ 1 − γα t1 Γ α2α
−θ t α−θ 1
(−μα,θ,γ t) α
c1 −i∞ 1 Γ 1 − 2α t1

The Green function (14) can be extended to the negative values of the argument x due to the
symmetry relation:
gα,θ,γ (− x, t) = gα,−θ,γ ( x, t). (15)

As an extension of the pricing formula for the exponential models (10), we now define the price
of an option driven by the fractional diffusion model as follows:

+∞

Cα,θ,γ (S, K, r, μα,θ,γ , τ ) = e−rτ P (Se(r+μα,θ,γ )τ + x , K ) gα,θ,γ ( x, τ ) dx. (16)
−∞

Similarly to Equation (7), the risk-neutral parameter now reads:

+∞

μα,θ,γ = − log e x gα,θ,γ ( x, 1) dx. (17)
−∞

It follows from (17) that the Green function has to admit an exponential decay on the positive
semi-axis to ensure the existence of the risk-neutral parameter. This is the case as soon as the maximal
negative asymmetry (or skewness) holds (see the next section for calculations of μα,γ,θ in the space- and
time-fractional cases):
θ = α − 2, α > 1. (18)

105
Mathematics 2019, 7, 796

In what follows, we will denote the corresponding Green functions, the risk neutral parameters,
and the call option prices as follows:

gα,γ := gα,α−2,γ , μα,γ := μα,α−2,γ , Cα,γ := Cα,α−2,γ . (19)

2.2.2. Financial Interpretation of the Parameters

The fractional diffusion models incorporate two new degrees of freedom (the order of the spatial
derivative α and the order of the time derivative γ) that act as the risk redistribution parameters.
More precisely, while the parameter σ in the model (3) represents the volatility of the returns of
the underlying asset and, as such, has an equal impact on all kinds of options (an increase of σ leads to
an increase of the price of both call and put options), any changes of the parameters α and γ do not
affect all options in the same direction.
For instance, as observed in [45], if γ < 1, then the short-term options become more expensive
and the long-term ones less expensive compared to the non-fractional case γ = 1. This situation is
observable when the market conditions are far from equilibrium (dramatic price jumps, exceptional
events impacting the markets, etc.) and can be interpreted as a manifestation of memory [24]. The
parameter γ plays therefore the role of a temporal redistribution. In this paper, we will show how it
affects the profit and loss of a portfolio.
The parameter α represents a spatial redistribution because it controls the heavy-tails of the
probability distributions (see the discussion thereafter). Its impact has been extensively discussed
in [44], where it was shown to be an excellent candidate for the long-term volatility modeling. Indeed,
when the maturity increases, it is known that the volatility smirk does not flatten out as expected if the
Gaussian hypothesis would be true. By letting α vary between one and two, the negative slope of the
smirk can be controlled. It is flat when α = 2 (Gaussian case) and becomes steeper when α decreases,
thus generating any observable slope in equity index options markets (the impact of γ on the volatility
structure for at-the-money option was discussed in [47]). In this paper, we will also prove that the
parameter α governs the delta-hedging policy of the portfolios, that is the way of constructing portfolios
whose values are independent of the fluctuations in the stock price St .
Finally, let us mention that a calibration of the model parameters from the traded options for
a one-month trading period was suggested in [45] for the European calls and puts (together and
separately). The results of calibration showed that α fluctuated around 1.5 and 1.6 (therefore, rather
far from the Gaussian hypothesis, but close to the Mandelbrot estimates) and that γ, although close
to one, varied simultaneously with and in the same direction as α. This leads to a relative stability for
the diffusion scaling exponent Ω := γα .

2.3. Particular Cases

In this subsection, we discuss some well-known exponential models (3) that are particular cases
of the fractional diffusion model (16).

2.3.1. Finite Moment Log Stable model

In the framework of the Finite Moment Log Stable (FMLS) model [44], the stochastic process X (t)
driving the stock price dynamic (3) is given by:

X (t) = Lα,−1 (t), (20)

where Lα,β is a standardized Lévy-stable process (see [43]) with stability α and asymmetry (or skewness)
β, whose probability distribution is the so-called stable distribution gα,β [54]. As already mentioned, this
model was introduced by Carr and Wu in order to capture the behavior of the volatility smirk (the
phenomenon that, for a given maturity, implied volatility is higher for out-of-the-money puts than
for out-of-the-money calls): as a function of moneyness, it is widely observed that the smirk does not

106
Mathematics 2019, 7, 796

flatten out for longer observable horizons (i.e., τ greater than two years), and this can only be achieved
if one violates the Gaussian hypothesis. The maximal negative hypothesis β = −1 ensures the existence
of a risk-neutral parameter: indeed, the two-sided Laplace transform of the stable distribution exists
only if β = +1, with the result [55]:
  α πα
EP e− px = e− p sec 2 . (21)

Therefore, from a symmetry argument, the risk-neutral parameter exists and is finite only if
β = −1. It follows from (21) that its value is given by the formula:
 α
√σ
2
μα := μα,1 = , (22)
cos πα
2
√
where 2 was introduced as a normalization constant to recover the Black-Scholes factor for α = 2:

σ2
μ2 : = − . (23)
2
Moreover, under the condition β = −1, the probability distribution gα,−1 := gα has a fat tail on
the negative real axis as soon as 1 < α < 2 (that is, it decays as | x |−α ) and decays exponentially on
the positive real axis. This highly-skewed behavior could not be captured by a traditional Brownian
motion and by any symmetric distribution. However, it is a very plausible assumption as large drops
are far more commonly observed in financial markets than large rises.
The FMLS model is actually a particular case of the fractional diffusion model: indeed, the choice
β = −1 is the probabilistic equivalent of the choice θ = α − 2 (18), and the probability distribution gα
is the Green function associated with the space-fractional diffusion equation (γ = 1):

∂gα
( x, τ ) + μα α−2 D xα gα ( x, τ ) = 0. (24)
∂τ
From the pricing Formula (16), the option price in the framework of the FMLS model reads:

+∞

Cα,1 (S, K, r, μα , τ ) = e−rτ P (Se(r+μα )τ + x , K ) gα ( x, τ ) dx, (25)
−∞

where the Green function is given by the formula:

c1+i∞
1 1 Γ (1 − t1 ) t1
gα ( x, τ ) = x t1 −1 (−μα τ )− α dt1 . (26)
2πi α Γ (1 − t1
c1 −i∞ α)

Recently, an analytic option pricing formula for the FMLS model in the case of European options
was derived in [48] in the form of a fast convergent double series. In the present paper, we extend
this formula to the case of a time-fractional derivatives in (24), that is for γ = 1, and to other types
of options.

2.3.2. Black-Scholes Model

The celebrated Black-Scholes model [40] assumes that the stochastic process driving the
exponential model (3) is a standard Brownian motion, i.e., X (t) = W (t). This model turns out
to be a particular case of the FMLS model (and therefore, of the fractional diffusion model) because for
α = 2, the Lévy-stable process L2,β degenerates into W (t) for any β ∈ [−1, 1]. It is well known that

107
Mathematics 2019, 7, 796

the probability distribution of the Wiener process is the Green function associated with the diffusion
equation:
∂g σ 2 ∂2 g
( x, τ ) − ( x, τ ) = 0, (27)
∂τ 2 ∂x2
which is known to be the Gaussian kernel:
2
1 − x
g( x, τ ) = √ e 2σ2 τ . (28)
σ 2πτ

As expected, the diffusion Equation (27) is a particular case of the space-fractional diffusion
Equation (24) for α = 2 and with the Gaussian risk-neutral parameter (23). Using the pricing
Formula (16), the price of the option in the Black-Scholes model reads:

+∞
 2
σ2 ) τ + x 1 − x
C2,1 (S, K, r, σ, τ ) = e−rτ P (Se(r− 2 , K) √ e 2σ2 τ dx. (29)
σ 2πτ
−∞

For P (S, K ) = [S − K ]+ , some elementary manipulations on the integral (29) lead to the Black-Scholes
formula for the European options, which, in our system of notations, reads as follows:
 √   √ 
( E) log KS + rτ σ τ log KS + rτ σ τ
C2,1 (S, K, r, σ, τ ) = SN √ + − Ke−rτ N √ −
σ τ 2 σ τ 2
     (30)
k z k z
= F ek N + −N − .
z 2 z 2

where N (.) is the standard normal cumulative distribution function, and the forward strike price F and
the log-forward moneyness k are defined by the expressions:

S S
F := Ke−rτ , k := log = log + rτ. (31)
F K

3. Pricing Formulas

3.1. Risk-Neutral Parameter

In general, the risk-neutral parameter μα,γ is not known in the analytic form as soon as γ < 1.
However, it has been shown in [46] that μα,γ can be expressed in terms of a Mellin-Barnes integral
involving the FMLS risk-neutral parameter μα (which is known in explicit form):

Proposition 1. Let 1 < α ≤ 2 and 0 < γ < α. Then, for any c ∈ (0, 1), the formula:
⎡ ⎤
+i∞
c
1 1 Γ ( s ) Γ ( 1−
α )
s s −1
μα,γ = − log ⎣ μα ds⎦
α
(32)
2πi α Γ(γs + 1 − γ)
c−i∞

holds true.

For the proof, see [46] or [47].

Under some conditions on α and γ, the right-hand side of the integral representation (32) can be
expressed as a series:

Theorem 1. Let 1 < α ≤ 2. If 1 − 1

α < γ < α, then the series representation:
∞
(−1)n Γ(1 + αn) n
μα,γ = − log ∑ μ
n! Γ(1 + γαn) α
(33)
n =0

108
Mathematics 2019, 7, 796

is valid.

Proof. The Stirling asymptotic formula for the Gamma function (see [56]) leads to the following
statement (ak , bk , c j , d j ∈ R):
 
 Π Γ( a s + b )  |s|→∞  π π
 k k 
∑ k ∑ j
a − c < 0 ⇒ 
k
 −→ 0 when arg s ∈ − ,
 Π j Γ(c j s + d j )  2 2
. (34)
k j

Applying it to the integrand in the integral representation (32) under the condition 1 − α1 − γ < 0,
we can evaluate the integral along the vertical line by closing the contour in the right-half plane
because it will not contribute when |s| → ∞. In this region, the function Γ( 1−
α ) is singular every time
s
α(−1)n
its argument equals a negative integer −n, that is when s = 1 + αn, n ∈ N, with residue n! (see
the details in [56] or [57]). Therefore, we get the formula:
!
1 Γ ( s ) Γ ( 1−
α )
s s −1 (−1)n Γ(1 + αn) n
ress=1+αn μα α = μ . (35)
α Γ(γs + 1 − γ) n! Γ(1 + αγn) α

Summing all residues and applying the residue theorem complete the proof.

An important approximation to the risk-neutral series (33) is easily obtained via the Taylor series
for log(1 + u) :
Γ (1 + α )
μα,γ = μα + O(μ2α ). (36)
Γ(1 + γα)
Now, it follows from the reflection formula for the Gamma function that one can re-write the
FMLS risk-neutral parameter (22) as follows:
    
1 1−α 1+α σ α
μα = Γ Γ √ . (37)
π 2 2 2

Plugging the last formula into Formula (36), we have the following result:

Corollary 1. Let 1 < α < 2 and 1 − 1

α < γ < α. Then, the formula:
   
1− α 1+ α    
1 Γ (1 + α ) Γ 2 Γ 2 σ α
μα,γ = √ + O σ2α (38)
π Γ(1 + γα) 2

holds true.

In particular, in the case of the fractional Black-Scholes model (α = 2), we get the expression:

σ2
μ2,γ = − + O ( σ 4 ), (39)
Γ(1 + 2γ)
2
which resumes to the well-known Gaussian parameter − σ2 when γ = 1.

3.2. European Options

( E)
Let us denote by Cα,γ (S, K, r, μα,γ , τ ) the price of the European call option in the fractional
diffusion model with the payoff P ( E) (S, K ) = [S − K ]+ .

109
Mathematics 2019, 7, 796

Proposition 2. Let P be the polyhedron P := {(t1 , t2 ) ∈ C2 , Re(t2 − t1 ) > 1 , 0 < Re(t2 ) < 1} . Then,
for any vector c = (c1 , c2 ) ∈ P,

( E)
Cα,γ (S, K, r, μα,γ , τ ) =

Ke−rτ Γ(t2 )Γ(1 − t2 )Γ(−1 − t1 + t2 ) t1 dt dt
(−1)−t2 (−k − μα,γ τ )1+t1 −t2 (−μα,γ τ γ )− α 1
∧ 2.
α Γ(1 − γα t1 ) 2iπ 2iπ
c + i R2
(40)

Proof. First, we mention that under the maximal negative asymmetry hypothesis and on the positive
real axis, the Green function (14) has the form:

c1+i∞   t1
1 Γ (1 − t1 ) x dt1
gα,γ ( x, t) =   , 0 < c1 < 1. (41)
αx Γ 1 − γα t1 (−μα,γ t)
1
α 2iπ
c1 −i∞

On the other hand, using the notations (31), we can rewrite the payoff function in the form:
 +
P (E) (Se(r+μα,γ )τ + x , K ) = K ek+μα,γ τ + x − 1 (42)

then integrate by parts over the variable x, and get the expression:

c1+i∞ +
∞
( E) Ke−rτ Γ (1 − t1 ) t1 dt
Cα,γ (S, K, r, μα,γ , τ ) = − e(k+μα,γ τ + x) x1t dx   (−μα,γ τ γ )− α 1
. (43)
α t1 Γ 1 − γα t1 2iπ
c1 −i∞ −(k+μα,γ τ )

The Mellin-Barnes representation for the exponential term (see [58] or any other monograph on
integral transforms) reads:

c2+i∞
dt2
ek+μα,γ τ + x = (−1)−t2 Γ(t2 ) (k + μα,γ τ + x )−t2 , c2 > 0. (44)
2iπ
c2 −i∞

Inserting Formulas (41) and (44) into Formula (43) for the call price leads to the Beta integral:

+∞

Γ(1 − t2 )Γ(−1 − t1 + t2 )
(k + μα,γ τ + x )−t2 x t1 dx = (−k − μα,γ τ )1+t1 −t2 . (45)
Γ(−t1 )
−(k+μα,γ τ )

The relation t1 Γ(−t1 ) = −Γ(1 − t1 ) along with some elementary simplifications yields the integral
representation (40).

Let us now represent the Mellin-Barnes double integral (40) in terms of a double series by means
of residue summation.

Theorem 2 (Pricing formula: European call). Let 1 < α ≤ 2 and 1 − α1 < γ ≤ α. Then, under the maximal
negative asymmetry hypothesis (θ = α − 2), the price of the European call driven by the space-time fractional
diffusion equation is as follows:
∞
( E) Ke−rτ 1 γ m− n
Cα,γ (S, K, r, μα,γ , τ ) =
α ∑ n!Γ(1 + γ m− )
n
n ( k + μα,γ τ ) (− μα,γ τ )
α . (46)
n =0 α
m =1

110
Mathematics 2019, 7, 796

Proof. Let ω denote the differential form in the integral at the right-hand side of (40). If we perform
the change of the variables: 
u1 : = −1 − t1 + t2 ,
(47)
u2 : = t2
then ω reads:

Γ ( u1 ) Γ ( u2 ) Γ (1 − u2 ) −1− u1 + u2 du1 du2

ω = (−1)−u2 (−k − μα,γ τ )−u1 (−μα,γ τ γ )− α ∧ . (48)
Γ(1 − γ −1−uα1 +u2 ) 2iπ 2iπ

In the region { Re(u1 ) < 0, Re(u2 ) < 0}, ω has a simple pole at every point (u1 , u2 ) = (−n, −m),
n, m ∈ N with the residue:

1 1+ m − n
Res(u1 =−n,u2 =−m) ω = (k + μα,γ τ )n (−μα,γ τ γ ) α . (49)
n!Γ(1 + γ 1+mα −n )

Performing the change of indexation m → m + 1 and summing all residues yields the double
sum (46). The fact that one can close a contour in this region of C2 is a consequence of a two-dimensional
generalization of the Stirling Formula (34) (see [46,47] for technical details).

The pricing Formula (46) is a simple and efficient way for the calculation of the pricing of the
European call options. The convergence of the partial sums in the double series (46) is very fast,
and therefore, only a few terms are needed to obtain an excellent level of precision (see numerical
applications and convergence tests in [46,47]). As a particular case of the pricing Formula (46) with
γ = 1, we recover the pricing formula for the FMLS model that was established in [48]:
∞
( E) Ke−rτ 1 m−n
Cα,1 (S, K, r, μα , τ ) =
α ∑ n!Γ(1 + m−n (k + μα τ )n (−μα τ ) α . (50)
n =0 α )
m =1

If we set α = 2 in the last formula, then we obtain the series expansion for the Black-Scholes
European call:

Ke−rτ ∞
1  n
( E)
C2,1 (S, K, r, σ, τ ) = ∑ m−n k − Z2 Z m−n , (51)
α n =0 n!Γ(1 + 2 )
m =1
√
where Z := σ√ τ . In [59], using the change of variables j = m + n and the properties of the Gamma
2
function, it was proven that (51) can be ordered in odd powers of Z:
 n
( E) 1 Ke−rτ (−1)n k
C2,1 (S, K, r, σ, τ ) = (S − Ke−rτ ) + ∑ Z2j+1 1− . (52)
2 2 j ≥0 n!Γ( 32 + j − n) Z2
n≤2j

A particularly interesting situation occurs when the asset is At-The-Money (ATM) forward, that
is when S = Ke−rτ or equivalently with our notations k = 0 in Equation (46). In this case, we get the
following result:

111
Mathematics 2019, 7, 796

Corollary 2 (At-the-money price: European call). For S = Ke−rτ , the price of the European call option
driven by the space-time fractional diffusion equation is given by the formula:
∞
( E,ATM) S (−1)n m+(α−1)n
Cα,γ (S, μα, γ , τ ) = ∑ γ
n (− μα,γ τ )
α τ (1− γ ) n (53)
α n =0 n!Γ(1 + γ m−
α )
m =1
!
S
1 γ
(−μα,γ ) α τ α
2 2γ
(−μα,γ ) α τ α  
1+ α1 1+ γα
= + μγ,α τ + + O (−μα,γ ) τ .
α Γ(1 + γα ) Γ(1 + 2γ α )

As a particular case of (53), the ATM price in the FMLS model reads:
∞
( E,ATM) S (−1)n m+(α−1)n
Cα,1 (S, μα , τ ) =
α ∑ n!Γ(1 + m− n (− μτ )
)
α (54)
n =0 α
m =1

as was already established in [48]. When α = 2, we get:

√ !
( E,ATM) S 1 σ τ √ 3 1 √ √
C2,1 (S, σ, τ ) = √ + O((σ τ ) ) = √ Sσ τ + O((σ τ )3 ) . (55)
2 Γ( 32 ) 2 2π

As √1 0.4, we have thus recovered the well-known Brenner–Subrahmanyam approximation

2π
for the European Black-Scholes call (which was first introduced in [60]):

( E,ATM) √
C2,1 (S, σ, τ ) 0.4Sσ τ . (56)

3.3. Binary (or Digital) Options

3.3.1. Cash-or-Nothing
(C/N )
Let us denote by Cα,γ (S, K, r, μα,γ , τ ) the price of the binary cash-or-nothing option in the
fractional diffusion model with the payoff P (C/N ) (S, K ) = H (S − K ).

Proposition 3. For any c1 > 0, the formula:

c1+i∞
(C/N ) e−rτ Γ(−t1 ) t1 dt
Cα,γ (S, K, r, μα,γ , τ ) = γ (−k − μα,γ τ )t1 (−μα,γ τ γ )− α 1
(57)
α Γ (1 − α t1 ) 2iπ
c1 −i∞

holds true.

Proof. Inserting the terminal payoff P (C/N ) (Se(r+μα,γ )τ + x , K ) into the call price (16) yields:

c1+i∞ +∞

(C/N ) e−rτ Γ (1 − t1 ) t1 dt
Cα,γ (S, K, r, μα,γ , τ ) = x t1 −1 dx γ (−μα,γ τ γ )− α 1
. (58)
α Γ (1 − α t1 ) 2iπ
c1 −i∞ −(k +μα,γ τ )

Performing the integration with respect to the x-variable and using the relation −t1 Γ(1 − t1 ) =
Γ(−t1 ) yield the representation (57).

112
Mathematics 2019, 7, 796

Theorem 3 (Pricing formula: cash or nothing call). Let 1 < α ≤ 2 and 1 − α1 < γ ≤ α. Then, under
the maximal negative asymmetry hypothesis (θ = α − 2), the price of the cash-or-nothing call option driven by
the space-time fractional diffusion equation is given by the formula:
∞
(C/N ) e−rτ 1
(k + μα,γ τ )n (−μα,γ τ γ )− α .
n
Cα,γ (S, K, r, μα,γ , τ ) =
α ∑ n!Γ(1 − γα n)
(59)
n =0

Proof. As −1 + γα ≤ 0, it follows from Formula (34) that the line integral in (57) can be expressed as a
sum of the residues induced by the poles of the Γ(−t1 ) function. These poles are located at the points
t = +n, and the associated residues are as follows:

(−1)n
(−k − μα,γ τ )n (−μα,γ τ γ )− α .
n
(60)
n!Γ(1 − γα n)

Summing up all residues completes the proof.

In the particular case of the Black-Scholes model (α = 2,γ = 1), the series (59) has only the odd
terms n = 2p + 1 because of the divergence of the denominator at negative integers. Using the known
values of the Gamma function at the half-integers, we then obtain the formula
"   !
(C/N ) e−rτ 2 +∞ (−1) p k 1 √ 2p+1
π p∑
C2,1 (S, K, r, σ, τ ) = 1+ √ − σ τ
α =0 2 p! (2p + 1) σ τ
p 2
 
k 1 √
= e−rτ N √ − σ τ . (61)
σ τ 2

(C/N )
Let us now consider the cash-or-nothing put option Pα,γ (S, K, r, μα,γ , τ ).

Proposition 4. For any c1 ∈ (−α, 0), the integral representation:

(C/N )
Pα,γ (S, K, r, μα,γ , τ ) =
c1+i∞
e−rτ Γ( tα1 )Γ(1 − tα1 )Γ(−t1 ) t1 dt
 γ  α −1 α −1
(k + μα,γ τ )t1 (−μα,γ τ γ )− α 1
(62)
α Γ 1 − α t1 Γ ( α t1 ) Γ (1 − α t1 ) 2iπ
c1 −i∞

holds true.

Proof. The proof is similar to the one given for the proposition 3. The only difference is that one has to
replace the payoff by P (C/N ) (S, K ) = H (K − S) and to consider the Green function on the negative
real axis. Using the symmetry property (15), it reads as follows:

c1+i∞   t1
1 Γ( tα1 )Γ(1 − tα1 )Γ(1 − t1 ) x dt1
gα,γ ( x, t) =   . (63)
αx Γ 1 − γα t1 Γ( α− α −1
α t1 ) Γ (1 − α t1 )
1 1
(−μα,γ t) α 2iπ
c1 −i∞

The integral for the put price (62) needs more efforts than the integral for the call price (57).
Indeed, it possesses two distinct series of poles in the positive half-plane:
t1
• The poles of Γ(1 − α) at the points t1 = α(1 + n), n ∈ N, whose residues are given by the formula:

(−1)n Γ(−α(1 + n))

(k + μα,γ τ )α(1+n) (−μα,γ τ γ )−(1+n) . (64)
n!Γ(1 − γ(1 + n))Γ((α − 1)(1 + n))Γ(1 − (α − 1)(1 + n))

113
Mathematics 2019, 7, 796

• The poles of Γ(−t1 ) at the points t1 = n, n ∈ N, whose residues are given by the formula:

(−1)n Γ( nα )Γ(1 − nα )
(k + μα,γ τ )n (−μα,γ τ γ )− α .
n
(65)
n!Γ(1 − γα n)Γ( α−
α n ) Γ (1 −
1 α −1
α n)

However, in the non-time-fractional case, it is possible to derive a simple representation for the
put price.

Theorem 4 (Pricing formula: space fractional cash or nothing put). Let 1 < α ≤ 2. Then, under the
maximal negative asymmetry hypothesis (θ = α − 2), the price of the cash-or-nothing call option driven by the
space fractional diffusion equation is as follows:
!
∞
(C/N ) e−rτ 1 − nα
Pα,1 (S, K, r, μα , τ ) =
α
( α − 1) − ∑ n!Γ(1 − nα )
n
(k + μα τ ) (−μα τ ) . (66)
n =1

Proof. In the space-fractional case, the parameter γ is equal to one. Then, we have the following
simplifications:

• In the denominator of (64), Γ(1 − γ(1 + n)) reduces to Γ(−n), which is infinite for any n ∈ N.
Therefore, (64) is equal to null for any n ∈ N.
• (65) simplifies to the form:

(−1)n Γ( nα )
(k + μα τ )n (−μα τ )− α .
n
α −1 α −1
(67)
n!Γ( α n)Γ(1 − α n)

Both the numerator and the denominator of (67) are singular for n = 0, but their quotient is not
singular:
α
Γ( nα )
∼ n
α = α − 1. (68)
Γ ( α−
α n)
1 n →0
( α −1) n
Γ( n )
• Finally, using the functional relation Γ(1 − n + nα ) = (−1)n−1 (1− n )(2− n )α...(n−1− n ) , we can simplify
α α α
Expression (67):

(−1)n Γ( nα ) 1
(k + μα τ )n (−μα τ )− α (k + μα τ )n (−μα τ )− α
n n
= − (69)
n!Γ( α n)Γ(1 − α−
α −1
α
1
n ) n!Γ(1 − nα )

and the proof is complete.

Note that in the space fractional case, the call Formula (59) can be written in the form:
!
∞
(C/N ) e−rτ 1
(k + μα τ )n (−μα τ )− α
n
Cα,1 (S, K, r, μα , τ ) =
α
1+ ∑ n!Γ(1 − γα n)
(70)
n =1

and therefore, the sum of the space fractional call and put options is given by the formula:

(C/N ) (C/N ) e−rτ

Cα,1 (S, K, r, μα , τ ) + Pα,1 (S, K, r, μα , τ ) = [1 + (α − 1)] = e−rτ . (71)
α

114
Mathematics 2019, 7, 796

The relation (71) is an example of a call-put parity relation. The fact, that it is independent of α
is not a surprise since, as already mentioned in the previous section, the space fractional model is
equivalent to the FLMS model for which the option prices admit a risk-neutral representation:

(C/N ) (C/N )
Cα,1 (S, K, r, μα , τ ) + Pα,1 (S, K, r, μα , τ ) = e−rτ EQ [ H (S − K ) + H (K − S)] (72)
−rτ Q
= e E [1] (73)
−rτ
= e . (74)

Of course, it does not hold true for γ = 1 because no probabilistic interpretation exists in this case.
Finally, we note that it follows from the Formulas (71) and (61) that in the Black-Scholes case (α = 2,
γ = 1), we have the relation:
  
(C/N ) k 1 √
P2,1 (S, K, r, σ, τ ) = e−rτ 1 − N √ − σ τ (75)
σ τ 2
 
k 1 √
= e−rτ N − √ + σ τ (76)
σ τ 2

as predicted by the conventional Black-Scholes theory.

3.3.2. Asset-or-Nothing
( A/N )
Let us denote by Cα,γ (S, K, r, μα,γ , τ ) the price of the binary cash-or-nothing option in the
fractional diffusion model with the payoff P ( A/N ) (S, K ) = S × H (S − K ).

Proposition 5. Let P be the polyhedron P := {(t1 , t2 ) ∈ C2 , Re(t2 − t1 ) > 1 , 0 < Re(t2 ) < 1} . Then, for
any vector c = (c1 , c2 ) ∈ P, the integral representation:

( A/N )
Cα,γ (S, K, r, μα,γ , τ ) =

Ke−rτ Γ(t2 )Γ(1 − t2 )Γ(−t1 + t2 ) t1 dt dt
(−1)−t2 (−k − μα,γ τ )t1 −t2 (−μα,γ τ γ )− α 1
∧ 2 (77)
α Γ(1 − γα t1 ) 2iπ 2iπ
c + i R2

holds true.

Proof. To prove the formula, we replace the payoff function in the option price Formula (16) by:

P ( A/N ) (Se(r+μα,γ )τ + x , K ) = S e(r+μα,γ )τ + x H (Se(r+μα,γ )τ + x − K ) (78)

k +μα,γ τ + x
= Ke 1{ x≥−k−μα,γ τ } (79)

and then proceed exactly as in the proof of Proposition 2.

Theorem 5 (Pricing formula: asset or nothing call). Let 1 < α ≤ 2 and 1 − α1 < γ ≤ α. Then, under the
maximal negative asymmetry hypothesis (θ = α − 2), the price of the asset-or-nothing call option driven by the
space-time fractional diffusion equation is given by the formula:
∞
( A/N ) Ke−rτ 1 γ m− n
Cα,γ (S, K, r, μα,γ , τ ) =
α ∑ n!Γ(1 + γ m− )
n
n ( k + μα,γ τ ) (− μα,γ τ )
α . (80)
n =0 α
m =0

Proof. The series (80) is obtained by summing the residues associated with the poles of the functions
Γ(t2 ) and Γ(−t1 + t2 ) exactly in the same manner as in the proof of Theorem 2.

115
Mathematics 2019, 7, 796

As a consequence of Theorems 2, 3 and 5, the European call can be represented as a difference

between an asset-or-nothing call and a cash-or-nothing call:

( E) ( A/N ) (C/N )
Cα,γ (S, K, r, μα,γ , τ ) = Cα,γ (S, K, r, μα,γ , τ ) − K Cα,γ (S, K, r, μα,γ , τ ). (81)

4. Risk Sensitivities and Portfolio Hedging

In quantitative finance, the risk sensitivities are also known as the “Greeks”. They quantify
the dependence of an option on the market parameters such as the asset (spot) price or volatility
and are essential tools for portfolio management. In this section, we derive for them some efficient
representations and study how the orders of the time- and space-fractional derivatives impact the
hedging policies.

4.1. First Order Sensitivity (Delta)

4.1.1. European Call

∂k
From the definition of k, we have the relation ∂S = S1 . Thus, by differentiation of (46) with respect
to S and re-arranging the terms, we obtain the formula:

( E)
∂Cα,γ
Δ ( E) (S, K, r, μα,γ , τ ) := (S, K, r, μα,γ , τ )
Cα,γ ∂S
e−k ∞ 1 γ m− n

α n∑
n
= m−n ( k + μγ,α τ ) (− μγ,α τ )
α , (82)
=0 n!Γ (1 + γ α )
m =0

−rτ
where we have used the definition (31) for the moneyness to deduce the relation KeS = e−k . When
the asset is ATM forward (S = Ke−rτ , and therefore, k = 0), Formula (82) can be simplified:
∞
1 (−1)n (α−1)n+m (α−γ)n+γm
Δ ( E,ATM) (S, K, r, μα,γ , τ ) =
α ∑ n!Γ(1 + γ m− n (− μα,γ )
α τ α

α )
Cα,γ
n =0
m =0
1 γ γ !
(−μα,γ )1− α τ 1− α
1
1 (−μα,γ ) α τ α
= 1+ − + O (− μ α,γ τ ) . (83)
α Γ(1 + γα ) Γ(1 − γα )

Equation (83) shows that the Delta of at-the-money calls is driven by α1 (see also Figure 1), which
generalizes the well-known feature of the Black-Scholes model, where at-the-money call options have
a Delta of 12 . In particular, (83) demonstrates that, at first order, only the space fractional parameter α
influences the delta-hedging policy of a portfolio. Indeed, it is sufficient to be long of one unit of the
asset S and short of α units of an European call to offset the impact of the variations of S:

( E,ATM) ∂Π 
Π := S − αCα,γ (S, K, r, μα,γ , τ ) =⇒ = 0, (84)
∂S S=S0

where S0 = Ke−rτ is the ATM forward price.

116
Mathematics 2019, 7, 796

Figure 1. In the left figure, we plot the evolution of the Delta of an European call option as a function of
S for various values of γ. We observe that, as expected from (83), the At-The-Money (ATM) call price is
approximately equal to α1 for any γ, and moreover, the Delta value is the same on a wide range of prices
around S0 , independently of γ. The situation is very different on the right figure (plot of the Gamma
of the European call option (93)), where sensitivities vary greatly in dependence of γ. To produce the
figures, the following values of parameters were used: K = 4000, r = 1%, σ = 20%, τ = 1Y, α = 2.

4.1.2. Cash-or-Nothing Call

Differentiation of Expression (59) with respect to S leads to the formula:

∞
e−k 1 n +1
Δ (C/N ) (S, K, r, μα,γ , τ ) = ∑ (k + μα,γ τ )n (−μα,γ τ γ )− α . (85)
Cα,γ αK n =0 n!Γ(1 − γα (n + 1))

In the at-the-money forward situation, we then get:

1
(−μα,γ τ γ )− α ∞
(−1)n α −1 α−γ
Δ (C/N,ATM)
Cα,γ
(S, K, r, μα,γ , τ ) =
αK ∑ n!Γ(1 − γα (n + 1))
(−μα,γ ) α n τ α n
n =0
⎡ ⎤
1
(−μα,γ τ γ )− α ⎣ 1
1 γ
(−μα,γ )1− α τ 1− α  
2− α2 2− 2γ ⎦.
= − + O (−μα,γ ) τ α (86)
αK Γ(1 − γα ) Γ(1 − 2γ α )

This formula can be simplified when γ → α (the so-called neutral diffusion). Using the known
asymptotic behavior of the Gamma function at the point 0, we then get:

− 1
α − γ (−μα,γ τ γ ) α
Δ (C/N,ATM) (S, K, r, μα,γ , τ ) ∼ . (87)
Cα,γ γ→α α 2 K

Another prominent particular case of Formula (86) is for α = 2 and γ = 1 (the Black-Scholes
model). In this case, (86) reduces to the form:

1 1 √
Δ (C/N,ATM) (S, K, r, σ, τ ) = √ √ + O ( σ τ ). (88)
C2,1 K 2π σ τ

4.1.3. Cash-or-Nothing Put

In the space-fractional case (γ = 1), it follows from the parity relation (71) that the cash-or-nothing
call and put options have opposite Deltas:

Δ (C/N ) (S, K, r, μα , τ ) = − Δ (C/N ) (S, K, r, μα , τ ) (89)

Pα Cα
e−k ∞ 1 n +1
= − ∑ (k + μα,γ τ )n (−μα,γ τ γ )− α . (90)
αK n =0 n!Γ(1 − γα (n + 1))

117
Mathematics 2019, 7, 796

4.1.4. Asset-or-Nothing Call

Differentiation of the relation (81) with respect to S along with Formulas (82) and (85) lead to the
following expression for the Delta of the asset-or-nothing call:

Δ ( A/N ) (S, K, r, μα,γ , τ ) = Δ ( E) (S, K, r, μα,γ , τ ) + K Δ (C/N ) (S, K, r, μα,γ , τ ) (91)

Cα,γ Cα,γ Cα,γ
∞
e−k 1 m − n −1
= ∑ (k + μγ,α τ )n (−μγ,α τ γ ) α . (92)
α n =0 n!Γ(1 + γ m−αn−1 )
m =0

4.2. Second Order Sensitivity (Gamma, Dollar Gamma)

−k −k
By the definition of k, ∂e∂S = ∂S
∂k −k
e = e S . Thus, differentiation of (82) with respect to S along
with a re-arrangement of the terms give the formula:

( E)
∂2 Cα,γ
Γ ( E) (S, K, r, μα,γ , τ ) := (S, K, r, μα,γ , τ )
Cα,γ ∂S2
m − n −1 m−n
!
∞ (−μγ,α τ γ ) α (−μγ,α τ γ ) α
e−k (k + μγ,α τ )n
=
αS ∑ m − n −1
−
Γ (1 + γ m −
. (93)
α )
n
n =0 Γ (1 + γ α ) n!
m =0

In the ATM forward situation, (93) is reduced to the form:

Γ ( E,ATM) (S, K, r, μα,γ , τ ) =

Cα,γ
⎡ ⎤
( α −1) n + m −1 ( α − γ ) n + γ ( m −1) (α−1)n+m (α−γ)n+γm
∞
1 (−1)n ⎣ (−μα,γ ) α τ α (−μα,γ ) α τ α
⎦ . (94)
αS ∑ Γ(1 + γ m−αn−1 )
−
Γ (1 + γ m −
α )
n! n
n =0
m =0

The leading term in the expression at the right-hand side of (94) is the following one:

1 (−μα,γ τ γ )− α
1

Γ ( E,ATM) (S, μα,γ , τ ) = . (95)
Cα,γ αS Γ(1 − γα )

Let us assume that we are long of one European call option and that we have delta-hedged our
portfolio when S = S0 (according to (84), this is made by being short of α1 units of the asset S). We can
then employ the Taylor formula to approximate the value of the portfolio when S varies:

( E) ( E,ATM) 1
Cα,γ (S, K, r, μα,γ , τ ) − Cα,γ (S0 , K, r, μα,γ , τ ) =
Γ (E,ATM) (S0 , σ, τ )2 (S − S0 )2 + O(S − S0 )3 .
2 Cα,γ
(96)
The left-hand side of (96) is the Profit and Loss, or P& L of the portfolio around the money. As we
have delta-hedged our position, this P&L is therefore essentially driven by the option’s convexity, and
one speaks of Gamma P& L. For α = 2, we can use the approximation (39) of the risk-neutral parameter
and its Taylor expansion with respect to γ to obtain the following easy-to-use P& L formulas (γE
denotes the Euler–Mascheroni constant):

• If γ → 0:
 2  
1 − 3γE γ S − S0
P&L(γ → 0) = + O γ 2 ( S − S0 ) 3 . (97)
8σ2 τ γ S0
• If γ → 1:
 2  
1 − (−3 + 3γE + log 4)(γ − 1) S − S0
P&L(γ → 1) = + O ( γ − 1 ) 2 ( S − S0 ) 3 . (98)
4πσ2 τ γ S0

118
Mathematics 2019, 7, 796

• If γ → 2:
 2  
3( γ − 2)2 S − S0
P&L(γ → 2) = + O ( γ − 2 ) 3 ( S − S0 ) 3 . (99)
4σ2 τ γ S0
 2
The term S− S0
S0 is the square of the underlying asset’s returns and, as such, is interpreted as
the realized variance. Its multiplicative factor is often called the dollar Gamma ($Γ). By definition, with
our notations, it reads:
1 
$Γ = S02 Γ ( E,ATM) ( S0 , μ α,γ , τ ) .
2
(100)
2 Cα,γ

The Gamma P&L can therefore be written down in the form:

P&L = $Γ × realized variance. (101)

From Equations (97)–(99), we obtain the following particular values for the dollar Gamma:
⎧
⎪ 1
⎪
⎪ γ = 0: $Γ = + O ( γ ),
⎪
⎨ 8σ2
1 (102)
⎪
⎪ γ = 1: $Γ = (the Black-Scholes Dollar Gamma),
⎪
⎪ 4πσ2 τ
⎩
γ =2: $Γ = 0.

We can observe that, for any γ, the graph of the dollar Gamma remains positive and therefore
works in favor of the long call position. In other words, the P&L of the call option will outperform the
one of the hedge, because the option price is a convex function of the asset price. Thus, this well-known
feature of the conventional Black-Scholes theory is preserved in the framework of the time-fractional
diffusion model. The presence of the parameter γ, however, significantly affects the impact of option’s
maturity to the Gamma P&L. As seen in (102), the Black-Scholes dollar Gamma is equal to 4πσ1 2 τ and
therefore is maximal for short-term options (i.e., for small τ). However, when γ tends to zero, then
$Γ tends to 8σ1 2 and becomes independent of τ. Similarly, when γ → 2, the option will no longer
outperform the hedge, and the P&L of the portfolio will remain flat, whatever the option’s maturity.
This phenomenon illustrates the temporal redistribution induced by the time-fractional parameter γ.
Finally, Figure 2 demonstrates that the Gamma P&L is a decreasing function of γ as soon as τ ≥ π2 ,
and there is an inflection at the point γ = 1:

∂2 ($Γ) 
= 0. (103)
∂γ γ=1
2

119
Mathematics 2019, 7, 796

Figure 2. The dollar Gamma for the time-fractional diffusion model (α = 2,γ ∈ (0, 2]) maximizes the
P&L when γ → 0 and offsets the impact of maturity τ at extremal values (γ = 0, γ = 2). The graph
of the Dollar Gamma possesses an inflection point at the point γ = 1 (the Black-Scholes model).
To produce the figure, the following values of parameters were used: σ = 20%, τ = 1Y.

5. Conclusions
In this article, we presented a theory of option pricing based on the fractional diffusion
equation and showed that it constituted a generalization of the well-known class of the exponential
market models. In particular, we extended the pricing formulas that were previously established for
the vanilla options to a basic class of exotic options, computed the related risk sensitivities, and applied
the results to delta-hedging and P&L calculations.
The pricing formulas were derived in the form of fast converging series of powers of the
log-forward moneyness and of the risk-neutral parameter that generalizes the volatility parameter to
the non-Gaussian case and also admits a convenient series representation. These series can be easily
used for calculations in practice without the help of any sophisticated numerical tools. Moreover,
they clearly exhibit the impact of the model parameters to risks and hedging: the order α of the space
derivative governs the delta-hedging of the portfolio, while the order γ of the time derivative drives
the P&L of this portfolio.
Other important problems include the development and analysis of similar analytic tools for
other types of commonly-traded payoffs (especially for the path-dependent options) and other types
of financial derivatives. We also are going to apply these models to the real market data. In order
to verify the effectiveness of our approach, we first need to determine the reliable estimations of the
model parameters and then to compare the numerical results produced by the model with the market
data. Another interesting problem concerns relaxing of the maximal asymmetry hypothesis. However,
this would imply divergence of the risk-neutral expectations, and thus, some suitable re-normalization
procedures such as, say, the cut-offs of the probability densities, would be probably needed to make
progress in the solution of this problem.

Author Contributions: J.-P.A. did the theoretical analysis and calculations. All authors wrote the paper.
Funding: J.K. was supported by the Austrian Science Fund (FWF) under Project I 3073 and by the Czech Science
Foundation under Grant 19-16066S.
Conflicts of Interest: The authors declare no conflict of interest.

120
Mathematics 2019, 7, 796

References
1. Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives: Theory and Applications;
Gordon and Breach: New York, NY, USA, 1993.
2. Kochubei, A.; Luchko, Y. (Eds.) Handbook of Fractional Calculus with Applications. Volume 1: Basic Theory;
De Gruyter: Berlin, Germany, 2019.
3. Kochubei, A.; Luchko, Y. (Eds.) Handbook of Fractional Calculus with Applications. Volume 2: Fractional
Differential Equations; De Gruyter: Berlin, Germany, 2019.
4. Tarasov, V. (Ed.) Handbook of Fractional Calculus with Applications. Volume 4: Applications in Physics, Part A;
De Gruyter: Berlin, Germany, 2019.
5. Tarasov, V. (Ed.) Handbook of Fractional Calculus with Applications. Volume 5: Applications in Physics, Part B;
De Gruyter: Berlin, Germany, 2019.
6. Petras, I. (Ed.) Handbook of Fractional Calculus with Applications. Volume 6: Applications in Control; De Gruyter:
Berlin, Germany, 2019.
7. Baleanu, D.; Mendes Lopes, A. (Eds.) Handbook of Fractional Calculus with Applications. Volume 7: Applications
in Engineering, Life and Social Sciences, Part A; De Gruyter: Berlin, Germany, 2019.
8. Baleanu, D.; Mendes Lopes, A. (Eds.) Handbook of Fractional Calculus with Applications. Volume 8: Applications
in Engineering, Life and Social Sciences, Part B.; De Gruyter: Berlin, Germany, 2019.
9. Mainardi, F. Fractional Calculus and Waves in Linear Viscoelasticity; Imperial College Press: London, UK, 2010.
10. Klages, R.; Radons, G.; Sokolov, I.M. Anomalous Transport: Foundations and Applications; Wiley-VCH: Hoboken,
NJ, USA, 2008.
11. Jin, B.; Rundell, W. A tutorial on inverse problems for anomalous diffusion processes. Inverse Probl. 2015,
31, 035003. [CrossRef]
12. Luchko, Y. Anomalous Diffusion: Models, Their Analysis, and Interpretation. In Advances in Applied Analysis;
Rogosin, S.V., Koroleva, A.A., Eds.; Birkhäuser: Basel, Switzerland, 2012; pp. 115–146.
13. Metzler, R.; Klafter, J. The restaurant at the end of the random walk: Recent developments in the description
of anomalous transport by fractional dynamics. Phys. A Math. Gen. 2004, 37, R161–R208. [CrossRef]
14. Schneider, W.R.; Wyss, W. Fractional diffusion and wave equations. J. Math. Phys. 1989, 30, 134–144.
[CrossRef]
15. Mainardi, F.; Luchko, Y.; Pagnini, G. The fundamental solution of the space-time fractional diffusion equation.
Fract. Calc. Appl. Anal. 2001, 4, 153–192.
16. Helbing, D.; Brockmann, D.; Chadefaux, T. Saving Human Lives: What Complexity Science and Information
Systems can Contribute. J. Stat. Phys. 2015, 158, 735. [CrossRef]
17. Perc, M.; Ozer, M.; Hojnik J. Social and juristic challenges of artificial intelligence. Palgrave Commun. 2019,
5, 61. [CrossRef]
18. Fallahgoul, H.; Focardi, S.; Fabozzi, F. Fractional Calculus and Fractional Processes with Applications to Financial
Economics: Theory and Application; Academic Press: Cambridge, MA, USA, 2016.
19. Gorenflo, R.; Mainardi, F.; Moretti, D.; Paradisi, P. Time fractional diffusion: a discrete random walk approach.
Nonlinear Dyn. 2002, 29, 129–143. [CrossRef]
20. Kerss, A.; Leonenko, N.N.; Sikorskii, A. Fractional Skellam processes with applications to finance. Fract. Calc.
Appl. Anal. 2014, 17, 532–551. [CrossRef]
21. Scalasa, E.; Gorenflo, R.; Mainardi, F. Fractional calculus and continuous-time finance. Phys. A 2000, 284,
376–384. [CrossRef]
22. Tarasov, V.E.;Tarasova, V.V. Macroeconomic models with long dynamic memory: Fractional calculus approach.
Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
23. Tarasov, V.E. On history of mathematical economics: Application of fractional calculus. Mathematics 2019,
7, 509. [CrossRef]
24. Tarasova, V.V.; Tarasov, V.E. Concept of dynamic memory in economics. Commun. Nonlinear Sci. 2018, 55,
127–145. [CrossRef]
25. Tejado, I.; Pérez, E.; Valério, D. Fractional calculus in economic growth modeling of the group of seven. Fract.
Calc. Appl. Anal. 2019, 22, 139–157. [CrossRef]
26. Elliott, R.J.; van der Hoek, J. A general fractional white noise theory and applications to finance. Math. Financ.
2003, 13, 301–330. [CrossRef]

121
Mathematics 2019, 7, 796

27. Longjin L.; Ren, F.-Y.; Qiu, W.-Y. The application of fractional derivatives in stochastic models driven by
fractional Brownian motion. Phys. A 2010, 389, 4809–4818. [CrossRef]
28. Vilela Mendes, R. A fractional calculus interpretation of the fractional volatility model. R. Nonlinear Dyn.
2009, 55, 395. [CrossRef]
29. Necula, C. Option Pricing in a Fractional Brownian Motion Environment. 2002. Available online: https:
//ssrn.com/abstract=1286833 (accessed on 29 August 2019).
30. Akrami, M.H.; Erjaee, G.H. Examples of analytical solutions by means of Mittag-Leffler function of fractional
Black-Scholes option pricing equation. Fract. Calc. Appl. Anal. 2015, 18, 38–47. [CrossRef]
31. Gong, X.; Zhuang, X. American option valuation under time changed tempered stable Lévy processes. Phys.
A 2017, 466, 57–68. [CrossRef]
32. Wang, X.T. Scaling and long-range dependence in option pricing I: Pricing European option with transaction
costs under the fractional Black-Scholes model. Phys. A 2010, 389, 438–444. [CrossRef]
33. Yavuz, M.; Necati, Ö. European Vanilla Option Pricing Model of Fractional Order without Singular Kernel.
Fractal Fract. 2018, 2, 3. [CrossRef]
34. Cartea A.; del-Castillo-Negrete, D. Fractional diffusion models of option prices in markets with jumps.
Phys. A 2007, 374, 749–763. [CrossRef]
35. Chen, W.; Xu, X.; Zhu, S.P. Analytically pricing double barrier options based on a time-fractional
Black-Scholes equation. Comput. Math. Appl. 2015, 69, 1407–1419. [CrossRef]
36. Xiao, W.L.; Zhang, W.G.; Zhang, X.L.; Wang, Y.L. Pricing currency options in a fractional Brownian motion
with jumps. Econ. Model. 2010, 27, 935–942. [CrossRef]
37. Koleva, M.N.; Vulkov, L.G. Numerical solution of time-fractional Black-Scholes equation. Comput. Appl.
Math. 2017, 36, 1699–1715. [CrossRef]
38. Song, L.; Wang, W. Solution of the fractional Black-Scholes option pricing model by finite difference method.
Abstr. Appl. Anal. 2013, 2013, 194286. [CrossRef]
39. Funahashi, H.; Kijima, M. A solution to the time-scale fractional puzzle in the implied volatility. Fractal Fract.
2017, 1, 14. [CrossRef]
40. Black, F.; Scholes, M. The Pricing of Options and Corporate Liabilities. J. Polit. Econ. 1973, 81, 637–654.
[CrossRef]
41. Heston, S.L. A Closed-Form Solution for Options with Stochastic Volatility with Applications to Bond and
Currency Options. Rev. Financ. Stud. 1993, 6, 327–343. [CrossRef]
42. Duan, J.C.; Popova, I.; Ritchken, P. Option pricing under regime switching. Quant. Financ. 2002, 2, 209.
[CrossRef]
43. Cont, R.; Tankov, P. Financial Modelling with Jump Processes; Chapman & Hall: New-York City, NY, USA, 2004.
44. Carr, P.; Wu, L. The Finite Moment Log Stable Process and Option Pricing. J. Financ. 2003, 58, 753–778.
[CrossRef]
45. Kleinert, H.; Korbel, J. Option pricing beyond Black-Scholes based on double-fractional diffusion. Phys. A
2016, 449, 200–214. [CrossRef]
46. Aguilar, J.-P.; Coste, C.; Korbel, J. Series representation of the pricing formula for the European option driven
by space-time fractional diffusion. Fract. Calc. Appl. Anal. 2018, 21, 981–1004. [CrossRef]
47. Aguilar, J.-P.; Korbel, J. Option pricing models driven by the space-time fractional diffusion: series
representation and applications. Fractal Fract. 2018, 2, 15. [CrossRef]
48. Aguilar, J.-P.; Korbel, J. Simple formulas for pricing and hedging European options in the Finite Moment
Log Stable model. Risks 2019, 7, 36. [CrossRef]
49. Korbel, J.; Luchko, Y. Modeling of financial processes with a space-time fractional diffusion equation of
varying order. Fract. Calc. Appl. Anal 2016, 19, 1414–1433.
50. Fama, E.F. The behavior of stock market prices. J. Bus. 1965, 38, 34–105 [CrossRef]
51. Mandelbrot, B. The variation of certain speculative prices. J. Bus. 1963, 36, 394–419 [CrossRef]
52. Wilmott, P. Paul Wilmott on Quantitative Finance; Wiley & Sons: Hoboken, NJ, USA, 2006.
53. Gerber, H.U.; Shiu, E.S.W. Option Pricing by Esscher Transforms. Trans. Soc. Actuar. 1994, 46, 99–191.
54. Zolotarev, V.M. One-Dimensional Stable Distributions; American Mathematical Society: Providence, RI,
USA, 1986.
55. Samorodnitsky, G.; Taqqu, M.S. Stable Non-Gaussian Random Processes: Stochastic Models with Infinite Variance;
Chapman & Hall: New York, NY, USA, 1994.

122
Mathematics 2019, 7, 796

56. Abramowitz, M.; Stegun, I. Handbook of Mathematical Functions; Dover Publications: Mineola, NY, USA, 1972.
57. Flajolet, P.; Gourdon, X.; Dumas, P. Mellin transform and asymptotics: Harmonic sums. Theor. Comput. Sci.
1995, 144, 3–58. [CrossRef]
58. Bateman, H. Tables of Integral Transforms (Volume I&II); McGraw & Hill: New-York, NY, USA, 1954.
59. Aguilar, J.-P. On expansions for the Black-Scholes prices and hedge parameters. J. Math. Anal. Appl. 2019,
478, 9739–9989. [CrossRef]
60. Brenner, M.; Subrahmanyam, M.G. A simple approach to option valuation and hedging in the
Black-Scholes Model. Financ. Anal. J. 1994, 50, 25–28. [CrossRef]

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

123
 mathematics
Article
Econophysics and Fractional Calculus: Einstein’s
Evolution Equation, the Fractal Market Hypothesis,
Trend Analysis and Future Price Prediction
Jonathan Blackledge 1,2,3,4,5,6, *, Derek Kearney 6,7 , Marc Lamphiere 6,8,9 , Raja Rani 10,11 and
Paddy Walsh 6,12
1 Stokes Professor, Science Foundation Ireland, Three Park Place, Dublin 2, Ireland
2 Honorary Professor, School of Electrical and Electronic Engineering, Technological University, Kevin Street,
Dublin 8, Ireland
3 Professor Extraordinaire, Department of Computer Science, University of Western Cape, Bellville 7535,
Cape Town, South Africa
4 Honorary Professor, School of Mathematics, Statistics and Computer Science, University of KwaZulu-Natal,
Westville Campus, University Road, Westville 3630, Durban, South Africa
5 Visiting Professor, Faculty of Arts, Science and Technology, University of Wales (Wrexham Glyndwr
University), Mold Road, Wrexham LL11 2AW, UK
6 Dublin Energy Laboratory, Technological University Dublin, Kevin Street, Dublin 8, Ireland;
[email protected] (D.K.); [email protected] (M.L.); [email protected] (P.W.)
7 Lecturer, School of Electrical and Electronic Engineering, Technological University Dublin, Kevin Street,
Dublin 8, Ireland
8 Research Associate, School of Electrical and Electronic Engineering, Technological University Dublin, Kevin
Street, Dublin 8, Ireland
9 Country Director, The Natural Power Consultants (Ireland) Limited, Beacon Court Sandyford, Dublin 18,
Ireland
10 Research Fellow, School of Engineering, University of Portsmouth, University House, Winston Churchill
Avenue, Portsmouth PO1 2UP, UK; [email protected]
11 Deputy Head of General Studies Department, Military Technological College, Al Matar Street,
Muscat 111, Oman
12 Senior Product Manager, Amazon, County Dublin, Burlington Road, Dublin 4, Ireland
* Correspondence: [email protected]; Tel.: +44-7582-676960

Received: 6 September 2019; Accepted: 19 October 2019; Published: 4 November 2019

Abstract: This paper examines a range of results that can be derived from Einstein’s evolution
equation focusing on the effect of introducing a Lévy distribution into the evolution equation.
In this context, we examine the derivation (derived exclusively from the evolution equation) of
the classical and fractional diffusion equations, the classical and generalised Kolmogorov–Feller
equations, the evolution of self-affine stochastic fields through the fractional diffusion equation,
the fractional Poisson equation (for the time independent case), and, a derivation of the Lyapunov
exponent and volatility. In this way, we provide a collection of results (which includes the derivation
of certain fractional partial differential equations) that are fundamental to the stochastic modelling
associated with elastic scattering problems obtained under a unifying theme, i.e., Einstein’s evolution
equation. This includes an analysis of stochastic fields governed by a symmetric (zero-mean) Gaussian
distribution, a Lévy distribution characterised by the Lévy index γ ∈ [0, 2] and the derivation of two
impulse response functions for each case. The relationship between non-Gaussian distributions and
fractional calculus is examined and applications to financial forecasting under the fractal market
hypothesis considered, the reader being provided with example software functions (written in
MATLAB) so that the results presented may be reproduced and/or further investigated.

Mathematics 2019, 7, 1057; doi:10.3390/math7111057 125 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 1057

Keywords: Einstein’s evolution equation; Kolmogorov–Feller equation; diffusion equation; fractional

diffusion equation; self-affine stochastic fields; random market hypothesis; efficient market
hypothesis; fractal market hypothesis; financial time series analysis; evolutionary computing.

1. Introduction
We study one of the principal field equations in statistical mechanics, namely, Einstein’s evolution
equation (EEE or E3 ). This is done in order to derive mathematical models and thereby specific financial
indices in a unified manner, an approach which includes the use of fractional calculus.
E3 models the random motion and (elastic) interactions of a canonical ensemble of particles.
It provides a description for the time evolution of the spatial density field that represents the
concentration of such particles in a macroscopic sense. In an n-dimensional space, each particle
is taken to be undergoing a random walk in which the direction that a particle “propagates” after a
“scattering event” (in which energy and momentum are conserved) is random together with the length
of propagation. The scattering angle θ is taken to be conform to a distribution of angles Pr[θ (r)], r ∈ Rn
and the (free) propagation length is taken to conform to some distribution of lengths Pr[ L(r)] whose
mean value defines the mean free path (MFP). This was the basis for Albert Einstein’s original study of
Brownian motion in 1905 [1], albeit for the one-dimensional case.
In addition to the work of Josiah Gibbs, the evolution equation that Einstein derived is one the
foundations of statistical mechanics [2,3]. The approach can, for example, be applied equally well to
modelling the diffusion of light propagating through a complex of scatterers. In this case the light
is taken to be a ray-field where each ray (reflected from one particle to another) has a random path
length and scattering angle.

1.1. Focus and Context

The focus of this paper is to derive a range of equations and metrics via an n-dimensional
version of E3 in order to demonstrate an inherent connectivity and association in a unified
sense. These equations include the classical diffusion equation, the classical and generalised
Kolmogorov–Feller equations and the evolution of self-affine stochastic fields through the fractional
diffusion equation. The fractional form of these equations is shown to be a direct consequence
of introducing non-Gaussian distributions as "governors" for the statistical characteristics under
which random processes occur, subject to the condition that all such processes involve independent
elastic interactions.
For certain non-Gaussian models such as Lévy processes, this leads naturally to the use of
fractional calculus to develop solutions to the evolution equation as studied in this paper. Further, it is
shown that such solutions are fundamental to the application of the fractal market hypothesis [4] for
analysing financial time series and thereby in developing trading strategies based on this hypothesis.
This approach represents an Econophysics methodology in which a fundamental model used to describe
stochastic processes, originally developed in the study of Brownian motion, is used to solve problems
in economics. In this paper, following developments published previously by Blackledge et al.
(e.g., [5–14]), it is shown that this approach is inclusive of the application of fractional calculus.

1.2. Structure and Organisation

The structure of the paper is as follows. Section 2 provides a brief overview of the principal
mathematical results used in this paper including basic definitions and notation. This section also
includes a short introduction to fractional calculus, specifically some of the conventional definitions
of a fractional integral and a fractional derivative. Section 3 presents E3 upon which all the results
derived in this paper are ultimately dependent, thereby providing a unifying framework for the work

126
Mathematics 2019, 7, 1057

reported as discussed in Section 4, which provides a brief introduction to financial time series analysis
in the context of E3 .
Two equations, that are a conditional representation of E3 , are considered in Section 5, namely the
Classical Kolmogorov–Feller and the Generalised Kolmogorov–Feller equations which are studied
later on in the paper, specifically in Section 14. In the context of E3 , Section 6 provides statements on the
random walk hypothesis and the efficient and fractal market hypotheses coupled with a brief history
associated with the development of such hypotheses for interpreting and analysing financial time
series. As discussed in Section 3, E3 is predicated on a model for the probability density function of a
stochastic system using a continuous random walk model and Section 7 therefore introduces density
functions whose basic properties are important to appreciate in the context of the work reported here.
Sections 8 and 9 study the derivation from the E3 of two metrics, namely, the Lyapunov exponent
and the volatility, respectively. These metrics are then combined into a Lyapunov-to-volatility ratio
(LVR) to develop a trend analysis algorithm which is presented in Section 10, the idea being to provide
an indicator that flags when a financial time series changes its trending behaviour. This is based
on a change in the polarity of the LVR and it is shown, for example, that in order to obtain suitable
accuracies appropriate for algorithmic trading, both pre- (of the financial signal) and post-filtering (of
the LVR) is required. This is quantified in Section 10 using a back-testing strategy. In addition to being
bi-polar, the amplitude of the LVR has values that reflect periods of relative stability in the dynamic
behaviour of a financial signal and in Section 11, a method is proposed to exploit this indication and
provide short term predictions on future prices using the principles of evolution computing (EC).
In this paper, EC is implemented using an online resource and applications package called ‘Eureqa’.
The remaining sections of the paper deal with the classical and fractional diffusion equations,
both of which are derived from E3 in Sections 12 and 13 using Gaussian and non-Gaussian (Lévy)
distributions, respectively. In the latter case, and, using the principles of fractional calculus
established in Section 2, a time series model is developed that depends upon the Lévy index.
Section 14 then provides a complementary approach to deriving similar results using the Generalised
Kolmogorov–Feller equation and an orthonormal memory function which yields the same scaling
properties compounded in the impulse response function. The application of this index for financial
trend analysis is provided in Section 15, illustrating that the Lyapunov exponent and the Lévy index
have similar predictive power providing the data is pre- and post-filtered. Section 16 provides a review
and discussion of the results presented followed by a general conclusion and some open questions to
direct future research.

1.3. Original Contributions

Judging from the open literature, and, to the best of the authors’ knowledge, the approach taken
in this paper is original as are the numerical results presented. In regard to the latter case, an effort
has been made by the authors to integrate important numerical functions with the derivation of
certain important metrics associated with the theoretical models used and the mathematical analysis
presented. These functions are given in Appendix A and their aim is to provide the reader with the
opportunity to reproduce the results presented (the online data sources being referenced throughout)
and investigate their performance for different financial data.

2. Mathematical Preliminaries
In this section, we provide a short overview of some of the mathematical results that are of
importance to the material developed in this paper, specifically the short introduction to fractional
calculus provided in Section 2.3.

2.1. Fourier Transformation and the Convolution Integral

The mathematical models developed in this paper rely on the properties of the Fourier transform
coupled with the convolution and correlation integrals in n-dimensions. For a square integrable

127
Mathematics 2019, 7, 1057

function f (r) ∈ L2 (Rn ) : C → C, we define the Fourier and inverse Fourier transforms in the
“non-unitary form” as
∞
F (k) = Fn [ f (r)] ≡ f (r) exp(−ik · r)dn r
−∞

and
∞
1
f (r) = Fn−1 [ F (k)] ≡ F (k) exp(ik · r)dn k
(2π )n
−∞
1
respectively. Here, r is the n-dimensional spatial vector where r ≡| r |= (r12 + r22 + ... + rn2 ) 2 . Similarly, k
is the spatial frequency vector where k ≡| k |= 2π/λ for wavelength λ and k · r = k1 r1 + k2 r2 + ... +
k n rn . These integral transforms define a Fourier transform pair which, in this paper, we write using
the notation
F ( k ) ↔ f ( r ).

We define the (n-dimensional) Dirac delta function as

∞
1
δn (r) = Fn−1 [1] ≡ exp(ik · r)dn k (1)
(2π )n
−∞

where, with ⊗ denoting the convolution integral,

f ( r ) = δ n ( r ) ⊗ f ( r ),

the convolution of two functions f (r) and g(r) being given by

∞
s(r) = g(r) ⊗ f (r) ≡ g(r − s) f (s)dn s
−∞

and their correlation by

∞
s(r) = g(r)  f (r) ≡ g(r + s) f (s)dn s
−∞

where [s(r), g(r), f (r)] ∈ L2 (Rn ) : C → C. Note that the dimension associated with the integral
operators ⊗ and  is taken to be inferred from the dimension of the functions to which these
operators are applied. In addition, note that, strictly speaking, the Fourier transform is taken over
a Schwartz tempered distributional space, and, in this context, the following theorems are fundamental:

(i) Convolution Theorem

g(r) ⊗ f (r) ↔ G (k) F (k)

where G (k) ↔ g(r) and F (k) ↔ f (r).

(ii) Correlation Theorem

g(r)  f (r) ↔ G∗ (k) F (k)

(iii) Product Theorem

1
g(r) f (r) ↔ G (k) ⊗ F (k)
(2π )n

128
Mathematics 2019, 7, 1057

We note the following Fourier transform relationships [15]

 πα 
2 sin 2 Γ (1 + α )
| x |α ↔ − (2)
| k | 1+ α

1 2π H (±k )
↔ , 0 < α < 1, (3)
(∓ix )α Γ(α) (±k )1−α
where H (k ) is the Heaviside step function

1, k ≥ 0; d
H (k) = ⇒ H (k) = δ( x )
0, k < 0. dx

and for r ∈ Rn α

1 (2π )n 1 n α Γ
↔ , 0 < Re [ α ] < n; c n,α = π 2 2  2 
(4)
| r |α cn,α | k |n−α Γ n−2
α

where Γ is the Gamma function,

∞
Γ(z) = x z−1 exp(− x )dx, Re[z] > 0.
0

2.2. The p- and Uniform-Norm

We define the p-norm as
⎛ ⎞1
 p

 f (r) p ≡ ⎝ | f (r) | p dn r⎠ , 1 ≤ p ≤ ∞
Rn

with the uniform norm being given by

 f (r)∞ = sup{| f (r) |, r ∈ Rn }

and principal properties

 f (r) + g(r) p ≤  f (r) p +  g(r) p ,  f (r) g(r) p ≤  f (r) p  g(r) p

and
 f (r) ⊗ g(r) p ≤  f (r) p  g(r) p .

2.3. Fractional Integrals and Differentials

Since, for n = 0, 1, 2, ...,
d±n
f ( x ) ↔ (ik )±n F (k )
dx ±n
we can, in principal, generalise this result to the case when n is non-integer. Thus, suppose we wish to
fractionally integrate the differential equation

dα
f ( x ) = g ( x ), 0 < α < 1
dx α

to obtain a solution for f ( x ) in terms of g( x ). Fourier transforming,

G (k)
F (k) = , F ( k ) ↔ f ( x ), G ( k ) ↔ g ( x )
(ik)α

129
Mathematics 2019, 7, 1057

and thus, using the convolution theorem, we can write

∞
1 exp(ikx ) H (x) 1
f ( x ) = h ( x ) ⊗ g ( x ), h ( x ) = dk =
2π (ik)α Γ ( α ) x 1− α
−∞

using Relationship (3).

This important result is easily derived by expressing the inverse Fourier transform in terms of a
Bromwich integral so that, with p = ik, we can write h( x ) in terms of the inverse Laplace transform

i∞
1 exp( px )
h( x ) = dp.
2πi pα
−i∞

Generalising the Laplace transform of the function x n (for positive integer n) given by

∞ ∞
n! Γ(n)
x n exp(− px )dx = ⇒ x n−1 exp(− px )dx = , Γ ( n ) = ( n − 1) !
p 1+ n pn
0 0

to
∞ ∞
Γ(α) 1 exp(− px ) 1
x α−1 exp(− px )dx = ⇒ dx = α
pα Γ(α) x 1− α p
0 0

it is then clear that

∞
1 1 H (x)
= exp(−ikx )dx.
(ik)α Γ(α) x 1− α
−∞

This expression for f ( x ) in terms of the convolution h( x ) ⊗ g( x ) is the basic fractional integral
known as the Riemann–Liouville integral which, specifying the limits of integration, takes the form

x
−α 1 f (y)
a Dx f ( x ) ≡ dy (5)
Γ(α) ( x − y )1− α
a

thereby expressing the integral in terms of an inverse differential operator D −α over the limits a and x.
This allows us to express a fractional differential denoted by the operator a Dxα ≡ dα /dx α in terms of a
fractional integral by noting that

−(1−α)
α
a Dx f ( x ) = Dx1 a Dxα−1 f ( x ) = Dx1 a Dx f (x)
x x
1 d f (y) −α f (y)
= dy = dy (6)
Γ(1 − α) dx ( x − y)α Γ (1 − α ) ( x − y )1+ α
a a

When α is a negative value and noting that Γ(1 + α) = αΓ(α),

x x
−α α f (y) 1 f (y)
a Dx f ( x ) = dy = dy
Γ (1 + α ) ( x − y )1− α Γ(α) ( x − y ) 1− α
a a

thereby recovering the expression for a fractional integral given by Equation (5). Thus, combining the
results, we can write
x
±α 1 f (y)
a Dx f ( x ) = dy. (7)
Γ(∓α) ( x − y ) 1± α
a

130
Mathematics 2019, 7, 1057

Since, for some scaling value λ (and with z = λy),

x x x
1 f (λy) 1 f (z) λ±α f (z)
dy = d(z/λ) = dz,
Γ(∓α) ( x − y ) 1± α Γ(∓α) [ x − (z/λ)]1±α Γ(∓α) (λx − z)1±α
a a a

this operator has the self-affine scaling characteristic

±α
a Dx f ( λx ) = λ±α a Dx±α f ( x ). (8)

Another related approach to defining a fractional differential is through application of the delta
function. For r ∈ R1

dn
f ( x ) = δ( x ) ⊗ f ( x ) and f (n) ( x ) = δ(n) ( x ) ⊗ f ( x ) where f (n) ( x ) ≡ f (x)
dx n
Generalising this result to the non-integer case, we write

f ( α ) ( x ) = δ ( α ) ( x ) ⊗ f ( x ),

where, from Equation (1),

∞
1
δ(α) ( x ) = (ik)α exp(ikx )dk.
2π
−∞

We can then write

∞
d ( α −1) d 1 1
δ(α) ( x ) = δ (x) = exp(ikx )dk
dx dx 2π (ik)1−α
−∞
1 d H (x) 1 δ( x ) α H (x)
= = − . (9)
Γ(1 − α) dx x α Γ (1 − α ) x α Γ (1 − α ) x 1+ α

A further definition of a fractional differential can be obtained using the sign function
sgn( x ) where ⎧
⎪
⎪+1, x > 0;
x ⎨
sgn( x ) = = −1, x < 0;
|x| ⎪ ⎪
⎩0, x = 0.
and
2
sgn( x ) ↔ ,
ik
when we can write
1
f (α) ( x ) = f ( x ) ⊗ sgn(1+α) ( x ), ∀α.
2
This result becomes clear if we note that
1 1 1
f ( x ) ⊗ sgn(1+α) ( x ) = f ( x ) ⊗ sgn(1+α) ( x ) ⊗ δ( x ) = f ( x ) ⊗ sgn( x ) ⊗ δ(1+α) ( x )
2 2 2
and therefore that
1 1 2
f ( x ) ⊗ sgn( x ) ⊗ δ(1+α) ( x ) ↔ F (k ) (ik )1+α = (ik)α F (k ).
2 2 ik

Defining a fractional differential and integral in terms of the operators a Dxα and a Dx−α , respectively,
is based on a generalisation of the Fourier transform under differentiation and integration, respectively.
Traditional (integer) calculus goes hand-in-hand with a geometrical interpretation of the associated

131
Mathematics 2019, 7, 1057

operations, starting with a differential defining the gradient of a function at a point (at least for a
piecewise continuous function). With fractional calculus, generalisations of this type do not easily
lend themselves to a geometrical interpretation. However, geometric and physical interpretation
of fractional derivatives have been developed (e.g., [16,17]) including the connectivity between
fractional calculus and fractal geometry [18] which is based on the scaling relationship compounded
in Equation (8). An important characteristic of these interpretations that is relevant to the remit of
this paper, is that the operator a Dx−α operating on a stochastic function is characterised by this scaling
property, a property that yields self-affine stochastic fields or random scaling fractals.
As discussed later on in this paper, many financial signals can be classified as random scaling
fractal signals (with a fractal dimension D ∈ [1, 2]). This is the basis for the fractal market hypothesis
in mathematical economics and hence, the applications of fractional calculus. Note however, that
the "process" of generalising the Fourier transform used above for defining fractional differentials
and integrals is just one such generalisation that can be applied. Thus, the operators defined by
Equations (5) and (6), for example, are not unique and there are many definitions and generalisations
of a fractional derivative that have been developed [19] and continue to be so [20].
Although there are, in principle, an unlimited number of definitions that may be “designed” to
define a fractional derivative, there is a common theme to all of them which is that they are expressed
in terms of a convolution. For example, the Caputo fractional derivative is given by

x
−α ( x − y ) n − α −1
a Dx f ( x ) = Kα ( x − y) f (n) (y)dy where Kα ( x − y) = ,
Γ(n − α)
a

which is easily formulated via application of the inverse Fourier transform given that if

∞
1
f (x) = exp(ikx ) F (k)dk
2π
−∞

and
∞ ∞
1 exp(ikx ) 1 (ik)n exp(ikx )
D −α f ( x ) = F (k )dk = F (k)dk
2π (ik)α 2π (ik)n+α
−∞ −∞

then from Relationship (3),

∞
1 dn exp(ikx ) dn H ( x ) 1
D −α f ( x ) = F (k)dk = n ⊗ f (x)
2π dx n (ik)n+α dx Γ(n + α) x1−(n+α)
−∞

and hence

H ( x ) dn 1 H (x) 1
Dα f ( x) = ⊗ f (x) = ⊗ f ( n ) ( x ).
Γ(n − α) dx n x1−(n−α) Γ(n − α) x1−(n−α)

The results considered here are fundamental to the implementation of fractional calculus in
econophysics (and physics in general) as they are predicated on the Fourier transform which arguably
plays the most pivotal role of all in so many aspects of physics and especially in the analysis and
processing of signals, e.g., [21,22], including financial signals.
Irrespective of the non-unique definition of a fractional derivative, there is one fundamental
difference between a classical and a fractional derivative which is characterised by Equation (7),
for example. A nth order derivative of a piecewise continuous function f ( x ) can be defined at a
single point on x at x0 say, and is independent of any other values of f ( x ) for x < x0 or x > x0 .
However, given that a fractional derivative involves the convolution of the function f ( x ) with 1/x1+α ,
for example, its value at a point x0 depends on prior values of f ( x ) for x < x0 . Thus the value

132
Mathematics 2019, 7, 1057

of a fractional derivative of f ( x ) depends on its “history” and thus, unlike an integer derivative, a
fractional derivative therefore incurs “memory”. This “memory effect” is another way of approaching
the analysis of financial signals using fractional calculus as financial signals are influenced by the
memory of past financial conditions, albeit within a stochastic context. This is a key element to the
analysis of financial signals using fractional calculus and a fundamental component to applications of
the fractal market hypothesis (as discussed later on in this paper).

3. Einstein’s Evolution Equation

Let p(r) denote a probability density function (PDF) where

∞
p(r)dn r = 1,
−∞

which characterises with the position of particles in an n-dimensional space r ∈ Rn where, at any
instant in time t, the particles exist as a result of some “random walk” generated by a sequence of
“elastic scattering” processes (with other like particles in the same n-dimensional space) that have
occurred over some period of time < t. Further, let u(r, t) denote the density function associated with
a canonical assemble of particles all undergoing the same random walk process (i.e., the number of
particles per unit space, e.g., per unit volume for n = 3).
Consider the initial condition where we have an infinitely small concentration of such particles at
a time t = 0 located at the origin r = 0. The density function at t = 0 is then given by u(r, 0) = δn (r)
where δn (r) is the n-dimensional Dirac delta function. At some short time later t = τ << 1, it can
be expected that the density function will be determined by the PDF governing the distribution of
particles after a (short duration) random walk. Thus we can write

u(r, τ ) = p(r) ⊗ u(r, 0) = p(r) ⊗ δn (r) = p(r),

where ⊗ denotes the convolution integral over all r. The PDF p(r) therefore represents the response
(in a statistical sense) associated with a short time random walk process, and, in this context, can be
considered to be a statistical impulse response function (IRF). Thus for any time t, the density field at
some later time t + τ will be given by

u(r, t + τ ) = p(r) ⊗ u(r, t). (10)

For any instant in time t, Equation (10) shows that the spatial behaviour of the density field at
some future time τ is given by the convolution of the density of particles at a previous time with the
PDF of the system that governs its “statistical evolution”. In this sense, p(r) is analogous to the IRF of
a linear stationary system when, for an initial condition u0 (r) ≡ u(r, t = 0), say,

u(r, t) = g(r, t) ⊗ u0 (r, t)

where g(r, t) is the characteristic Green’s function of the system. However, in this case u(r, t)
denotes a deterministic function associated with the behaviour of a deterministic system, whereas
in Equation (10), u(r, t) is the density function associated with the evolution of a statistical system.
This “system” is taken to be stationary in a statistical sense because it is assumed that p(r) does not
vary in time and the time evolution model given by Equation (10) is referred to as being “Ergodic”.
Further, we note that if the PDF is symmetric, then p(r) ≡ p(r ).
Equation (10) is Einstein’s evolution equation (E3 ). It is a “master equation” for elastic scattering
processes in statistical mechanics and is an example of a continuous time random walk model. On the
basis of Equation (10), one can derive a variety of stochastic field equations as shall be shown later on
in this paper.

133
Mathematics 2019, 7, 1057

In regard to the continuous time random walk model given by Equation (10), p(r) is the PDF for
the displacement r of a particle’s position over time interval τ. The equivalent discrete time random
walk model, Equation (10) takes the form

u(rm , tn + τ ) = p(rm ) ⊗ u(rm , tn )

where rm and tn are discrete vector and scalar arrays, respectively, and, ⊗ denotes the convolution
sum. In this case, τ is fixed time step and, in the context of the work reported in this paper, may be
considered to be a time-unit for financial markets, i.e., a minute, hour, day or week associated with a
price value u(tn ).
For a source function s(r, t) (a source density), which may be a stochastic function, the evolution
equation is
u(r, t + τ ) = p(r) ⊗ u(r, t) + s(r, t). (11)

This equation describes the evolution of of the density function u(r, t) when the initial particle
concentration is replenished in space and/or time and can be extended further to include a decay factor
over time when it is required to consider an evolution equation of the type (for decay rate factor R)

u(r, t + τ ) = p(r) ⊗ u(r, t) + s(r, t) − Ru(r, t) (12)

The financial time series models and metrics that are considered in this paper are all derived from
Equation (11) and for this reason, in the following section, a short introduction to financial time series
analysis is provided. This is necessary for readers to appreciate the focus of the application that is
considered in this paper.

4. Financial Time Series Analysis

A financial time series is a discrete set of price values that are most commonly regular samples
over a specific time interval (minutes, hours, days, etc.) which depend on the financial price index
available (e.g., world-wide indices such as FTSE100, S & P 500, FOREX, etc.). Over longer time
intervals, the price index is usually an average of the samples taken over the next smallest time interval.
Most financial data is available as a time series and therefore developing mathematical models (both
linear and non-linear) of time series data is an essential component underpinning many aspects of
mathematical finance leading to algorithms for day-to-day trading, forecasting and econometrics
in general.
There are numerous internet resources that provide up-to-date and historical data of different
indices over different time scales such as the data available at [23] which is the internet source used
to access the data presented in this paper. Similarly, there are numerous “metrics” (also called a
financial index) which are the result of processing samples of data over a look-back window of a
specified length usually known as the “period”. Such metrics range from statistical metrics based on
an autoregressive moving average and nonlinear locally non-constant variance models (applicable
to volatile financial returns, interest, exchange rates and futures) through to descriptive techniques
for various features, such as long term level fluctuations and distributions, short and long memory
dependence, directionality and volatility.
Methods of fitting time series models to time series data and their statistical validation determine
the application to which they can (or otherwise) be successfully applied to forecasting, systematic
trading, fund manager evaluation, hedging and simulation for example. The online resource
‘Investopedia’ [24] provides descriptions, computational algorithms and examples of the numerous
metrics, indices and other parameters that have, and are continuing to be, developed for financial time
series analysis.

134
Mathematics 2019, 7, 1057

In this paper, continuous time series models for a financial signal denoted by u(t) are derived
exclusively from Equation (11), the associated discrete time series model being denoted by un , n =
1, 2, ..., N which is taken to describe a digital financial signal consisting of N elements.

5. Einstein’s Evolution Equation and the Kolmogorov–Feller Equations

The Classical and Generalised Kolmogorov–Feller Equations can be derived directly from E3
through application of a Taylor series in time and a memory function (in time), respectively. They are
in fact representations of E3 for the case when τ << 1 and otherwise, respectively, as shall now be
shown, both equations being studied later on in this paper.

5.1. The Classical Kolmogorov–Feller Equation

Consider the following Taylor series for the function u(r, t + τ ) in Equation (10):

∂ τ 2 ∂2
u(r, t + τ ) = u(r, t) + τ u(r, t) + u(r, t) + ...
∂t 2! ∂t2
For τ << 1
∂
u(r, t + τ ) u(r, t) + τ u(r, t)
∂t
and from Equation (10), we obtain the classical Kolmogorov–Feller equation (CKFE), [25,26]

∂
τ u(r, t) = −u(r, t) + u(r, t) ⊗ p(r ), (13)
∂t
which is essentially a representation of Equation (10) for τ << 1.
Equation (13) is based on a critical assumption which is that the time evolution of the density
field u(r, t) is influenced only by short term events and that longer term events have no influence on
the behaviour of the field at any time t, i.e., the “system” described by Equation (13) has no “memory”.
This statement is the physical basis upon which the condition τ << 1 is imposed, thereby facilitating
the Taylor series expansion of the function u(r, t + τ ) to first order alone.

5.2. The Generalised Kolmogorov–Feller Equation

Given that Equation (13) is memory invariant, the question arises as to how longer term temporal
influences can be modelled, other than by taking an increasingly larger number of terms in the Taylor
expansion of u(r, t + τ ) which is not of practical analytical value, i.e., writing Equation (10) in the form

∂ τ 2 ∂2
τ u(r, t) + u(r, t) + ... = −u(r, t) + u(r, t) ⊗ p(r ).
∂t 2! ∂t2
The key to solving this problem is to express the infinite series on the left hand side of the equation
above in terms of a “memory function” m(t) and write

∂
τm(t) ⊗ u(r, t) = −u(r, t) + u(r, t) ⊗ p(r ).
∂t

This is the generalised Kolmogorov–Feller equation (GKFE) which reduces to the CKFE when m(t) =
δ ( t ).
A characteristic time spectrum M(ω ) for m(t) can be obtained by noting that we have, in effect,
considered the result

∂ τ 2 ∂2 ∂
u(r, t + τ ) = u(r, t) + τ u(r, t) + u(r, t) + ... = u(r, t) + τm(t) ⊗ u(r, t)
∂t 2! ∂t2 ∂t

135
Mathematics 2019, 7, 1057

so that, after taking the Fourier transform with respect to t, we obtain

1
U (r, ω ) exp(iωτ ) = U (r, ω ) + iωτU (r, ω ) + (iωτ )2 U (r, ω ) + ... = U (r, ω ) + iωτM(ω )U (r, ω )
2!

where U (r, ω ) ↔ u(r, t) and M (ω ) ↔ m(t), from which it follows that we can write M(ω ) as
∞
1 exp(iωτ ) − 1
M(ω ) = ∑ (iωτ )n−1 =
n =1
n! iωτ

5.3. Orthonormal Memory Functions

For any inverse function or class of inverse functions of the type n(t), say, such that

n ( t ) ⊗ m ( t ) = δ ( t ),

the GKFE can be written in the form

∂
τ u(r, t) = −n(t) ⊗ u(r, t) + n(t) ⊗ u(r, t) ⊗ p(r), (14)
∂t

where the GKFE is again recovered when n(t) = δ(t) given that δ(t) ⊗ δ(t) = δ(t). The function n(t)
is an orthonormal function of m(t).

6. The Random Walk, the Efficient and the Fractal Market Hypotheses
From Equation (11) we can generate a simple (continuous) financial time series model by
integrating over r to obtain
u ( t + τ ) = u ( t ) + s ( t ), (15)

where
∞ ∞
u(t + τ ) = u(r, t + τ )dn r, s(t) = s(r, t)dn r
−∞ −∞

and, for p(r) = δn (r),

∞ ∞
u(t) = [δn (r) ⊗ u(r, t)]dn r = u(r, t)dn r.
−∞ −∞

If s(t) is taken to be a (bi-polar) stochastic function of time and u(t) is some price value (of some
commodity) then Equation (15) describes the case in which a future price at some future time t + τ is
given by the known price at time t plus some random price value s(t). Note that for any value of t, s(t)
may be a positive or negative value thereby giving a higher or lower price value at t + τ. The principal
point here is that although Equation (15) is the simplest of models for price variation, it can nevertheless
be seen to be the result of a spatial integration of E3 when p(r) = δn (r). Moreover, it is a model that
encompasses some of the earliest questions associated with the dynamics of a free market economy as
discussed in the following section.

6.1. The Random Walk Hypothesis

In 1900, Louis Bachelier [27] concluded that the price of a commodity today is the best estimate
of its price in the future (at least in the short term). The random behaviour of commodity prices was
again noted by Holbrook Working in 1934 [28] in an analysis of time series data. In the 1950s, Maurice
Kendall [29] attempted to find periodic cycles in the financial time series of various securities and
commodities but did not observe any. Prices appeared to be yesterday’s price plus some random change
(up or down); he suggested that price changes were independent and that they followed random
walks. Thus the first models conceived for price variation were based on the sum of independent

136
Mathematics 2019, 7, 1057

random variations often referred to as Brownian motion and quantified in Equation (15). This led to
the creation of the random walk hypothesis, and the closely related efficient market hypothesis which
states that random price movements indicate a well-functioning or efficient market.
An example of the type of time series that illustrates this effect is given in Figure 1. The figure
shows a signal obtained using a zero mean Gaussian random number generated to compute sn based
on the iteration
un+1 = un + sn , u1 = 100, n = 1, 2, 3, ..., 999.

Trivial though this model is, it nevertheless provides remarkably similar signals to those that
characterise many financial signals. However, it is an example of a stationary signal in the sense that
the scale of random deviations is invariant of time and the trends (up and down) are over similar
amplitude and time scales—characteristics that are not properties of financial signals in general, at least
over large time scales.

Figure 1. Simulation of a financial signal based on the sum of independent random walks; basis for the
Random Walk Hypothesis.

6.2. The Efficient Market Hypothesis

It is often stated that asset prices should follow Gaussian random walks because of the efficient
market hypothesis (EMH), e.g., [30–32] (and references therein). The EMH states that the current price
of an asset fully reflects all available information relevant to it and that new information is immediately
incorporated into the price. Thus, in an efficient market, models for asset pricing are concerned with
the arrival of new information which is taken to be independent and random.
The EMH implies independent price increments, but why should they be Gaussian distributed?
A Gaussian PDF is chosen because price movements are presumed to be an aggregation of smaller
ones and sums of independent random contributions have a Gaussian PDF due to the central limit
theorem. This is equivalent to arguing that all financial time series used to construct an “averaged
signal” such as the FTSE100 or Dow Jones Industrial Average are statistically independent. Such an
argument is not fully justified because it assumes that the reaction of investors to one particular stock
market is independent of investors in other stock markets which, in general, will not be the case as
each investor may have a common reaction to economic issues that transcend any particular stock.
In other words, asset management throughout the markets relies on a high degree of connectivity and
the arrival of new information can send "shocks" through the market as people react to it and then to
each other’s reactions.
The EMH assumes that there is a rational and unique way to use available information, that
all agents possess this knowledge and that any chain reaction produced by a “shock” happens
instantaneously. This is clearly not physically possible or financial viable and financial models that are
based on such a hypothesis have and will continue to fail.

137
Mathematics 2019, 7, 1057

6.3. The Fractal Market Hypothesis

One of the principal concerns with regard to the EMH relates to the issue of assuming that the
markets are Gaussian distributed. This is because it has long been known that financial time series
(specifically price changes) do not adhere to a Gaussian distribution and this is arguably the most
important of the shortcomings relating to the EMH model (i.e., the failure of the independence and the
Gaussian distribution of increments assumption). It is fundamental to the inability for EMH-based
analysis such as the Black-Scholes model [33] to explain the characteristics of financial signals such
as clustering, flights and failure to explain “boom-bust” events, and, in particular, financial “crashes”
leading to recession.
More recently, financial time series have been shown to be random self-affine signals which has
led to the related development of the fractal market hypothesis in which price variations are in effect
random walks whose statistical distribution of values is similar over different time scales. Ralph Elliott
(a professional accountant) first reported on the apparent self-affine properties of financial data in
1938 [34,35]. He was the first to observe that segments of financial time series data of different sizes
could be scaled in such a way that they were statistically the same, producing so-called Elliot waves.
He proposed that trends in financial prices resulted from investors’ predominant psychology and
found that swings in mass psychology always seemed to be a manifestation of the same recurring
self-affine patterns in financial markets.
A primary goal of an investor is to attempt to obtain information that can provide some confidence
in the immediate future of a commodity’s price, based on patterns of the past. One of the principal
components of this goal is based on the observation that there are “waves within waves” that appear
to permeate financial signals when studied in sufficient detail and imagination. It is these repeating
self-affine wave patterns that occupy both the financial investor and the financial systems modeller
alike and it is clear that although economies have undergone many changes in the last 100 years,
the dynamics of market data does not appear to have changed significantly (ignoring scale).
The Elliott wave principal developed in the late 1930s and the fractal market hypothesis developed
in the late 1990s provide data consistent models for the interpretation and analysis of financial signals
and investment theory. In turn, and, as discussed in this paper, fractal signals and fields can be cast
in terms of solutions to certain fractional differential equations for which an understanding of the
fractional calculus is a pre-requisite. Hence, the application of fractional calculus is and is likely to
continue to have a primary role in mathematical economics.
In this context, and, on the basis of Equation (11), an overview of the contents of this paper and
its subject connectivity is quantified in terms of the flow diagram given in Table 1 where the discrete
time dependent behaviour of u(t) is taken to represent a digital financial time series un , n = 1, 2, ..., N.
This flow diagram highlights the relationship between the E3 and the applications of fractional calculus
in mathematical economics which is a theme of this paper. It is illustrative of the unified approach that
has been taken in order to produce a coherent exposition for the development of three fundamental
indices that are used to analyse financial signals, namely, the Lyapunov exponent, the volatility and
the Lévy index. As shall be studied later on in this paper, these indices are used to undertake a trend
analysis which, in turn, provides a confidence criterion for the application of evolutionary computing
to predict future prices.

6.4. Principal Properties of Financial Signals

Whatever the hypothesis that is considered in regard to understanding and analysing financial
signals, there are some basic characteristics of such signals that are common. These include
the following:

• financial signals are stochastic signals;

• they are non-stationary signals;
• their distributions (specifically the price differences) are non-Gaussian;

138
Mathematics 2019, 7, 1057

• they are often characterized by long term historical correlations;

• they have random repeating patterns at different scales—they are statistically self-affine
(random fractals);
• they have instabilities at all scales—sometimes referred to a “Lévy flights”.

The models, metrics and computation algorithms reported in this paper attempt to take each of
the above properties into account while maintaining adherence to E3 as a unifying theme.

Table 1. Fow diagram illustrating the connectivity between Einstein’s Evolution Equation (E3 ), two well
known financial indices (i.e., the volatility σ and Lyapunov exponent λ) and the classical and fractional
diffusion equations both of which can be derived from the evolution equation using the Characteristic
Functions (CFs) shown (where c is a constant, k is the spatial frequency and γ is the Lévy index).
The flow diagram also illustrates the relationship between the evolution equation and two principal
market hypotheses: the efficient market hypothesis and the fractal market hypothesis, the latter
hypothesis being a concomitant of the fractional calculus. The asterisk (∗ ) denotes the connection
between the Generalised KFE and the introduction of a memory function which allows E3 to be written
in a different form without loss of generality.

Kolmogorov–Feller Equation (KFE) ← Taylor Series Analysis → Generalised KFE∗

↑ ↓
Lyapunov Exponent (λ) ← E3 Volatility (σ)
↓
Gaussian Distribution with ← Probability Density → Lévy Distribution with
CF exp(−c | k |2 ) Function CF exp(−c | k |γ ), γ ∈ [0, 2]
↓ ↓
Classical Diffusion Equation Fractional Diffusion Equation
 
Classical Calculus ↔ Memory Function∗ ↔ Fractional Calculus
 
Efficient Market Hypothesis Fractal Market Hypothesis
↓ ↓
Time Series Model → Financial Trend Analysis ← Time Series Model
based on time variations
in σ, λ & γ
↓
Evolutionary Computing
↓
Future Price Prediction

7. Density Function Distributions

Suppose that the one-dimensional density function u( x, t) is ergodic and has a PDF p( x ) ≡
Pr[u( x, t)]∀t where
∞
p( x )dx = 1.
−∞

If, for all time t > 0, the distributions of u(y, t) and u(z, t) are identical, what is the (symmetric)
distribution of the density functions in the plane r ∈ R2 and the volume r ∈ R3 ?
It is clear that the cumulative distribution function of u( x, t) is given by

x
c( x ) = p( x )dx,

and hence, from the fundamental theorem of calculus

d
p( x ) = c ( x ).
dx

139
Mathematics 2019, 7, 1057

Thus, for r ∈ R2 , when p( x )'

= p(y), the (circularly symmetric) cumulative distribution is (using
polar coordinates (r, θ ) with r = x2 + y2 )

 2πr r
c (r ) = p (r ) d2 r = p(r )rdrdθ = 2π p(r )rdr
0 0 0

and so the PDF p2 (r ), say, is given by

r
d
p2 (r ) = 2π p(r )rdr = 2πrp(r ), r ∈ R2 .
dr
0

Similarly for r ∈ R3 ,
when p( x ) = p(y) '= p(z), then for the spherically symmetric case (using
spherical polar coordinates (r, θ, φ) with r = x2 + y2 + z2 ),

 r 1 2π r
c (r ) = p (r ) d3 r = p(r )r2 drd(cos θ )dφ = 4π p(r )r2 dr
0 −1 0 0

so that
r
d
p3 (r ) = 4π p(r )r2 dr = 4πr2 p(r ). r ∈ R3 .
dr
0

7.1. Gaussian and Rayleigh Distributions

In the case when u( x, t), u(y, t) and u(z, t) are (zero mean) Gaussian distributed and

exp[− x2 /(2σ2 )]
p( x ) = √
2πσ2

where σ is the standard deviation and when the characteristic function (CF) is given by [36]

P(k ) = Fn [ p( x )] = exp(−σ2 k2 /2),

then
r
p2 (r ) = exp[−r2 /(2σ2 )], r ∈ R2 ,
σ2
which is a standard Rayleigh distribution with characteristics function [36]
√
P2 (k ) = −i 2πσk exp(−σ2 k2 /2).

For the three dimensional case

2 r2
p3 (r ) = exp[−r2 /(2σ2 )], r ∈ R3 ,
π σ3
which has the CF [36]
P3 (k ) = −2(σ2 k2 − 1) exp(−σ2 k2 /2).

The distributions p2 (r ) and p3 (r ) represent the random length of the two- and three-vectors
respectively.

140
Mathematics 2019, 7, 1057

The case associated with p2 (r ) frequently occurs when a random time signal u(t) has a distribution
p( x ). By computing the Hilbert transform of this signal, we obtain the quadrature component w(t)
which has the same distribution as u(t). The analytic signal s(t), is then given by

1
s(t) = u(t) + iw(t) where w(t) = ⊗ u(t) (16)
πt
'
and the amplitude modulations given by A(t) = u2 (t) + w2 (t) are therefore 2πrp(r ) distributed.

7.2. Lévy and Associated Distributions

The symmetric Lévy distribution features in material considered later on in this work and is
a key to the connectivity between E3 and the fractional diffusion equation. We therefore take the
opportunity at this point in the paper to consider some of the basic definitions and results associated
with this distribution. The CF of a (zero-mean) Gaussian distribution can be written as P(k ) = exp(−c |
k |2 ) where c ≥ 0 is a real constant (=σ2 /2). The Lévy distribution is one whose CF is based on a
generalisation of the CF of a Gaussian distribution to

P(k) = exp(−c | k |γ ), γ ∈ [0, 2] (17)

where γ is the Lévy index. It is then clear that γ = 2 recovers a Gaussian PDF, γ = 1 generates a
Cauchy distribution given that
 
1 c2 1
exp(−c | k |) ↔ ∼ , x→∞
πc c2 + x 2 x2

and it is noted that

lim exp(−c | k |γ ) ↔ δ( x ).
γ →0

For γ ∈ (0, 2) it possible to derive the asymptotic result [37]

1
exp(−c | k |γ ) ↔ , x → ∞.
| x | 1+ γ

A simple derivation of this result can be obtained by noting that

∞
cn
p( x ) = F1−1 [exp(−c | k |γ )] = F1−1 [1] + ∑ (−1)n n! F1−1 [| k |nγ ]
n =1
 πnγ  !
∞ n sin Γ(1 + nγ)
1 nc 1
= δ( x ) −
π ∑ (−1) n!
2
| x |1+nγ
∼
| x | 1+ γ
, | x |→ ∞
n =1

using Relationship (2). The non-asymptotic Lévy distribution for arbitrary values of γ can easily be
evaluated numerically through application of a discrete Fourier transform. Figure 2 shows examples
of the Lévy distribution p( x ) for different values of γ (with c = 2) and associated distributions xp( x )
for the same values of γ but for c = 1/2. It is noted that the tails of each distribution for γ < 2 are
longer than those for the case when γ = 2, thereby representing stochastic processes in which rare
but extreme events are more likely to occur than with a Gaussian distributed process. These events
include Lévy flights which, in financial time series analysis, mark positions in time when the value
of a price may increase or decrease in a way that is inconsistent with the statistical signature of the
series in a more general sense. An example of this is given in Figure 3 which shows Lévy flights in the
complex plane associated with a FTSE 100 signal, the data having been obtained from [23].

141
Mathematics 2019, 7, 1057

Identifying metrics that can flag the positions in time at which a Lévy flight may occur are an
important feature of financial trading. The first of these that we consider in this paper is the Lyapunov
exponent which, in the context E3 , is discussed in the following section.

Figure 2. Left: Lévy distributions p( x ) for γ = 2 (solid line), γ = 1.5 (dotted line), γ = 1 (dot-dashed
line), and γ = 0.5 (dashed line) for c = 2; Right: Plots of associated distributions xp( x ) for γ = 2 (solid
line), γ = 1.5 (dotted line), γ = 1 (dot-dashed line) and γ = 0.5 (dashed line) for c = 1/2.

Figure 3. Example of Lévy flights. Left: Plot of the FTSE100 daily prices from 14/03/2006 to
12/04/2019—u(t); Right: Complex plane plot of s(t) given by Equation (16). Both time functions u(t)
and s(t) are taken to be uniformly sampled discrete functions un and sn , respectively; the analytic
signal is computed using a fast Fourier transform (FFT) and the algorithm presented in [38].

8. The Lyapunov Exponent

The Lyapunov exponent is a quantity that characterises the rate of separation of infinitesimally
close trajectories, a trajectory being a time-ordered set of states of a dynamical system.

142
Mathematics 2019, 7, 1057

Specifically, a trajectory is a sequence f n ( x ), n ∈ N calculated by the iterated application of a mapping

f to an element x of its source. In this section, we illustrate the derivation of this exponent from E3 .
Consider some dynamical system that is modelled by an iterative equation characterised by
a function f (t, x0 ) which produces a solution at some time t given by x (t) for an initial condition
x0 ≡ x (t = 0). The system is such that it may be stable or unstable depending on the initial condition
x0 and system parameters, i.e., the numerical value of the parameters that characterise the function
f (t, x0 ). For stability, we can expect the solution x (t) to be characterised by convergence to a specific
value (which could be zero) so that as t → ∞, x (t) = x (t + τ ) where τ << 1. If the solution is unstable
we can expect x (t) might increase in value with t and/or have some chaotic behaviour where x (t)
becomes a chaotic variable of time. In this case, a fundamental “diagnostic” is associated with asking
the following question: Is a given system, characterised by the function f (t, x0 ), unstable, and, if so,
how unstable is it? The answer to this question is compounded in the Lyapunov exponent, whose
value is typically taken to be a measure of how sensitive x (t) is to the initial condition x0 . If we denote
δx (t) to be some change to the solution which depends on a change to the initial condition denoted by
δx0 , then this sensitivity is compounded in the following equation

δx (t) p ∼ exp(λt)δx0  p , (18)

where λ is referred to as the (leading) Lyapunov exponent and has the solution
 
1 δx (t) p
λ∼ ln . (19)
t δx0  p

This exponent represents the mean rate of separation of trajectories of the system where the
term “trajectory” refers to the time evolution of x (t) subject to the initial condition x0 . Thus, any two
trajectories x (t) = f (t, x0 ) and x (t) + δx (t) = f (t, x0 + δx0 ), say, that are close to each other for t << 1
and consequently separate exponentially with time, will represent a system defined by function f (t, x0 )
that has a large value of λ. On the other hand, if all values of x (t) and x (t) + δx (t) converge to the
same value in some neighbourhood of time, then δx (t) must approach zero, and, from Equation (18),
this implies that λ < 0. Thus, on the basis of Equation (18), a positive value of λ defines a system with
chaotic behaviour in time and a negative value of λ characterises a stable system which convergences
in time. Moreover, the larger the value of λ becomes the faster the rate of convergence (for λ < 1) or
the “route to chaos” (for λ > 1), [39–41].
Given the description above as to what the Lyapunov exponent is and what it characterises,
we consider a derivation of this exponent within the context of Equation (10) for r ∈ R3 and uniform
discretisation in time so that we can write

u(r, tn+1 ) = p(r) ⊗ u(r, tn ), n = 0, 1, 2, ...., N. (20)

Suppose that after many time steps, this iteration converges to the function u(r, t∞ ), say. We can
then represent the iteration in the form

u(r, tn ) = u(r, t∞ ) + (r, tn ), (21)

where (r, tn ) denotes the error at any time step n. Convergence to the function u(r, t∞ ) then occurs if
(r, tn ) → 0 as n → ∞. If we now consider a model for the error at each time step given by (for some
real constant ε)
(r, tn+1 ) = ε exp(λtn ) (22)

with tn = nτ (where τ defines the time sampling interval) it is clear that we can then write

(r, tn+1 ) = (r, tn ) exp(λτ ),

143
Mathematics 2019, 7, 1057

or, after taking the p = 1-norm of both sides,

¯ (tn+1 ) = ¯ (tn ) exp(λτ )

where
¯ (tn ) = (r, tn )1 .

Thus we can consider an expression for λ given by

1 N
¯ (tn+1 )
λ = lim
N →∞ Nτ ∑ ln ¯ (tn )
n =1

If λ is negative, then the iterative process is stable since we can expect that for n >> 1,
¯ (tn+1 )/¯ (tn ) < 1 and thus ln ¯ (tn+1 )/¯ (tn ) < 0. If λ is positive then the iterative process will
diverge, such a criterion for convergence/divergence being dependent on the exponential model given
in Equation (22) used to represent the error function at each iteration. This result applies for any
time iteration process. However, in the case of Equation (20), we note that, if u(r, t0 ) = δ3 (r) then
u(r, t1 ) = p(r), u(r, t2 ) = p(r) ⊗ u(r, t1 ) = p(r) ⊗ p(r), ... so that, through application of the Central
Limit Theorem, we have
∞
u(r, t∞ ) = ∏
⊗ pn (r) ≡ p(r) ⊗ p(r) ⊗ ... = Gauss(r)
n =1

where Gauss(r) is a normalised three-dimensional Gaussian function such that

∞
Gauss(r)d3 r = 1
−∞

From Equation (21) we can now consider the equation u(r, tn ) = Gauss(r) + (r, tn ), or,
after taking p = 1 norms, ū(tn ) ≤ 1 + ¯ (tn ) where ū(tn ) = u(r, tn )1 . For a discrete time series
un > 0∀n, say, we compute the Lyapunov exponent using the relatively simple formula

N  
1 u n +1
λ=
Nτ ∑ ln un
. (23)
n =1

Hence, for a time series which is assumed to be predicated on Equation (10), we can compute
the corresponding Lyapunov exponent using Equation (23), albeit, in practice, for a finite array of
size N. This includes financial time series data when λ can be computed for a moving look-back
window to generate a signal composed of Lyapunov exponents. In this context, the product Nτ merely
scales the computed value of the exponent but if un+1 > un , ∀n = 1, 2, ..., N then λ > 0 and if
un+1 < un , ∀n = 1, 2, ..., N then λ < 0. Hence, irrespective of the scale used, a change of polarity in
the value of λ is a signature of a change in the gradient of the time series. For this reason a change in
polarity of the Lyapunov exponent can be used to quantify the transition between the growth or decay
of a financial series.
An example of this is given in Figure 4 which shows a financial signal (the first 1000 elements of
the FTSE 100 prices given in Figure 3)—from 14/03/2006–26/02/2010—which has been normalised for
display purposes, i.e., un := un /un ∞ . The associated Lyapunov exponent has been computed using
function Lyapunov given in Appendix A.2 and re-scaled for values τ = 0.01 and N = 32 according to
Equation (23). Note that the first N values in Figure 3 are missing which is due to the window being a
look-back window holding data that contributes to the first computation of the exponent at point N.

144
Mathematics 2019, 7, 1057

Figure 4. Example of computing the time varying Lyapunov exponent (lower signal) for a financial
signal (upper signal after normalisation) through application of Equation (23) for a look-back window
of size N = 32 and with τ = 0.01.

As with other metrics computed from financial time series, the Lyapunov exponent obtained
depends critically on the look-back window that is applied. However, subject to a delay which is
proportional to the size of the look-back window used (determined by the time scale of the analysis),
the polarity (and continuity thereof) of the signal can be used to estimate the macroscopic trends in a
financial time series as illustrated in Figure 4. This is discussed further later on in this paper.
Since we can write Equation (23) in the form

N
1
λ=
Nτ ∑ (ln un+1 − ln un )
n =0

and that
ln un+1 − ln un d
∼ ln u(t),
τ dt
then, for the continuous case with time series function u(t), t ∈ [0, T ], we can write

T  
1 d 1 u( T )
λ∼ ln u(t)dt = ln
T dt T u (0)
0

giving a result that is analogous to Equation (19).

9. The Evolution Equation, Volatility and Risk

For a stochastic source term s(r, t), as given in Equation (11), Equation (14) becomes

∂
τ u(r, t) = −n(t) ⊗ u(r, t) + n(t) ⊗ u(r, t) ⊗ p(r) + n(t) ⊗ s(r, t)
∂t

Consider the case when p(r) = δn (r). Integrating over r ∈ Rn , we can then write the rate equation

t t
dt u(t) = σ[n(t) ⊗ s(t)] ⇒ u(t) = σ [n(t) ⊗ s(t)]dt = σs(t) ⊗ n(t)dt (+constant)

where
∞ ∞
d 1
dt u(t) ≡ u ( t ), u ( t ) = u(r, t)dn r, s(t) = s(r, t)dn r and σ = .
dt τ
−∞ −∞

145
Mathematics 2019, 7, 1057

Suppose we write this equation in the form

1 d σ
dt ln u(t) ≡ u(t) = [n(t) ⊗ s(t)] , u(t) > 0 ∀ t
u(t) dt u(t)

and consider an iterative solution for u(t) given by

σ
dt ln uk (t) = [n(t) ⊗ s(t)] , k = 0, 1, 2, ...
uk (t)

so that the first iterate u(t) := u1 (t) becomes the solution to the rate equation

dt ln u(t) ∼ σ f (t) (24)

where, for u0 (t) = 1,

f ( t ) = n ( t ) ⊗ s ( t ).

Equation (24) then shows that the volatility is a measure of the randomness of ln u(t) through the
convolution of s(t) with the time integration of n(t). If, for example, | s(t) |≤ 1, then, in the term σs(t),
σ determines the amplitude of s(t).
Equation (24) does not provide a practically useful formula for σ as it relies on defining the
functions n(t) and s(t) when what is ideally required is a definition for σ that relies on knowledge of
u(t) alone. To do this we are required to derive a formula for σ in terms of the function u(t) through
the elimination f (t) and this requires a condition to be applied. In this context, suppose we assume
that f (t) is a phase only function (with unit amplitude) of compact support T and with a bandwidth
Ω. This requires that both s(t) and n(t) are phase only functions of the same compact support and
bandwidth. In this case F (ω ) = exp[iθ (ω )] where θ (ω ) is the "Phase Spectrum" and using Parseval’s
Theorem, we have

T/2 Ω/2

1 Ω
| f (t) |2 dt = | F (ω ) |2 dω = .
2π 2π
− T/2 −Ω/2

Hence, we obtain an expression for the volatility given by

" ⎛ ⎞ 12

T/2
2π
σ= dt ln u(t)2 , dt ln u(t)2 := ⎝ | dt ln u(t) |2 dt⎠ .
Ω
− T/2

For a uniformly sampled discrete time series un , n = 1, 2, 3, ..., N, application of a forward

differencing scheme for a time interval Δt when
 
ln un+1 − ln un 1 u n +1
dt ln u(t) → = ln
Δt Δt un

gives
" (  ( (  (   2 ! 12
2π ( (
(ln un+1 ( ,
( (
(ln un+1 ( =
N  
ln un+1 
σ=
ΩΔt ( u n (2 ( un ( ∑  un 
.
2 n =1

The sampling interval Δt of un is related to the sampling interval Δω of the discrete Fourier
transform of un by the equation
2π
ΔtΔω =
N

146
Mathematics 2019, 7, 1057

and since the bandwidth of the discrete spectrum of un is NΔω is is clear that ΔtΩ = 2π. Thus we
derive a simple formula for the volatility given by
(  (
( u n +1 (
σ=(
(ln
( . (25)
un ( 2

Comparing Equation (25) with Equation (23), we observe similarities in regard to the commonality
of the quotient un+1 /un and the logarithmic function but where λ < 0 or λ ≥ 0 and where σ ≥ 0∀n.
An example of the short time volatility is given in Figure 5 which shows a financial signal (the
first 1000 elements of the FTSE 100 prices given in Figure 3), normalised for display purposes. In this
example σ in Equation (25) was computed using function volatility given in Appendix A.3 for N = 32.
In financial time series modelling, the volatility is a measure of the noise in the signal. For data
that has a stable trend (up or down) the volatility is relatively low, and, in this context, trading is best
undertaken working with financial signals that have a low volatility other than options trading where
there may have been a “bet” of a move of a certain magnitude. In this sense, the volatility of a signal
provides a measure of the risk, a low risk loosely equating to a low volatility. In the derivation of the
volatility provided in this section: σ = 1/τ where τ is a coefficient in Equation (14). In this respect, and,
in context of the evolution equation, τ is a measure of risk, the greater the value of τ the lower the risk
associated with an investment. For short δ(t) type memory functions the GKFE reduces to the classical
Kolmogorov–Feller equation which, in terms of its relationship to the evolution equation requires that
τ << 1. Thus low risk requires that a financial time series is characterised by long memory functions,
at least in terms of the model compounded in Equation (14)—a result that makes intuitive sense.

Figure 5. Example of computing the time dependent volatility (lower signal) using Equation (25) for a
normalised financial signal (upper signal) with a look-back window of N = 32.

10. Trend Analysis Using the Lyapunov Exponent to Volatility Index Ratio
The changes in polarity or “zero-crossings” associated with the Lyapunov exponent (computed
on a moving window basis) as discussed in Section 8 provide the positions in time where there is a
transition in the type of trend (growth leading to decay and decay leading to growth). The value of the
volatility indicates the “stability” of the time series, the temporal characteristics of all indicators being
dependent of the size of the window or "period" used. This suggests scaling the Lyapunov exponent
with the inverse of the volatility, i.e., computing the quotient

λ
λσ = (26)
σ

147
Mathematics 2019, 7, 1057

where σ is defined by Equation (25) and λ is defined by Equation (23) with τ = 1/N thereby making
λσ scale independent. This index then assesses not only changes in the direction of a trend but
the corresponding stability of that trend. This idea has obvious applications to a range of time
series but especially in regard to financial time series analysis where forecasting both the type and
characteristics of a trend is of fundamental importance, a positive trend with low volatility indicating a
good investment horizon, for example. We define λσ as the Lyapunov-to-volatility ratio (LVR). Figure 6
shows the time varying LVR of a financial signal (the first 1000 elements of the FTSE 100 prices given
in Figure 3 after normalisation) for N = 32.

Figure 6. Example of computing the time dependent LVR λσ (solid line —) for a normalised financial
signal (bold solid line —) with a look-back window of N = 32.

10.1. Pre- and Post-Filtering

As shall be discussed later, the numerical accuracy of results obtained in predicting a trend and its
longevity, is critically dependent on the filtering of both the input data and λσ —pre- and post-filtering,
respectively.

10.1.1. Pre-Filtering
The positions in time at which the zero crossings are evaluated using Equation (26) depend
on the accuracy of the algorithm used to compute λσ which in turn, depends on the intrinsic noise
associated with the time series data. This can yield errors in the positions at which the zero-crossings
are computed especially in regard to changes associated with very short time micro-trends.
In the context of longer term macro-trends, such micro-trends may legitimately be interpreted as
noise although, in the context of financial times series analysis, for example, the term "noise" must be
understood to reflect legitimate price values. To overcome this effect, un is filtered using a moving
average filter defined by:
u(t) := w(t) ⊗ u(t)

where 
1/W, | t |≤ W
w(t) =
0, | t |> W
and W defines the length of the “moving window”. The function given in Appendix A.4 provides a
moving average filter for pre-filtering the data un using a window of size W.

148
Mathematics 2019, 7, 1057

10.1.2. Post-Filtering
In addition to pre-filtering the time series data, an option for post-filtering the λσ is required to
further control the dynamic behaviour of this index. We therefore again consider a moving average
filter given by
λσ (t) := w(t) ⊗ λσ (t)

where 
1/T, | t |≤ T
w(t) =
0, | t |> T
and T defines the length of the “moving window”, T = W.

10.2. Zero-Crossings Analysis

On the basis of the ideas considered in the previous section, the critical points at which a trend
forecasting decision is made are the zero crossing points associated with λσ . By computing λσ (t) where
t is the position in time of the window, identification of the zero crossings denoted by the function
zc (t) involves the follow basic procedure:
⎧
⎪
⎨+1, λσ (t) < 0 & λσ (t + ε) ≥ 0;
⎪
zc (t) = −1, λσ (t) > 0 & λσ (t + ε) ≤ 0;
⎪
⎪
⎩0, otherwise.

where ε is a small perturbation in time. This procedure generates a series of Kronecker delta functions
whose polarity determines the position(s) in time at which a trend is expected to be positive or negative.
Thus the function zc (t) identifies the zero crossings associated with the end of an upward trend and
the start of a downward trend when zc (t) = −1 and the end of downward trend and the start of an
upward trend when zc (t) = +1. This is therefore a "critical indicator" in regard to forecasting the
trending behaviour of a time series.

10.3. Back-Testing Evaluator

Back-testing algorithms are designed to “gauge” the accuracy of results in terms of trend
predictions, for example, and, are usually, but not exclusively, related to testing a strategy for forecasting
the behaviour of a financial time series. They are usually designed to assess the overall accuracy of
some trading strategy based on historical data when the future outcomes of such a strategy can be
evaluated. In this context, the function given in Appendix A.5 evaluates the performance associated
with the zero-crossings analysis discussed in the previous section. This evaluation operates on the basis
that the price differences should reflect the interval between the start and end points of a predicted
trend if the prediction is correct. Thus, in the case when zc (t) > 0 and the trend is positive, the price
difference between this point in the time series and the next point in time series when zc (t) < 0 should
be positive, thereby representing a net price gain between the two zero crossings. Similarly, when
zc (t) < 0 and the trend is negative, the price difference between this point in the time series and the
next point in time series when zc (t) > 0 should be negative, thereby representing a net price loss
between the two zero crossings. In those cases where this occurs throughout the duration of the time
series considered, the predicted entry and exits points are taken to be correct, or else, they are taken
to be incorrect. The accuracy associated with this evaluation is computed as a percentage in terms of
successful entries and exits, i.e., going “long” (when an investment might be made because the price
of a commodity is increasing) and going “short” (when an investment would be held or sold at the
start of a downward trend), respectively.

149
Mathematics 2019, 7, 1057

10.4. Example Results

A function called Backtester is provided in Appendix A.6 which gives the user options on the
sizes of the look-back window length W and T and the size L of the data stream that is used from the
available input data. This data is provided in the form of a column vector via a .txt file. The function
normalises this data so that it can be plotted on a scale that is consistent with the scale of λσ [n].
The function provides a plot which shows the evolution of un (normalised), λσ [n] and zc [n] and then
evaluates the results using function evaluator as discussed in the previous section. Note that both the
Lyapunov exponent and the volatility are evaluated from the original data (and not the normalised
data used for the plot) using a look-back window of T.
Figure 7 shows some example results of running Backtester for the first 1000 elements of the FTSE
100 prices given in Figure 3. The three examples provided are for function Backtester(10,10,1000),
Backtester(20,10,1000) and Backtester(30,10,1000) for which the combined entry/exit (long/short)
accuracy is 36.55% , 64.58% and 72.73%, respectively. From these results it is clear that the accuracy
improves significantly with the extent of the pre-filtering that is applied to the time series before
computation of the LVR. This is to be expected as pre-filtering reduces the noise associated with the
time series prior to the computation of the LVR.

Figure 7. Example graphical outputs from function Backtester. Each plot shows the post-filtered (for
T = 10) LVR λσ [n] (dotted line), the pre-filtered time series un after normalisation (dot-dashed line)
and zc [n] (solid line) which identifies the zero-crossings of λσ [n]. The plots provided are for the case
when pre-filtering is undertaken for look-back window sizes of W = 10 (top), W = 20 (centre) and
W = 30 (lower plot). The financial time series data used in this case is the FTSE100 daily prices from
14/03/2006 to 26/02/2010.

In order to quantify both the pre- and post-filtering effect on the combined accuracy of the
long/short predictions, Figure 8 shows a surface (mesh) plot of the combined accuracy as a function
of the pre- and post-filtering look-back window sizes W and T, respectively. The maximum value
associated with this ‘WT-map’ is 87.5% which occurs at (W, T ) coordinates (40, 10). From Figure 8
it can be seen that the highest combined accuracies (>70%) are obtained for approximate values
of W ∈ [30, 50] and T ∈ [10, 20]. However, it should be noted that WT-maps of this kind are
data dependent and will vary with the type of financial time series that is processed and on the
non-stationary characteristics that occur over the length of the data series that is chosen (i.e., the
input parameter L in function Backtester). Hence, WT-maps of the type given in Figure 8 provide a

150
Mathematics 2019, 7, 1057

“signature” for a financial signal from which optimal values of the pre- and post filtering windows can
be established. This optimisation is based on finding the smallest values of W and T that will maintain
a combined accuracy compatible with an expected return over a given time scale.
As a comparative example, Figure 9 shows the equivalent WT-map for 1000 elements of the
Eruo-Dollar (USA) daily (averaged) exchange rates from 29/04/2008 to 27/02/2012 as given in
Figure 10. In this case, a maximum value of 83.33% occurs at WT coordinates with minimum values of
(38, 12). Although the quantitative details of this WT-map are unique to the data used, in qualitative
terms, it is similar to the WT-map given in Figure 8 revealing that greater accuracy is achieved for
large values of W relative to T which is intuitively to be expected. Clearly, for any specific financial
date series, a WT-map is required to provide an optimal accuracy associated with the trend analysis
of that series under the assumption that the stochastic behaviour of the series is stationary, i.e., the
financial signal is Ergodic.

Figure 8. Surface (mesh) plot of the combined long/short predictive accuracy as a function of the
pre- and post-filtering look-back window sizes W and T, respectively, for FTSE100 daily prices from
14/03/2006 to 26/02/2010.

Figure 9. Surface (mesh) plot of the combined long/short predictive accuracy as a function of the
pre- and post-filtering look-back window sizes W and T, respectively, for Eruo-Dollar (USA) daily
(averaged) exchange rates from 29/04/2008 to 27/02/2012 as given in Figure 10.

151
Mathematics 2019, 7, 1057

Figure 10. Eruo-Dollar (USA) daily (averaged) exchange rates from 29/04/2008 to 27/02/2012.

11. Price Prediction Using Evolutionary Computing

The trend analysis considered in the previous sections provides evidence of being able to predict
a positive or negative trend in a financial signal over a period of time that yields a net positive of
negative gradient, respectively, before the trend is reversed. This is based on pre-filtering the financial
signal and post-filtering the LVR in which the accuracy achieved is based on the look-back windows of
both filters. Such an approach only provides a statement on the expected future trend of a financial
signal; it does not provide an estimate of the actual future price.
On the basis of Equation (15) and the random walk hypothesis it represents, it is not possible to
determine a future price with 100% accuracy whatever the time scale, given that most financial signals
are known to be self-affine stochastic fields which exhibit the same statistical distributions over all
time scales. Thus, it is well known and understood (but not always appreciated) that in economics,
only an estimate (essentially an informed guess) of a future price is possible. However, in principle,
the lower the volatility of the signal, the less likely it is to exhibit large random variation at some
future (short) time and hence, the larger the LVR the more likely it is that an estimate of a future price
will be a more accurate prediction. In terms of Equation (15), this means that u(t + τ ) ∼ u(t) given
that s(t) ∼ 0, i.e., ‘tomorrow’s price is likely to be close to today’s price. This provides the basis for
using evolutionary computing to estimate short time price values by using the LVR to flag when the
approach can be used effectively, i.e., when the LVR reaches a maximum or minimum above or below
a certain threshold, respectively—as illustrated in Figure 7 for a threshold of 2, for example.

11.1. Evolutionary Computing

Evolutionary computing (EC) involves “applying the Darwinian principles of natural selection
to algorithmic problem solving” [42] and has its origins in the 1960s with the introduction of
“evolutionary programming” [43], “genetic algorithms” [44], and “evolutionary strategies” [45].
Following independent developments in the 1990s these areas merged to form the discipline of
genetic programming known today as EC in which a correlation exists between natural evolution and
evolution by computational problem solving [46].
In the context of a local environment that has a population striving for survival and to reproduce,
with natural evolution, the success (fitness) of each individual is dependent on their environment and
how well they meet their goals. Similarly, with a trial-and-error mathematical process, a candidate
solution is judged in the context of the problem that it is trying to solve and how well the candidate
solves the problem which determines whether or not it is kept as a candidate solution. A common
theme in EC is the idea of taking a population of individuals “operating” according to environmental
pressures causing natural selection and thereby the growth of a fitter population. Many aspects of EC

152
Mathematics 2019, 7, 1057

are stochastic and the starting point of candidate solutions can be either deterministic or stochastic.
In either case, the aim is to produce a “solution” that minimises some fitness function.

11.2. Eureqa
Eureqa is an EC tool originally developed by the Cornell Creative Machine Laboratory (Cornell
University) and commercialised by Nutonian Inc. ( Boston, MA, USA) [47]. The underlying principle
is to use genetic programming to generate equations, each of which provides an increasingly better
fitness function to model a given dataset. The system iteratively generates a sequence of non-linear
functions to describe input (digital) signals which may include stochastic signals [48]. It is a modelling
engine predicated on artificial intelligence using evolutionary searches to determine an equation that
represents a set of data [49]. The system automatically discovers formulae through evolutionary
algorithms requiring no human intervention starting by randomly creating equations via sequences
of mathematical building blocks based on a combination of common functions. The content of these
formulae is ordered only by a basic syntax (e.g., two addition signs cannot appear one after the other).
Beyond this basic syntax, the sequences generated by the program are entirely random.

11.3. Application to Financial Forecasting

With a little data “Eureqa generates fundamental laws of nature” [50]. However, there have been
few applications of EC to financial forecasting. This is partly due to the significance of Equation (15)
and the basic random walk hypothesis which financial signals adhere to, albeit as self-affine stochastic
fields. Thus although EC can be used to generate a non-linear equation for some short time financial
signal, no fundamental significance in terms of a “law of nature” can be inferred by such an equation
due to the random walk nature of the data that is used. To date, the only “law of nature” that can be
used to describe financial signals is that they are statistically self-affine fields to which the fractal market
hypothesis is thereby applicable. Nevertheless, EC can be used to provide short time predictions
including the performance of equity markets [51] and energy commodities [52], for example. This is
done by using EC to generate representative equations for existing prices over a look-back window
and can, in principle, be applied successive for a moving (look-back) window especially for time
periods where the volatility of the time series is low and future prices can be expected to be random
but locally similar to past prices.

11.4. An Example Result

With reference to Figure 7, we consider the daily prices for array values between 870 and 900
(inclusively) which correspond to days 24/08/2009 to 08/09/2009 when the LVR is ∼3 and relatively
flat. With these 30 price values, Eureqa provides the following formula:

f (t) = 5025.73417939762 + 8.96527863946579t2 + 1.52597679067939 × 10−6 t6

+ cos(8.96527863946579t) − 76.8453768284695t − 0.253321938783733t3
− 48.5578781261177 sin(0.96446841878421 + 7.16244232473996t) (27)

obtained after 51,056 generations giving a correlation coefficient of 0.98362871, an R2 (coefficient of

determination) goodness of fit of 0.96684623, a mean absolute error of 16.623419 and a complexity
of 51.
Figure 11 shows a comparison of the true price values with the estimates obtained using a
discretised version of Equation (27) given by

f n = 5025.73417939762 + 8.96527863946579t2n + 1.52597679067939 × 10−6 t6n

+ cos(8.96527863946579tn ) − 76.8453768284695tn − 0.253321938783733t3n
− 48.5578781261177 sin(0.96446841878421 + 7.16244232473996tn ), n = 1, 2, ..., N, tn = n (28)

153
Mathematics 2019, 7, 1057

for N = 30, and, additionally, for n = 31, 32 and 33 thereby providing future price prediction for three
days ahead.

Figure 11. FTSE 100 price values (dots) for 30 days between 24/08/2009 and 08/09/2009 (inclusively)
and the estimate (solid line) obtained from the evolutionary computed formula given by Equation (28).
The graph compares the actual prices (circled dots) and predicted future values for three days after
08/09/2009 using the same formula.

A comparison of the numerical values for these price predictions is given in Table 2.

Table 2. Table comparing the actual and predicted prices for three days of the FTSE100 using
Equation (28).

Day 09/09/2009 10/09/2009 11/09/2009

tn 31 32 33
Predicted price value f n 5022.9 5066.6 5087.4
Actual price value un 5138.0 5108.9 5154.6

11.5. Discussion
With reference to Figure 11, the local trend in prices before and inclusive of element 30 (i.e.,
elements 26–30) is downward and so based on the principle of Equation (15) for s(t) ∼ 0 and
application of exponential smoothing for time series forecasting [53], for example, continuation of
this trend will lead to inaccurate predictions that are inconsistent with the local increase in prices for
elements 31, 32 and 33—the circled dots in Figure 11. However, the equivalent future predictions given
by Equation (28) are consistent with the actual values which represent a short time up-ward trend
as shown in Table 2. The predictive ability of EC can only be considered for very short future time
increments (a look-forward prediction window) but this example result does provide evidence for the
success of using EC exercised on a moving look-back window basis.
A quantitative study on the accuracy of this approach in terms of the look-back window and
the look-forward (prediction) window relative to the local LVR lies beyond the scope of this work.
However, it is to be expected that the success of this approach will be predicated on the size of the
amplitude of | λσ | when the volatility is low. Hence, based on the results given in Figure 7, an EC
moving window approach may be used when | λσ |≥ 2 where λσ is given by Equation (26). In this

154
Mathematics 2019, 7, 1057

context, the LVR not only provides a method of predicting trends (subject to appropriate pre- and
post-filtering) based on its change in polarity but also flags when to apply EC to generate future
price estimates.

12. Derivations of the Diffusion Equation from the Evolution Equation

So far in this paper, we have developed a predictive indicator that is based on combining the
Lyapunov exponent and the volatility into a ratio (the LVR), both parameters having been derived
from E3 and computed on a moving window basis. We have then used the amplitude of this ratio
to gauge the likelihood of using EC to successfully predict short term future price values. We have
not yet studied the effect of applying specific models for the PDF associated with E3 which is the
subject of later sections. In particular, we show how the classical diffusion equation is a result of
considering a Gaussian PDF in E3 and the non-classical fractional diffusion equation is the result of
considering a non-Gaussian PDF, in particular, a Lévy distribution and undertaken using the associated
characteristic functions.
In this section we consider three approaches to deriving the classical diffusion equation in order
to show the connectivity between this equation and E3 in terms of applying different conditions and
approximations. We start with Einstein’s original approach which is independent of the specific PDF
but on the condition that the PDF is symmetric.

12.1. Einstein’s Derivation for r ∈ R1

In his 1905 paper [1], Einstein considered the one-dimensional case, when r ∈ R1 , and where the
PDF is taken to be symmetric so that p( x ) = p(− x ). In this case, Equation (10) can be written as

∞ ∞ ∞
u( x, t + τ ) = p( x − λ)u(λ, t)dλ = p( x + λ)u(λ, t)dλ = p(λ)u( x + λ, t)dλ
−∞ −∞ −∞

Taylor expanding u( x, t) to first order in time, and, to second order in space, we then obtain

∞  
∂ ∂ λ2 ∂2
u( x, t) + τ u( x, t) = dλp(λ) u( x, t) + λ u( x, t) + u ( x, t )
∂t ∂x 2 ∂x2
−∞

∞ ∞ ∞
∂ ∂2 λ2
= u( x, t) p(λ)dλ + u( x, t) λp(λ)dλ + u( x, t) p(λ)dλ
∂x ∂x2 2
−∞ −∞ −∞

∞
∂2 λ2
= u( x, t) + u( x, t) p(λ)dλ
∂x2 2
−∞

since
∞ ∞
p(λ)dλ = 1 and λp(λ)dλ = 0.
−∞ −∞

We can thus write the equation

∂ ∂2
u( x, t) = D 2 u( x, t) (29)
∂t ∂x
where
∞
λ2
D= p(λ)dλ. (30)
2τ
−∞

155
Mathematics 2019, 7, 1057

which is the one-dimensional diffusion equation for diffusivity D with dimensions of Length2 /Time.
This derivation of the diffusion equation relies on the conditions τ << 1 and λ2 << 1 which are
required in order to truncate the Taylor series expansion of u( x, t + τ ) in time and u( x + λ, t) in space.
However, this derivation of the diffusion equation is independent of the PDF (subject to the condition
that the PDF is symmetric) which determines the diffusivity D through Equation (30).

12.2. Einstein’s Derivation for r ∈ R3

A similar approach can be used to deriving of the diffusion equation for r ∈ R3 as shall now be
demonstrated. In this case

u(r, t + τ ) = p(r ) ⊗ u(r, t), p(r ) = p(−r )

can be written out in the form

∞
u(r, t + τ ) = u(r + λ, t) p(λ)dλ
−∞

where λ is a scalar with dimensions of length and components λ x , λy and λz . Expanding u(r + λ, t) in
terms of a three-dimensional Taylor series,

∂u(r, t) λ2x ∂2 u(r, t) λy ∂2 u(r, t) λ2z ∂2 u(r, t)

2
∂u(r, t) ∂u(r, t)
u(r + λ, t) = u(r, t) + λ x + λy + λz + + +
∂x ∂y ∂z 2! ∂x2 2! ∂y2 2! ∂z2

∂2 u(r, t) ∂2 u(r, t) ∂2 u(r, t)

+λ x λy + λ x λz + λy λz + ...
∂x∂y ∂x∂z ∂y∂z
so that, for τ << 1,

∞ ∞  
∂u(r, t) ∂u(r, t) ∂u(r, t) ∂u(r, t)
u(r, t) + τ = u(r, t) p(λ)dλ + λx + λy + λz p(λ)dλ
∂t ∂x ∂y ∂z
−∞ −∞

∞
 
λ2x ∂2 u(r, t) λy ∂2 u(r, t) λ2z ∂2 u(r, t)
2
+ + + p(λ)dλ
2! ∂x2 2! ∂y2 2! ∂z2
−∞

∞  
∂2 u(r, t) ∂2 u(r, t) ∂2 u(r, t)
+ λ x λy + λ x λz + λy λz p(λ)dλ
∂x∂y ∂x∂z ∂y∂z
−∞

We then obtain
∞ ∞ ∞
∂ λ2x ∂2 u(r, t) λ x λy ∂2 u(r, t) λ x λz ∂2 u(r, t)
τ u(r, t) = p(λ)dλ + p(λ)dλ + p(λ)dλ
∂t 2 ∂x2 2 ∂x∂y 2 ∂x∂z
−∞ −∞ −∞
∞ ∞ λ2 2 ∞
λy λ x ∂2 u(r, t) y ∂ u ( r, t ) λy λz ∂2 u(r, t)
+ p(λ)dλ + p(λ)dλ + p(λ)dλ
2 ∂y∂x 2 ∂y 2 2 ∂y∂z
−∞ −∞ −∞
∞ ∞ ∞
λz λ x ∂2 u λz λy ∂2 u(r, t) λ2z ∂2 u(r, t)
+ p(λ)dλ + p(λ)dλ + p(λ)dλ
2 ∂z∂x 2 ∂z∂y 2 ∂z2
−∞ −∞ −∞
∞ ∞ ∞
∂u(r, t) ∂u(r, t) ∂u(r, t)
+ λx p(λ)dλ + λy p(λ)dλ + λz p(λ)dλ
∂x ∂y ∂z
−∞ −∞ −∞

156
Mathematics 2019, 7, 1057

which can be written as

∂
u(r, t) = ∇ · D∇u(r, t) + V · ∇u(r, t)
∂t
where D is the diffusion tensor given by
⎛ ⎞
Dxx Dxy Dxz ∞
⎜ ⎟ λi λ j
D = ⎝ Dyx Dyy Dyz ⎠ , Dij = p(λ)dλ
2τ
Dzx Dzy Dzz −∞

and V is a flow vector which describes any drift velocity that the particle ensemble may have and is
given by ⎛ ⎞
Vx ∞
⎜ ⎟ λi
V = ⎝Vy ⎠ , Vi = p(λ)dλ.
τ
Vz −∞

Note that as λi λ j = λ j λi , the diffusion tensor is diagonally symmetric (i.e., Dij = D ji ).

For isotropic diffusion when λi λ j = 0 for i = j and λi λ j = λ2 for i = j and with a zero
drift velocity when V = 0,
⎛ ⎞
D 0 0 ∞ 2
∂ ⎜ ⎟ λ
u(r, t) = ∇ · ⎝ 0 D 0 ⎠ ∇u(r, t) = D ∇2 u(r, t), D = p(λ)dλ
∂t 2τ
0 0 D −∞

12.3. PDF Dependent Derivation of the Diffusion Equation

Consider the case when, for r ∈ R1 , p( x ) is a zero-mean normal (Gaussian) distribution with
Standard Deviation σ and Variance σ2 , i.e.,
 
1 x2
p( x ) = √ exp − 2 ,
2πσ2 2σ

Taylor expansion to first order of Equation (10) followed by application of the convolution theorem
yields the Fourier space equation

∂
U (k, t) + τ U (k, t) = P(k )U (k, t) (31)
∂t
where
∞
U (k, t) = u( x, t) exp(−ikx )dx
−∞

and
∞  
σ2 k2
P(k) = p( x ) exp(−ikx )dx = exp − ,
2
−∞

P(k ) being the Characteristic Function.

Suppose we now consider the case when the variance is small, i.e., σ2 << 1. Then
 2
σ2 k2 1 σ2 k2 σ2 k2
P(k) = 1 − + + ... ∼ 1 − , (σk )2 << 1
2 2! 2 2

and Equation (31) can be written as

∂ σ2 k2
U (k, t) = −U (k, t)
∂t 2τ

157
Mathematics 2019, 7, 1057

through which we again obtain the diffusion equation

∂ ∂2 σ2
u( x, t) = D 2 u( x, t) where D =
∂t ∂x 2τ
given that
∂2
−k2 U (k, t) ↔ u( x, t).
∂x2
In this case, the “key” to the derivation of the diffusion equation is the assumption that the
variance of a normal distribution is small and that τ << 1. We note that an identical analysis in the
two- and three-dimensional domains yields the the two- and three-dimensional diffusion equation

∂
u(r, t) = D ∇2 u(r, t), r ∈ Rn , n = 2, 3
∂t

12.4. Generalisation
We can generalise this approach further by writing the evolution equation in Fourier space using
the convolution theorem and in expanded form as

∂ τ 2 ∂2 τ2
U (k, t) + τ U (k, t) + U (k, t) + ... = U (k, t) − τDk2 U (k, t) + D2 k4 U (k, t) − ...
∂t 2! ∂t 2 2!
so that upon inverse Fourier transformation we have, for r ∈ Rn , n = 1, 2, 3

∂ τ 2 ∂2 τ2
τ u(r, t) + u(r, t) + ... = τD ∇2 u(r, t) − D2 ∇4 + ...
∂t 2! ∂t2 2!
Equating terms with the same coefficients in regard to powers of τ, we have (for any positive
integer m)

∂ ∂2 ∂m
u(r, t) = D ∇2 u(r, t), 2 u(r, t) = D2 ∇4 u(r, t), ..., m u(r, t) = D m ∇m+2 u(r, t)
∂t ∂t ∂t
Since all such equations can be constructed from the diffusion equation, i.e.,

∂2 ∂
u(r, t) = D ∇2 u(r, t) = D ∇2 [ D ∇2 u(r, t)], ...
∂t2 ∂t

this analysis confirms that the diffusion equation is E3 for the case when the PDF is a Gaussian
distribution.

12.5. Green’s Function Solution

For the initial condition u0 (r) ≡ u(r, t = 0), r ∈ Rn , n = 1, 2, 3 and in the infinite domain,
the Green’s function solution to the homogeneous diffusion equation is [54]

u(r, t) = G (r, t) ⊗ u0 (r)

where G (r, t) is the Green’s function given by

 n  
1 2 r2
G (r, t) = exp − , t > 0; n = 1, 2, 3.
4πDt 4Dt

158
Mathematics 2019, 7, 1057

12.6. The Black–Scholes Model

There is a synergy associated with the diffusion equation and the Black–Scholes model for a call
premium which is compounded in the partial differential equation [55]

∂ 1 ∂2 ∂
c( x, t) + σ2 s2 2 c( x, t) + rs c( x, t) − rc( x, t) = 0
∂t 2 ∂s ∂s

where c( x, t) is the call premium, s is the stock price, σ is the volatility and r is the risk. Subject to
specific initial and boundary conditions, this equation can be transformed into the classical diffusion
equation through application of a change of variables when it can be written in the form

∂ 1 ∂2
u( x, τ ) + σ2 2 u( x, t)
∂τ 2 ∂x
which has the same Green’s function solution as given in the previous section, for n = 1 and initial
condition u( x, t = 0). Thus, just as the classical diffusion equation is a manifestation of the PDF
associated with E3 being normal, so the Black-Scholes model may be taken to be predicated on
Gaussian processes.

13. The Fractional Diffusion Equation

The fractional diffusion equation (FDE) can be derived by generalising the Gaussian characteristic
 
function P(k ) = exp −σ2 k2 /2 to the form

P(k ) = exp (−c | k |γ )

where γ ∈ [0, 2] is the Lévy index and c is a constant with dimensions of Lengthγ as previously
discussed in Section 7.2.
Using the Reisz definition of the fractional Laplacian operator ∇γ , r ∈ Rn , namely,

∇γ ↔ − | k |γ

with Dγ = c/τ, repetition of the analysis given in Section 12.4 yields the homogeneous FDE

∂
u(r, t) = Dγ ∇γ u(r, t)
∂t

where Dγ is the fractional diffusivity with dimensions of Lengthγ /Time, and, for r ∈ R3 with Cartesian
coordinates ( x, y, z),
∂γ ∂γ ∂γ
∇γ ≡ + +
∂ | x |γ ∂ | y |γ ∂ | z |γ
Thus, we obtain a fundamental connectivity between between Einstein’s evolution equation and
fractional calculus, i.e., application of the Lévy distribution in Equation (10) yields the FDE.

13.1. Continuity Equation

For the case when γ = 1, we can use the FDE to construct the transport equation

∂
u(r, t) + D1 n̂ · ∇u(r, t) = 0
∂t
where n̂ is the unit vector. This is a continuity equation, and, in the context of the evolution equation,
illustrates the connectivity between the concept of flux (the flow of an ensemble of particles) and the
Cauchy distribution (as discussed in Section 7.2).

159
Mathematics 2019, 7, 1057

13.2. Time-Independent Analysis

If we consider Equation (11), then for the time dependent case the FDE becomes

∇ γ u ( r ) = s ( r ), r ∈ Rn

where u(r) is a stochastic function. Since ∇γ u(r) ↔ − | k |γ we can construct the solution
 
S(k)
u(r) = Fn−1 γ
, S(k) ↔ s(r)
|k|

Using Equation (4) and the convolution theorem, we can then write u(r) as
 
cn,γ 1 n n−γ
Γ n−2 γ
u(r) = ⊗ s(r), 0 < Re[γ] < n; cn,γ = π 2 2   .
(2π )n | r |n−γ Γ γ2

This solution for u(r) defines the Riesz potential and has a fundamental scaling property obtained
by considering the convolution of the source function for a scaling factor λ when it is simple to
show that
cn,γ 1 1
⊗ s(λr) = γ u(λr)
(2π )n | r |n−γ λ
and hence u(r) is a scale-invariant stochastic function defined by the relationship

Pr[u(λr)] = λγ Pr[u(r)]. (32)

Thus, for a stochastic source, the Riesz potential u(r) is a random scaling self-affine field—a
random scaling fractal. In this context, Appendix B develops the relationship between the topological
dimension n, the fractal dimension D and the Lévy index γ which is given by

3n + 2 − 2γ
D= . (33)
2

Thus, for example, a Mandelbrot surface, which has a fractal dimension D = 4 − γ ∈ [2, 3], can be
defined in terms of the solution to the two-dimensional fractional Poisson equation (FPE)

∇γ u(r) = s(r), r ∈ R2 , γ ∈ [2, 1]

and if s(r) has a white spectrum, i.e., a spectrum whose power spectral density function (PSDF) is a
constant, then the PSDF of u(r) is determined by 1/ | k |4−γ .
We note that by Taylor expanding Equation (11) for the time-independent case, then in Fourier
space, we obtain U (k) = P(k)U (k) + S(k) and with P(k) = exp(−c | k |γ ) it can be shown that [37]
 
S(k) 1
u(r) = Fn−1 γ
∼ n+γ ⊗ s(r), r → ∞; r ∈ Rn
1 − exp(−c | k | ) r

This asymptotic result yields a similar inverse power law but with the scaling law,

Pr[u(r)] = λγ Pr[u(λr)],

a result that is a characteristic of scale-invariant field theory when the field equations are scale invariant
so that for any solution φ(r), say, of the field equations, there exist other solutions of the form λΔ φ(r)
for an exponent Δ (not necessarily related to γ).

160
Mathematics 2019, 7, 1057

13.3. Time-Dependent Analysis

We study the FDE with r ∈ R1 for a stochastic source, namely,
 
∂γ ∂
Dγ − u( x, t) = −s( x, t) (34)
∂ | x |γ ∂t

and consider the generic Green’s function solution

u( x, t) = g( x, t) ⊗ s( x, t)

where the convolution operation is taken to apply to both x and t. We are then required to compute
the Green’s function in this case.

13.4. Green’s Function for the Fractional Diffusion Equation

We consider an evaluation of the Green’s function for the fractional diffusion equation which is
defined as the solution to
 
∂γ ∂
Dγ − g(| x |, t) = −δ(| x |)δ(t), t ≥ 0. (35)
∂ | x |γ ∂t

Writing the Green’s function in terms of the Fourier transformation

∞
1
g(| x |, t) = G (k, ω ) exp(ik | x |) exp(iωt)dkdω
(2π )2
−∞

noting that
∞ ∞
1 1
δ(| x |) = exp(ik | x |)dk, and δ(t) = exp(iωt)dω
2π 2π
−∞ −∞

and using the Reisz definition of a fractional derivative, Equation (35) becomes

[− Dγ | k |γ −(iω )] G (k, ω ) = −1

which can be written in the factored form

1
[(| k |γ/2 +(−iω/Dγ )1/2 )(| k |γ/2 −(−iω/Dγ )1/2 )] G (k, ω ) = . (36)
Dγ

It is well known that for the equation

 
∂2 ∂2
Dγ − 2 g(| x |, t) = −δ(| x |)δ(t),
∂x2 ∂t

1 + + 1
(k2 − ω 2 /Dγ ) G (k, ω ) = or [(k + ω/ Dγ )(k − ω/ Dγ )] G (k, ω ) =
Dγ Dγ
and the outgoing Green’s function is given by

i +
g( x | x0 , ω ) = ' exp[i (ω/ Dγ ) | x |]. (37)
2ω Dγ

Generalising this result for Equation (36), we therefore consider the expression

i
g(| x |, ω ) = exp[i (−iω/Dγ )1/γ | x |], (38)
2Dγ (−iω/Dγ )1/γ

161
Mathematics 2019, 7, 1057

given that when γ = 2 and −iω/Dγ := ω 2 /Dγ , Equation (37) is recovered.

To find the time evolution of the Green’s function, we are required to take the inverse Fourier
transform of g(| x |, ω ), and evaluate the integral

∞
1 i
g(| x |, t) = exp[i (−iω/Dγ )1/γ | x |] exp(iωt)dω. (39)
2π 2Dγ (−iω/Dγ )1/γ
−∞

This can be achieved by writing the exponential function exp[i (−iω/Dγ )1/γ | x |] as a series
which yields the series solution
!
∞ n
i 1 i
g(| x |, t) = Fn−1 + i | x | + ∑ (− iω/D γ ) (n−1)/γ
| x | n
2Dγ (−iω/Dγ )1/γ n =2
n!
  !
in −1 (n−1)/γ
1/γ ∞
i Dγ
= Fn−1 + i | x | + ∑ n! Dγ ( iω ) (n−1)/γ
| x | n
2Dγ (−1)1/γ (iω )1/γ n =2
!
∞ n  
−1 (n−1)/γ
1/γ
i Dγ H (t) i n [(n−1)/γ]
=
2Dγ (−1)1/γ Γ(1/γ) t1−1/γ
+ i | x | δ ( t ) + ∑ n! Dγ
| x | δ ( t ) (40)
n =2

where
t
H (t) = δ(s)ds
−∞

is the Heaviside step function. This result comes from noting that for 0 < α < 1

1 1 1 dα
↔ and (iω )α ↔ δ(α) (t) ≡ α δ(t),
Γ ( α ) t 1− α (iω )α dt

the function δ[(n−1)/γ] (t) being defined in terms of Equation (9).

Note that from Equation (39) when γ = 2

∞ '
1 1
g(| x |, t) = √ √ exp(− iω/D2 | x |) exp(iωt)dω
2π D2 2 iω
−∞
i∞ √
1 1 exp(− s | x |)
= √ √ √ exp(st)ds
2 D2 2πi s D2
−i∞
1
= √ exp(− | x |2 /4D2 t), t > 0, (41)
2 πD2 t

which is the Green’s function for the classical diffusion equation where D2 is the classical diffusivity.

13.5. Asymptotic Solution

From Equation (40), it is clear that we can define the time dependent Green’s function for the case
when x → 0 as
cγ H (t) i
g(0, t) = 1−1/γ , cγ = 1−1/γ
. (42)
t 2Dγ (−1)1/γ Γ(1/γ)
The Green’s function solution to Equation (34) is then given by

u(t) = g(t) ⊗ s(t) (43)

162
Mathematics 2019, 7, 1057

where u(t), g(t) and s(t) represent the functions u(0, t), g(0, t) and s(0, t). We note that as x → 0,
Equation (41) reduces to
1
g(0, t) = √ , t>0 (44)
2 πD2 t
√
and that this result is consistent with Equation (42) given that for γ = 2, Γ(1/2) = π and we have

i 1
g(0, t) = √ √ = √ , t > 0.
2 D2 −1Γ(1/2)t1/2 2 πD2 t

The scaling relationship associated with Equation (43) is given by (c.f. Equation (32))

Pr[u(λx )] = λ1/γ Pr[u( x )]

and from Equation (33), the relationship between Fractal Dimension D and the Lv́ey index in this case
is D = (5 − 2/γ)/2, ∈ [1, 2]; ⇒ γ ∈ [2/3, 2]. Figure 12 shows examples of the function uγ (t) for
γ = 2/3, 1 and 3/2 using the same stochastic source function s(t). Comparing these results with the
example given in Figure 3, it is clear that the case of γ ∼ 1 provides a time series that (through visual
inspection) better matches that of the financial signal. This is verified through application of regression
applied to the data given in Figure 3 which yields γ = 1.1455 based on assuming that the data has an
amplitude spectrum | U (ω ) | with the following spectral power law:

1
| U (ω ) |∼ , | ω |> 0 (45)
| ω |1/γ

This value of γ is the one associated with the data given in Figure 3 in its entirety, and, like the
Lyapunov exponent and the volatility, it can be computed on a moving window basis to obtain a
(short) time dependent signature which is explored further in Section 15.

Figure 12. Examples of the function un given by the discretised form of Equation (43) for γ = 1/2, 1,
and 3/2 (left plots, respectively) and the associated complex plane representations obtained using
Equation (16)—right plots, respectively.

163
Mathematics 2019, 7, 1057

The regression algorithm used to achieve this result is given in Appendix A.7 and is based on
computing the exponent α associated with the power law U (ω ) ∼| ω |α using the least squares method
(LSM). For a uniformly sampled frequency (or time) series un > 0∀n, n = 1, 2, ..., N α is given by

N N N
∑ ln Un ∑ ln n − N ∑ ln Un ln n
n =1 n =1 n =1
α=  2 .
N N
∑ ln n − N ∑ ln n2
n =1 n =1

The Lv́ey index is then related to α by the equation γ = −1/α. Note that to compute γ using
the LSM requires computation of the amplitude spectrum using a discrete (fast) Fourier transform.
The data used is that in the positive half-space of the amplitude spectrum with the DC component
removed, thereby adhering to the condition | ω |> 0 in the spectral power law defined by Equation (45).
Thus, the LSM is applied for | Un |, n = 2, 3, ..., N/2.

13.6. Discussion: Impulse Response Functions for Classical and Fractional Diffusion
Given Equation (43), it is clear that, in the asymptotic limit x → 0, the difference between classical
and fractional diffusion is compounded in the different Green’s function given by Equations (42)
and (44). Thus, ignoring the scaling parameters in Equations (42) and (44) as well as those of their
Fourier transforms, we can compare the asymptotic solutions as follows:

• Classical Diffusion

1 | S(ω ) |
u(t) = √ ⊗ s(t), t > 0 ⇒| U (ω ) |= ' (46)
t |ω|

• Fractional Diffusion

1 | S(ω ) |
u(t) = ⊗ s(t), t > 0 ⇒| U (ω ) |= , γ ∈ [1, 2] ⇒ D ∈ [1.5, 2] (47)
t1−1/γ | ω |1/γ
Unlike classical diffusion, fractional diffusion is characterised by a range of values of the Lévy
index. The efficient market hypothesis is predicated on classical diffusion processes, based on E3 for
a Gaussian distribution. The ramifications of this is that the time series model for u(0, t) given by
√
Equation (46) is characterised by the impulse response function (IRF) 1/ t. By comparison, the fractal
market hypothesis is predicated on fractional diffusion processes based on E3 for a Lévy distribution.
The consequence of this is that the time series model for u(0, t) given by Equation (47) is characterised
by the IRF 1/t1−1/γ . Since financial signals tend to be non-stationary random fractals, variations in
γ as a function of time are informative. However, before we study this, we consider another way to
derive what is, in effective, the same basic result but via a different approach, an approach that is also
based on E3 but obtained via the GKFE subject to application of an appropriate memory function.
This is discussed in the following section.

14. Solution to the GKFE for an Orthonormal Memory Function

In this section, we show that the temporal power law which characterises
Equation (43)—i.e., 1/t1−1/γ —can be obtained from Equation (14) for a specific orthonormal
memory function. The purpose of this is to show another route to deriving the power law which is
informative in that it is based on the application of a memory function alone and does not involve
specific application of the FDE as presented in the previous section. In this case, and, for r ∈ R1 ,
by writing Equation (14) in the form

∂
τ u( x, t) + u( x, t) = u( x, t) − n(t) ⊗ u( x, t) + n(t) ⊗ u( x, t) ⊗ p( x )
∂t

164
Mathematics 2019, 7, 1057

we can construct a Green’s function solution is given by

u( x, t) = g(t) ⊗ u( x, t) − g(t) ⊗ n(t) ⊗ u( x, t) + g(t) ⊗ n(t) ⊗ u( x, t) ⊗ p( x ) (48)

where g(t) is the Green’s function given by

1
g(t) = exp(−t/τ ), t > 0
τ
which is the solution to
∂
τ g ( t ) + g ( t ) = δ ( t ).
∂t
Provided the Laplace transform of the function n(t) exists, we can write this Green’s function
solution as
u( x, t) = h(t) ⊗ u( x, t) ⊗ p( x ) (49)

where
n̄(s)
h(t) ↔
τs + n̄(s)
and ↔ denotes the Laplace transformation, i.e., the mutual transformation from t-space to s-space.
This result is obtained by using the convolution theorems for the Fourier and Laplace transforms,
when Equation (14) can be written as

u ¯ (k, s) − ḡ(s)n̄(s)u
¯ (k, s) = ḡ(s)u ¯ (k, s) + ḡ(s)n̄(s)u
¯ (k, s) p(k)

where
∞ ∞ ∞
¯ (k, s) =
u u( x, t) exp(−ikx )dx exp(−st)dt, ḡ(s) = g(t) exp(−st)dt,
0 −∞ 0

∞ ∞
n̄(s) = n(t) exp(−st)dt and p(k) = p( x ) exp(−ikx )dx
0 −∞

Thus, noting that ḡ(s) = (1 + τs)−1 , we can write

¯ (k, s) = − ḡ(s) ¯ (k, s) + ḡ(s) n̄(s)u

¯ (k, s) p(k ) = − n̄(s) u
¯ ( x, t) + n̄(s) u
¯ (k, s) p(k )
u n̄(s)u
1 − ḡ(s) 1 − ḡ(s) τs τs

leading to the equation

¯ (k, s) = h̄(s)u
u ¯ (k, s) p(k ).

Inverse Fourier-Laplace transformation then gives Equation (49).

Equation (49) supports an iterative solution of the form

um+1 ( x, t) = h(t) ⊗ um ( x, t) ⊗ p( x ), m = 0, 1, 2, ...

and we may therefore consider an approximation based on the first iterate, i.e.,

u( x, t) = h(t) ⊗ u0 ( x, t) ⊗ p( x )

The condition required for this approximation to apply can be obtained as follows: Given that

u( x, t)1 ≤ h(t)1 u0 ( x, t)1  p( x )1 = h(t)1 u0 ( x, t)1

165
Mathematics 2019, 7, 1057

then
u( x, t)1
≤ h(t)1
u0 ( x, t)1
and hence we required that
h(t)1 << 1. (50)

Further, if we consider the case when u0 ( x, t) = δ( x )s(t), then we can write

∞ ∞
u(t) = u( x, t)dx = h(t) ⊗ s(t) p( x )dx = h(t) ⊗ s(t).
−∞ −∞

If we now choose a memory function m(t) whose Laplace transform is s β−1 then the
orthonormality property n(t) ⊗ m(t) = δ(t) is satisfied if the Laplace transform of n(t) is s1− β given
that from the convolution theorem for Laplace transforms n̄(s)m̄(s) = 1. In this case

s 1− β 1 1
h̄(s) = = ∼ β , τ >> 1.
τs + s1− β 1 + τs β τs

Since
1 H (t)
↔
sβ Γ ( β ) t 1− β
we obtain the solution
u(t) = h(t) ⊗ s(t)

where
1
h(t) = , t > 0.
τΓ( β)t1− β
This solution is characterised by Riemann–Liouville (fractional) integral which has self-affine
properties, i.e., properties that exhibit "stochastic trending characteristics". In other words, u(t) defines
a random scaling fractal function whose impulse response function is 1/t1− β , a result that is, in light of
the above analysis, been shown to be a PDF independent first order solution to the GKFE for a memory
function
1
m(t) = .
Γ (1 − β ) t β
In order to comply with Condition (50), we require that

τ
1 1 1
h(t)1 = = << 1,
τΓ( β) t 1− β βΓ( β)τ 1− β
0

which is satisfied for the case when τ >> 1, β ∈ [0, 1).

Clearly, ignoring differences in scaling, compatibility of this solution for u(t) with Equation (43)
is obtained when β = 1/γ. Thus, subject to the conditions imposed in each case we have shown that
there exist temporal solutions to the FDE and the GKFE that exhibit a fundamental power law of
1/t1−1/γ for Lévy index γ. In the former case, the solution is predicated on defining the PDF in the
evolution equation (a Lévy distribution) whereas in the latter case, the result is independent of the
PDF but predicated on the definition of the memory function (with power law 1/t β ). In both cases,
the solution is characterised by a fractional integral which is self-affine, a property that is fundamental
to the analysis and interpretation of financial signals and underpins the fractal market hypothesis.

166
Mathematics 2019, 7, 1057

15. Time Varying Lévy- and α-Indices

As with the other indices considered in this paper, the time dependence of γ for a financial signal
can be obtained by computing it over a moving (look-back) window. Figure 13 shows an example of
this short time signature. for a financial signal (the first 1000 elements of the FTSE 100 prices given in
Figure 3), normalised for display purposes. In this example γ has been computed using the function
given in Appendix A.7.

Figure 13. Example of computing the time dependent Lévy index (solid line —) for a normalised
financial signal (bold solid line —) based on the application of Equation (45) using function Levy with
a look-back window of 50 data elements.

−1
This result assumes that the short-time amplitude spectrum adheres to the scaling law | ω | γ ,
and, strictly within the context of this spectral model, the numerical range of γ is only limited by the
original definition of the of a Lévy distribution, i.e., γ ∈ [0, 2] as given in Equation (17). The statement
γ ∈ (1, 2] given in Equation (47) is a result of imposing the condition that 0 < 1/γ < 1 in order that
the Fourier transform pair relationship given by Equation (3) is satisfied. However, if we arbitrarily
consider a modified IRF given by 1/t1−1/γ , γ ∈ [0, 2], then it is clear that we can consider a short time
scaling function given by (for t > 0)
⎧
⎪
⎨α > 0,
⎪ γ < 1;
u(t) = tα where α = 0, γ = 1;
⎪
⎪
⎩α < 0, γ > 1.

for the case when s(t) = δ(t). This result has similar properties to the Lyapunov exponent in terms
of providing an ‘α-index’ that reflects up-ward (for α > 0) and down-ward (for α < 0) trends.
Further, as with the LVR considered in Section 10 and compounded in Equation (26) we can scale the
α-index by the inverse of the volatility to produce Alpha-to-volatility ratio (AVR) index given by
α
ασ = (51)
σ
In practice, the value of the α can easily be computed using the LSM which is compounded in the
function Alpha given in Appendix A.8.

167
Mathematics 2019, 7, 1057

Following the same procedure to that discussed in Section 10.4 (specifically Figure 7), Figure 14
shows example results of running Backtester for the first 1000 elements of the FTSE 100 prices given in
Figure 3 but for the AVR index ασ [n] instead of λσ [n]. The example given is for Backtester (30,10,1000)
which yields a combined entry/exit (long/short) accuracy of 60.98%. Note that this results is obtained
by replacing the code

L(m)=Lyapunov(s,T,1);%Compute the Lyapunov Exponent.

V(m)=Volatility(s,T);%Compute the Volatility.
R(m)=L(m)/V(m);%Compute the Lyapunov to Volatility Ratio (LVR).

with

A(m)=Alpha(s,T);%Compute the Alpha Index.

V(m)=Volatility(s,T);%Compute the Volatility.
R(m)=A(m)/V(m);%Compute the Alpha-To-Volatility Ratio (AVR).

in function Backtester given in Appendix A.6.

Apart from the scale in amplitude, the signature of the ARV is very similar to the LVR (comparing
Figures 14 and 7). However, the trend prediction accuracy is relatively low and the computational time
greater (due to the repeated application of the LSM) which suggests that the LVR is a more reliable
and computationally efficient index. However, this statement must be understood within the context
of the limited data that was used and demonstrated for this publication and must be quantified further
using WT-maps for a range of financial signals and the functions given in Appendix A, a study that
lies beyond the scope of this work.

Figure 14. Example graphical output from function Backtester. The plot shows the post-filtered (for
T = 10) AVR ασ [n] (dotted line), the pre-filtered time series un (dot-dashed line) and zc [n] (solid line)
which identifies the zero-crossings of ασ [n]. The plot is for the case when pre-filtering is undertaken
for a look-back window size of W = 30 and post-filtering using a look-back window size of T = 10.
The financial time series data used in this case is FTSE100 daily prices: 14/03/2006–26/02/2010.

16. Summary, Conclusions and Open Questions

One of the principal themes of this paper has been to develop financial indices that in all
cases can been traced back to a fundamental field equation of statistical physics, namely, Einstein’s

168
Mathematics 2019, 7, 1057

evolution equation—Equation (10). In this context, we have developed expressions for the following
financial indices:

• the Lyapunov Exponent;

• the Volatility;
• the Lévy Index.

16.1. Summary
We have explored the ability for the time varying Lyapunov-to-volatility ratio (LVR) to predict
the trend of a financial signal in terms of a change in polarity and the period over which that polarity
is sustained subject to pre- and post-filtering as discussed in Sections 10.1.1 and 10.1.2, respectively.
The filtering processes are critically dependent on the values of the look-back windows that are
applied and a quantification of the values required to optimise the predictive power has been explored
in Section 10.4 in terms of the WT map. Application of the LVR provides a time signature whose
maximum and minimum values correlate with regions of a financial signal that have up-ward and
down-ward trends with low volatility, respectively. In Section 11, a short study has been presented to
use this result as a criterion for the application of EC to predict short term future prices. In this context,
computing the time varying LVR has two primary uses:

• predicting the entry points in time for making, holding or withdrawing an investment;
• assessing the position in time when application of EC can be expected to yield optimally accurate
short term price predictions.

It is noted that in regard to the application of EC, the volatility alone can be used as an assessment
criterion, low volatilities providing a flag for the use of EC on a moving window basis to update
previous price predictions.
While the derivation and the application of the LVR is predicated on the evolution equation (at
least, as demonstrated in this paper), it does not rely on the application of fractional calculus which
has been a focal issue in regard to the composition of this paper. Thus, the latter half of this paper was
devoted to an analysis of fractional calculus with the aim of showing how, in particular, the classical
diffusion and fractional diffusion equations are both directly related to the evolution equation and can
be derived directly from it, the difference between the two equations being compounded in the PDF
that "governs" the spatial distribution of the density field.
We have shown that the classical diffusion equation is predicated on a Gaussian distribution and
that the fractional diffusion equation is predicated on a (symmetric) Lévy distribution. In turn, it has
been shown that at the spatial origin (i.e., as x → 0), the temporal impulse response functions for
√
these two cases are given by 1/ t and 1/t1−1/γ , respectively, functions that underpin the efficient and
fractal market hypotheses, respectively. In deriving these functions, we have attempted to show the
intrinsic connectivity between the application of Lévy statistics to the evolution equation, the fractional
diffusion, the application of fractional calculus for solving this equation and the analysis of the solution
leading directly to the description of a stochastic self-affine field—a random scaling fractal signal.
In addition to the theoretical concepts presented in this paper, we have provided a set of numerical
algorithms that allows the reader to reproduce the results given. These algorithms are based on the
m-code given in Appendix A. They have been designed to give interested readers the facility to study
the methods used for the wide variety of financial time series available online and to develop the
algorithms further as required. Their development has been based on maintaining consistency with the
theoretical analysis derived at the expense of any further and more sophisticated software engineering.
Hence, issues such as error checks on input/output data, processing parameters and data/processor
compatibility have not been considered.

169
Mathematics 2019, 7, 1057

16.2. Conclusions
The application of fractional calculus in mathematical finance is well known and in this paper we
have provided a unified approach to showing that this is the case using Einstein’s evolution equation
as a fundamental field equation. This approach has the potential for the development of a range of
new models for a financial signal by introducing different PDFs in Equation (11) to those that have
been considered here, the categorisation of such models for different time series lying beyond the
scope of this publication.
The primary results are given in Section 10 which shows that a relatively high accuracy for
predicting up-ward and down-ward trends can be obtained, thereby providing the potential for a
profitable trading strategy to be implemented. However, it must be noted that the quantitative results
given in Section 10 in regard to this statement are strictly applicable only to the data used (i.e., the
daily FTSE100 and Euro-USA dollar exchange rate). Application of the algorithms presented must
therefore be fully quantified and characterised for any and all specific financial time series data used,
"quantification" being compounded in the associated WT map.
The use of EC discussed in Section 11 verifies that short time price prediction can be exercised
if the LVR has reached a maximum or minimum threshold in excess of +2 or −2, respectively.
However, as pointed out in Section 11, the material presented in this respect has only been introduced
to complement the main theme of this paper. Further studies are required to assess the accuracy of EC
prediction on a moving window basis in terms of the number of future projected price values which
maintain an appropriate forecasting accuracy and the associated look-back window used to generate
short time forecasting equations of the type given by Equation (27), for example.

16.3. Open Questions

There are a number of open questions which this paper has raised that are the subject of further
investigation. The reader is invited to consider the following examples:

• The specific form of the evolution equation used in this work has been based on Equation (11)
and it may be of value to consider the affect of the decay term − Ru(r, t) given in Equation (12).
• Given that the critical step in deriving the IRF 1/t1−1/γ (from which γ can be computed) is the
asymptotic condition x → 0, what are the consequences of developing a numerical algorithm to
compute γ when this condition is negated?
• What is the impact of the LVR and AVR in terms of their possible inclusion into machine learning
algorithms that use sets of more conventional financial indices and other statistical metrics
for forecasting?

In regard to more generic questions, the following examples may be of interest:

• In regard to E3 , the PDFs considered in this work are the delta function, Gaussian function
and Lévy distribution which provide models associated with the random walk, efficient and
fractal hypothesis, respectively. An investigation into the models for u(r, t) and metrics
thereof, associated with the application of different PDF (including non-symmetric distributions),
is therefore warranted.
• Similarly, what is the effect of introducing different memory functions into the generalised
Kolmogorov–Feller equation, i.e., E3 in all but name, expressed in terms of memory function m(t),
and, further, is it possible to develop an inverse solution in which a financial signal u(t) can be
used to derive a estimate of m(t) for a known distribution p(r).
• What is the relationship/connectivity (or otherwise) between fractional and Itô calculus in regard
to E3 ?

170
Mathematics 2019, 7, 1057

16.4. Final Remarks

One of the primary aims of this paper was to realise the connectivity compounded in Table 1,
and, in this broader context, to show the relationship between E3 and fractional calculus through
the application of a non-Gaussian distribution, specifically a symmetric Lévy distribution whose
characteristic function is a generalisation of the Gaussian function (for a real constant c) exp(−c | k |2 )
to exp(−c | k |γ ), 0 < γ < 2. The effect of this has been to show that there is a close relationship
between non-Gaussian processes of this type and the self-affine characteristics of stochastic signals
modelled in terms of the solution to a fractional differential equation, i.e., the fractional diffusion
equation. This approach provides the basis for a more general study that transcends the specific
distributions considered in order to derive stochastic models that are a more complete and accurate
description associated with the varied properties of financial signals in which the applications of
fractional calculus is a central theme.
In terms of the computational methods presented, a primary aim is to classify the WT maps for
a range of different financial data in terms of the LVR and AVR and to further quantify the accuracy
of these two indices in regard to different data types. The purpose of this is to categorise the type of
financial times series that are best suited to the trend analysis proposed in terms of a robust predictive
accuracy. In turn, this exercise will inform a quantification of the use of EC for predicting short
term prices with the aim of obtain a quantitative relationship between the look-back window used,
the number of future prices that can be predicted with a specified accuracy and the amplitude of the
LVR and/or ALR for a specific financial signal.

Author Contributions: Concept, supervision and writing original manuscript, J.B.; Data curation, P.W.; Theoretical
analysis, R.R.; Data processing and review, M.L.; Project management and administration, D.K.
Funding: This research was funded by the Science Foundation Ireland (SFI) Stokes Professorship, SFI Reference
Number: 07/SK/11196b.
Acknowledgments: The authors acknowledge the support of the Technological University Dublin and the Science
Foundation Ireland, Republic of Ireland, the University of Western Cape, and the University of KwaZulu-Natal,
Republic of South Africa, the University of Wales, the University of Portsmouth, UK and the Military Technological
College, Oman.
Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations
The following abbreviations are used in this manuscript:

AVR Alpha-to-Volatility Ratio

CF Characteristic Function
CGI Computer Generated Imagery
DC Direct Current
EC Evolutionary Computing
E3 Einstein’s Evolution Equation
FDE Fractional Diffusion Equation
FFT Fast Fourier Transform
FMH Fractal Market Hypothesis
FPE Fractional Poisson Equation
GKFE Generalised Kolmogorov–Feller Equation
IRF Impulse Response Function
KFE Kolmogorov–Feller Equation
LSM Least Squares Method
LVR Lyapunov-to-Volatility Ratio
MFP Mean Free Path
PDF Probability Density Function
PSDF Power Spectral Density Function

171
Mathematics 2019, 7, 1057

Appendix A. Prototype MATLAB Functions

The functions given in this appendix have not been exhaustively tested and no data/parameter
error checks or processing anomalies, for example, have been implemented. The functions are provided
to give the reader a guide to the basic numerical solutions required to implement the computational
procedures discussed in this paper, and, in turn, to help the reader appreciate the theoretical models
presented. It is expected that interested readers will use the functions provided as a guide to extending
their operational characteristics and software engineer their functionality. Where possible, the notation
used for array variables and constants are based on the mathematical notation used in this paper or
are acronyms for the function names. The software was developed and implemented using (64-bit)
MATLAB R2017b with double precision floating point arithmetic.

Appendix A.1. Software Development and Usage

The MATLAB function given in this appendix are provided to give readers access to prototype
source code that implements the algorithms discussed in this paper using m-code. In both cases,
copyright is attributed to the authors and all rights are reserved. Redistribution and use in source
and binary forms, with or without modification, are permitted provided that the following conditions
are met:

• Redistributions of source code must retain the above copyright notice, this list of conditions and
the following disclaimer.
• Redistributions in binary form must reproduce the above copyright notice, this list of conditions
and the following disclaimer in the documentation and/or other materials provided with
the distribution.
• Neither the name of the organisation nor the names of its contributors may be used to endorse or
promote products derived from this software without specific prior written permission.

The software listed in this appendix is provided by the copyright holders and contributors as
is and any express or implied warranties, including, but not limited to, the implied warranties of
merchantability and fitness for a particular purpose are disclaimed. In no event shall the copyright
holders be liable for any direct, indirect, incidental, special, exemplary, or consequential damages
(including, but not limited to, procurement of substitute goods or services; loss of use, data, or profits;
or business interruption) however caused and on any theory of liability, whether in contract, strict
liability, or tort (including negligence or otherwise) arising in any way out of the use of this software,
even if advised of the possibility of such damage.

Appendix A.2. Function Lyapunov

function lambda=Lyapunov(data,N)
% Function to compute the Lyapunov Exponent - lambda -
%for a data stream of length N and time period defined by tau.
%
%Compute the log differences of the data.
for n=1:N-1
d(n)=log(data(n+1)/data(n)); end
d(N)=d(N-1);%Set end point value.
%Return the exponent.
lambda=sum(d);

Appendix A.3. Function Volatility

function sigma=Volatility(data,N)
%Function to compute the Volatility - sigma - for data of size N.

172
Mathematics 2019, 7, 1057

%
%Compute the log price differences.
for n=1:N-1
d(n)=log(data(n+1)/data(n)); end
d(N)=d(N-1);%Set end point value.
%Return the Volatility
sigma=sqrt(sum(abs(d).^2));

Appendix A.4. Function Movav

function Fdata=Movav(data,N,W)
%Function to compute the moving average of data of length N
%using a period (a look-back window) of size W
for n=W:N
%Window data.
for m=1:W
D(m)=data(n-W+m); end
%Compute the mean.
Fdata(n-W+1)=mean(D); end

Appendix A.5. Function Evaluator

function Evaluator(ZC,G,M,T)
%FUNCTION:
%Evaluates the accuracy of a short time trend analysis
%indicator in terms of the actual price differences that
%occurred. This~provides a measure of the accuracy in
%terms of the long and short positions identified that were
%successful in terms of the short time series dynamic
%relative to a net price difference.
%
%INPUTS:
%ZC - Array composed of zeros crossing point indicators;
% ZC = +1 flags the start of a positive trend,
% ZC = -1 flags the start of a negative trend.
%G - Array composed of the values of the time series at
% which the start of a trend is identified by a
% zero crossings.
%M - Data size.
%T - Period (moving window size used for data analysis).
%
%Read non-zero entries of input array ZC and G to vector arrays
%P and Q, respectively, thereby extracting all non-zero values.
n=1; N=1;%Initiate counters.
%Start process
for m=T+1:M-T-1
if G(m)>0.0
P(n)=G(m);n=n+1; Q(N)=ZC(m);N=N+1; end
end
%Count the number of times that an indication of a future upward
%trend led to a net positive price increase (up_good)and then the
%number of times that this failed to be the case (up_bad).

173
Mathematics 2019, 7, 1057

up_good=0; up_bad=0; %Initiate counters

%Start process.
for n=1:N-2
if Q(n)>0 & P(n+1)-P(n)>0
up_good=up_good+1; end
if Q(n)>0 & P(n+1)-P(n)<0
up_bad=up_bad+1; end
end
%Count the number of times that an indication of a future downward
%trend led to a net negative price decrease (down_good)and then the
%number of times that this failed to be the case (down_bad).
down_good=0; down_bad=0;%Initiate counters
%Start process
for n=1:N-2
if Q(n)<0 & P(n+1)-P(n)<0
down_good=down_good+1;
end
if Q(n)<0 & P(n+1)-P(n)>0
down_bad=down_bad+1;
end
end
%Provide outputs on the percentage accuracy of:
% - Successfully predicted upward trend - ’Entries_Accuracy’
% - Successfully predicted downward trend -’Exits_Accuracy’
% - The combine accuracy of both success rates - ’Combined_Accuracy’.
if (double(up_good)+double(up_bad))>0
Entries_Accuracy=100*double(up_good)/(double(up_good)+double(up_bad))
else
Entries_Accuracy=0.0
end
if (double(down_good)+double(down_bad))>0
Exists_Accuracy=100*double(down_good)/(double(down_good)+double(down_bad))
else
Exists_Accuracy=0
end
Combined_Accuracy=(Exists_Accuracy+Entries_Accuracy)/2

Appendix A.6. Function Backtester

function Backtester(W,T,L)
%FUNCTION: Back-testing procedure to compute accuracy of trend analysis
%INPUT PARAMETERS:
%(int) W > 0 - Size of window for pre-filtering using moving average of data
%(int) T > 1 - Size of window for computing financial indices.
%(int) L > 4 - Size of data stream to be processed (must be less than or
%equal length of data read from file).
%
%Read financial time series from Data.txt file into data array where ... denotes
%the path to the folder containing the file (for a windows operating system).
fid=fopen(’...\Data.txt’,’r’);%Open file
[series M]=fscanf(fid,’%g’,[inf]);%Read time series data.

174
Mathematics 2019, 7, 1057

fclose(fid); series=flip(series); %Flip order of data (as required).

%Note: Some historical financial time series data which is available
%on the Internet is often given in terms in reverse time and for this
%reason the flip function is used.
%
%Extract L data components from the time series where L<=M.
for n=1:L
data(n)=series(n);
end
data=data./max(data);%Normalise the series for comparative
%display purpose involving the plotting of multiple data sets.
M=size(data,2);%Reset M to data size
%Filter the time series data using a moving average filter
Fdata=Movav(data,M,W);
%figure(100), plot(data); Figure~(200), plot(Fdata);
M=size(Fdata,2);%Reset M to size of filtered data
%Start moving window process.
for m=T:M-T
%Window the data.
for n=1:T
s(n)=Fdata(n-1+m);
end
%
L(m)=Lyapunov(s,T,1);%Compute the Lyapunov Exponent
V(m)=Volatility(s,T);%Compute the Volatility.
R(m)=L(m)/V(m);%Compute the Lyapunov to Volatility Ratio (LVR).
D(m)=Fdata(m-1+T);%Assign value of Fdata to D (for later use)
x(m)=m;%Set counter to x (for later use).
%
%Compute zero crossings
if m>T
k=1;
%Compute mean of LVRs - post filtering.
for n=m-T:m
Data(k)=R(n);
k=k+1;
end
F(m)=mean(Data);
%Evaluate zero crossings from negative to positive half-space
if F(m)>0 & F(m-1)<=0
ZC(m)=1;%Zero-crossing given positive flag.
G(m)=D(m);%Assignment for later evaluation
else
ZC(m)=0;%Set value to zero
G(m)=0;%set value to zero
end
%Evaluate zero crossings from positive to negative half-space
if F(m)<0 & F(m-1)>=0
ZC(m)=-1;%Zero-crossing given negative flag.
G(m)=D(m);%Assignment for later evaluation

175
Mathematics 2019, 7, 1057

end
%Plot filtered data D, filtered LVR and Zero-crossing flags
%using black dashdot, dotted and solid lines, respectively.
Figure~(1), plot(x,D,’k-.’,x,F,’k:’,x,ZC,’k-’);
%For colour plots, plotting filtered data D, filtered LVR and
%Zero-crossing flags using red, green and blue lines, respectively,
%use Figure~(1), plot(x,D,’r-’,x,F,’g-’,x,ZC,’b-’);
else
end
grid on%Display grid
pause(0.01);%Retain plot ffor 0.01 seconds
end%Repeat process and update plot
%Evaluate accuracy of strategy.
Evaluator(ZC,G,M,T);
clear;%Remove all variables from the workspace.

Appendix A.7. Function Levy

function gamma=Levy(data,N)
%Computation of the Levy Index using the least squares algorithm.
%Compute the Amplitude Spectrum
data=abs(fft(data));
%Compute the logarithm of the data for half-space data with DC
%component removed.
for n=2:round(N/2)
ydata(n)=log(data(n)); xdata(n)=log(n); end
%Compute each term of the least squares formula.
%associated with log scaling law gamma*log(data)
term1=sum(ydata).*sum(xdata); term2=sum(ydata.*xdata);
term3=sum(xdata)^2; term4=sum(xdata.^2);
%Compute alpha
gamma=(term1-(N*term2))/(term3-(N*term4));
%Compute Levy Index
gamma=-1/gamma;

Appendix A.8. Function Alpha

function alpha=Alpha(data,N)
%Computation of the Alpha Index using the least squares algorithm.
%Compute the logarithm of the input data
for n=1:N
ydata(n)=log(data(n)); xdata(n)=log(n); end
%Compute each term of the least squares formula.
%associated with log scaling law alpha*log(data)
term1=sum(ydata).*sum(xdata); term2=sum(ydata.*xdata);
term3=sum(xdata)^2; term4=sum(xdata.^2);
%Compute alpha
alpha=(term1-(N*term2))/(term3-(N*term4));

176
Mathematics 2019, 7, 1057

Appendix B. Relationship between the Lévy Index and the Fractal Dimension
Consider a simple Euclidean straight line  of length L() over which we ‘walk’ a shorter ‘ruler’
of length δ. The number of steps taken to cover the line N [ L(), δ] is then L()/δ which is not always
an integer for arbitrary L and δ. Since

L()
N [ L(), δ] = = L()δ−1
δ
 
ln L() − ln N [ L(), δ] ln N [ L(), δ] − ln L()
⇒1= =−
ln δ ln δ
which expresses the topological dimension n = 1 of the line. In this case, L() is the Lebesgue measure
of the line and if we normalize by setting L() = 1, the latter equation can then be written as
 
ln N (δ)
1 = − lim
δ →0 ln δ

and, in the asymptotic limit

1
N (δ) = , δ→0 (A1)
δ
For extension to a fractal line f , the essential point is that the fractal dimension should satisfy an
equation of the form
N [ F ( f ), δ ] = F ( f ) δ − D

where N [ F ( f ), δ] is ‘read’ as the number of rulers of size δ needed to cover a fractal set f whose
measure is F ( f ) which can be any valid suitable measure of the curve. Normalising, for F () = 1,
we can then define the fractal dimension as
 
ln N (δ)
D = − lim
δ →0 ln δ

and, in the asymptotic limit

1
N (δ) = , δ → 0. (A2)
δD
Consider the scaling relationship between the amplitude A(t) of a signal at a time t ∈ [0, 1]
given by
A(t) = t H , H ∈ [0, 1]

where H is the Hurst dimension. If the time period is divided up into N = 1/Δt equal intervals, then
the amplitude increments ΔA are given by

ΔA = Δt H = N − H

The number of boxes of size δ required to cover the area ΔAΔt is, using Equation (A1), given by
N − H /δ2 = N 2− H . Thus we can write

1
N (δ) = , δ→0
δ 2− H
and, given Equation (A2), by inspection,

D = 2 − H.

177
Mathematics 2019, 7, 1057

Thus, for example, a signal where H = 1/2 has a fractal dimension of 1.5. For higher topological
dimensions n, using a similar box counting measure, we have

D = n + 1 − H, r ∈ Rn (A3)

Consider a random scaling fractal signal defined by a time dependent function f (t). Let f T (t)
denote a component of the function which is of finite support

f ( t ), 0 < t < T;
f T (t) =
0, otherwise.

where
FT (ω ) ↔ f T ( x )

which has a power spectrum defined by

1
PT (ω ) = | FT (ω ) |2 , P(ω ) = lim PT (ω ).
T T →∞

Let the function g(t) be the result of scaling the function f (t) by 1/a H for a real constant a > 0.
Then we can write

g(t) = a1H f ( at), 0 < t < T;
gT (t) =
0, otherwise.

where
GT ( ω ) ↔ g T ( x )

with power spectrum

1
QT (ω ) = | GT (ω ) |2 , Q(ω ) = lim Q T (ω ).
T T →∞

We can therefore construct the equation

 T  T  
1 iωτ
GT ( ω ) = gT (t) exp(−iωt)dt = f (τ ) exp − dτ, τ = at
0 a H +1 0 a

showing that
1 ω
GT ( ω ) = FT .
a H +1 a
The power spectrum of gT (t) is therefore given by

1 1   ω 2 1 ω
QT (ω ) = FT  ⇒ Q(ω ) = 2H +1 P , T→∞
a2H +1 aT a a a
and setting ω = 1 and then replacing 1/a by ω we obtain

1
Q(ω ) ∝ , β = 2H + 1.
| ω |β

The corresponding amplitude spectrum A(ω ) is therefore characterised by

1
A(ω ) ∝ , γ = β/2.
| ω |γ

The result β = 2H + 1 applies to case when r ∈ R1 and for r ∈ Rn generalises to β = 2H + n so

that from Equation (A3) we obtain Equation (33).

178
Mathematics 2019, 7, 1057

References
1. Einstein, A. On the Motion of Small Particles Suspended in Liquids at Rest Required by the Molecular-Kinetic
Theory of Heat. Annalen der Physik 1905, 17, 549–560. [CrossRef]
2. Navarro, L. Gibbs, Einstein and the Foundations of Statistical Mechanics. Arch. Hist. Exact Sci. 1998, 53,
147–180. [CrossRef]
3. Tindel, S.; Tudor, C.A.; Viens, F. Stochastic Evolution Equations with Fractional Brownian Motion.
Probab. Theory Relat. Fields 2003, 127, 186–204. [CrossRef]
4. Peters, E.E. Fractal Market Analysis: Applying Chaos Theory to Investment and Economics; Wiley Finance; Wiley:
Hoboken, NJ, USA, 1994; ISBN-13 978-0471585244.
5. Blackledge, J.M. The Fractal Market Hypothesis: Applications to Financial Forecasting; Lecture Notes Series 4;
Centre for Advanced Studies; Polish Academy of Sciences: Warsaw, Poland, 2010; ISBN 978-83-61993-01-8.
6. Blackledge, J.M.; Lamphiere, M.; Murphy, K.; Overton, S. Financial Forecasting using the Kolmogorov-Feller
Equation. In AENG Transactions on Engineering Technologies: Special Volume of the World Congress on
Engineering 2012; Lecture Notes in Electrical Engineering; Yang, G.-C., Ao, S.-l., Gelman, L., Eds.; Springer:
Berlin/Heidelberg, Germany, 2013; Volume 229, ISBN 978-94-007-6190-2.
7. Blackledge, J.M.; Lamphiere, M.; Murphy, K.; Overton, S. Stochastic Volatility Analysis using the Generalised
Kolmogorov-Feller Equation. In Proceedings of the International Conference of Financial Engineering, World
Congress on Engineering (WCE2012), London, UK, 4–6 July 2012; pp. 453–458.
8. Blackledge, J.M. A New Forex Currency Exchange Indicator. In European Success Stories in Industrial
Mathematics; Springer: Berlin/Heidelberg, Germany, 2011; p. 131, ISBN 978-3-642-23847-5.
9. Blackledge, J.M.; Murphy, K. Forex Trading Using MetaTrader4 with the Fractal Market Hypothesis,
Risk Management Trends; Giancarlo, N., Ed.; InTech Publishing: London, UK, 2011; pp. 129–146,
ISBN 978-953-307-154-1.
10. Blackledge, J.M. Application of the Fractal Market Hypothesis for Modelling Macroeconomic Time Series.
ISAST Trans. Electron. Signal Process. 2008, 2, 89–110.
11. Blackledge, J.M. Application of the Fractional Diffusion Equation for Predicting Market Behaviour.
IAENG Int. J. Appl. Math. 2010, 40, 130–158.
12. Blackledge, J.M.; Rebow, M. Economic Risk Assessment using the Fractal Market Hypothesis. In Proceedings
of the 5th International Conference on Internet Monitoring and Protection, Valencia, Spain, 22–26 May 2010;
pp. 41–47.
13. Blackledge, J.M.; Murphy, K. Forex Trading using MetaTrader 4 with the Fractal Market Hypothesis.
In Proceedings of the Third International Conference on Resource Intensive Applications and Services,
INTENSIVE 2011, Venice, Italy, 22–27 May 2011; pp. 1–9, ISBN 978-1-61208-006-2
14. Blackledge J.M.; Murphy, K. Predicting Currency Pair Trends using the Fractal Market Hypothesis.
In Proceedings of the IET ISSC2011, Dublin, Ireland, 23–24 June 2011.
15. The Fourier Transform, Tables of Important Fourier Transforms, Wikipedia. 2019. Available online: https:
//en.wikipedia.org/wiki/Fourier_transform (accessed on 23 July 2019).
16. Podlubny, I. Geometric and Physical Interpretation of Fractional Integration and Fractional Differentiation.
Fract. Calc. Appl. Anal. 2002, 5, 367–386.
17. Tavassoli, M.H.; Tavassoli, A.; Ostad Rahimi, M.R. The Geometric and Physical Interpretation of Fractional
Order Derivatives of Polynomial Functions. Differ. Geom.-Syst. 2013, 15, 93–104.
18. Butera, S.; Di Paola, M. A Physically based Connection Between Fractional Calculus and Fractal Geometry.
Ann. Phys. 2014, 350, 146–158. [CrossRef]
19. Kiryakova, V. Generalised Fractional Calculus and Applications; Longman: Harlow, UK, 1994.
20. Zheng, Z.; Zhao, W.; Dai, H. A New Definition of Fractional Derivative. Int. J. Non-Linear Mech. 2019, 108,
1-6. [CrossRef]
21. Hermann, R. Fractional Calculus: An Introduction for Physicists; World Scientific: Singapore, 2014.
22. Zhou, Y. Basic Theory of Fractional Differential Equations; World Scientific: Singapore, 2014.
23. Investing.com. 2019. Available online: https://uk.investing.com/indices/uk-100-historical-data (accessed
on 2 August 2019).
24. Investopedia, 2019. Available online: https://www.investopedia.com (accessed on 24 July 2019).

179
Mathematics 2019, 7, 1057

25. Kolmogorov, A.N. On Analytic Methods in Probability Theory. In Selected Works of A.N. Kolmogorov, Volume
II: Probability Theory and Mathematical Statistics; Shiryaev, A.N., Ed.; Kluwer: Dordrecht, The Netherlands,
1992; pp. 61–108. (Original: Uber die Analytischen Methoden in der Wahrscheinlichkeitsrechnung. Math.
Ann. 1931, 104, pp. 415–458).
26. Feller W. On Boundaries and Lateral Conditions for the Kolmogorov Differential Equations. Ann. Math. 1957
65, 527–570. [CrossRef]
27. Bachelier, L. Théorie de la spéculation. Annales Scientifiques de l’École Normale Supérieure 1900, 3, 21–86.
[CrossRef]
28. Working, H. A Random-Difference Series for Use in the Analysis of Time Series. J. Am. Stat. Assoc. 1934, 29,
11–24. [CrossRef]
29. Kendall, M.G.; Bradford Hill, A. The Analysis of Economic Time-Series-Part I: Prices. J. R. Stat. Soc. Ser. A
1953, 116, 11–34. [CrossRef]
30. Fama, E.F. Efficient Capital Markets: A Review of Theory and Empirical Work. J. Financ. 1970, 25, 383–417.
[CrossRef]
31. Campbell, J.Y.; Lo, A.W.; Mackinlay, A.C. The Econometrics of Financial Markets; Princeton University Press:
Princeton, NJ, USA, 1996.
32. Sewell, M. History of the Efficient Market Hypothesis, Research Not RN/11/04, UCL Department of
Computer Science. 2011. Available online: http://www.cs.ucl.ac.uk/fileadmin/UCL-CS/images/Research_
Student_Information/RN_11_04.pdf (accessed on 6 July 2019).
33. Investopedia, 2019. Available online: https://www.investopedia.com/terms/b/blackscholes.asp (accessed
on 25 July 2019).
34. Elliott, R.N. The Wave Principle, Amazon. 2012. Available online: https://www.amazon.co.uk/Wave-
Principle-Ralph-Nelson-Elliott/dp/1607964961 (accessed on 28 July 2019).
35. Frost, A.J.; Prechter, R.R. Amazon 2017. Available online: https://www.amazon.co.uk/Elliott-Wave-
Principle-Market-Behavior/dp/1616040815 (accessed on 28 July 2019).
36. WolframAlpha Computational Intelligence. 2019. Available online: https://www.wolframalpha.com/input/
?i=Fourier+transform (accessed on 28 July 2019).
37. Blackledge, J.M.; Rani, T.R. Stochastic Modelling for Lévy Distributed Systems. Math. Aeterna 2017, 7,
193–210.
38. Blackledge, J.M. Digital Signal Processing, 2nd ed.; Horwood Publishing: Cambridge, UK, 2006; p. 418, ISBN
1-904275-26-5. Available online: https://arrow.dit.ie/engschelebk/4/ (accessed on 19 June 2019).
39. Bryant, P.; Brown, R.; Abarbanel, H. Lyapunov Exponents From Observed Time Series. Phys. Rev. Lett. 1990,
65(13), 1523 -1526. [CrossRef]
40. Brown, R.; Bryant, P.; Abarbanel, H. Computing the Lyapunov Spectrum of a Dynamical System From an
Observed Time Series. Phys. Rev. A 1991, 43, 2787–2806. [CrossRef]
41. Abarbanel, H.; Brown, R.; Kennel, M.B. Local Lyapunov Exponents Computed from Observed Data.
J. Nonlinear Sci. 1992, 2, 343–365. [CrossRef]
42. Eiben, A.E.; Schoenauer, M. Evolutionary Computing. Inf. Process. Lett. 2002, 82, 1–6. [CrossRef]
43. Fogel, L.J.; Owens, A.J.; Walsh, M.J. Artificial Intelligence through Simulated Evolution; Wiley:
Hoboken, NJ, USA, 1966.
44. Holland, J.H. Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology,
Control, and Artificial Intelligence; University of Michigan Press: Ann Arbor, MI, USA, 1975.
45. Rechenberg, I. Evolutionsstrategien, Simulationsmethoden in der Medizin und Biologie; Springer:
Berlin/Heidelberg, Germany, 1978; pp. 83–114.
46. Eiben, A.E.; Smith, J.E. Introduction to Evolutionary Computing; Springer: Berlin/Heidelberg, Germany, 2003.
47. Schmidt, M.; Lipson, H. Eureqa. 2014. Available online: www.nutonian.com (accessed on 23 August 2019).
48. Schmidt, M.; Lipson, H. Distilling Free-form Natural Laws from Experimental Data, Science. 2009.
Available online: https://science.sciencemag.org/content/324/5923/81 (accessed on 24 August 2019).
49. Dubcakova, R. Eureqa: Software Review. Genet. Program. Evolvable Mach. 2011, 12, 173–178. [CrossRef]
50. Ehrenberg, R. Software Scientist: With a Little Data, Eureqa Generates Fundamental Laws of Nature’.
Sci. News 2012, 181, 20–21. [CrossRef]
51. Abbott, R. Equity Markets and Computational Intelligence. In Proceedings of the 2010 UK Workshop on
Computational Intelligence (UKCI), Colchester, UK, 8–10 September 2010; pp. 1–6.

180
Mathematics 2019, 7, 1057

52. Walsh, P. An Econophysics Approach to the Trading of Energy Commodities. Ph.D. Thesis, Technological
University Dublin, Dublin, Ireland, 2018.
53. Wheelwright, S.C.; Makridakis, S.G. Forecasting Nethods for Management; Wiley: Hoboken, NJ, USA, 1973.
54. Evans, G.; Blackledge, J.M.; Yardley, P. Analytic Solutions to Partial Differential Equations; Springer:
Berlin/Heidelberg, Germany, 1999.
55. Crack, T.F. Basic Black-Scholes: Option Pricing and Trading; Amazon: Seattle, WA, USA, 2017; ISBN-13
978-0994138682.

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

181
 mathematics
Article
The Mittag-Leffler Fitting of the Phillips Curve
Tomas Skovranek
BERG Faculty, Technical University of Kosice, Nemcovej 3, 04200 Kosice, Slovakia; [email protected];
Tel.: +421-55-602-5143

Received: 23 May 2019; Accepted: 23 June 2019; Published: 1 July 2019 

Abstract: In this paper, a mathematical model based on the one-parameter Mittag-Leffler function
is proposed to be used for the first time to describe the relation between the unemployment rate
and the inflation rate, also known as the Phillips curve. The Phillips curve is in the literature
often represented by an exponential-like shape. On the other hand, Phillips in his fundamental
paper used a power function in the model definition. Considering that the ordinary as well as
generalised Mittag-Leffler function behave between a purely exponential function and a power
function it is natural to implement it in the definition of the model used to describe the relation
between the data representing the Phillips curve. For the modelling purposes the data of two different
European economies, France and Switzerland, were used and an “out-of-sample” forecast was done
to compare the performance of the Mittag-Leffler model to the performance of the power-type and
exponential-type model. The results demonstrate that the ability of the Mittag-Leffler function to fit
data that manifest signs of stretched exponentials, oscillations or even damped oscillations can be of
use when describing economic relations and phenomenons, such as the Phillips curve.

Keywords: econometric modelling; identification; Phillips curve; Mittag-Leffler function

1. Introduction
It is because of, or thanks to, Paul Anthony Samuelson and Robert Merton Solow [1], that the
economists all around the world call the negative correlation between the rate of wage change (or the
price inflation rate) and the unemployment rate the Phillips curve (PC). It is lesser-known that the idea
occurred in the work by Irving Fisher [2] more than 30 years before publishing the famous paper of
Alban William Housego Phillips [3]. Fisher was not the only one who would deserve such an important
discovery be named after him. Three years before Phillips paper, Arthur Joseph Brown [4] precisely
described the inverse relation between the wage and price inflation and the rate of unemployment. Also
Richard George Lipsey [5] played an important role by the birth, creation of the theoretical foundations
and popularisation of the PC. In the empirical studies the inverse relationship between the rate of
wage change and the unemployment rate was proven, e.g., for the United States of America [1,6,7]
or United Kingdom [3–5]. The policy implications were for the first time mentioned by Samuelson
and Solow [1]. The PC was in its beginnings widely used by the policy-makers to benefit from the
trade-off to decrease the unemployment at a cost of minor increase of the inflation–the “sacrifice ratio”.
Since then the PC has been studied, extended and re-formulated by many authors. For example, the
model representing the New Keynesian theory of the output-inflation trade-off allows the expectations
to jump based on the current and anticipated changes in policy. The new Keynesian Phillips curve
(NKPC) model uses the ideas coming from the models of staggered contracts [8,9] and the quadratic
price adjustment cost model of Rotemberg and Woodford [10], all of which have a similar formulation
as the expectations-augmented PC of Friedman and Phelps [11,12]. The work of Clarida et al. [13]
illustrates the widely usage of this model in theoretical analysis of monetary policy. Shifting the focus
from the unemployment rate to the output gap, the Phillips’ relationship has become an aggregate

Mathematics 2019, 7, 589; doi:10.3390/math7070589 183 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 589

supply curve. The NKPC stayed popular also in the late 1990s and at the beginning of the 21st century
as a theory for understanding inflation dynamics (e.g., [14]).
When Magnus Gustaf Mittag-Leffler, in his works [15,16] proposed a new function Eα ( x ),
he surely did not expect how important generalisation of the exponential function e x he developed.
The Mittag-Leffler (ML) function and its generalisations interpolate between a purely exponential
law and a power-law-like behaviour, and they arise naturally in the solution of fractional-order
integro-differential equations, random walks, Lévy flights, the study of complex systems, and in
other fields. In numerous works the properties, generalisations and applications of the ML-type
functions were studied e.g., [17–26], and computation procedures for evaluating the ML function
were developed e.g., [27–29]. The ML function become of great use and importance not only for
mathematicians, but thanks to its special properties and huge potential for solving applied problems it
found its applicability also in the fields such as psychorheology [19], electrotechnics [30,31], modeling
of processes (diffusion [32], combustion [33], universe expansion [34]), etc. The ML function is also
widely used in the numerical methods for solving ordinary and partial fractional-order differential
equations, and in the the field of “fractal calculus” [35]. The idea to use the fractional-order calculus
and the ML function for modelling phenomenons from the fields of economics and econophysics was
elaborated by several authors [36–42].
In this paper the one-parameter ML function is for the first time used to model the relation
between the unemployment rate and the inflation rate - the Phillips curve, and its performance is
compared to the power-type model and the exponential-type model. French and Swiss econometric
data are taken for the period of time 1980–2017 from the portal EconStatsTM [43] to identify the PC of
these economies. The dataset is split into two subsets, the “modelling” subset is used to identify the
model parameters, and a shorter “out-of-sample” subset serves for evaluating the forecast-performance
of the models. The performance of all three models is evaluated based on the fitting-criterion, i.e., the
sum of squared errors (SSE). The results are presented in the form of figures and tables, where the SSE
of the fitting curve to the “modelling” subset, SSE of the fitting curve to the “out-of-sample” subset,
and SSE of the fitting curve to the complete dataset, as well as some other quality criterions for the
goodness-of-fit are listed.
The paper is organised as follows. Section 2 gives an overview of Mittag-Leffler function and its
generalisations. The original Phillips curve as well as the Mittag-Leffler model for fitting the Phillips
curve is described in Section 3. The numerical results and the discussion on the experiments can be
found in Section 4. Finally, concluding remarks are given in Section 5.

2. Preliminaries: Mittag-Leffler Function and Its Generalisations

In 1903 M. G. Mittag-Leffler [15,16] introduced a new function Eα ( x ), a generalisation of the
classical exponential function e x , which is till today known as the one-parameter ML function. Using
Erdélyi’s notation [44], where z is used instead of x, the function can be written as:

∞
zk
Eα (z) = ∑ Γ ( αk + 1)
, α ∈ C, Re(α) > 0, z ∈ C, (1)
k =0

where Γ denotes the (complete) Gamma function, having the property Γ(n) = (n − 1)!. The
one-parameter ML function and its properties were further investigated [45–48] followed by the
generalisation to a two-parameter function of the ML-type, by some authors called the Wiman’s
function (some give the credit to Agarwal). Following the Erdélyi’s handbook the formula has the
form [44]:
∞
zk
Eα,β (z) = ∑ Γ(αk + β)
, α, β ∈ C, Re(α) > 0, Re( β) > 0, z ∈ C. (2)
k =0

184
Mathematics 2019, 7, 589

The main properties of the above mentioned functions, and other ML-type functions, can be
found in the book by Erdélyi et al. [44], and a detailed overview in the book by Dzhrbashyan [17].
To demonstrate the concept of generality of the ML-type functions let us point out, that the ML
function for one parameter (1), is a special case of the two-parameter ML function, i.e., if we substitute
β = 1 in (2). Accordingly, the classical exponential function is a special case of the one-parameter ML
function, where α = 1:
∞
zk
Eα,1 (z) = ∑ Γ(αk + 1)
≡ Eα (z),
k =0
∞ ∞
zk zk
E1 (z) = ∑ Γ ( k + 1)
= ∑ k!
= ez .
k =0 k =0

Also, other authors introduced and investigated further generalisations of the ML function, but
as these are not used in the following experiments, they are not discussed in details here.

3. Modelling the Phillips Curve

As in many fields of science and applications, so in economics, to describe a relation between
two variables, regression analysis is often used. One can use different regression models from simple
linear-type, throughout exponential- and power-type models, to polynomial ones, and many other
more complex and sophisticated. The discussion on the linearity or nonlinearity, and on the convex or
concave shape of the PC, if it is supposed to be nonlinear, is still ongoing. Some authors are in favour
of convex shape [49–52], some of concave [53], and some of their combination [54]. The application of
the ML-type function to describe the PC perfectly fits into this discussion.

3.1. The “Original” Phillips Curve

Phillips in [3] used British econometric data–the rate of change of money wage rates, provided by
the Board of Trade and the Ministry of Labour (calculated by Phelps Brown and Sheila Hopkins [55]),
and corresponding percentage employment data, quoted in [56]. But, for a simpler evaluation, the
data were first preprocessed, i.e., the average values of the rate of change of money wage rates and of
the percentage unemployment for six different levels of the unemployment (0–2, 2–3, 3–4, 4–5, 5–7,
7–11) were calculated. The crosses in the Figure 1 refer to these average values. Phillips then fitted a
curve to the crosses using a model in the form:

y + a = b xc ⇒ (3)
log (y + a) = log b + c log x,

where y stands for the rate of change of wage rates and x for the percentage unemployment. The
parameters b and c were estimated using the least squares to fit four crosses laying between 0–5% of
unemployment, and the parameter a was chosen to fit the remaining two crosses laying in the interval
5–11% of unemployment. Based on this “fitting criterion” Phillips identified the parameters of the
model (3) as follows:

y + 0.900 = 9.638 x −1.394 ⇒

log (y + 0.900) = 0.984 − 1.394 log x.

185
Mathematics 2019, 7, 589

Figure 1. “Original” Phillips curve [3] (with permission from the John Wiley and Sons publisher).

3.2. The Mittag-Leffler Model for Fitting the Phillips Curve

The idea to use an ML-type function to describe the econometric data (representing the Phillips
curve) results naturally from the observation of two facts:

• the simplicity of the model used by Phillips in his paper [3] given in (3), where a power-type
regression is used to fit the data, and where the model can be defined in the form:

y( x ) = b x c − a, a, b, c ∈ R, (4)

• the usual shape of the PC, used in the literature, which reminds on the exponential-type function:

y( x ) = b e c x − a, a, b, c ∈ R, (5)

where for both cases, (4) and (5), x stands for the unemployment rate and y for the inflation rate.
Based on these facts, the one-parameter ML function appears to be a general model to fit the PC
relation, as it behaves between a purely exponential function and a power function. The one-parameter
ML function defined in (1), which includes the special case when α = 1, i.e., the classical exponential
function, is used to model the econometric data under study. Generally, the proposed fitting model
can be written as follows:

y( x ) = c Eα (b x α ), α ∈ C, Re(α) > 0, b, c ∈ R, (6)

where the parameters α, b, c are subject to optimisation procedure minimising the squared sum of the
vertical offsets between the data points and the fitting curve. Some of the possible manifestations of
the ML model given in (6) are shown in Figure 2 (figures generated using the Matlab demo published
by Igor Podlubny [57]), with the identified parameters listed in Table 1.

186
Mathematics 2019, 7, 589

1.2 1
original function original function
noised data noised data
1 ML fitting
0.8 ML fitting

0.6
0.8
0.4
0.6
0.2
0.4
0

0.2 -0.2

0 -0.4
0 5 10 15 20 0 5 10 15 20

√
(a) y( x ) = e x erfc ( x ). (b) y( x ) = c Eα (b x α ).

1.5 1
original function original function
noised data 0.8 noised data
1 ML fitting ML fitting
0.6
0.5
0.4

0 0.2

0
-0.5
-0.2
-1
-0.4

-1.5 -0.6
0 5 10 15 20 0 5 10 15 20

(c) y( x ) = cos ( x ). (d) y( x ) = e−dx cos ( x ).

Figure 2. Mittag-Leffler fitting using different functions y( x ) for generating data [57].

Table 1. Identified parameters of the Mittag-Leffler fitting model: y( x ) = c Eα (b x α ).

√
y( x) = e x erfc ( x) y( x) = c Eα (b x α ) y( x) = cos ( x) y( x) = e−dx cos ( x)
c - 0.8 - -
generating α - 1.5 - -
parameters b - −0.2 - -
d - - - 0.2
c 0.9982 0.7869 1.0045 0.9722
identified α 0.5008 1.4999 2.0000 1.7538
parameters b −0.9974 −0.1988 −0.9999 −1.0327

4. Numerical Results and Discussion

To evaluate the performance of the proposed ML model (6) in comparison to the power-type
model (4), and the exponential-type model (5) the econometric data of two European countries (France
and Switzerland) were used, that were obtained from the EconStatsTM portal [43]. The unemployment
rate and the inflation rate were taken for the period of time 1980–2017. The whole list of the processed
data can be found in the Table A1.

4.1. Goodness-of-Fit Statistics and Data Preprocessing

The sum of squared errors (SSE) between the fitting models and the used data serves as the
fitting-criterion, with values closer to 0 indicating a smaller random error component of the model.

187
Mathematics 2019, 7, 589

Also some other quality measures were evaluated, i.e., the R-square from interval [0, 1], that indicates
the proportion of variance satisfactory explained by the fitting-model (e.g., R-square = 0.7325 means
that the fit explains 73.25% of the total variation in the data about the average); the adjusted R-square
statistic, with values smaller or equal to 1, where values closer to 1 indicate a better fit; the root mean
squared error (RMSE), with values closer to 0 indicating a fit more useful for prediction [58].
The used dataset, where the unemployment rate corresponds to the x-coordinate and the inflation
rate corresponds to the y-coordinate (each sample represents the state of these two indicators for
each year from the period under study), is first split into two subsets, the “modelling” subset
is used to identify the model parameters, the “out-of-sample” subset serves for evaluating the
forecast-performance of the models. For both economies, French and Swiss, all three models were
first fitted to the data from the “modelling” subset (composed of 31 samples), by minimising SSE,
identifying the optimal parameters. The obtained parameters were then used to compute SSE of the
identified models to the “out-of-sample” subset (composed of seven samples with the greatest values
of unemployment rate) and SSE of the fitting model to the complete dataset.

4.2. Experiments
The first experiment was conducted using the French econometric data. The modelling subset
of 31 samples, was used for the identification purposes. All three models, the power-type model (4),
the exponential-type model (5), and the ML model (6), were fitted to these data minimising the SSE
obtaining so the optimal parameters. The identified models were then used to compute the SSE to the
complete dataset of 38 samples (including the “out-of-sample” subset). SSE results to the modelling
subset as well as SSE to the “out-of-sample” subset and SSE to the complete dataset for the French
Phillips curve are shown in Table 2, alongside the values of R-square, adjusted R-square, and RMSE.
The ML model outperformed the compared models in all listed statistic indicators, with SSE to the
“out-of-sample” subset double smaller than the exponential-type model, and almost three-times smaller
than the power-type model (see Table 2, where bold stands for better result).

Table 2. The statistical results of the French Phillips curve fitting.

Power-Type Model Exponential-Type Model ML Model

SSE to “modelling” subset 157.8422 155.8276 149.6035
SSE to “out-of-sample” subset 10.9024 8.0347 3.9904
SSE to complete dataset 168.7446 163.8623 153.5939
R-square 0.5634 0.5690 0.5862
adjusted R-square 0.5322 0.5382 0.5567
RMSE 2.3740 2.3590 2.3110
Model y( x ) = b x c − a y( x ) = b e c x − a y( x ) = c Eα (b x α )
definition
a = 1.552 a = −0.0187 α = 1.358
Identified b = 1.009e + 04 b = 563.6 b = −0.6378
parameters c = −3.471 c = −0.571 c = −149.9

The result of the French Phillips curve fitting is also shown in Figure 3, where it is possible to
observe a similar behaviour of all three models, i.e., with the increase of the unemployment rate the
inflation rate exponentially decreases. However, for the last samples, the decrease of the ML model
slows down in comparison to the power-type and exponential-type models. This behaviour of the ML
model is obviously better describing the trend of the “out-of-sample” subset. This is also confirmed by
the smallest SSE value of the ML fitting curve to the “out-of-sample” subset (see Table 2).

188
Mathematics 2019, 7, 589

16
"modelling" subset
14 "out-of-sample" subset
power-type model
12 exponential-type model
Mittag-Leffler model

inflation rate [%]

10

0
6 7 8 9 10 11 12
unemployment rate [%]
Figure 3. Fitting the French Phillips curve.

Identically as in the French case, the Swiss econometric data (unemployment rate and inflation
rate) were first preprocessed. The complete dataset was split into the “modelling” subset composed
of 31 samples, that was used to identify the optimal parameters of the power-type model (4), the
exponential-type model (5), and the ML model (6). The SSE between the “modelling” subset and the
fitting curves was again used as the fitting criterion. Using the identified model parameters the SSE of
the evaluated models to the complete dataset of 38 samples (including the “out-of-sample” subset)
was computed. In order to compare the forecast-performance of the models SSE to the “out-of-sample”
subset, as well as the SSE values for the “modelling” subset fitting, and SSE to the complete dataset for
the Swiss Phillips Curve are shown in Table 3, alongside the values of R-square, adjusted R-square,
and RMSE.

Table 3. The statistical results of the Swiss Phillips curve fitting.

Power-Type Model Exponential-Type Model ML Model

SSE to “modelling” subset 39.6506 40.0588 39.2992
SSE to “out-of-sample” subset 6.8961 4.6826 5.0041
SSE to complete dataset 46.5466 44.7414 44.3033
R-square 0.6389 0.6351 0.6420
adjusted R-square 0.6131 0.6091 0.6165
RMSE 1.1900 1.1960 1.1850
Model y( x ) = b x c − a y( x ) = b e c x − a y( x ) = c Eα (b x α )
definition
a = −551.3 a = −0.7809 α = 0.7733
Identified b = −548.6 b = 6.364 b = −1.468
parameters c = 30.85e − 04 c = −1.376 c = 8.823

Observing the result of the Swiss Phillips curve fitting shown in Figure 4, one can see an interesting
case, where although all the compared models are exponentially decreasing, the curve representing the
proposed ML model proceeds in-between the power-type and the exponential-type models, that form
a kind of scissors. In respect to the “out-of-sample” subset it is possible to conclude that two points
of that subset deviate, having higher inflation rate value then the others. This strongly influenced
the fitting results. In this case the exponential-type model visually represents the “out-of-sample”
subset slightly better then the ML model, that is also demonstrated by a smaller value of SSE of the
exponential-type model to the “out-of-sample” subset (see Table 3). In spite of this, the ML model
outperforms the compared models in all other used statistic indicators, including smaller SSE to the

189
Mathematics 2019, 7, 589

complete dataset, proving it’s capability. Moreover, in case of filtering these two outliers from the
“out-of-sample” subset, the ML model better fits the data-trend.

7
"modelling" subset
6 "out-of-sample" subset
power-type model
5 exponential-type model
Mittag-Leffler model
inflation rate [%]

-1
0 1 2 3 4 5
unemployment rate [%]
Figure 4. Fitting the Swiss Phillips curve.

5. Conclusions
The ability of the Mittag-Leffler function to behave between the power-type and the
exponential-type function, and moreover to fit data that manifest signs of stretched exponentials,
oscillations or damped oscillations is demonstrated in this paper, with application to fitting the
econometric data (Phillips curve) of two European economies, where the proposed ML model
outperforms the compared fitting-models in terms of the chosen performance criterions. Exploiting the
full potential of the Mittag-Leffler function and it’s generalisations, as well as associating the model
parameters with the corresponding economic indicators will be the topic of further work.

Funding: This research was funded in part by the Slovak Research and Development Agency under Grants
APVV-14-0892, SK-SRB-18-0011, SK-AT-2017-0015, APVV-18-0526; in part by the Slovak Grant Agency for Science
under Grant VEGA 1/0365/19; and in part by the framework of the COST Action CA15225.
Conflicts of Interest: The author declares no conflict of interest.

Appendix A. The Econometric Dataset

Table A1. The complete dataset: Econometric data for years 1980–2017 [43].

France Switzerland
Year Unemployment Inflation Unemployment Inflation
Rate [%] Rate [%] Rate [%] Rate [%]
1980 6.3490 13.7300 0.1970 4.4260
1981 7.4380 13.8900 0.1810 6.6370
1982 8.0690 9.6910 0.4040 5.4850
1983 8.4210 9.2920 0.8010 2.1000
1984 9.7710 6.6900 1.0590 2.9040
1985 10.2300 4.7030 0.8970 3.2380
1986 10.3600 2.1210 0.7440 0.0400
1987 10.5000 3.1150 0.6970 1.8870
1988 10.0100 3.0810 0.6130 1.9490
1989 9.3960 3.5630 0.4690 5.0220
1990 8.9750 3.2120 0.4720 5.2760
1991 9.4670 3.0630 0.9550 5.2270

190
Mathematics 2019, 7, 589

Table A1. Cont.

France Switzerland
Year Unemployment Inflation Unemployment Inflation
Rate [%] Rate [%] Rate [%] Rate [%]
1992 9.8500 1.9180 2.2190 3.4210
1993 11.1200 2.0700 3.8970 2.4820
1994 11.6800 1.4690 4.1020 0.4200
1995 11.1500 2.1720 3.6950 1.9480
1996 11.5800 2.0860 4.0510 0.7810
1997 11.5400 1.2820 4.5050 0.3860
1998 11.0700 0.6680 3.3380 −0.1680
1999 10.4600 0.5620 2.3620 1.6680
2000 9.0830 1.8270 1.7190 1.4930
2001 8.3920 1.7810 1.5810 0.3250
2002 8.9080 1.9380 2.3300 0.8910
2003 8.9000 2.1690 3.3530 0.5940
2004 9.2330 2.3420 3.5090 1.3320
2005 9.2920 1.9000 3.3840 1.0060
2006 9.2420 1.9120 2.9490 0.6210
2007 8.3670 1.6070 2.4000 2.0040
2008 7.8080 3.1590 2.5760 0.7010
2009 9.5000 0.1030 3.7090 0.2830
2010 9.8020 1.7360 3.8500 0.6860
2011 9.6750 2.2930 3.1100 0.2280
2012 9.9290 1.9520 3.3790 −0.5000
2013 10.0600 1.6300 3.5850 0.5000
2014 9.8010 1.8480 3.3150 1.0000
2015 9.4430 1.9040 3.2780 1.0000
2016 9.1440 1.9490 3.2590 1.0000
2017 8.8350 2.0150 3.2620 1.0000

References
1. Samuelson, P.A.; Solow, R.M. Analytical aspects of anti-inflation policy. Am. Econ. Rev. 1960, 50, 177–194.
2. Fisher, I. A Statistical Relation between Unemployment and Price Changes. Int. Labour Rev. 1926, 13, 785–792.
3. Phillips, A.W. The relation between unemployment and the rate of change of money wage rates in the
United Kingdom, 1861–1957. Economica 1958, 25, 283–299. [CrossRef]
4. Brown, A.J. Great Inflation 1939–1951; Oxford University Press: London, UK, 1955.
5. Lipsey, R.G. The relation between unemployment and the rate of change of money wage rates in the United
Kingdom, 1862–1957: A further analysis. Economica 1960, 27, 1–31. [CrossRef]
6. Bowen, W.G. Wage Behavior in the Postwar Period: An Empirical Analysis; Princeton University: Princeton, NJ,
1960.
7. Bodkin, R.G. The Wage-Price-Productivity Nexus; University of Pennsylvania Press: Philadelphia, PA, USA,
1966.
8. Taylor, J.B. Aggregate dynamics and staggered contracts. J. Political Econ. 1980, 88, 1–23. [CrossRef]
9. Calvo, G.A. Staggered Prices In A Utility-maximizing Framework. J. Monetary Econ. 1983, 12, 383–398.
[CrossRef]
10. Rotemberg, J.J.; Woodford, M. An Optimization-Based Econometric Framework for the Evaluation of
Monetary Policy. In NBER Macroeconomics Annual; MIT Press: Cambridge, MA, USA, 1997; pp. 297–346.
11. Friedman, M. The role of monetary policy. Am. Econ. Rev. 1968, 58, 1–17.
12. Phelps, E.S. Money-Wage Dynamics and Labor Market Equilibrium. J. Political Econ. 1968, 76, 678–711.
[CrossRef]
13. Clarida, R.; Galí, J.; Gertler, M. The science of monetary policy: A new Keynesian perspective. J. Econ. Lit.
1999, 37, 1661–1707. [CrossRef]
14. Galí, J.; Gertler, M.; López-Salido, J.D. Robustness of the Estimates of the Hybrid New Keynesian Phillips
Curve. J. Monetary Econ. 2005, 52, 1107–1118. [CrossRef]

191
Mathematics 2019, 7, 589

15. Mittag-Leffler, M.G. Une generalisation de l’integrale de Laplace-Abel. Comptes Rendus de l’Académie des
Sciences 1903, 137, 537–539.
16. Mittag-Leffler, M.G. Sur la nouvelle fonction Eα ( x ). Comptes Rendus de l’Académie des Sciences 1903,
137, 554–558.
17. Dzhrbashyan, M.M. Integral Transforms and Representations of Functions in Complex Plane; Nauka: Moscow,
Russia, 1966.
18. Caputo, M.; Mainardi, F. Linear models of dissipation in anelastic solids. La Rivista del Nuovo Cimento 1971,
1, 161–198. [CrossRef]
19. Blair, G.W.S. Psychorheology: links between the past and the present. J. Texture Stud. 1974, 5, 3–12. [CrossRef]
20. Torvik, P.J.; Bagley, R.L. On the appearance of the fractional derivative in the behaviour of real materials.
J. Appl. Mech. 1984, 51, 294–298. [CrossRef]
21. Samko, S.G.; Kilbas, A.A.; Marichev, O.I. Fractional Integrals and Derivatives: Theory and Applications; Gordon
and Breach: New York, NY, USA, 1993.
22. Gorenflo, R.; Kilbas, A.A.; Mainardi, F.; Rogosin, S.V. Mittag-Leffler Functions, Related Topics and Applications;
Springer Monographs in Mathematics; Springer: Berlin/Heidelberg, Germany, 2014. [CrossRef]
23. Kilbas, A.A.; Saigo, M.; Saxena, R.K. Generalized Mittag-Leffler function and generalized fractional calculus
operators. Integral Transform. Spec. Funct. 2004, 15, 31–49. [CrossRef]
24. Mainardi, F.; Gorenflo, R. On Mittag-Leffler-type functions in fractional evolution processes. J. Comput. Appl.
Math. 2000, 118, 283–299. [CrossRef]
25. West, B.; Mauro Bologna, M.; Grigolini, P. Physics of Fractal Operators; Springer-Verlag: New York, NY, USA,
2003.
26. Haubold, H.J.; Mathai, A.M.; Saxena, R.K. Mittag-Leffler functions and their applications. J. Appl. Math.
2011, 2011, 51. [CrossRef]
27. Chen, Y. Generalized Generalized Mittag-Leffler function; MathWorks, Inc. Matlab Central File Exchange.
2008. Available online: http://www.mathworks.com/matlabcentral/fileexchange/21454 (accessed on 21
May 2019).
28. Podlubny, I. Mittag-Leffler Function; MathWorks, Inc. Matlab Central File Exchange. 2005. Available online:
http://www.mathworks.com/matlabcentral/fileexchange/8738 (accessed on 21 May 2019).
29. Garrappa, R. The Mittag-Leffler function; MathWorks, Inc. Matlab Central File Exchange. 2015. Available
online: http://www.mathworks.com/matlabcentral/fileexchange/48154 (accessed on 21 May 2019).
30. Capelas de Oliveira, E.; Mainardi, F.; Vaz, J., Jr. Models based on Mittag-Leffler functions for anomalous
relaxation in dielectrics. Eur. Phys. J. Spec. Top. 2011, 193, 161–171. [CrossRef]
31. Petras, I.; Sierociuk, D.; Podlubny, I. Identification of Parameters of a Half-Order System. IEEE Trans. Signal
Process. 2012, 60, 5561–5566. [CrossRef]
32. Maindardi, F. Fractional Calculus and Waves in Linear Vescoelasticity: An Introduction to Mathematical Models
(Second Revised Edition); Imperial College Press: London, UK, 2018.
33. Samuel, S.; Ricon, J.; Podlubny, I. Modelling combustion in internal combustion engines using the
Mittag-Leffler function. In Proceedings of the ICFDA 16: International Conference on Fractional
Differentiation and its Applications, Novi Sad, Serbia, 18–20 July 2016; pp. 551–552.
34. Zeng, C.; Chen, Y.; Podlubny, I. Is our universe accelerating dynamics fractional order? In Proceedings of
the 2015 ASME/IEEE International Conference on Mechatronic and Embedded Systems and Applications,
Boston, MA, USA, 2–5 August 2015; Paper No. DETC2015-46216.
35. Atangana, A.; Baleanu, D. New Fractional Derivatives With Non-Local And Non-Singular Kernel: Theory
and Application to Heat Transfer Model. Therm. Sci. 2016, 20-2, 763–769. [CrossRef]
36. Scalas, E.; Gorenflo, R.; Mainardi, F. Fractional calculus and continuous-time finance. Phys. A Stat. Mech.
Its Appl. 2000, 284, 376–384. [CrossRef]
37. Cartea, A.; del Castillo-Negrete, D. Fractional diffusion models of option prices in markets with jumps.
Phys. A Stat. Mech. Its Appl. 2007, 374, 749–763. [CrossRef]
38. Garibaldi, U.; Scalas, E. Finitary Probabilistic Methods in Econophysics; Cambridge University Press: Cambridge,
UK, 2010.
39. Skovranek, T.; Podlubny, I.; Petras, I. Modeling of the national economies in state-space: A fractional calculus
approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]

192
Mathematics 2019, 7, 589

40. Tarasov, V.E.; Tarasova, V.V. Macroeconomic models with long dynamic memory: Fractional calculus
approach. Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
41. Tarasov, V.E.; Tarasova, V.V. Phillips model with exponentially distributed lag and power-law memory.
Comput. Appl. Math. 2019, 38, 13. [CrossRef]
42. Tarasov, V.E.; Tarasova, V.V. Dynamic Keynesian Model of Economic Growth with Memory and Lag.
Mathematics 2019, 7, 178. [CrossRef]
43. Econstats.com. EconStats: World Economic Outlook (WEO) data, IMF. 2019. Available online: http:
//www.econstats.com (accessed on 21 May 2019).
44. Erdélyi, A.; Magnus, W.; Oberhettinger, F.; Tricomi, F.G. Higher Transcendental Functions; McGraw-Hill:
New York, NY, USA, 1955; Volume 3.
45. Mittag-Leffler, M.G. Sopra la funzione Eα ( x ). Comptes Rendus de l’Académie des Sciences 1904, 13, 3–5.
46. Wiman, A. Über den Fundamentalsatz in der Teorie der Funktionen Eα ( x ). Acta Math. 1905, 29, 191–201.
[CrossRef]
47. Pollard, H. The completely monotonic character of the Mittag-Leffler function Ea (− x ). Bull. Am. Math. Soc.
1948, 54, 1115–1116. [CrossRef]
48. Humbert, P.; Agarwal, R.P. Sur la fonction de Mittag-Leffler et quelques-unes de ses généralisations. Bulletin
des Sciences Mathématiques 1953, 77, 180–185.
49. Cover, J.P. Asymmetric effects of positive and negative money-supply shocks. Q. J. Econ. 1992, 107, 1261–1282.
[CrossRef]
50. Karras, G. Are the output effects of monetary policy asymmetric? Evidence from a sample of European
countries. Oxf. Bull. Econ. Stat. 1996, 58, 267–278. [CrossRef]
51. Nobay, A.R.; Peel, D.A. Optimal monetary policy with a nonlinear Phillips curve. Econ. Lett. 2000,
67, 159–164. [CrossRef]
52. Schaling, E. The nonlinear Phillips curve and inflation forecast targeting: Symmetric versus asymmetric
monetary policy rules. J. Money Credit Bank. 2004, 36, 361–386. [CrossRef]
53. Stiglitz, J. Reflections on the natural rate hypothesis. J. Econ. Perspect. 1997, 11, 3–10. [CrossRef]
54. Filardo, A. New evidence on the output cost of fighting inflation. Fed. Reserve Bank Kansas City Econ. Rev.
1998, 83, 33–61.
55. Brown, P.; Hopkins, S. The Course of Wage-Rates in Five Countries, 1860–1939. Oxf. Econ. Pap. 1950, 2,
226–296. [CrossRef]
56. Beveridge, H.W. Full Employment in a Free Society; W. W. Norton & Co., Inc.: New York, NY, USA, 1945.
57. Podlubny, I. Fitting Data Using the Mittag-Leffler Function; MathWorks, Inc. Matlab Central File Exchange.
2011. Available online: http://www.mathworks.com/matlabcentral/fileexchange/32170 (accessed on
21 May 2019).
58. MathWorks, Inc. Evaluating Goodness of Fit. 2019. Available online: https://www.mathworks.com/help/
curvefit/evaluating-goodness-of-fit.html (accessed on 21 May 2019).

c 2019 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

193
 mathematics
Article
Stability and Bifurcation of a Delayed
Time-Fractional Order Business Cycle Model
with a General Liquidity Preference Function
and Investment Function
Yingkang Xie, Zhen Wang and Bo Meng *
College of Mathematics and Systems Science, Shandong University of Science and Technology,
Qingdao 266590, China; [email protected] (Y.X.); [email protected] (Z.W.)
* Correspondence: [email protected]

Received: 7 August 2019; Accepted: 12 September 2019; Published: 13 September 2019 

Abstract: In this paper, the business cycle (BC) is described by a delayed time-fractional-order
model (DTFOM) with a general liquidity preference function and an investment function. Firstly,
the existence and uniqueness of the DTFOM solution are proven. Then, some conditions are presented
to guarantee that the positive equilibrium point of DTFOM is locally stable. In addition, Hopf
bifurcation is obtained by a new method, where the time delay is regarded as the bifurcation
parameter. Finally, a numerical example of DTFOM is given to verify the effectiveness of the proposed
model and methods.

Keywords: business cycle model; stability; time delay; time-fractional-order; Hopf bifurcation

1. Introduction
Macroeconomics is an essential economic field that analyzes the general law of economics through
macroeconomic indicators such as national income, market investment, and money supply [1]. As one
of the most important issues of macroeconomics, business cycle (BC) theory has been studied by many
economists because of its realistic meaning and practical value [2].
To obtain the factors involved in fluctuations in the business cycle (BC), many mathematical
models are founded on nonlinear dynamics and relative theories that improve the development of
BC theory. Using graph analysis, Kaldor [3] proved that a BC exists when the investment and saving
function are time-varying nonlinear. Chang and Smyth [4] proved the results shown in [3] by using
mathematical theory and gave the conditions required for the existence of limit cycles. J. R.Hicks and
A. H. Hansen proposed the IS-LM model, which is an important tool for describing the macroeconomic
analysis of the interlinked theoretical structure between the product market and the money market [5].
These early studies are of great significance to current and future BC.
Time delay is a phenomenon existing in practice that often appears in engineering
applications [6–8]. Time delay is an important factor that also widely exists in economics. Since
economics is not only affected by the present state, but also by the past state, the delayed mathematical
model is more suitable for describing economic systems, and some significant results have been drawn
in recent years [9–11]. Considering expectation and delay, Liu and Cai [12] studied a BC model and
gave some conditions of stability and bifurcation. Hu and Cao [13,14] studied the Kaldor–Kalecki
model of delayed BC. To summarize, time delay is an important reason for fluctuations in BC.
In recent years, the theory of fractional calculus (FC) has not only rapidly developed, but has
also been widely applied in many fields [15–19]. In fact, most economic systems have long-term
memories. Compared with integer derivatives, since fractional derivatives are related to the entire

Mathematics 2019, 7, 846; doi:10.3390/math7090846 195 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 846

time domain of the economic process, the fractional-order systems are more suitable for describing
economic systems. In the last few years, fractional calculus equations have been widely used to describe
a class of economic processes with power law memory and spatial nonlocality. Some continuous-time
mathematical models describing economic dynamics with long memory have been proposed [20,21],
and some interesting results were obtained. Wang and Huang [22] studied a delayed fractional-order
financial system and obtained some conditions of stability and chaos. Ma and Ren [23] studied a
fractional-order macroeconomic system and obtained some conditions of stability and Hopf bifurcation.
Considering negative parameters, Tacha and Munoz-Pacheco [24] studied a fractional-order finance
system, and the cause of chaos was found. Motivated by the above considerations, a new delayed
fractional-order model (DFOM) for BC with a general liquidity preference function and an investment
function is considered in this paper.
The arrangements of the article are as follows: In Section 2, the model description and some
definitions and lemmas are provided. Section 3 shows the main results. The existence and
uniqueness of the solution and the local stability and bifurcation of the positive equilibrium point of
DFOM for the BC are presented. Numerical simulations and conclusions are respectively presented
in Sections 4 and 5.

2. Preliminaries and Model Descriptions

In this paper, the Caputo form of the fractional-order derivative is used.

Definition 1. [25] For a continuous function f (t) and a positive integer n, if α ∈ (n − 1, n) is satisfied,
the fractional-order Caputo’s derivative is defined as:
 t
1 f (n) ( τ )
Dα f (t) = dτ, (1)
Γ(n − α) t0 ( t − τ ) α +1− n

where t0 ∈ R is the initial time and Γ(·) is the Gamma function.

Definition 2. [26] For 0 < α < 1, under the initial condition x (t0 ) = xt0 , x ∗ is called an equilibrium point of
system D α x (t) = f (t, x ) if and only if f (t, x ∗ ) = 0.

Lemma 1. [26] For α ∈ (0, 1], under the initial condition x (t0 ) = xt0 , if f (t, x ) satisfies the local Lipschitz
condition, then system D α x (t) = f (t, x ) has a unique solution for t > t0 .

Lemma 2. [27] Suppose that x ∗ is the equilibrium point of system D α x (t) = f ( x ); if | arg(λi )| > απ 2 is
satisfied, then x ∗ is asymptotically locally stable, where λi is any one of the eigenvalues for the Jacobian matrix
J = ∂ f /∂x evaluated at x ∗ .

We consider the augmented IS-LM BC model given by Gabisch and Lorenz [28],

Ẏ (t) = a[ I (Y (t), K (t), R(t)) − S(Y (t), R(t))],

Ṙ(t) = b[ L(Y (t), R(t)) − M̄], (2)
K̇ (t) = I (Y (t), K (t), R(t)) − δK (t),

where Y (t), R(t), and K (t) represent the gross product, interest rate at time t, and capital stock,
respectively. I (Y (t), K (t), R(t)) is the investment function, L(Y (t), R(t)) is the liquidity preference
function, S(Y (t), R(t)) is the saving function, and M̄ is a constant money supply. a > 0 and b > 0
denote the adjustment coefficient of goods and the monetary market, respectively. 0 < δ < 1 represents
the depreciation rate of the capital stock.
We consider an investment delay τ > 0, which is always encountered in capital stock, and suppose
that S(Y (t), R(t)) = s1 Y (t) + s2 R(t). The DFOM for the BC with a general liquidity preference function

196
Mathematics 2019, 7, 846

and an investment function under the initial conditions Y (θ ) = φ1 (θ ), R(θ ) = φ2 (θ ), K (θ ) = φ3 (θ ) is

described as follows:
D α Y (t) = a[ I (Y (t), K (t), R(t)) − s1 Y (t) − s2 R(t)],
D α R(t) = b[ L(Y (t), R(t)) − M̄ ], (3)
D α K (t) = I (Y (t − τ ), K (t − τ ), R(t − τ )) − δK (t),
where −τ ≤ θ ≤ 0, 0 < α ≤ 1 and s1 and s2 are positive constants.
Suppose that I (Y (t), K (t), R(t)), and L(Y (t), K (t)) are differentiable. X (t) = K (t + τ ) is denoted
as the expected capital stock. By substituting it into System (3), one obtains:

D α Y (t) = a[ I (Y (t), X (t − τ ), R(t)) − s1 Y (t) − s2 R(t)],

D α R(t) = b[ L(Y (t), R(t)) − M̄], (4)
D α X (t) = I (Y (t), X (t − τ ), R(t)) − δX (t).

3. Main Results
We firstly prove that the solution of DFOM for BC exists and is unique. In addition, we give some
conditions to guarantee that the positive equilibrium point is stable and bifurcates.

3.1. Existence and Uniqueness of the Solution

Theorem 1. If C = C ([−τ, 0], R3 ) is the continuous function of the Banach space and Z0 (t) ∈ C is an initial
condition, then System (4) has a unique solution Z (t), where Z (t) = (Y (t), R(t), X (t)).

Proof. Consider a mapping H ( Z ) = ( H1 ( Z ), H2 ( Z ), H3 ( Z )), where:

H1 ( Z (t)) = −s1 aY (t) − s2 aR(t) + aI (Y (t), X (t − τ ), R(t)),

H2 ( Z (t)) = −b M̄ + bL(Y (t), R(t)), (5)
H3 ( Z (t)) = −δX (t) + I (Y (t), X (t − τ ), R(t)).

The following conclusion can be made:

 H ( Z (t1 )) − H ( Z (t2 ))

≤| H1 ( Z (t1 )) − H1 ( Z (t2 ))| + | H2 ( Z (t1 )) − H2 ( Z (t2 ))| + | H3 ( Z (t1 )) − H3 ( Z (t2 ))|
=| a{[ I (Y (t1 ), X (t1 − τ ), R(t1 )) − I (Y (t2 ), X (t2 − τ ), R(t2 ))] − s1 [Y (t1 ) − Y (t2 )] − s2 [ R(t1 ) − R(t2 )]}|
+ |[ I (Y (t1 ), X (t1 − τ ), R(t1 )) − I (Y (t2 ), X (t2 − τ ), R(t2 ))] − δ( X (t1 ) − X (t2 ))|
+ |b[ L(Y (t1 ), R(t1 )) − L(Y (t2 ), R(t2 ))]|
∂I (Y ( T1 ), X (t1 − τ ), R(t1 )) ∂L(Y ( T2 ), R(t1 )) (6)
≤|Y (t1 ) − Y (t2 )|[ a|s1 | + ( a + 1)| | + b| |]
∂Y ∂Y
∂I (Y (t2 ), X ( T3 ), R(t1 ))
+ max{| X (t1 − τ ) − X (t2 − τ )|, | X (t1 ) − X (t2 )|}[|( a + 1) | + δ]
∂X
∂I (Y (t2 ), X (t2 − τ ), R( T4 )) ∂L(Y (t2 ), R( T5 ))
+ | R(t1 ) − R(t2 )|[s2 + |( a + 1) | + b| |]
∂R ∂R
≤ L  Z ( t1 ) − Z ( t2 )  .

197
Mathematics 2019, 7, 846

The constants T1 , T2 , T4 , T5 ∈ [t1 , t2 ], T3 ∈ [t1 − τ, t2 − τ ] exist, satisfying:

∂I (Y ( T1 ), X (t1 − τ ), R(t1 )) ∂I (Y (t), X (t1 − τ ), R(t1 ))

| | = max | |,
∂Y t∈[t1 ,t2 ] ∂Y
∂L(Y ( T2 ), R(t1 )) ∂L(Y (t), R(t1 ))
| | = max | |,
∂Y t∈[t1 ,t2 ] ∂Y
∂I (Y (t2 ), X ( T3 ), R(t1 )) ∂I (Y (t2 ), X (t), R(t1 ))
| |= max | |, (7)
∂X t∈[t1 −τ,t2 −τ ] ∂X
∂I (Y (t2 ), X (t2 − τ ), R( T4 )) ∂I (Y (t2 ), X (t2 − τ ), R(t))
| | = max | |,
∂R t∈[t1 ,t2 ] ∂R
∂L(Y (t2 ), R( T5 )) ∂L(Y (t2 ), R(t))
| | = max | |,
∂R t∈[t1 ,t2 ] ∂R

where one chooses a positive constant,

∂I (Y (t2 ), X (t2 − τ ), R( T3 )) ∂L(Y (t2 ), R( T5 ))

L = max{|s2 | + |( a + 1) | + b| |,
∂R ∂R
∂I (Y (t2 ), X ( T3 ), R(t1 ))
|( a + 1) | + δ, (8)
∂X
∂I (Y ( T1 ), X (t1 − τ ), R(t1 )) ∂L(Y ( T2 ), R(t2 ))
a|s1 | + (α + 1)| | + b| |}.
∂Y ∂Y
It is quite clear that H ( Z ) satisfies the Lipschitz condition. According to lemma 1, System (4) with
Zt0 has a unique solution Z (t).

3.2. Stability and Bifurcation

Assume that System (4) contains positive equilibrium points, and let E∗ = (Y ∗ , R∗ , X ∗ ) be one of
them. For convenience, we define:

∂I (Y ∗ , X ∗ , R∗ ) ∂L(Y ∗ , R∗ )
= IY , = LY ,
Y Y
∗ ∗
∂I (Y , X , R ) ∗ ∗
∂L(Y , R )∗
= IR , = LR ,
R R (9)
∗ ∗
∂I (Y , X , R ) ∗
= IX , x (t) = X (t) − x∗ ,
X
y(t) = Y (t) − Y ∗ , r (t) = R(t) − R∗ .

By substituting (9) into (4) and using linearization methods for (4), we get:

D α y(t) = a[ IY y(t) + IR r (t) + IX x (t − τ ) − s1 y(t) − s2 r (t)],

D α r (t) = b[ LY y(t) + L R r (t) − M̄], (10)
α
D x (t) = IY y(t) + IR r (t) + IX x (t − τ ) − δx (t).

The Jacobian matrix of (10) is described as:

⎡ ⎤
a( IY − s1 ) a( IR − s2 ) aIX e−sτ
⎢ ⎥
J=⎣ bLY bL R 0 ⎦.
IY IR IX e − sτ −δ

Then, the characteristic equation can be expressed as:

s3α + a1 s2α + a2 sα + a3 + ( a4 s2α + a5 sα + a6 )e−sτ = 0, (11)

198
Mathematics 2019, 7, 846

where:
a1 = − a( IY − s1 ) − bL R + δ,
a2 = a( IY − s1 )(bL R − δ) − bL R δ − abLY ( IR − s2 ),
a3 = abδ[( IY − s1 ) L R − ( IR − s2 ) LY ],
(12)
a 4 = IX ,
a5 = a( IY − s1 ) IX + bL R IX − aIX IY ,
a6 = − ab( IY − s1 ) L R IX − abLY IR IX + abL R IY IX + ab( IR − s2 ) LY IX .

Theorem 2. For τ = 0 and System (4), E∗ = (Y ∗ , R∗ , X ∗ ) is one of the positive equilibrium points.
If the conditions:

(1) Δ ≤ 0, a1 + a4 > 0, a2 + a5 > 0, a3 + a6 > 0, and ( a1 + a4 )( a2 + a5 ) − ( a3 + a6 ) > 0,

(2) Δ > 0, ( A + B − a1 +3 a4 ) < 0, and | arg( Aω + Bω 2 − a1 +3 a4 )| > απ
2 ,

q2 p3 3( a2 + a5 )−( a1 + a4 )2 2( a1 + a4 )3 −9( a1 + a4 )( a2 + a5 )+27( a3 + a6 )

exist, where Δ = + 27 , p = , q = ,
" +
4 " +
3 27
q2 p3 q2 p3
then E∗ is locally asymptotically stable.
3 q 3
A = −2 + 4 + 27 and B = − 2q − 4 + 27 ,

Proof. If τ = 0, then (11) can be rewritten as:

s3α + ( a1 + a4 )s2α + ( a2 + a5 )sα + a3 + a6 = 0. (13)

Let sα = λ, and substitute it into (13). This produces:

λ3 + ( a1 + a4 )λ2 + ( a2 + a5 )λ + a3 + a6 = 0. (14)

a1 + a4
Let λ∗ = λ + 3 . Then, (14) can be transformed into:

(λ∗ )3 + pλ∗ + q = 0, (15)

3( a2 + a5 )−( a1 + a4 )2 2( a + a )3 −9( a1 + a4 )( a2 + a5 )+27( a3 + a6 )

where p = 3 ,q= 1 4 27 .
2 3
If Δ = q4 + 27 p
≤ 0, (15) has three real roots. Thus, (14) has three
real roots. Then, according to
the Routh–Hurwitz criterion and the theory of the equilibrium point [29,30], if:

a2 + a5 > 0, a1 + a4 > 0, a3 + a6 > 0,

(16)
( a1 + a4 )( a2 + a5 ) − ( a3 + a6 ) > 0,
απ
all three real roots are negative. Thus, | arg(λi )| = π > 2 , and E∗ is locally asymptotically stable.
q2 p3
If Δ = + > 0, then = A + B, λ1∗ λ2∗
= Aω + Bω 2 , and λ3∗ = Aω 2 + Bω, where A =
" +
4 27 " + √
. Thus, λ1 = A + B − a1 +3 a4 , λ2 = Aω +
2 p3 q2 p3
3 3 −1+ 3i
− 2q + q4 + 27 , B = −2 −
q
4 + 27 , and ω = 2
Bω 2 − a1 +3 a4 , and λ3 = Aω 2 + Bω − a1 + a4
3 . If | arg( Aω + Bω 2 − a1 + a4
3 )| > απ
2 and A + B − a1 + a4
3 < 0,
E∗ is locally asymptotically stable.

Next, we use the method in [31] to deal with the case where τ = 0. Assume that (11) has a purely
imaginary root s = iϕ ( ϕ > 0) and a4 s2α + a5 sα + a6 = 0. By substituting s = iϕ into Equation (11),
we get:

απ απ 3απ 3απ
a1 ϕ2α (cos απ + sin(απ )i ) + a2 ϕα (cos + sin i ) + a3 + ϕ3α (cos + sin i)
2 2 2 2 (17)
απ απ
+( a4 ϕ2α (cos απ + i sin απ ) + a5 ϕα (cos + i sin ) + a6 )(cos ϕτ − i sin ϕτ ) = 0.
2 2

199
Mathematics 2019, 7, 846

If we write the real and imaginary parts separately, then we have:

⎧
⎪ 3απ απ
⎪
⎪
⎪ ϕ3α cos + a1 ϕ2α cosαπ + a2 ϕα cos + a3
⎪
⎪ 2 2
⎪
⎪ απ
⎪
⎨ = −[ a4 ϕ2α cos(απ − ϕτ ) + a5 ϕα cos( − ϕτ ) + a6 cos ϕτ ],
2 (18)
⎪
⎪ 3απ απ
⎪
⎪ ϕ3α sin + a1 ϕ2α sin απ + a2 ϕα sin
⎪
⎪ 2 2
⎪
⎪
⎩ = −[ a4 ϕ2α sin(απ − ϕτ ) + a5 ϕα sin( απ − ϕτ ) − a6 sin ϕτ ].
⎪
2
By squaring the corresponding sides of (18) and adding them, we get:

απ 5α 3απ απ απ 3α
ϕ6α + 2a1 cos ϕ + ( a21 + 2a2 cos απ − a24 ) ϕ4α + (2a3 cos + 2a1 a2 cos − 2a4 a5 cos )ϕ
2 2 2 2 (19)
απ απ
+ ( a22 + 2a1 a3 cos απ − a25 − 2a4 a6 cos απ ) ϕ2α + (2a4 a3 cos − 2a5 a6 cos ) ϕα + a23 − a26 = 0.
2 2

If we let h( ϕ) = ϕ6α + 2a1 cos απ

2 ϕ + ( a1 + 2a2 cos απ − a4 ) ϕ + (2a3 cos 2 + 2a1 a2 cos 2 −
5α 2 2 4α 3απ απ
απ απ απ α
2a4 a5 cos 2 ) ϕ + ( a2 + 2a1 a3 cos απ − a5 − 2a4 a6 cos απ ) ϕ + (2a4 a3 cos 2 − 2a5 a6 cos 2 ) ϕ + a23 −
3α 2 2 2α

a26 , and suppose that a23 − a26 < 0, then ϕ0 is one of the positive roots of h( ϕ). Using (17), we get:

1 CE + DF
τj = {arccos − 2 + 2jπ }, j = 0, 1, 2, . . . , n, (20)
ϕ0 E + F2

where:
3απ α απ
C = ϕ3α
0 cos + a1 ϕ2α
0 cos απ + a2 ϕ0 cos + a3 ,
2 2
3απ α απ
D = ϕ3α
0 sin + a1 ϕ2α
0 sin( απ ) + a2 ϕ0 sin ,
2 2 (21)
α απ
E = a4 ϕ2α
0 cos απ + a5 ϕ0 cos + a6 ,
2
α απ
F = a4 ϕ0 sin απ + a5 ϕ0 sin
2α
.
2
Then, τ appears to be a bifurcation parameter. If we assume that (11) has an eigenvalue λ(τ ) =
ω (τ ) + iϕ(τ ), then we get ω (τ0 ) = 0 and ϕ(τ0 ) = ϕ0 , where τ0 = min{τj }.


Theorem 3. If we assume that a23 − a26 < 0, if h ( ϕ0 ) = 0 is satisfied, Hopf bifurcation occurs.

Proof. By using the implicit function theorem and differentiating (11) with respect to τ, we get:

ds sP2 (s)e−sτ
= −   , (22)
dτ −τP2 (s)e sτ + P2 (s)e−sτ + P1 (s)

where P1 (s) = s3α + a1 s2α + a2 sα + a3 , P2 (s) = a4 s2α + a5 sα + a6 . Therefore,

 
ds −1 −τ P (s) P1 (s)
= + 2 +
dτ s sP2 (s) sP2 (s)e−sτ
  (23)
−τ P (s) P (s)
= + 2 − 1 .
s sP2 (s) sP1 (s)

200
Mathematics 2019, 7, 846

Then,
 ds −1 
 
P (s) P (s)
Re  = Re[ 2 − 1 ]|
dτ s=iϕ0 sP2 (s) sP1 (s) s=iϕ0
2αa4 s2α + αa5 sα 3αs3α + 2αa1 s2α + αa2 sα
= Re[ − ]|s=iϕ0 (24)
s2 ( a4 s2α + a5 sα + a6 ) s2 (s3α + a1 s2α + a2 sα + a3 )

h ( ϕ0 )
= απ 3α απ α .
2ϕ0 ( a24 ϕ4α
0 + 2a4 a5 cos 2 ϕ0 + ( a5 + 2a4 a6 ) cosαπ ϕ0 + 2a5 a6 cos 2 ϕ0 + a6 )
2 2α 2

απ 3α
By assuming a4 s2α + a5 sα + a6 = 0, we get 2ϕ0 ( a24 ϕ4α
0 + 2a4 a5 cos 2 ϕ0 + ( a5 + 2a4 a6 ) cos απϕ0 +
2 2α

2a5 a6 cos απ ϕ α + a2 )  = 0. Thus, if h ( ϕ )  = 0, Re ds  = 0, then the transversality condition holds.

2 0 6 0 dτ
Therefore, Hopf bifurcation occurs at τ = τ0 .

In conclusion, we have the following theorem.


Theorem 4. If we assume that a23 − a26 < 0, h ( ϕ0 ) = 0 and the conditions in Theorem 2 are satisfied, we get
the following results:

(1) If τ < τ0 , E∗ is stable;

(2) If τ > τ0 , E∗ is unstable;
(3) A Hopf bifurcation exists at τ = τ0 .

Remark 1. According to (20), it is easy to see that τ0 relates to the order α. Therefore, if τ is selected, the order
α may be the cause of bifurcation.

4. Numerical Simulation
There are many numerical simulation methods like the Monte Carlo [19] and the predict-evaluate
and correct-evaluate (PECE) method [32,33], and the PECE method is used for numerical simulation
presented in this section. According to [12], the following Kaldor investment function is used:

eY ( t )
I (Y (t), R(t), K (t)) = − cR(t) − dK (t), (25)
1 + eY ( t )

where c, d > 0. According to the literature [12,34], the following liquidity preference function is chosen:
n
L(Y (t), R(t)) = mY (t) + , (26)
R(t) − R̂

where m, n, R̂ > 0. The following parameter values are chosen:

a = 2, b = 1.5, s1 = 0.2, s2 = 0.1, c = 0.1, d = 0.2,

(27)
m = 0.05, n = 0.0005, R̂ = 0.001, δ = 0.1, M̄ = 0.05,

one can get the following system:

eY ( t )
D α Y ( t ) = 2[ − 0.1R(t) − 0.2X (t − τ ) − 0.2Y (t) − 0.1R(t)],
1 + eY ( t )
0.0005
D α R(t) = 1.5[0.05Y (t) + − 0.05], (28)
R(t) − 0.001
eY ( t )
Dα X (t) = − 0.1R(t) − 0.2X (t − τ ) − 0.1X (t).
1 + eY ( t )

It is easy to see that E∗ = (0.97, 0.36, 2.30) is a positive equilibrium point of (28). In this case,

a23 − a26 < 0, h ( ϕ0 ) = 0, and the conditions in Theorem 2 are satisfied. By calculating, the critical value

201
Mathematics 2019, 7, 846

of System (28) is determined to be τ0 = 3.10 when α = 0.98. The initial values are chosen to be Y0 = 1,
R0 = 0.35, and X (θ ) = 2.3 where θ ∈ (−τ, 0). This is shown in Figures 1–3.

1.2 0.37

0.36
1.15
0.35

1.1 0.34

0.33
1.05
0.32

R
Y

1
0.31

0.95 0.3

0.29
0.9
0.28

0.85 0.27
0 200 400 600 800 1000 1200 0 200 400 600 800 1000 1200
t t

2.55

2.5
2.6

2.45
2.5

2.4
2.4

X
X

2.35
2.3

2.3 2.2
0.38
0.36
2.25 1.2
0.34 1.15
0.32 1.1
1.05
0.3 1
2.2 0.28 0.95
0.9
0 200 400 600 800 1000 1200 R 0.26 0.85
Y
t

Figure 1. E∗ is asymptotically stable, when α = 0.98, τ = 3.

2 0.5

1.8
0.45
1.6

1.4 0.4

1.2
0.35
1
R
Y

0.3
0.8

0.6 0.25

0.4
0.2
0.2

0 0.15
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
t t

2.5
2.5
X
X

2
1.5
0.5
0.4 2
0.3 1.5
1
0.2 0.5
1.5
0 100 200 300 400 500 600 700 800 900 1000 R 0.1 0
Y
t

Figure 2. Stable periodic orbit of System (28), when α = 0.98, τ = 3.2.

1.06 0.36

1.04
0.35

1.02

0.34
1

0.98 0.33
R
Y

0.96
0.32

0.94

0.31
0.92

0.9 0.3
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
t t

2.42

2.4
2.45
2.38
2.4
2.36

2.35
X

2.34
X

2.3
2.32

2.3 2.25
0.36
0.35
1.06
2.28 0.34 1.04
0.33 1.02
1
0.32 0.98
0.96
2.26 0.31 0.94
0.3 0.92
0 50 100 150 200 250 300 350 400 450 500 R 0.9
Y
t

Figure 3. E∗ is asymptotically stable, when α = 0.9, τ = 3.2.

In contrast with the results shown in Figures 1 and 2, if τ < τ0 , E∗ is stable, and if τ > τ0 , E∗ is
unstable, which conforms to Theorem 4. When comparing Figures 2 and 3, it can be seen that the order
α is an important factor in the stability of E∗ when τ is selected, which conforms to Remark 1.

202
Mathematics 2019, 7, 846

By virtue of this important discovery, it can be seen that it is necessary to establish a fractional
financial model with an appropriate fractional-order. Moreover, through analyzing the current capital
stock, predicting the future capital stock is beneficial to weaken the level of fluctuations.

5. Conclusions
In this paper, a DFOM for BC was established to describe the interaction of markets using
the interest rate. The existence and uniqueness of the proposed model were obtained. By mathematical
analysis, we found that the investment delay is an important factor in the stability and bifurcation of the
economic equilibrium in dynamic macroeconomics. Moreover, we also found that the fractional-order
affects the economic equilibrium’s stability and bifurcation.
In order to reduce the factors of macroeconomic instability and to promote stable development,
the government can adjust its investment activities from the following aspects, according to
the conclusions of this paper. On the one hand, through a series of measures, such as improving
the production equipment and working efficiency to reduce investment delay, the economic
fluctuations can be weakened. On the other hand, considering the output efficiency of various
industries under the current macroeconomic environment, by calculating the average investment
delay, the shortage of short-term capital stock in the future can be reasonably predicted, so investment
policy and the expected results can be combined to restrain the economic fluctuations caused by
investment delay.

Author Contributions: Y.X. studied the stability and the Hopf bifurcation analysis. Z.W. was in charge of the
fractional calculus theory and the simulation. B.M. provided guidance and recommendations for research; and,
lastly, Y.X. contributed to the contents and writing of the manuscript. All authors have read and approved the
final manuscript.
Funding: This work was supported by the National Natural Science Foundation of China (Nos. 61573008,
61973199), the Natural Science Foundation of Shandong Province (Nos. ZR2018MF005, ZR2016FM16), and the
Shandong University of Science and Technology graduate innovation project (SDKDYC190353).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Sargent, T.J. Dynamic Macroeconomic Theory; Harvard University Press: Cambridge, MA, USA, 2009.
2. Goodwin, R.M. The nonlinear accelerator and the persistence of business cycles. Econometrica 1951, 19, 1–17.
[CrossRef]
3. Kaldor, N. A model of the trade cycle. Econ. J. 1940, 50, 78–92. [CrossRef]
4. Chang, W.W.; Smyth, D.J. The existence and persistence of cycles in a nonlinear model: Kaldor’s 1940 model
re-examined. Rev. Econ. Stud. 1971, 38, 37–44. [CrossRef]
5. Hicks, J.R. Mr. Keynes and the “Classics”: A Suggested Interpretation. Econometrica 1937, 5, 147–159.
[CrossRef]
6. Hu, X.H.; Xia, J.W.; Wei, Y.L.; Meng, B.; Shen, H. Passivity-based state synchronization for semi-Markov
jump coupled chaotic neural networks with randomly occurring time delays. Appl. Math. Comput. 2019,
361, 32–41. [CrossRef]
7. Min, H.F.; Xu, S.Y.; Zhang, B.Y.; Ma, Q. Output-feedback control for stochastic nonlinear systems subject to
input saturation and time-varying delay. IEEE Trans. Autom. Control 2019, 64, 359–364. [CrossRef]
8. Min, H.F.; Xu, S.Y.; Zhang, B.Y.; Ma, Q. Globally adaptive control for stochastic nonlinear time-delay systems
with perturbations and its application. Automatica 2019, 102, 105–110. [CrossRef]
9. Riad, D.; Hattaf, K.; Yousfi, N. Dynamics of a delayed business cycle model with general investment function.
Chaos Soliton. Fract. 2016, 85, 110–119. [CrossRef]
10. Hattaf, K.; Riad, D.; Yousfi, N. A generalized business cycle model with delays in gross product and capital
stock. Chaos Soliton. Fract. 2017, 98, 31–37. [CrossRef]
11. Zhang, X.; Zhu, H. Hopf Bifurcation and Chaos of a Delayed Finance System. Complexity 2019, 6715036.
[CrossRef]

203
Mathematics 2019, 7, 846

12. Liu, X.; Cai, W.; Lu, J.; Wang, Y. Stability and Hopf bifurcation for a business cycle model with expectation
and delay. Commun. Nonlinear Sci. 2015, 25, 149–161. [CrossRef]
13. Hu, W.; Zhao, H.; Dong, T. Dynamic analysis for a Kaldor-Kalecki model of business cycle with time delay
and diffusion effect. Complexity 2018, 4, 1263602. [CrossRef]
14. Cao, J.Z.; Sun, H.Y. Bifurcation analysis for the Kaldor-Kalecki model with two delays. Adv. Differ. Equ. 2019,
2019, 107.
15. Fan, Y.; Huang, X.; Wang, Z.; Li, Y. Nonlinear dynamics and chaos in a simplified memristor-based
fractional-order neural network with discontinuous memductance function. Nonlinear Dyn. 2018, 9, 1–17.
[CrossRef]
16. Baitiche, Z.; Guerbati, K.; Benchohra, M.; Zhou, Y. Boundary value problems for hybrid Caputo fractional
differential equations. Mathematics 2019, 7, 282. [CrossRef]
17. Secer, A.; Altun, S. A new operational matrix of fractional derivatives to solve systems of fractional differential
equations via legendre wavelets. Mathematics 2018, 6, 238. [CrossRef]
18. Wang, Z.; Xie, Y.; Lu, J.; Li, Y. Stability and bifurcation of a delayed generalized fractional-order prey-predator
model with interspecific competition. Appl. Math. Comput. 2019, 347, 360–369. [CrossRef]
19. Fulger, D.; Scalas, E.; Germano, G. Monte Carlo simulation of uncoupled continuous-time random walks
yielding a stochastic solution of the space-time fractional diffusion equation. Phys. Rev. E 2008, 77, 021122.
[CrossRef] [PubMed]
20. Aguilar, J.P.; Korbel, J.; Luchko, Y. Applications of the Fractional Diffusion Equation to Option Pricing and
Risk Calculations. Mathematics 2019, 7, 796. [CrossRef]
21. Tarasov, V.E. Rules for Fractional-Dynamic Generalizations: Difficulties of Constructing Fractional Dynamic
Models. Mathematics 2019, 7, 554. [CrossRef]
22. Wang, Z.; Huang, X.; Shi, G. Analysis of nonlinear dynamics and chaos in a fractional-order financial system
with time delay. Comput. Math. Appl. 2011, 62, 1531–1539. [CrossRef]
23. Ma, J.; Ren, W. Complexity and Hopf bifurcation analysis on a kind of fractional-order IS-LM macroeconomic
system. Int. J. Bifurcat. Chaos 2016, 26, 1650181. [CrossRef]
24. Tacha, O.I.; Munoz-Pacheco, J.M.; Zambrano-Serrano, E.; Stouboulos, I.N.; Pham, V.-T. Determining the
chaotic behavior in a fractional-order finance system with negative parameters. Nonlinear Dyn. 2018,
94, 1303–1317. [CrossRef]
25. Podlubny, I. Fractional Differential Equations; Academic Press: New York, NY, USA, 1999.
26. Li, Y.; Chen, Y.; Podlubny, I. Stability of fractional-order nonlinear dynamic systems: Lyapunov direct
method and generalized Mittag-Leffler stability. Comput. Math. Appl. 2010, 59, 1810–1821. [CrossRef]
27. Yan, Y.; Kou, C. Stability analysis for a fractional differential model of HIV infection of CD4+ T-cells with
time delay. Math. Comput. Simulat. 2012, 82, 1572–1585. [CrossRef]
28. Gabisch, G.; Lorenz, W.H. Lorenz Business Cycle Theory-a Survey of Methods and Concepts, 2nd ed.; Springer:
New York, NY, USA, 1989.
29. Meng, B. Existence and convergence results of meromorphic solutions to the equilibrium system with
angular velocity. Bound. Value Probl. 2019, 2019, 88. [CrossRef]
30. Meng, B. Minimal thinness with respect to the Schrodinger operator and its applications on singular
Schrodinger-type boundary value problems. Bound. Value Probl. 2019, 2019, 91. [CrossRef]
31. Deng, W.; Li, C.; Lu, J. Stability analysis of linear fractional differential system with multiple time delays.
Nonlinear Dyn. 2007, 48, 409–416. [CrossRef]
32. Wang, Z. A numerical method for delayed fractional-order differential equations. J. Appl. Math. 2013,
2013, 256071. [CrossRef]
33. Wang, Z.; Huang, X.; Zhou, J. A numerical method for delayed fractional-order differential equations: Based
on G-L definition. Appl. Math. Inform. Sci. 2013, 7, 525–529. [CrossRef]
34. Cesare, D.L.; Sportelli, M. A dynamic IS-LM model with delayed taxation revenues. Chaos Soliton. Fract.
2005, 25, 233–244. [CrossRef]

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

204
 mathematics
Article
Fractional Dynamics and Pseudo-Phase Space
of Country Economic Processes
José A. Tenreiro Machado 1,† , Maria Eugénia Mata 2,† and António M. Lopes 3, *,†
1 Department of Electrical Engineering, Institute of Engineering, Polytechnic of Porto, R. Dr. António
Bernardino de Almeida, 431, 4249-015 Porto, Portugal; [email protected]
2 Nova SBE, Nova School of Business and Economics (Faculdade de Economia, Universidade Nova de
Lisboa), Rua da Holanda, 1, 2775-405 Carcavelos, Portugal; [email protected]
3 UISPA-LAETA/INEGI, Faculty of Engineering, University of Porto, Rua Dr. Roberto Frias,
4200-465 Porto, Portugal
* Correspondence: [email protected]
† These authors contributed equally to this work.

Received: 29 November 2019; Accepted: 30 December 2019; Published: 3 January 2020 

Abstract: In this paper, the fractional calculus (FC) and pseudo-phase space (PPS) techniques are
combined for modeling the dynamics of world economies, leading to a new approach for forecasting
a country’s gross domestic product. In most market economies, the decline of the post-war prosperity
brought challenging rivalries to the Western world. Considerable social, political, and military
unrest is today spreading in major capital cities of the world. As global troubles including mass
migrations and more abound, countries’ performance as told by PPS approaches can help to assess
national ambitions, commercial aggression, or hegemony in the current global environment. The 1973
oil shock was the turning point for a long-run crisis. A PPS approach to the last five decades
(1970–2018) demonstrates that convergence has been the rule. In a sample of 15 countries, Turkey,
Russia, Mexico, Brazil, Korea, and South Africa are catching-up to the US, Canada, Japan, Australia,
Germany, UK, and France, showing similarity in many respects with these most developed countries.
A substitution of the US role as great power in favor of China may still be avoided in the next decades,
while India remains in the tail. The embedding of the two mathematical techniques allows a deeper
understanding of the fractional dynamics exhibited by the world economies. Additionally, as a
byproduct we obtain a foreseeing technique for estimating the future evolution based on the memory
of the time series.

Keywords: fractional calculus; pseudo-phase space; economy; system modeling

1. Introduction
The last crisis (2007–2008) was severe, and it came without adequate warning to markets and
policymakers. High unemployment rates could not be avoided, and the austerity programs had
enormous impacts on standards of living, and caused extensive suffering. While sound democracies
have been implanted in the Western world, sacrifice associated with austerity throughout the 2008
financial crisis could not be alleviated, and great discredit has befallen economics as a social science.
The use of gross domestic product (GDP), and GDP per capita, to assess economic performance
and prosperity has long been discussed. The indicator does not include any out-of-market
production, as self-consumption escapes accounting efforts and methods altogether. At the same
time, negative externalities of economic growth, such as inequality, resource depletion, pollution,
environmental degradation, and effects on climate, have been forgotten, although they always affect
future economic growth [1].

Mathematics 2020, 8, 81; doi:10.3390/math8010081 205 www.mdpi.com/journal/mathematics

Mathematics 2020, 8, 81

The claim for better prosperity indicators and welfare measurement includes political fears about
the capacity of democratic political regimes to implement them [2]. While there are no better metrics to
reflect people’s lives and aspirations that will be able to inspire better policies for better lives after the
2008–2009 crisis, the US and the European economies have been growing slightly more slowly than in
earlier years, according to GDP metric.
As slower economic growth has been coupled with urban riots, strikes, warring, and mass
migrations around the world, analysts have been led to doubt the possibility of overcoming social
problems. Pessimistic views from Fogel 2007 even discuss the US capacity to go on performing as
the world great power [3]. They predict an acceleration of convergence, and suggest that by 2040 the
world may experience a global geopolitical turnover related with the relative economic decline of the
US. Beyond alternative techniques to be invented to measure prosperity, the relative share of the US
GDP in the global GDP will fall from 22% to 14%, according to Fogel [3]. For Europe, the relative share
of the European GDP in the global GDP will fall from 21% to 5%, it is said.
The contrast of post-war prosperity with post 2007–2008 crisis is evident. Social, political, and military
conflicts, including urban riots, terrorism, guerrilla actions, and outright warfare, are afflicting all
continents and regions of the world. Independently of the construction of other indicators to assess
economic performance and social progress, as suggested by Stiglitz et al. [1], the successful catching-up
of Asian partners to core countries, coupled with their expansionary demography, will bring about
fear for the survival of democracy, as this will depend on the political options of those Asian countries,
according to Fogel.
Experts have recalled other aspects related with past crisis indicators, and concluded that the
ongoing crisis has a new scale in comparison with any others in the past: Banking systems in Europe
have faced episodes of instability several times in the nineteenth and twentieth centuries, but those
crises were much less severe. The new millennium has brought monetary policies based on low interest
rates that do not compensate for individuals’ saving efforts, which is consistent with the ongoing
banking crisis and the financial instability, which have increased in this period [4–7].
From the perspective of market economies, crises are normal episodes. Modern markets in
capitalism have developed frequent episodes of booms and busts [8]. Surely, booms mean prosperity
(sometimes with bubbles) and give origin to busts (or even crises, recessions, and depressions),
that historically were followed by new prosperity [9]. Marxian views on the end of capitalism never
materialized [10]. The failure of the less-adapted firms during crises can explain Capitalism’s resilience
and survival (until now). Business cycles alternating between prosperity and crisis are intrinsic features
of market economies.
Independently of the invention of new measures for human welfare and happiness, economic
convergence of national economies deserves generalized approval, because, in improving the lowest
standards of living around the world, convergence contributes to global welfare.
In the 1960s Alexander Gerschenkron, economic historian at Harvard, devoted his attention to
economic growth in a historical perspective, and identified the meaning of industrialization for prosperity
to recall the higher rates of economic growth in the late-comers. According to Gerschenkron’s hypothesis,
the adoption of new technologies commands industrialization, helping imitators to grow faster [11].
This observation from the post-WWII perspective is remarkable, even if governments assumed strong
roles in national economies. In many countries, such as France or the UK, the government’s policies
did not stimulate capital markets in the 1950s and 1960s because of nationalization of important sectors
of economic activity [12]. However, the American role in joining European countries to the Marshall
Plan offer in the OEEC, in 1948, guided them to considerable co-operation, putting Western partners
into an openness attitude towards global capital markets, again.
The 1948–1973 period experienced the most successful economic growth of the entire history
of humankind. It allowed the defeated Japan and late-industrialized countries to catch up to the
developed European countries. Genuine economic modernization and urbanization occurred [13].
Individual aggressive competitive behavior was extended to more traditional societies. Co-operation

206
Mathematics 2020, 8, 81

with governments in strategies of growth, which included the formation of conglomerates and
cross-participation of multinational firms, provided safety and more robust opportunities in Asian
markets. In adopting new technologies for productive investment, Asian countries were successful
in catching-up with the most-developed nations. A convergence process took place in each country
in industrialization.
Fractional calculus (FC) is a mathematical formalism that models efficiently phenomena with
non-locality and long term memory. On the other hand, the pseudo-phase space (PPS) is a tool for
studying dynamics while avoiding the calculation of derivatives. Recently, various authors applied
the tools of classical dynamical systems to the economy. Petráš and Podlubny [14] used the state
space for describing national economies, while adopting GDP, inflation, and unemployment rates as
state variables. Machado and Mata [15,16] analyzed the Economic and Monetary Union countries
and their similarities during the integration process, and investigated the Portuguese short-run
business cycles over the last 150 years, using the multidimensional scaling method for visualizing
the results. Škovránek et al. [17] proposed an approach to macroeconomic modeling based on the
state space, fractional-order differential equations, and orthogonal distance fitting. The GDP, inflation,
and unemployment rates were adopted as state variables. Machado and Mata [18] presented a
bond-graph approach to model economy. The generalization of the principle of conservation of
power and the assignment of causalities were circumvented by means of a variable fractional-order
element. Machado et al. [19] investigated the economic growth using the multidimensional scaling
method and state space portrait analysis. The GDP per capita was adopted as the main indicator for
economic growth and prosperity, and the long-run perspective from 1870 to 2010 for identifying the
main similarities among countries’ economic growth. Machado and Mata [20] proposed the PPS and
FC for modeling the Western global economic downturn. Tarasova and Tarasov [21] introduced a
generalization of the economic model of logistic growth by considering the effects of memory and
crises. The memory effects are modeled with fractional order derivatives. Using the equivalence of
fractional differential equations and the Volterra integral equations they obtained discrete maps with
memory that were exact discrete analogs of fractional differential equations of economic processes.
Tarasov and Tarasova [22] designed a model of economic growth with fading memory and continuous
distribution of delay time. Their approach can be considered as a generalization of the standard
Keynesian macroeconomic model. Tejado et al. [23] presented models of economic growth for the
countries of the Group of Seven (G7) for the period 1973–2016. Such models consisted of differential
equations of both integer and fractional order, where the GDP was a function of the country’s land
area, arable land, population, school attendance, gross capital formation, exports of goods and services,
general government final consumption expenditure, and broad money. Ming et al. [24] applied the
Caputo fractional derivative to simulate China’s GDP growth, while comparing the effectiveness of
both fractional and integer order models. Škovránek [25] proposed a mathematical model based on
the one-parameter Mittag–Leffler function to describe the relation between the unemployment and the
inflation rates, known as the Phillips curve. For a comprehensive literature review see [26] and the
references therein.
Hereafter, the aforementioned questions are analyzed using the FC and PPS in the scope
of the complex dynamics of world economies and the perspective of 15 important countries.
Furthermore, a new approach for forecasting the GDP is proposed. The common adoption of the
two approaches leads to (i) a clear visualization and a straightforward interpretation of the dynamic
effects, and (ii) foreseeing the future dynamics of the time series (TS). The 15 selected countries are
Australia, Brazil, Canada, China, France, Germany, India, Japan, Korea, Mexico, Russia, South Africa,
Turkey, the United Kingdom, and the United States.
The paper has the following organization. Section 2 introduces the fundamentals of the FC and the
PPS representation. Section 3 develops the methodology and the analytical formulation for studying
the dynamics of the TS. Section 4 discusses the proposed estimators and evaluates their efficiency.
Finally, Section 5 summarizes the conclusions.

207
Mathematics 2020, 8, 81

2. Fundamental Concepts

2.1. The Fractional Calculus

The FC generalizes the derivative of a function f , D α f ( x ), to orders α ∈ R [27,28]. During the
last decades FC has become a popular tool [29–31], and new areas of application have emerged.
The modeling finance [32] and economy [18,21,22,26,33–35] phenomena became of particular relevance.
The Riemann–Liouville and Caputo are classical definitions of a fractional derivative (FD), being
given by [36]:

RL α 1 dn t x (τ )
a Dt x (t) = dτ, n − 1 < α < n, (1)
Γ(n − α) dtn a (t − τ )α−n+1
 t
C α 1 x (n) ( τ )
a Dt x (t) = dτ, n − 1 < α < n, α > 0, (2)
Γ(n − α) a ( t − τ ) α − n +1
where Γ (·) denotes the gamma function and {t, a} ∈ R (t > a) are the upper and lower limits of the
calculation interval, respectively.
Another classical expression is the Grünwald–Letnikov (GL) definition. The GL formulation has
the advantage of leading to a straightforward digital implementation and is given by [36]:

[ t−h a ]
GL α 1
a Dt x (t) = lim α
h →0+ h
∑ γ (α, k ) x (t − kh) , t > a, α > 0 (3a)
k =0
Γ ( α + 1)
γ (α, k ) = (−1)k , (3b)
k!Γ (α − k + 1)
 
so that t−h a is the integer part of t−h a and h stands for the time increment.
The calculation of Equation (3) can be approximated by the truncated series
r
1
D α [ x (t)] ≈ ∑ γ (α, k) x (t − kT ) , (4)
Tα k =0

where r represents the truncation order and T is the sampling period. Therefore, in the Z -domain we have:
 
1 r
α
γ(α, k)z−k
T α k∑
Z { D [ x (t)]} ≈ Z { x (t)} , (5)
=0

with Z {·} and z denoting the Z transform and variable, respectively.

Equation (5) results from adopting the Euler (or first backward difference) in the discrete
approximation. However, other approximations are possible, such as the Tustin (or bilinear) and
the Simpson rules, as well as their combinations [37]. To obtain rational expressions the approximants
need to be expanded into Taylor series and the final algorithm corresponds either to a truncated series,
or to a rational Padé fraction.
In the follow-up we shall adopt Equations (4) and (5) to approximate the FD, due to their direct
applicability in TS analysis.

2.2. The Pseudo-Phase Space

The PPS is useful for studying complex and non-linear dynamics. The PPS is more robust against
noise than the classical phase space (PS) method and allows representations using a small number of
measurements. The PPS reconstruction follows the Takens’ embedding theorem [38]. If a TS represented

208
Mathematics 2020, 8, 81

by x (t) is an attractor component represented by a smooth d-dim manifold, then the topological
properties of the TS are equivalent to those of the embedding n-dim vector:

v (t) = [ x (t) x (t + τ ) x (t + 2τ ) · · · x [t + (n − 1) τ ]] , (6)

where t stands for time, d, n ∈ N, n > 2d + 1, and τ > 0. Moreover, the symbols n and τ represent
the embedding dimension and time delay, respectively. For n = 2 or n = 3, the vector v (t) can be
represented in an n-dim plot, so that the vectors [ x (t) x (t + τ )] and [ x (t) x (t + τ ) x (t + 2τ )] in the
PPS reflect the classical dynamic described by [ x (t) ẋ (t)] and [ x (t) ẋ (t) ẍ (t)] in the PS.
An aspect of utmost importance in the PPS method is the time delay τ. Let us assume, for example,
that the signal x (t) has a limited superimposed noise. For small values of τ, the variables x (t), x (t + τ )
and x (t + 2τ ) have close values for the same sample and we obtain a straight line in the PPS. On the
other hand, for large values of τ the TS are almost independent, but their intersection almost vanishes.
Indeed, for a TS of length L ∈ N, the intersection becomes L − (n − 1) τ, when considering an n-dim
representation. Therefore, some kind of compromise needs to be established to obtain the time delay,
τm , that minimizes a given index, that is, for calculating min { J [v (t)]}.
τ
One possible way for selecting J is to adopt the autocorrelation function between x (t) and
x (t + τ ), for τ = 0, 1, 2, . . . . The value of τm is given by the first minimum. The fractal dimension
and the mutual information [19,39] were also adopted, but it is not clear which index is superior and
we often find several difficulties, such as low precision or small sensitivity, the influence of noise,
or problems due to the limited length L.

3. The Description of the Time Series Dynamics

We consider the dynamics of the TS characterizing a set of 15 countries during the years 1970–2018.
The set is made up of Australia, Brazil, Canada, China, France, Germany, India, Japan, Korea, Mexico,
Russia, South Africa, Turkey, the United Kingdom and the United States, herein denoted by {AUS,
BRA, CAN, CHN, FRA, DEU, IND, JPN, KOR, MEX, RUS, ZAF, TUR, GBR, USA}. The performance
of the economy is captured through the GDP per capita. The data were obtained at the World Bank
website on 15 November, 2019. In the following, each TS is represented by xi (t), where i = 1, . . . , 15.
A classical method for analyzing the dynamics consists of the PS. For example, Figures 1
and 2 depict the GDP of China versus time, x4 (t), and the corresponding 2- and 3-dim PS during
years 1970–2018.

8000

7000

6000

5000

4000

3000

2000

1000

0
1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020

Figure 1. The gross domestic product (GDP) of China, x4 (t), during 1970–2018.

209
Mathematics 2020, 8, 81

500

18
450
17

10 16
400 11 15
14
12 13
07 09
350 06
08

300 05

250 04

03
200
02

150 01
00
93 95
99
100 92 94 9798
96
86 91
84 8785
50 80 83 90
82 88
77 78 89
76 79 71 81
7574 73
72 70
0
0 1000 2000 3000 4000 5000 6000 7000 8000

80

60
05

40 04
09
91
90 06
20 03
82 92 02
83 01
76 77
75 81 99 00
0 7473 80
86
84
93 18
78 98 17
79
72 89 85 95
71 96 94 08 10 16
70 87 97
-20 13 15
12 14
88
07
-40 11

0
2000
4000
500
6000 400
300
200
8000 100
0

Figure 2. The 2- and 3-dim PS of China during 1970–2018.

For the numerical calculation of ẋ4 (t) and ẍ4 (t) we adopt the algorithm proposed in [40], since it
mitigates the effect of noise:

1
ẋ4 (t) = {2 [ x4 (t + T ) − x4 (t − T )] + x4 (t + 2T ) − x4 (t − 2T )} , (7a)
8T
1
ẍ4 (t) = {[ x4 (t + 2T ) + x4 (t − 2T )] − 2x4 (t)} , (7b)
4T 2
where T stands for the sampling period.
We note in the PS considerable variations in the derivatives ẋ4 (t) and ẍ4 (t). We can adopt
other filtering techniques, but they introduce undesirable delay and reveal limited performance.
Consequently, the PPS technique discussed in Section 2 emerges as a relevant strategy to solve
such issues.

210
Mathematics 2020, 8, 81

Two important limitations imposed by the TS are the large sampling period of one year, and the
limited length of the TS of 49 years. To have a smaller sampling period, but avoiding artificial numerical
artifacts, we implement a half-year piecewise cubic interpolation. Such an approach is consistent with
the TS evolution, without revealing large fluctuations among consecutive samples, and allows passing
from a TS with L = 49 to another one with L = 97 samples.
China’s economic growth after 1970 was very successful, as Figure 1 illustrates [24]. While
extensive literature calls into question the veracity and accuracy of Chinese GDP data based on
officially published information, often suggesting that the declared economic growth rate of Chinese
statistics greatly overstates the Chinese real rate of economic growth, it is true that China could adopt
technological innovation and move to a position of strong participation in international markets.
Thanks to institutional adjustments and entrepreneurial initiative, industrial growth based largely
on bank loans, and increasing consumption of electricity and freight, Chinese economic growth
has been remarkable, and clearly confirms resilience to adverse economic shocks [41]. Moreover,
China has experienced structural transformations that encourage productivity growth in producing
new commodities and services (especially those for electronic delivery), which explain an annual
average accumulated growth rate above 10%. While quick ongoing economic growth may be an
important reason to believe that China will not be able to continue indefinitely at such a pace, it is
a fact that Chinese expansion consistently out-performed most analysts’ expectations over the past
50 years [42].
To determine the delay τ, the cosine correlation [43] is adopted:

L−τ
∑ x (t) x (t + τ )
t =1
r (τ ) = . . (8)
L−τ L−τ
∑ x (t) ∑ x (t + τ )
2 2
t =1 t =1

The first minimum of r versus τ provides the value τm . Several numerical experiments confirmed
its good performance when compared with the mutual information and the Pearson or the Kendall tau
rank correlations.
Both the FD and the PPS capture the memory of past dynamics. Indeed, the FD implicitly includes
the past in Equation (4) by means of the series of signal samples at the time instants kT, k = 1, · · · , r.
On the other hand, the PPS captures memory through the delay τm in Equation (4). Therefore,
we can compare D α [ x (t)] with τm by some kind of average, γav , representing the TS. Nonetheless,
the arithmetic mean reveals difficulties due to the slow convergence of the GL coefficients γ (α, k ).
On the other hand, the geometric mean was tested numerically, revealing good convergence for
obtaining γav . We “average” the signal samples in Equation (4) by the geometric average of their terms,
yielding [20]:
 1
r r
r (r +1)
γav (α, r ) z−τm = ∏ γ (α, k) z− 2 , (9)
k =1

in the Z domain.
r (r +1)
Equation (9) leads directly to τm = 2 . Figure 3 depicts the evolution of γav versus (α, r ),
where the line connects the points with maximum value γav = max (γav ).

211
Mathematics 2020, 8, 81

Figure 3. Locus of γav versus (α, r ) and the points with maximum value γav = max (γav ).

For obtaining the geometric average of the GL series Equation (9) we interpolate γav = max (γav )
at τm and determine the corresponding α. The values of τm versus α are listed in Table 1 for the set of
15 countries.

Table 1. List of τm versus α for the set of 15 countries.

AUS BRA CAN CHN FRA DEU IND JPN KOR MEX RUS ZAF TUR GBR USA
τm (years) 13 9.5 10 8 17.5 20 24.5 20.5 22 10 15 19 6 12 13
α 0.430 0.454 0.450 0.470 0.406 0.400 0.390 0.398 0.395 0.449 0.419 0.402 0.492 0.435 0.430

After calculating the values of τm for each TS we can plot the corresponding PPS. Figure 4
represents the PPS [ x (t) x (t + τm )] of the GDP for the 15 countries during the period 1970–2018.
We observe the emergence of three main clusters, namely S1 = {CHN, IND}, S2 = {BRA, KOR, MEX,
RUS, TUR, ZAF} and S3 = {AUS, CAN, DEU, FRA, GBR, JPN, USA}. We note a clear trend towards
the 45 degree line, since the TS values do not vary quickly and, in general, recent values are higher than
previous ones, due to the global economic progress. Another detail is the clusters’ location, with S1
representing countries with a fast and steady growth, S2 standing for economies both with progress
and recession, and S3 for countries with a small but sustained improvement.

212
Mathematics 2020, 8, 81

10 5
AUS
USA
JPN CAN
DEU
KOR FRA

GBR
TUR
RUS
BRA
4
10 MEX
CHN
ZAF

IND

10 3

10 2
10 2 10 3 10 4 10 5

Figure 4. The pseudo-phase space (PPS) [ x (t) x (t + τm )] of GDP per capita for the set of 15 countries
during 1970–2018.

In a different perspective, we approximate the PPS [ x (t) x (t + τm )] by means of power law

functions, x (t + τm ) ≈ c · [ x (t)]b , using a non-linear least-squares fit algorithm. We verify that the
parameters b are correlated with α, as shown in Figure 5, reflecting the fractional behavior of the
economy dynamics. With the exception of IND, RUS and ZAF, we verify an almost linear correlation
given by b = −1.9737 + 6.2083 α for the rest of the countries.

2.5

IND
2

1.5

CHN TUR
AUS
1 CAN
GBR
USA
KOR DEU FRA BRA
0.5 MEX
JPN
0

-0.5

-1 RUS

-1.5
ZAF

-2
0.38 0.4 0.42 0.44 0.46 0.48 0.5

Figure 5. The locus of parameters b versus α. The markers’ size is proportional to the coefficient of
determination R2 .

The usual PPS portrait requires the calculation of τm , but lacks giving assertive information about
the memory embedded in the TS. Since a heuristic relationship between τm and α was formulated
previously, we expand the PPS by including the fractional dynamics information. Figure 6 shows
the 3-dim locus [ x (t) x (t + τm ) α] for the 15 countries during 1970–2018. This 3-dim locus extends
the usual 2-dim PPS, by placing x (t) and x (t + τm ) in the x- and y-axes, respectively, and α in the
z-axis. We find countries exhibiting slow dynamics (i.e., low α) in the bottom (e.g., IND, JPN, and
KOR), and countries with fast dynamics (i.e., high α) in the top (e.g., BRA, CHN, and TUR).

213
Mathematics 2020, 8, 81

0.5
TUR
0.48

CHN CAN
0.46
BRA
GBR
AUS
0.44 USA
MEX

0.42 FRA
RUS
DEU
JPN
0.4 KOR

ZAF
0.38
10 5
IND

10 4

10 3 10 5
10 4
10 3
10 2 10 2

Figure 6. The locus [ x (t) x (t + τm ) α] of GDP per capita for the set of 15 countries during 1970–2018.

The sample is made up of the G5 partners (UK, France, Germany, US, and Japan) and the three
new successful Asian partners (India, China, and Korea), and also includes the heirs of old empires
(Turkey and Russia), and four old European offshores (Australia, Canada, Mexico, Brazil, and South
Africa). Figure 4 proves that in their long-term economic growth, gaps have been blurred. This means
that globalization has brought acculturation and a contagious spread of economic growth. All over the
world, a process of diffusion of technology has brought an extension of capitalism:

• The European partners and Japan could catch-up to the US and two other old European offshores
(Australia and Canada). Before the late 1980s and the fall of the Berlin Wall, fighting communism
may be considered to have been crucial to the national political strategies in most Western
European countries and in East Asia. The anti-communist strategies clearly stimulated national
policies in drawing them toward stock markets after the late 1980s. Many decision makers
working at the World Bank and other international development agencies have even criticized
codification of capital markets as meaning overregulation for the purpose of extending capitalism
to communist-socialist areas.
• The heirs of old empires (Turkey and Russia), Korea, and three of the European offshores
(Mexico, Brazil, and South Africa) also converged. In spite of cultural differences, genetic specificity,
and climatic influences, they experimented consumption uniformization, with barriers that result
from inequality in the distribution of revenue. Russia’s political change has made its transition to
convergence with the most-developed economies difficult, and the country exhibits more erratic
economic growth behavior, comprising periods of strong rates of growth separated by frequent
and severe crises.
• The two Asian historical civilizations, China and India, have proceeded at fast and regular
economic growth rates; China and India’s comparatively less-modern sectors have been catching
up disproportionately faster to the world productivity frontier [44].

4. Estimation
In the last years the interest on models and algorithms for economic data forecasting
has been growing [45,46]. The available methods include regression analysis [47], moving
average [48], artificial neural networks [49], evolutionary computing [50], and empirical analysis [51],
among others [52,53].
The proposed strategy leads implicitly to one estimation method based on the PPS. While our
main objective was to establish a relationship between the time delay and the fractional order, it is
relevant to explore this method and to compare results with other standard schemes.
After obtaining the value of τm for each economy and the corresponding 2-dim PPS, we have
distinct alignments between the two TS, x (t) and x (t + τm ). This effect is negative, because it reduces

214
Mathematics 2020, 8, 81

the size of the intersection between the two vectors. Nonetheless, we can use the time shift between x (t)
and x (t + τm ) for foreseeing purposes, so that the future values [ x̂ ( L + 1) · · · x̂ ( L + τm )] are estimated
on the basis of the old values [ x ( L + 1 − τm ) · · · x ( L)]. In this perspective, the new values are those that
maximize the index of Equation (8) for the pair [ x̂ ( L + 1) · · · x̂ ( L + τm )] and [ x ( L + 1 − τm ) · · · x ( L)],
as shown in the diagram of Figure 7.

Figure 7. Estimation of future GDP values.

The values of the delay depend on the dataset and the estimation period varies accordingly.
Consequently, the estimation horizon is different for each country.

4.1. Assessing the Estimation Method

For assessing the proposed scheme we divide the available data into two subsets: 1970–2013 for
prediction and 2013.5–2018 for testing. First, we calculate the delays τm∗ for the prediction TS, since
they are slightly different from the ones resulting for the complete dataset, and we predict the future
values. Second, we compare the real and predicted values for the period 2013.5–2018 by means of the
metrics E1 and E2 , given by: /
0 T
01
E1 = 1 ∑ [ x (t) − x̂ (t)]2 , (10)
T t =1

1 T
| x (t) − x̂ (t)|
E2 =
T ∑ x (t) + x̂ (t)
, (11)
t =1

where T = 10 corresponds to the number of estimated points in the interval 2013.5–2018. These expressions
closely follow the so-called Euclidean and Canberra distances [54] often adopted to assess the
differences between numerical data.
Table 2 includes the complete list of values {τm∗ , α∗ } of the estimation TS and the errors {E1 , E2 }
for all countries. Since the prediction algorithm is supported by the past, we do not obtain unrealistic
values often proposed in the literature.

Table 2. The values of τm∗ , α∗ , E1 , and E2 for the set of 15 countries: Real data for 1970–2013 and
comparison between real and estimated values for 2013.5–2018.

AUS BRA CAN CHN FRA DEU IND JPN KOR MEX RUS ZAF TUR GBR USA
τm∗ (years) 13 9.5 8.5 8.5 9 19 20.5 20 21 10 13.5 17 5 8 12.5
α∗ 0.430 0.454 0.464 0.464 0.459 0.402 0.398 0.399 0.397 0.449 0.427 0.409 0.534 0.469 0.433
E1 3759.3 848.4 3170.7 315.3 2879.7 3278.0 325.0 6031.7 3995.9 374.3 1950.5 947.6 3764.0 2216.6 4089.5
E2 0.028 0.031 0.026 0.019 0.029 0.032 0.083 0.050 0.060 0.017 0.086 0.061 0.095 0.022 0.031

The errors E1 and E2 are compared with those obtained with common regression analysis. Usually,
regression methods perform adequately with economic data, since the TS evolves smoothly in time [55].
We adopt the nonlinear least-squares to fit 62 distinct models to the prediction TS, while discarding

215
Mathematics 2020, 8, 81

those that have a large number of parameters or that depict clear divergent behavior outside the fitting
interval. The best fit is obtained with the rational and the third degree polynomial functions:

a + bt
x̂ R (t) = , (12)
1 + ct + dt2

x̂ P (t) = a + bt + ct2 + dt3 , (13)

where a, b, c, d ∈ R are parameters to be estimated for each time series.

Equations (12) and (13) are then used for predicting the countries’ GDP per capita for the period
2013.5–2018 and for calculating E1 and E2 .
Figure 8 depicts the locus of the errors obtained with the PPS method and Equations (12) and (13).
We verify that the PPS method does not lead always to the best prediction. This result was expected
since Equations (12) and (13) were chosen as the best between a large number of possible functions,
while the PPS-based estimation is just a byproduct of the relationship between the time delay and
the fractional order. Nonetheless, we must note that the PPS method yields values of E1 and E2 in the
narrower interval, being more robust to the variations between the distinct country GDP per capita.

10 0
Rational model
Third degree polynomial
PPS method

10 -1

10 -2

10 -3
10 1 10 2 10 3 10 4 10 5 10 6 10 7

Figure 8. The locus of E1 and E2 obtained with the PPS method and Equations (12) and (13) for the set
of 15 countries.

Figure 9 depicts E2 versus α∗ , revealing three clusters that comprise {IND, RUS, TUR}, {JPN,
KOR, ZAF}, and {AUS, BRA, CAN, CHN, FRA, DEU, MEX, GBR, USA}, respectively. We verify the
small values of E2 particularly for the third (and larger) set of countries.

216
Mathematics 2020, 8, 81

0.1
TUR

0.09
RUS
IND
0.08

0.07

ZAF
0.06 KOR
JPN
0.05

0.04
DEU USA BRA
0.03 FRA
AUS CAN

MEX GBR
0.02
CHN

0.01
0.38 0.4 0.42 0.44 0.46 0.48 0.5 0.52 0.54

Figure 9. The locus of E2 versus α∗ for the set of 15 countries.

4.2. Complete Estimation of GDP per Capita

We estimate GDP per capita of the set of 15 countries up to the maximum accomplished by the
PPS method. This means that we have distinct prediction horizons for each country, going from 2024
for TUR up to 2042.5 for IND, as listed in Table 3. Figure 10 depicts the GDP per capita of China versus
time. For this country we have τm = 8 years and, therefore, we have real and estimation data for the
periods 1970–2018 and 2018.5–2026, respectively. Figure 11 illustrates the real and estimated PPS for
all countries. Again, we verify that the PPS representation produces good evolution without abrupt
changes or uncommon behavior.






















   
 
   

Figure 10. The GDP per capita of China versus time: Real data for 1970–2018 and estimated values
for 2018.5–2026.

217
Mathematics 2020, 8, 81

0.5
Real values TUR
Predicted values
0.48

CHN CAN
0.46
BRA
GBR
MEX AUS
0.44 USA

FRA
0.42
RUS DEU
JPN
KOR
0.4
ZAF

0.38
10 5 IND

4
10

10 3 10 5
10 4
3
10
10 2 10 2

Figure 11. The real and predicted PPS for the set of 15 countries.

Table 3. Estimation of GDP per capita for the set of 15 countries.

AUS BRA CAN CHN FRA DEU IND JPN KOR MEX RUS ZAF TUR GBR USA
α 0.430 0.454 0.450 0.469 0.406 0.400 0.390 0.398 0.395 0.449 0.419 0.402 0.492 0.435 0.430
2018.5 57,908.8 11,509.6 52,625.9 8026.6 44,457.5 49,163.0 2084.8 50,188.3 27,743.1 10,273.7 10,915.1 7221.4 14,975.7 43,929.1 55,651.7
2019 58,898.2 11,992.9 53,387.6 8298.1 45,251.4 50,824.2 2065.3 51,457.0 28,724.3 9928.3 10,385.6 7002.8 15,469.5 44,872.2 56,761.7
2019.5 59,887.6 12,476.2 53,769.9 8569.7 46,045.3 52,449.0 2045.9 52,725.7 29,705.5 10,071.7 10,728.3 6784.3 15,741.7 45,815.3 57,871.6
2020 60,877.1 12,959.6 54,424.7 8841.3 46,839.2 53,163.3 2026.5 53,994.4 30,686.7 10,286.1 11,092.5 6565.7 15,993.1 46,758.4 58,981.6
2020.5 61,866.5 13,362.1 55,079.4 9112.9 47,633.1 53,701.7 2007.1 55,263.2 31,667.9 10,406.9 11,545.3 6629.8 16,368.9 47,701.5 60,091.6
2021 62,855.9 13,431.3 55,593.1 9384.5 48,427.0 53,973.0 1992.4 56,531.9 32,649.1 10,515.7 12,041.0 6710.8 16,686.6 47,397.5 61,000
2021.5 63,845.3 13,497.0 55,785.7 9656.1 49,220.9 53,947.0 2012.9 57,800.6 33,630.3 10,662.1 12,566.2 6765.2 16,809.6 47,562.3 59,875.8
2022 64,834.7 13,662.4 55,963.7 9927.7 50,014.8 53,884.5 2049.3 58,994.7 34,611.5 10,752.8 13,086.0 6824.4 16,938.8 47,831.8 58,935
2022.5 65,058.1 13,781.4 56,268.1 10,199.3 50,808.7 53,658.1 2098.5 58,961.4 35,592.7 10,754.4 13,533.4 6924.8 17,436.8 48,037.5 59,360
2023 65,271.7 13,768.1 56,665.0 10,470.9 51,602.6 53,474.2 2172.2 59,226.3 36,573.9 10,756.0 13,774.1 7049.4 17,912.0 48,238.8 59,946.1
2023.5 65,553.7 13,732.4 57,290.6 10,742.5 52,169.3 53,727.3 2243.0 59,732.2 37,555.1 10,813.4 13,233.9 7183.7 18,045.4 48,408.1 60,196.4
2024 65,956.3 13,493.9 57,706.9 11,014.1 52,749.4 54,107.9 2267.5 60,358.5 38,536.3 10,916.1 12,696.3 7330.1 18,100.0 48,595.7 60,437.7
2024.5 66,662.2 13,135.1 57,690.3 11,285.7 53,105.6 54,294.2 2288.3 61,030.6 39,517.5 11,024.9 12,907.5 7485.4 48,883.4 60,871.6
2025 67,345.2 12,795.3 57,673.7 11,557.2 53,060.8 54,523.1 2322.7 61,559.4 40,498.7 11,136.6 13,261.0 7643.9 49,262.1 61,345.8
2025.5 67,656.5 12,596.3 57,661.9 11,828.8 52,942.5 55,401.3 2357.7 62,036.6 41,479.9 11,239.3 13,544.7 7810.5 49,786.5 61,685.6
2026 67,925.5 12,608.7 57,655.0 12,100.4 52,069.1 56,601.2 2380.1 62,462.1 42,461.1 11,323.9 13,820.3 7949.7 50,333.6 62,043.4
2026.5 68,282.5 12,628.1 58,110.3 51,195.6 57,679.8 2406.4 62,881.9 43,442.3 11,378.4 14,090.5 8046.8 50,753.3 62,526.5
2027 68,639.5 12,647.4 58,672.6 51,428.2 58,527.7 2469.8 63,473.8 44,423.5 11,425.2 14,307.1 8094.7 51,104.7 63,099.1
2027.5 68,931.3 12,670.1 58,855.4 51,901.3 59,035.7 2543.5 63,847.7 45,404.7 11,473.6 14,459.3 7977.9 51,387.8 63,819.6
2028 69,239.4 58,936.1 52,441.3 59,270.6 2617.1 63,630.8 46,385.9 11,522.0 14,533.6 7860.3 51,652.7 64,441.5
2028.5 69,687.7 52,780.7 57,903.2 2690.8 63,115.6 47,367.1 14,481.8 7905.0 51,949.5 64,718.7
2029 70,106.7 52,736.0 56,535.8 2764.5 61,522.1 48,348.3 14,376.4 7982.0 52,223.6 64,977.9
2029.5 70,330.0 52,691.2 56,940.7 2838.2 59,928.6 49,329.5 14,168.9 8058.1 52,438.4 65,425.2
2030 70,550.1 52,699.9 58,474.4 2911.8 60,950.7 50,310.7 14,015.1 8119.8 52,612.4 65,993.4
2030.5 70,923.4 52,721.9 60,135.6 2985.5 62,207.9 51,291.9 14,019.7 8145.9 66,664.6
2031 71,417.3 52,817.5 61,742.2 3059.2 62,231.9 52,273.1 14,035.0 8168.8 67,461.1
2031.5 52,973.2 61,861.7 3132.8 62,250.2 53,254.3 14,114.7 8207.0
2032 53,145.6 61,932.8 3206.5 62,629.9 54,235.5 14,248.1 8238.7
2032.5 53,338.1 61,991.7 3279.4 63,284.8 55,216.7 14,394.9 8252.7
2033 53,500.4 62,068.8 3353.1 64,043.1 56,197.9 14,571.2 8258.6
2033.5 53,709.6 62,546.4 3426.8 64,640.7 57,179.1 8250.0
2034 54,259.6 63,151.6 3500.5 64,813.8 58,160.3 8230.9
2034.5 54,916.5 63,429.0 3574.1 64,969.8 59,049.7 8185.2
2035 55,383.0 63,700.3 3647.8 65,383.8 59,799.7 8149.9
2035.5 55,762.5 64,111.2 3721.5 65,832.3 60,609.0 8150.2
2036 64,601.0 3795.1 66,062.8 61,409.5 8149.6
2036.5 65,194.0 3868.8 66,310.5 62,106.2 8135.4
2037 65,750.6 3942.5 66,982.7 62,791.0 8103.3
2037.5 66,155.4 4016.2 67,700.8 63,547.0
2038 66,469.2 4089.8 68,114.8 64,341.5
2038.5 4163.5 68,374.0 65,183.4
2039 4237.2 66,028.3
2039.5 4310.8 66,813.9
2040 4384.5 67,566.9
2040.5 4458.2
2041 4531.9
2041.5 4605.5
2042 4679.2
2042.5 4752.9

5. Discussion and Conclusions

A novel approach for analyzing the dynamics of economies was developed. The method embeds
the concepts of FC and PPS for processing information recorded in the TS.

218
Mathematics 2020, 8, 81

From the economic history perspective it is possible to say that by 2030 the world ranking of
countries’ GDPs per capita will not present significant differences from the ranking that prevails
today. Australia and the US will be at the top of the ladder, followed by Canada and Japan. However,
the three European countries (Germany, France, and UK), will be closer to them. Clearly, in a recovery
period, optimism has prevailed in Western economies [56], but commercial warring means that
some competition worries are in the air. Catching-up also expresses Western lifestyles convergence,
confirming Gerschenkron [11].
Among the newly successful partners, their relative positions will change quite clearly: Today
the Brazilian GDP per capita is about the same as Russia’s, much above China’s, while India remains
on the tail. In the same way, distances among Brazil, Russia, and China will disappear, to make
a pool of competitors. Global development of stock markets has allowed even the less-developed
Russian regions to improve and ameliorate poverty [57], while the economic growth expectations in the
Brazilian case will continue slowly, and growth from China relies on intensive diplomatic commercial
dealings with the US [56].
Looking again to the presented forecasts, 2030 will begin a global turnover. Germany will grow
and catch up to Japan. These two countries, both of which were defeated in WWII, will move ahead of
France and the UK, both of which were victors in that conflict. The results from the PPS methodology
indicate that India will not catch up, although it will benefit from the stimulus of strong external
demand from developing economies such as China, and fast dynamics. By 2030 and even 2040 India
will remain at the back, on the tail. The reason, as Battisti et al. [58] say, is that, although a country
converges to its long-run growth path, such a path may be not enough and can even diverge from
those countries at the global economic growth frontier. This means that for long-run performance,
low levels of wealth at departure do really matter.
In a modeling perspective, exploring the combination of FC and PPS led to a new strategy for
describing the dynamics of the economies while highlighting their fractional order. As a byproduct of
the PPS approach, a robust algorithm for estimating the future of the TS was also obtained.

Author Contributions: J.A.T.M. and A.M.L. conceived and designed the experiments; J.A.T.M. and A.M.L.
performed the experiments; J.A.T.M., M.E.M. and A.M.L. analyzed the data; J.A.T.M., M.E.M. and A.M.L. wrote
the paper. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Acknowledgments: We thank John Huffstot for correcting our English.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Stiglitz, J.E.; Fitoussi, J.P.; Durand, M. Measuring What Counts: The Global Movement for Well-Being; The New
Press: New York, NY, USA; London, UK, 2019.
2. Stiglitz, J. It’s Time to Retire Metrics Like GDP. They Don’t Measure Everything that Matters. Available
online: https://www.theguardian.com/commentisfree/2019/nov/24/metrics-gdp-economic-performance-
social-progress?utm_campaign=todays_worldview&utm_medium=Email&utm_source=Newsletter&
wpisrc=nl_todayworld&wpmm=1 (accessed on 26 November 2019).
3. Fogel, R.W. Capitalism and Democracy in 2040: Forecasts and Speculations; NBER Working-Paper No. 13184;
National Bureau of Economic Research: Cambridge, MA, USA, 2007.
4. Bordo, M.; Eichengreen, B.; Klingebiel, D.; Martinez-Peria, M.S.; Rose, A.K. Is the Crisis Problem Growing
More Severe? Econ. Policy 2001, 16, 53–82. [CrossRef]
5. Schularick, M.; Taylor, A.M. Credit Booms Gone Bust: Monetary Policy, Leverage Cycles and Financial
Crises, 1870–2008. Am. Econ. Rev. 2012, 102, 1029–1061. [CrossRef]
6. Taylor, A.M. The Great Leveraging; NBER Working-Paper No. 18290; National Bureau of Economic Research:
Cambridge, MA, USA, 2012.
7. Maito, E.E. The Historical Transience of Capital: The Downward Trend in the Rate of Profit Since XIX Century;
MPRA Paper No. 55894; National Bureau of Economic Research: Cambridge, MA, USA, 2014.

219
Mathematics 2020, 8, 81

8. Schumpeter, J.A. Business Cycles, a Theoretical, Historical and Statistical Analysys of the Capitalist Process;
Martino Pub.: New York, NY, USA, 1939.
9. Hayek, F.A. The Road to Serfdom: With the Intellectuals and Socialism; Routledge: London, UK, 1944.
10. Rostow, W.W. Stages of Economic Growth, A Non-Communist Manifesto; Cambridge University Press:
Cambridge, UK, 1960.
11. Gerschenkron, A. Economic Backwardness in Historical Perspective: A Book of Essays; Belknap Press of Harvard
University Press: Cambridge, MA, USA, 1962.
12. Roe, M.J. Political Determinants of Corporate Governance: Political context, Corporate Impact; Oxford University
Press on Demand: Oxford, UK, 2006.
13. Maddison, A. Development Centre Studies the World Economy: A Millennial Perspective; OECD Pub.: Paris,
France, 2001.
14. Petráš, I.; Podlubny, I. State Space Description of National Economies: The V4 Countries. Comput. Stat.
Data Anal. 2007, 52, 1223–1233. [CrossRef]
15. Tenreiro Machado, J.; Mata, M.E. Europe at the Crossroads of Economic Integration: Multidimensional
Scaling Analysis of the Period 1970–2010. Écon. Soc. 2012, 45, 1553–1577.
16. Tenreiro Machado, J.; Mata, M.E. Multidimensional Scaling Analysis of the Dynamics of a Country Economy.
Sci. World J. 2013, 2013, 594587. [CrossRef] [PubMed]
17. Škovránek, T.; Podlubny, I.; Petráš, I. Modeling of the National Economies in State-space: A Fractional
Calculus Approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]
18. Machado, J.A.T.; Mata, M.E. A Fractional Perspective to the Bond Graph Modelling of World Economies.
Nonlinear Dyn. 2014, 80, 1839–1852. [CrossRef]
19. Machado, J.; Mata, M.; Lopes, A. Fractional State Space Analysis of Economic Systems. Entropy 2015, 17, 5402–5421.
[CrossRef]
20. Machado, J.T.; Mata, M.E. Pseudo-phase Plane and Fractional Calculus Modeling of Western Global
Economic Downturn. Commun. Nonlinear Sci. Numer. Simul. 2015, 22, 396–406. [CrossRef]
21. Tarasova, V.V.; Tarasov, V.E. Logistic Map with Memory from Economic Model. Chaos Solitons Fractals 2017,
95, 84–91. [CrossRef]
22. Tarasov, V.E.; Tarasova, V.V. Dynamic Keynesian Model of Economic Growth with Memory and Lag. Mathematics
2019, 7, 178. [CrossRef]
23. Tejado, I.; Pérez, E.; Valério, D. Fractional Calculus in Economic Growth Modelling of the Group of Seven.
Fract. Calc. Appl. Anal. 2019, 22, 139–157. [CrossRef]
24. Ming, H.; Wang, J.; Fečkan, M. The Application of Fractional Calculus in Chinese Economic Growth Models.
Mathematics 2019, 7, 665. [CrossRef]
25. Skovranek, T. The Mittag-Leffler Fitting of the Phillips Curve. Mathematics 2019, 7, 589. [CrossRef]
26. Tarasov, V.E. On History of Mathematical Economics: Application of Fractional Calculus. Mathematics 2019,
7, 509. [CrossRef]
27. Samko, S.; Kilbas, A.; Marichev, O. Fractional Integrals and Derivatives: Theory and Applications; Gordon and
Breach Science Publishers: Amsterdam, The Netherlands, 1993.
28. Kilbas, A.; Srivastava, H.; Trujillo, J. Theory and Applications of Fractional Differential Equations; North-Holland
Mathematics Studies; Elsevier: Amsterdam, The Netherlands, 2006; Volume 204.
29. Tarasov, V. Fractional Dynamics: Applications of Fractional Calculus to Dynamics of Particles, Fields and Media;
Springer: New York, NY, USA, 2010.
30. Baleanu, D.; Diethelm, K.; Scalas, E.; Trujillo, J.J. Fractional Calculus: Models and Numerical Methods; Series on
Complexity, Nonlinearity and Chaos; World Scientific Publishing Company: Singapore, 2012.
31. Machado, J.T.; Lopes, A.M. Analysis of Natural and Artificial Phenomena Using Signal Processing and
Fractional Calculus. Fract. Calc. Appl. Anal. 2015, 18, 459–478. [CrossRef]
32. Scalas, E.; Gorenflo, R.; Mainardi, F. Waiting-times and Returns in High-frequency Financial Data: An
Empirical Study. Phys. A Stat. Mech. Its Appl. 2002, 314, 749–755.
33. Machado, J.T.; Lopes, A.M. Relative Fractional Dynamics of Stock Markets. Nonlinear Dyn. 2016,
86, 1613–1619. [CrossRef]
34. Lopes, A.M.; Machado, J.T.; Huffstot, J.S.; Mata, M.E. Dynamical Analysis of the Global Business-cycle
Synchronization. PLoS ONE 2018, 13, e0191491. [CrossRef]

220
Mathematics 2020, 8, 81

35. Tarasov, V.E.; Tarasova, V.V. Macroeconomic Models with Long Dynamic Memory: Fractional Calculus
Approach. Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
36. Valério, D.; Trujillo, J.J.; Rivero, M.; Machado, J.T.; Baleanu, D. Fractional Calculus: A Survey of Useful
Formulas. Eur. Phys. J. Spec. Top. 2013, 222, 1827–1846. [CrossRef]
37. Lopes, A.M.; Machado, J.T. Discrete-time Generalized Mean Fractional Order Controllers. IFAC-PapersOnLine
2018, 51, 43–47. [CrossRef]
38. Takens, F. Detecting Strange Attractors in Turbulence. In Dynamical Systems and Turbulence, Proceedings
of a Symposium Held at the University of Warwick 1979/80; Rand, D.; Young, L.S., Eds.; Lecture Notes in
Mathematics; Springer: Berlin, Germany, 1981; Volume 898, pp. 366–381.
39. Ali Shah, S.A.; Aziz, W.; Nadeem, A.; Sajjad, M.; Almaraashi, M.; Shim, S.O.; Habeebullah, T.M. A Novel
Phase Space Reconstruction-(PSR-) Based Predictive Algorithm to Forecast Atmospheric Particulate Matter
Concentration. Sci. Program. 2019, 2019, 6780379. [CrossRef]
40. Holoborodko, P. Smooth Noise Robust Differentiators. 2008. Available online: http://www.holoborodko.
com/pavel/numerical-methods/numerical-derivative/smooth-low-noise-differentiators/ (accessed on 26
November 2019).
41. Clark, H.; Pinkovskiy, M.; Sala-i Martin, X. China’s GDP Growth May Be Understated; Technical Report;
National Bureau of Economic Research: Cambridge, MA, USA, 2017.
42. Haltmaier, J. Challenges for the Future of Chinese Economic Growth. In FRB International Finance Discussion
Paper; Board of Governors of the Federal Reserve System: Washington, DC, USA, 2013.
43. Deza, M.M.; Deza, E. Encyclopedia of Distances; Springer: Berlin/Heidelberg, Germany, 2009.
44. di Giovanni, J.; Levchenko, A.A.; Zhang, J. The Global Welfare Impact of China: Trade Integration and Technological
Change; CEPR Discussion Paper; Centre for Economic Policy Research: London, UK, 2013.
45. Clements, M.P.; Hendry, D.F. Forecasting Non-Stationary Economic Time Series; MIT Press: Cambridge, MA,
USA, 2000.
46. Granger, C.W.J.; Newbold, P. Forecasting Economic Time Series; Academic Press: Cambridge, MA, USA, 2014.
47. Abeysinghe, T.; Rajaguru, G. Quarterly Real GDP Estimates for China and ASEAN4 with a Forecast
Evaluation. J. Forecast. 2004, 23, 431–447. [CrossRef]
48. Maity, B.; Chatterjee, B. Forecasting GDP Growth Rates of India: An Empirical Study. Int. J. Econ. Manag.
Sci. 2012, 1, 52–58.
49. Kaastra, I.; Boyd, M. Designing a Neural Network for Forecasting Financial and Economic Time Series.
Neurocomputing 1996, 10, 215–236. [CrossRef]
50. Claveria, O.; Monte, E.; Torra, S. Evolutionary Computation for Macroeconomic Forecasting. Comput. Econ.
2019, 53, 833–849. [CrossRef]
51. Gupta, M.; Minai, M.H. An Empirical Analysis of Forecast Performance of the GDP Growth in India.
Glob. Bus. Rev. 2019, 20, 368–386. [CrossRef]
52. Moiseev, N.A. Forecasting Time Series of Economic Processes by Model Averaging Across Data Frames of
Various Lengths. J. Stat. Comput. Simul. 2017, 87, 3111–3131. [CrossRef]
53. Bencivelli, L.; Marcellino, M.; Moretti, G. Forecasting Economic Activity by Bayesian Bridge Model Averaging.
Empir. Econ. 2017, 53, 21–40. [CrossRef]
54. Deza, E.; Deza, M.M. Dictionary of Distances; Elsevier: Amsterdam, The Netherlands, 2006.
55. Xavier, S.I.M. I Just Ran Two Million Regressions. Am. Econ. Rev. 1997, 87, 178–83.
56. Stovsky, R.; Esch, K. Trendsetter Barometer: Business Outlook; Technical Report; Pricewaterhousecoopers LLP:
London, UK, 2014.
57. Guriev, S.; Vakulenko, E. Breaking out of Poverty Traps: Internal Migration and Interregional Convergence in
Russia; DP9675; Center for Economic Policy Research: Washington, DC, USA, 2013.
58. Battisti, M.; di Vaio, G.; Zeira, J. Global Divergence in Growth Regressions; DP 96872013; Center for Economic
Policy Research: Washington, DC, USA, 2013.

c 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

221
 mathematics
Article
Fractional Derivatives for Economic Growth
Modelling of the Group of Twenty:
Application to Prediction
Inés Tejado 1, *, Emiliano Pérez 1 and Duarte Valério 2
1 Industrial Engineering School, University of Extremadura, 06006 Badajoz, Spain; [email protected]
2 IDMEC, Instituto Superior Técnico, Universidade de Lisboa, 1049-001 Lisbon, Portugal;
[email protected]
* Correspondence: [email protected]; Tel.: +34-924-289-300

Received: 16 December 2019; Accepted: 24 December 2019; Published: 1 January 2020 

Abstract: This paper studies the economic growth of the countries in the Group of Twenty (G20) in
the period 1970–2018. It presents dynamic models for the world’s most important national economies,
including for the first time several economies which are not highly developed. Additional care
has been devoted to the number of years needed for an accurate short-term prediction of future
outputs. Integer order and fractional order differential equation models were obtained from the
data. Their output is the gross domestic product (GDP) of a G20 country. Models are multi-input;
GDP is found from all or some of the following variables: country’s land area, arable land, population,
school attendance, gross capital formation (GCF), exports of goods and services, general government
final consumption expenditure (GGFCE), and broad money (M3). Results confirm the better
performance of fractional models. This has been established employing several summary statistics.
Fractional models do not require increasing the number of parameters, neither do they sacrifice
the ability to predict GDP evolution in the short-term. It was found that data over 15 years allows
building a model with a satisfactory prediction of the evolution of the GDP.

Keywords: fractional calculus; modelling; economic growth; prediction; Group of Twenty

1. Introduction
In this paper, models of economic growth are developed. The economies considered are those
of the countries members of the Group of Twenty (G20). The period under consideration consists
of the years from 1970 until 2018. The gross domestic product (GDP) is obtained as the output of a
dynamic system with eight input variables. The best models found employ derivatives of fractional
order. These models are compared with alternative versions with integer order derivatives only.
The comparison employs several statistical tools, commonly used to assess the quality of a model
to predict future outputs. In this manner, the ability of predicting the evolution of the GDP in the
short-term is demonstrated. This paper comes in the sequence of similar models obtained for Portugal,
Spain [1], France, Italy [2], all the EU member-states [3], the Group of Seven [4], and China [5].
The success of fractional order models for this purpose has been justified as consistent with the
mechanisms of economic growth, and is supported by the results.
Of the countries mentioned above, only China is not a highly developed country. Hence, this paper
has, for the first time, long term fractional order economic growth models for several countries which
are not highly developed, contributing to show that such models are suitable also for this particular
case. It also includes a study of the best number of years included in each model to optimise results.
The remainder of this paper is organised in the following manner. Section 2 describes the G20 and
presents a short state-of-the-art. Section 3 explains the methodology followed for economic growth

Mathematics 2020, 8, 50; doi:10.3390/math8010050 223 www.mdpi.com/journal/mathematics

Mathematics 2020, 8, 50

modelling of the G20. Section 4 gives and discusses the results obtained. The main conclusions of this
paper are drawn in Section 5.

2. Preliminaries

2.1. The G20

The G20 is a group of 19 countries and one international institution (the European Union),
9
that together account for over 10 of the world’s GDP. Membership comprises all the members of
the G7, which are the seven wealthiest advanced countries in the world (the European Union being
a permanent invitee, though not an eighth member). Other economies, which are not developed
economies (according to the criteria either of the United Nations or of the World Bank), play an
important role in the world economic and political scenes because of their size, or at least of their
regional importance. For this reason, the G7 conceived an extended international forum comprising
such countries, which became the G20. Membership was established by invitation upon foundation of
the G20 in 1999. Several summits have taken place every year since then. The most visible get together
the heads of state or of the heads of government of the members. Others take place at ministerial level,
or further business and trade partnerships.
The G20 members are, in alphabetical order: Argentina (ARG), Australia (AUS), Brazil (BRA),
Canada (CAN), China (CHI), the European Union (EUU), France (FRA), Germany (DEU), India (IND),
Indonesia (IDN), Italy (ITA), Japan (JPN), Mexico (MEX), Russian Federation (RUS), Saudi Arabia
(SAU), South Africa (ZAF), South Korea (KOR), Turkey (TUR), the United Kingdom (GBR), and the
United States of America (USA). Notice that France, Germany, Italy and the United Kingdom are
also member states of the European Union, and are thus represented both directly and indirectly at
meetings. (As of writing, the United Kingdom is about to leave European Union membership, and thus
to become only directly represented).
The final year for which data is available is 2018. While the G20 exists since 1999 only, it was
decided to extend the analysis further back to 1970. There are two reasons for this. First, data for less
than 20 years might suffice to establish some models, but not to validate them, verify their performance
over some extended period of time, or test their prediction abilities. Second, the period since 1970 is
one for which data is easily found for nearly all countries in the G20. Still, for four members, viz. China,
Russia, Saudi Arabia and Turkey, it was impossible to find data for this entire period. (This is not
surprising for Russia given that it was part of the Union of Soviet Socialist Republics until 1991. In the
other cases poor statistics may be due precisely to the lack of economic development). Restricting the
period so that there should be data for all countries would result in too short a period, as mentioned
above. Thus, only the remaining sixteen members of the G20 will be considered below. Despite this
fact, we will refer to those as the G20.

2.2. Fractional Calculus in Economic Modelling

Several financial and economic models of fractional order have been developed. A current review
can be found in [6]. An economic interpretation of fractional derivatives is given in [7]. A review of
methods for financial models is given in [4]. The particular case of economic growth was addressed
using fractional state-spaces [8–11], variable order derivatives [12], and pseudo-phase plane and state
space analysis [13,14]; the effect of memory obtained with fractional derivatives was studied in [15,16];
fractional diffusion models were used for economic crises [17] and financial markets [18,19].

3. Methodology
This section describes the methodology followed in this paper for economic growth modelling of
the G20.

224
Mathematics 2020, 8, 50

3.1. GDP Models

The models presented below rely on the following assumption: the evolution of the GDP is a
result of variables of two types. Variables of the first type reflect available resources; variables of the
second type reflect impacts on the economy. Consequently, the first model structure conceived for the
GDP is a linear integer order differential equation, which is, for each country of the G20, given by:
t
y(t) = C1 x1 (t) + C2 x2 (t) + C3 x3 (t) + C4 x4 (t) + C5 t0 x5 (t)dt + C6 x6 (t) + C7 x7 (t)
dx (t) dx (t)
+ C8 8 + C9 9 , (1)
dt dt
Variables are as follows:

• y(t): GDP, in 2010 US$;

• Ck : weights, constant in time, for each of the input variables xk ;
• t0 : first year considered—1970 in this case;
• xk : inputs of the model, viz.:
• x1 : land area, in km2 —measures the natural resources available;
• x2 : arable land, in km2 —measures of the quality of the natural resources;
• x3 : population—measures the human resources available;
• x4 : school attendance, in years—measures the quality of human resources;
• x5 : gross capital formation (GCF), in 2010 US$—measures manufactured resources (the model
considers the accumulated manufactured resources);
• x6 : exports of goods and services, in 2010 US$—measures external impacts on the economy;
• x7 : general government final consumption expenditure (GGFCE), in 2010 US$—measures
budgetary impacts on the economy;
• x8 : broad money (M3), in 2010 US$—measures monetary impacts on the economy (the model
considers the variation of monetary impacts);
• x9 : variation of x5 , in 2010 US$—the variation of GCF measures the impact of investment on
the economy.

Keynesian models for the dynamics of economies usually consider as inputs variables that have
short-term impacts in the economy. Growth accounting usually favors a more long-term approach.
(See examples in [20–22], and the discussion in [23] about the factors economic growth depends upon).
The variables above combine both. Notice that, to make the role of GCF clearer, since it appears twice
in the model with different roles, two different variables (x5 and x9 ) are used to denote it.
As explained below in Section 4.1, not all variables in (1) have the same importance for the
accuracy of the model. Their relative importance was found for each country for the whole time period.
In this manner simpler models could be obtained. In particular, a second integer order model, with five
variables only, was taken as an alternative:
 t
y(t) = ∑ Ck xk (t) + C5
t0
x5 (t)dt. (2)
k =1,3,6,7

Impacts on the economy have effects that are felt for an extended period of time. Of course,
this effect wanes away. Such a behavior can be modelled with fractional derivatives, since fractional
derivatives are operators with memory [15,24]. In other words, the fractional derivative of a function is
not a local operator, but its value depends on past values of the function. Depending on the particular
order employed, this memory of past values can correspond to weights of the said past values that
vanish for older time instants. This is the reason fractional derivatives are used to model phenomena
such as distributions corresponding to power laws, long tails in general, or chaotic systems [25].

225
Mathematics 2020, 8, 50

Hence, a fractional generalization of model (2) was considered. Rather than using more variables,
and thus recovering model (1), the variables of model (2) representing impacts, and only those,
are affected with a fractional derivative. Such variables are x5 , x6 and x7 , so the considered model was:

y(t) = ∑ Ck xk (t) + ∑ Ck Dαk xk (t). (3)

k=1,3 k =5,6,7

The sign of the differentiation orders α5 , α6 and α7 can be positive or negative. This type of
generalization has already been successfully used in our works [1–4].
Notice that the resulting fractional model has eight parameters. This is one parameter less than
the number of variables of the original integer model (1). As the number of variables is similar,
the comparison between the performance of the two is fair. The extra parameter of the integer model
gives it a slight advantage; consequently, should the performance turn out to be the same, the fractional
model will be considered better, since it achieves the same results with one parameter less.
The fractional differentiation operator Dαk was numerically implemented following the Caputo
α
definition [24], as 0 Dt k xk (t). Years are counted from 1970, which thus corresponds to the lower terminal
0. Terms for initial conditions were not included. Consequently, the effects of inputs are considered
only from 1970 on. This approximation reduces statistical data needed to develop models and was
used in previously published works, where it has provided acceptable results [4].

3.2. Optimizing and Assessing Performance

A fitting procedure implemented in MATLAB was used to find models (1)–(3) for each of
the G20 countries. This procedure relies on Nelder-Mead’s simplex search method. MATLAB’s
implementation from function fminsearch was used. The objective was the minimization of the mean
square error (MSE):
N
∑ (y j − ŷ j )2
j =1
MSE = , (4)
N
where N is the number of years—N = 49 in this case—and y j and ŷ j are the GDP and the model’s
GDP estimate, respectively. Several performance indexes other than the MSE were used from function
regstats to further evaluate the quality of the resulting models, viz.:

1. The mean absolute deviation (MAD):

N
∑ |y j − ŷ j |
j =1
MAD = . (5)
N

2. The coefficient of determination (R2 ∈ [0, 1]):

N
∑ (y j − ŷ j )2
j =1
R2 = 1 − , (6)
N
∑ (y j − ȳ)2
j =1

where ȳ is the mean of the GDP.

3. The t-values and p-values for each variable.

In Section 4 it will be shown that not all nine variables x1 , x2 , . . ., and x9 were necessary for every
single model given by (1). This was already the case in models for other countries [1–4]. This result was
established in three ways. First, from the t- and p-values for each variable. Second, from performance

226
Mathematics 2020, 8, 50

indexes MAD and R2 , that should not be significantly worst when one or more variables are removed
from the model, if they are indeed necessary. Third, from the Akaike information criterion (AIC):

N  2
∑ y j − ŷ j
j =1 2K (K + 1)
AIC = N log + 2K + , (7)
N N−K−1

where K is the number of model parameters. The value of the AIC itself does not give information
about the quality of a model. But comparing the AIC values of different models does. With such a
comparison, it is possible to find out which models have a higher probability be good models for
the data. In fact, a lower value of the AIC denotes a higher probability of a model being the best.
Assuming that there are M models, this probability of model i being the best can be normalized as the
Akaike weight wi , i = 1, . . . , M, by:
 
AICi − min AIC
M
exp −
2
wi =  . (8)
M AIC j − min AIC
∑ exp
M
−
j =1
2

In this way, models given by (2) and (3) were developed.

3.3. Models Found from Data for Different Numbers of Years

For each of the expressions (1)–(3), four models were obtained. The first uses the data for the
entire 1970–2018 period, so as to obtain a long term fit. This method, however, may lead to overfitting.
This can be caused by an excessive influence in parameters of data of too many years into the past.
Furthermore, it is impossible to assess the capability of this model of economic growth to predict
the future evolution of the economy, because there are no additional years of data for testing the
prediction ability.
To improve on this, three models for shorter time ranges were obtained. To find out which
numbers of years could be reasonable, trend lines were found for the GDP of each country. Both linear
and exponential trend lines were obtained; the former provided a better fit for some countries, and the
latter for others. Finally, a fast Fourier transform (FFT) was used to obtain, for the different tendencies,
the spectral content of the oscillations y(t) − ỹ(t). This was done in [4] to obtain the best time ranges
of models. In the present case, Figure 1 shows the spectral content of these oscillations for all countries,
normalized so that every curve peaks at 1. It can be seen that, in the G20, economies do not have similar
periods of oscillations around the corresponding tendencies. Within the frequencies where most peaks
take place, three reasonable values of time ranges were chosen: periods of 5, 10, and 15 years. In this
way, for each country, using (4) as cost function, 34 models were found for N = 15, for the periods
1970–1985, 1971–1986, 1972–1987, and so on, such as a moving average; and similarly for N = 10
(39 models) and N = 5 (44 models). And this was done separately for models given by (1)–(3). Each of
these models can be tested, using for this purpose the data of years in the future. In this manner, it is
possible to check how good the model is predicting GDP values which were not used to adjust its
parameters. The quality of the prediction was measured with performance indicators MSE, R2 , MAD,
AIC, and w for each country.
The GDP of different countries has different orders of magnitude. To make model performance
comparison easier, the figures below present the R2 performance index, to show the quality of
predictions obtained with each N–year model. The R2 is always in a normalized range, irrespective of
the magnitude of the variable under study, which makes it particularly suited for this visual purpose.
In this way, all the important characteristics of the different models in relation to the others can
be studied.

227
Mathematics 2020, 8, 50

Linear CAN
1 DEU
FRA
GBR
0.8 ITA
JPN
USA
0.6 EUU
ARG
AUS
0.4 BRA
IND
IDN
0.2 MEX
ZAF
KOR

0
0 0.1 0.2 0.3 0.4 0.5
Frequency / year -1
(a)

Exponential CAN
1
DEU
FRA
GBR
0.8 ITA
JPN
USA
0.6 EUU
ARG
AUS
0.4 BRA
IND
IDN
MEX
0.2
ZAF
KOR

0
0 0.1 0.2 0.3 0.4 0.5
Frequency / year -1
(b)

Best fit CAN

1 DEU
FRA
GBR
0.8 ITA
JPN
USA
0.6 EUU
ARG
AUS
0.4 BRA
IND
IDN
MEX
0.2
ZAF
KOR

0
0 0.1 0.2 0.3 0.4 0.5
Frequency / year-1
(c)

Figure 1. Spectral content of the oscillations y(t) − ỹ(t) for all countries, normalized so that every
curve has a peak with amplitude 1, where y(t) is the GDP, and ỹ(t) is a trend line: (a) Linear tendency
(b) Exponential tendency (c) The best of the previous two.

228
Mathematics 2020, 8, 50

4. Results
This section presents the models obtained, as well as their performance predicting the GDP of
G20 countries. Due to its extension, a full tabulation of results is not included in the paper, but is
available in [26]. Data sources are described in Appendix A.

4.1. Models for the 1970–2018 Period

Figures 2–4 show the results of the models obtained for each country from data for the
entire 1970–2018 period. The performance indices are tabulated in Tables 1–3. In those tables,
the t-values given in bold are those corresponding to variables which, assuming a 5% significance
level, are necessary for the model. This information is also given in Table 4. It turns out that variables
important for modelling six or more countries are x1 , x3 , x5 , x6 , and x7 . That is why model (1)
could be simplified into model (2), which is in its turn generalized to fractional orders by model (3),
only considering x5 , x6 , and x7 to have fractional influence.

11 ARG 11 AUS
10 10
5 16
Data Data
2 2
Integer 9var (R = 0.99055) Integer 9var (R = 0.9984)
4.5 14
Integer 5var (R 2 = 0.98024) Integer 5var (R 2 = 0.9975)
4 Fractional 5var (R 2 = 0.9916) 12 Fractional 5var (R 2 = 0.9994)
GDP (2010 US$)

GDP (2010 US$)

3.5 10

3 8

2.5 6

2 4

1.5 2
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(a) (b)
10
12 BRA 1012 CAN
2.5 2
Data Data
2
Integer 9var (R 2 = 0.98715) Integer 9var (R = 0.9983)
2 Integer 5var (R 2 = 0.98225) Integer 5var (R 2 = 0.9976)

Fractional 5var (R 2 = 0.99641) Fractional 5var (R 2 = 0.9984)

GDP (2010 US$)
GDP (2010 US$)

1.5
1.5

1
1

0.5

0 0.5
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(c) (d)

Figure 2. Cont.

229
Mathematics 2020, 8, 50

13 EUU 12 FRA
10 10
2 3
Data Data
Integer 9var (R 2 = 0.9975) Integer 9var (R 2 = 0.9984)
1.8
Integer 5var (R 2 = 0.9919) Integer 5var (R 2 = 0.9982)
2.5
1.6 Fractional 5var (R 2 = 0.9976) Fractional 5var (R 2 = 0.9985)
GDP (2010 US$)

GDP (2010 US$)

1.4
2
1.2

1
1.5

0.8

0.6 1
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(e) (f)

Figure 2. Results obtained with integer and fractional models for the following members of the G20:
(a) Argentina (b) Australia (c) Brazil (d) Canada (e) European Union (f) France. GDP estimations were
obtained with integer models (1) and (2) and with fractional model (3). R2 values are given to show the
quality of the results of each model. As GDPs have different orders of magnitude, different scales were
used in the y-axis for different countries.
12 12 IND
10 DEU 10
4 3
Data Data
2
Integer 9var (R 2 = 0.9949) Integer 9var (R = 0.9994)
2.5
3.5 Integer 5var (R 2 = 0.9920) Integer 5var (R 2 = 0.9987)

Fractional 5var (R 2 = 0.9958) Fractional 5var (R 2 = 0.9997)

GDP (2010 US$)
GDP (2010 US$)

2
3
1.5

2.5
1

2
0.5

1.5 0
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(a) (b)
1011 IDN 1012 ITA
12
Data 2.4 Data
Integer 9var (R 2 = 0.9917) Integer 9var (R 2 = 0.9968)
10 2.2
Integer 5var (R 2 = 0.9916) Integer 5var (R 2 = 0.9957)

Fractional 5var (R 2 = 0.9980) 2 Fractional 5var (R 2 = 0.9973)

GDP (2010 US$)
GDP (2010 US$)

8
1.8
6
1.6

4 1.4

1.2
2
1

0 0.8
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(c) (d)

Figure 3. Cont.

230
Mathematics 2020, 8, 50

12 JPN
10 1011 MEX
14
7 Data Data
Integer 9var (R 2 = 0.9898) Integer 9var (R 2 = 0.9950)
12
6 Integer 5var (R 2 = 0.98039) Integer 5var (R 2 = 0.9939)
Fractional 5var (R 2 = 0.9963) Fractional 5var (R 2 = 0.9977)
GDP (2010 US$)

GDP (2010 US$)

5 10

4 8

3 6

2 4

1 2
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(e) (f)

Figure 3. Results obtained with integer and fractional models for the following members of the G20:
(a) Germany (b) India (c) Indonesia (d) Italy (e) Japan (f) Mexico. GDP estimations were obtained with
integer models (1) and (2) and with fractional model (3). R2 values are given to show the quality of the
results of each model. As GDPs have different orders of magnitude, different scales were used in the
y-axis for different countries.

11 ZAF 11 KOR
10 10
4.5 15
Data Data
Integer 9var (R 2 = 0.9975) Integer 9var (R 2 = 0.9975)
4
Integer 5var (R 2 = 0.9951) Integer 5var (R 2 = 0.9952)
3.5 Fractional 5var (R 2 = 0.9979) Fractional 5var (R 2 = 0.9994)
GDP (2010 US$)
GDP (2010 US$)

10
3

2.5
5
2

1.5

1 0
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(a) (b)

1012 GBR 1013 USA

3 1.8
Data Data
Integer 9var (R 2 = 0.9966) Integer 9var (R 2 = 0.9972)
1.6
2.5 Integer 5var (R 2 = 0.99601) Integer 5var (R 2 = 0.99532)

Fractional 5var (R 2 = 0.9970) 1.4 Fractional 5var (R 2 = 0.9985)

GDP (2010 US$)
GDP (2010 US$)

2
1.2

1
1.5

0.8
1
0.6

0.5 0.4
1970 1978 1986 1994 2002 2010 2018 1970 1978 1986 1994 2002 2010 2018
Year Year
(c) (d)

Figure 4. Results obtained with integer and fractional models for the following members of the G20:
(a) South Africa (b) Korea (c) United Kingdom (d) United States of America. GDP estimations were
obtained with integer models (1) and (2) and with fractional model (3). R2 values are given to show the
quality of the results of each model. As GDPs have different orders of magnitude, different scales were
used in the y-axis for different countries.

231
Mathematics 2020, 8, 50

Table 1. Performance indices of the different models obtained for the G20 members in Figure 2; for an
explanation of performance assessment see Section 3.2.

Argentina Australia
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1019 ) 8.355 17.475 7.423 17.608 27.232 6.235
R2 0.9906 0.9802 0.9916 0.9984 0.9975 0.9994
MAD (×109 ) 7.761 10.101 7.614 10.208 11.986 5.467
x1 −0.385 3.393 −3.005 3.506 3.161 3.511
x2 2.154 − − 1.124 − −
x3 −1.000 −1.424 8.360 −1.319 −2.160 6.392
x4 2.065 − − −1.490 − −
t-values x5 0.910 3.128 9.086 2.193 1.316 −13.267
x6 1.694 3.794 2.214 2.194 3.222 4.974
x7 6.784 5.138 3.232 5.404 6.257 7.013
x8 1.139 − − 3.916 − −
x9 2.013 − − 0.646 − −
AIC (×103 ) 2.270 2.295 2.253 2.307 2.317 2.245
w (%) 0.02 0 99.98 0 0 100
Brazil Canada
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1020 ) 41.684 57.550 11.633 2.783 4.046 2.745
R2 0.9871 0.9823 0.9964 0.9983 0.9976 0.9984
MAD (×1010 ) 5.006 6.357 2.551 1.167 1.542 1.296
x1 −0.516 2.291 −8.401 −1.299 3.645 5.391
x2 1.583 − − 2.808 − −
x3 1.039 −0.063 10.666 −2.823 −3.194 −3.830
x4 −2.038 − − 2.150 − −
t-values x5 1.702 2.829 10.562 5.079 6.318 8.927
x6 1.067 1.816 8.516 8.195 11.611 14.040
x7 3.272 3.338 −6.579 5.487 8.069 6.094
x8 0.130 − − 2.200 − −
x9 2.509 − − 2.371 − −
AIC (×103 ) 2.462 2.466 2.388 2.329 2.336 2.317
w (%) 0 0 100 0.50 0 99.50
European Union France
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1020 ) 926.375 1122.995 334.785 4.942 5.812 4.852
R2 0.9975 0.9919 0.9986 0.9984 0.9982 0.9985
MAD (×1010 ) 25.326 29.613 12.679 1.816 1.897 1.731
x1 −1.385 0.697 −0.679 −2.704 −5.648 −4.586
x2 −1.384 − − −2.384 − −
x3 0.901 −0.805 2.200 5.069 5.968 5.294
x4 8.103 − − 1.606 − −
t-values x5 −3.213 −1.869 −11.728 −8.077 −9.564 −10.868
x6 3.636 2.915 7.922 9.667 14.155 14.625
x7 2.242 5.126 13.059 5.649 7.512 3.418
x8 0.994 − − 0.770 − −
x9 2.195 − − 0.592 − −
AIC (×103 ) 2.614 2.612 2.553 2.357 2.354 2.345
w (%) 0 0 100 0.22 1.18 98.60

232
Mathematics 2020, 8, 50

Table 2. Performance indices of the different models obtained for the G20 members in Figure 3; for an
explanation of performance assessment see Section 3.2.

Germany India
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1020 ) 25.579 39.739 21.073 2.911 6.925 1.395
R2 0.9949 0.9920 0.9958 0.9994 0.9987 0.9997
MAD (×1010 ) 4.086 4.265 3.675 1.231 2.051 0.958
x1 −1.204 −1.441 −4.573 1.527 −0.234 −3.781
x2 −1.993 − − −1.173 − −
x3 3.265 3.289 7.579 −0.764 4.915 7.666
x4 0.710 − − 1.670 − −
t-values x5 1.970 6.050 −10.136 16.351 10.091 −5.143
x6 −3.322 −3.310 2.629 −2.325 −0.570 5.745
x7 2.127 1.097 14.676 1.147 2.883 6.815
x8 2.385 − − 1.889 − −
x9 3.933 − − 3.630 − −
AIC (×103 ) 2.438 2.448 2.417 2.331 2.363 2.284
w (%) 0 0 100 0 0 100
Indonesia Italy
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1020 ) 7.362 7.166 1.741 6.567 6.716 5.670
R2 0.9917 0.9916 0.9980 0.9968 0.9957 0.9973
MAD (×1010 ) 2.057 2.036 0.896 1.918 1.999 1.887
x1 0.532 1.177 7.401 1.426 3.620 −4.646
x2 0.094 − − −0.143 − −
x3 −0.010 −1.134 −5.814 −3.038 −3.776 7.412
x4 −0.511 − − 3.588 − −
t-values x5 1.632 3.955 3.968 −5.028 −3.627 −15.949
x6 3.408 5.490 1.280 6.788 10.848 13.414
x7 6.959 7.320 40.381 10.987 32.791 18.617
x8 −0.031 − − 0.249 − −
x9 0.184 − − 0.829 − −
AIC (×103 ) 2.377 2.364 2.295 2.371 2.361 2.353
w (%) 0 0 100 0 1.56 98.44
Japan Mexico
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1020 ) 177.805 332.447 60.170 4.194 5.077 1.935
R2 0.9898 0.9804 0.9996 0.9950 0.9939 0.9977
MAD (×1010 ) 11.103 15.812 6.289 1.649 1.678 1.119
x1 −5.556 −3.962 10.164 1.582 −0.951 −3.354
x2 3.615 − − 0.388 − −
x3 0.456 3.324 −9.361 −1.214 3.162 5.822
x4 1.159 − − 2.077 − −
t-values x5 0.004 −0.027 −15.740 1.189 −0.446 8.484
x6 0.074 −0.138 0.892 3.199 3.515 7.318
x7 0.874 1.250 16.207 3.055 3.919 5.921
x8 −0.930 − − 0.794 − −
x9 0.878 − − 1.997 − −
AIC (×103 ) 2.533 2.552 2.469 2.349 2.347 2.303
w (%) 0 0 100 0 0 100

233
Mathematics 2020, 8, 50

Table 3. Performance indices of the different models obtained for the G20 members in Figure 4; for an
explanation of performance assessment see Section 3.2.

South Africa Korea

Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1019 ) 1.972 3.852 1.679 44.263 85.248 10.634
R2 0.9975 0.9951 0.9979 0.9975 0.9952 0.9994
MAD (×109 ) 3.201 4.548 3.163 16.928 24.418 7.162
x1 3.369 7.434 19.420 0.994 −4.833 −3.940
x2 −1.880 − − −4.284 − −
x3 0.874 −5.803 −12.655 0.715 4.116 6.034
x4 −4.015 − − −0.484 − −
t-values x5 1.017 1.711 4.929 0.353 2.794 −10.182
x6 9.658 9.529 17.246 1.659 −0.553 −18.575
x7 4.470 8.247 −14.366 1.252 1.230 13.575
x8 −0.639 − − 1.222 − −
x9 1.377 − − 2.531 − −
AIC (×103 ) 2.199 2.221 2.180 2.352 2.373 2.271
w (%) 0 0 100 0 0.03 99.97
United Kingdom United States of America
Integer Integer Fractional Integer Integer Fractional
Index/Statistic Variable
(1) (2) (3) (1) (2) (3)
MSE (×1021 ) 1.155 1.355 1.022 100.813 76.825 24.287
R2 0.9966 0.9960 0.9970 0.9972 0.9953 0.9985
MAD (×109 ) 2.828 2.917 2.578 235.854 221.266 127.878
x1 9.140 12.407 12.899 −4.909 −5.387 −14.723
x2 −0.976 − − 1.013 − −
x3 −6.393 −10.798 −11.453 4.969 5.474 19.974
x4 2.225 − − −2.622 − −
t-values x5 8.255 9.848 4.313 0.696 1.720 9.501
x6 3.738 5.963 7.597 −0.237 0.766 −5.742
x7 −1.567 −0.369 10.167 −0.017 1.458 6.522
x8 2.093 − − 3.485 − −
x9 0.692 − − 2.003 − −
AIC (×103 ) 2.399 2.396 2.382 2.618 2.593 2.537
w (%) 0.02 0.10 99.88 0 0 100

As can be observed, the MSE, R2 and MAD allow reaching the same conclusion: the performance
of models given by (3) is clearly better than the performance of integer models, in what concerns the
quality of the fit during the period used to build each model. This happens for all sixteen countries.
The Akaike weight, summarized in the last row of every country, also supports that models (3) are the
best of the three for this purpose.

234
Mathematics 2020, 8, 50

Table 4. Relevance of the independent variables of model (1), from which the GDP depends, for each
country; for an explanation of how variable relevance was determined, see Section 4.1.

x1 x2 x3 x4 x5 x6 x7 x8 x9
Argentina 
Australia   
Brazil  
Canada      
European Union   
France      
Germany    
India   
Indonesia  
Italy     
Japan  
Mexico  
South Africa    
Korea  
United Kingdom    
United States of America    

4.2. Models for N–Year Period

Figures 5–9 show the performance of integer and fractional models of a group of selected countries
(one per continent), namely Australia, the European Union, India, South Africa, and the United States
of America, for N = 5, 10 and 15 years, predicting the future evolution of the GDP. Showing results
for all countries would take too much space; then, results obtained with all models and performance
indices can be found in [26] for all countries.
Notice that models obtained with data from periods beginning in the 1970s can be used to predict
the GDP for many years until 2018. On the other hand, models developed with data from periods
ending in the 21st century can be used to predict the GDP for a few years only. Furthermore, predictions
for many years into the future have, as can be expected, a lower performance than those for years
close to the end of the data from which the model was got. In fact, the performances in Figures 5–9
deteriorate over time, but are quite good at prediction for a short period, and here again fractional
models show their better performance, as R2 values do not decrease so significantly.
As far as the number of years for prediction is concerned, it was observed that the smaller the
value of N, the better fitting—MSE obtained for every N–year period was really close to zero—but the
lower the ability to predict GDP in future: the values of R2 were the smallest of the three cases. This was
especially clear for integer model (1). Conversely, the largest the value of N, the lower the value of
MSE, but the better the prediction. Notice that the values of R2 for N = 15 were close to 1, especially
for predictions with fractional model (3).
Hence, in order to predict the economic growth of a country of G20 with certainty, it is necessary
to consider a relatively large period of years.

235
Mathematics 2020, 8, 50

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.8 0.95 0.95

R2

R2

R2
0.6 0.9 0.9

0.4 0.85 0.85

0.2 0.8 0.8

20 40 10 20 30 40 20 40
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(a)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.98 0.98 0.98

0.96 0.96 0.96

R2

R2

0.94 0.94 R2 0.94

0.92 0.92 0.92

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(b)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.995 0.995 0.995

R2

R2

R2

0.99 0.99 0.99

0.985 0.985 0.985

0.98 0.98 0.98

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(c)

Figure 5. R2 values of GDP estimates for Australia. Estimates were obtained with models (1) (left),
(2) (middle), and (3) (right). The models were obtained with data for different numbers of years:
(a) N = 5 (top) (b) N = 10 (center) (c) N = 15 (bottom). The models were used to estimate the GDP for
as many years as possible after the period for which they were built. The scale of the y-axis is not the
same for all models and all values of N.

236
Mathematics 2020, 8, 50

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.8 0.8 0.8

0.6 0.6 0.6

R2

R2

R2
0.4 0.4 0.4

0.2 0.2 0.2

0 0 0
10 20 30 40 10 20 30 40 10 20 30 40
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(a)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.99 0.99 0.99

0.98 0.98 0.98

R2
R2
R2

0.97 0.97 0.97

0.96 0.96 0.96

0.95 0.95 0.95

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(b)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.995 0.995 0.995

R2

R2

R2

0.99 0.99 0.99

0.985 0.985 0.985

0.98 0.98 0.98

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(c)

Figure 6. R2 values of GDP estimates for the European Union. Estimates were obtained with models
(1) (left), (2) (middle), and (3) (right). The models were obtained with data for different numbers of
years: (a) N = 5 (top) (b) N = 10 (center) (c) N = 15 (bottom). The models were used to estimate the
GDP for as many years as possible after the period for which they were built. The scale of the y-axis is
not the same for all models and all values of N.

237
Mathematics 2020, 8, 50

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.8
0.95 0.95

R2

R2
R2

0.6
0.9 0.9
0.4

0.2 0.85 0.85

10 20 30 40 10 20 30 40 10 20 30 40
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(a)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.99 0.99 0.99

R2

R2

R2
0.98 0.98 0.98

0.97 0.97 0.97

0.96 0.96 0.96

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(b)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.995 0.995 0.995

R2

R2

R2

0.99 0.99 0.99

0.985 0.985 0.985

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(c)

Figure 7. R2 values of GDP estimates for India. Estimates were obtained with models (1) (left),
(2) (middle), and (3) (right). The models were obtained with data for different numbers of years:
(a) N = 5 (top) (b) N = 10 (center) (c) N = 15 (bottom). The models were used to estimate the GDP for
as many years as possible after the period for which they were built. The scale of the y-axis is not the
same for all models and all values of N.

238
Mathematics 2020, 8, 50

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.8
0.9 0.9
0.6

R2
R2

R2
0.4
0.8 0.8
0.2

0 0.7 0.7
10 20 30 40 10 20 30 40 10 20 30 40
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(a)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.95
0.99 0.99
0.9
R2

R2
R2

0.98 0.98
0.85
0.97 0.97
0.8

0.96 0.75 0.96

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(b)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.996 0.98 0.996

0.992 0.96 0.992

R2

R2
R2

0.988 0.94 0.988

0.984 0.92 0.984

0.98 0.9 0.98

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(c)

Figure 8. R2 values of GDP estimates for South Africa. Estimates were obtained with models (1) (left),
(2) (middle), and (3) (right). The models were obtained with data for different numbers of years:
(a) N = 5 (top) (b) N = 10 (center) (c) N = 15 (bottom). The models were used to estimate the GDP for
as many years as possible after the period for which they were built. The scale of the y-axis is not the
same for all models and all values of N.

239
Mathematics 2020, 8, 50

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.8
0.95 0.95
0.6

R2

R2
R2

0.9 0.9
0.4
0.85 0.85
0.2

0 0.8 0.8
10 20 30 40 10 20 30 40 10 20 30 40
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(a)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.98 0.98 0.98

0.96 0.96 0.96

R2

R2
R2

0.94 0.94 0.94

0.92 0.92 0.92

0.9 0.9 0.9

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(b)

Int. (9 var) Int. (5 var) Fract. (5 var)

1 1 1

0.98 0.98 0.98

R2
R2

R2

0.96 0.96 0.96

0.94 0.94 0.94

10 20 30 10 20 30 10 20 30
Year-ahead-prediction Year-ahead-prediction Year-ahead-prediction
(c)

Figure 9. R2 values of GDP estimates for the United States of America. Estimates were obtained with
models (1) (left), (2) (middle), and (3) (right). The models were obtained with data for different numbers
of years: (a) N = 5 (top) (b) N = 10 (center) (c) N = 15 (bottom). The models were used to estimate the
GDP for as many years as possible after the period for which they were built. The scale of the y-axis is
not the same for all models and all values of N.

240
Mathematics 2020, 8, 50

5. Conclusions
The models of economic growth, of both integer and fractional order, presented in this paper
for countries of the Group of Twenty (G20), from 1970–2018, are satisfactory. The variables chosen to
predict variations of gross domestic product (GDP) prove to be suitable to the desired purpose.
It is clear from the results obtained that the performance of fractional models is superior.
This statement is qualitatively backed by several indexes. Fractional models do not require an
additional number of parameters, neither do they sacrifice the ability to predict the evolution of
the GDP in the short-term. As to the number of years needed to build acceptable models, results show
that N = 15 years lead to the best results.
The methodology followed in this paper can be further applied to more countries, and eventually
generalised to more variables. Database [27], for instance, includes many time series, usually of
good coherence, for all countries, that could be tested in the systematic manner described. The main
difficulty is the disparity in the number of years for which time series are available; while for the
G20 we could complete the missing values for eight variables, sixteen countries, and forty-nine years,
this would likely be very difficult or even impossible if the number of variables, countries or years
should be increased. So it would be necessary to improve this methodology in a manner that would
cope with missing data and still be able to find, validate and compare models.

Author Contributions: This work involved all coauthors. I.T. wrote the original draft of this manuscript and
contributed to the obtaining and the analysis of data and models. E.P. contributed to the obtaining of data and
the coding in MATLAB. D.V. supervised all the work and contributed to frequency analysis of the data and the
manuscript editing. All authors have read and agreed to the published version of the manuscript.
Funding: This research was supported in part by the Consejería de Economía, Ciencia y Agenda Digital (Junta
de Extremadura) under the grant “Ayuda a Grupos de Investigación de Extremadura” (no. GR18159), in part
by the European Regional Development Fund “A way to make Europe”, and in part by FCT, through IDMEC,
under LAETA, project UID/EMS/50022/2019.
Acknowledgments: The authors would like to thank José Emilio Traver for its help in editing the
graphical abstract.
Conflicts of Interest: The authors declare no conflict of interest.

Appendix A. Data Sources

This appendix lists the sources of data used in this paper, which is not tabulated in this paper
because of its size. It is available in [28].

• As in [4], variables for the EUU were the sum of the figures for its member states in each year.
The only exception was x4 , addressed below.
• The source for the GDP, x1 , x2 , x3 , x5 , x6 , and x7 was [27].
• Variable x2 was available until 2016 only. It was assumed that x2 (2017 : 2018) = x2 (2016).
For Belgium and Luxemburg, which are member-states of the EUU, there is no x2 data until 2000.
Thus, x2 was assumed constant until that year. This approximation corresponds, in the worst case,
to an error in x2 of 1.9% for the EUU during those years.
• The source for x1 and x3 for DEU until 1990 was [29]. In the same period, figures for x2 were
reduced in the same proportion.
• The source for x4 was [30] until 2010. Figures are available with a 5-year period only, and were
interpolated with a third-order spline. The figure for 2010 was extended into the future, using the
increase rate of the figures in [31], also interpolated with a third-order spline. However, Figures for
the following member-states of the EUU are not found in [30]: Croatia, Estonia, Latvia, Lithuania,
Slovenia, Slovak Republic. The source for x4 for these states was [27]. The EUU figure for x4 is
a weighted average of the figures for the member states in each year. The weight is the share of
each state in x3 .

241
Mathematics 2020, 8, 50

• Figures for x5 , x6 , and x7 for JPN and USA for 2018 are those of 2017, updated with the yearly
growth rate of the index in [32].
• The source for x7 for ARG until 1992 was [27]. In the 1993–2018 period, the figure for 1992 was
updated with the yearly growth rate of the index in [32].
• The source for x8 for ARG, AUS, BRA, IDN, IDN, JPN, MEX, ZAF, GBR, and USA was [27].
• The source for x8 for CAN until 2008 was [27]. In the 2009–2018 period, the figure for 2008 was
updated with the yearly growth rate of the index in [32].
• The source for x8 for DEU, FRA, ITA and other states of the EUU until 2015 was [33]. Figures
were converted to 2010 US$ using the price index in [27]. In the 2016–2018 period, the figure
for 2015 was updated with the growth rate in [34–36] for DEU, FRA, and ITA, respectively.
However, figures for x8 for Luxembourg and Romania in [33] are only available until 2011 and
2013, respectively. The figure for the last year was updated with the growth rate of [27].

References
1. Tejado, I.; Valério, D.; Pérez, E.; Valério, N. Fractional calculus in economic growth modelling. The Spanish
and Portuguese cases. Int. J. Dyn. Control 2017, 5, 208–222. [CrossRef]
2. Tejado, I.; Valério, D.; Pérez, E.; Valério, N. Fractional calculus in economic growth modelling: The economies
of France and Italy. In Proceedings of the International Conference on Fractional Differentiation and Its
Applications, Novi Sad, Serbia, 18–20 July 2016.
3. Tejado, I.; Pérez, E.; Valério, D. Economic growth in the European Union modelled with fractional derivatives:
First results. Bull. Pol. Acad. Sci.-Tech. Sci. 2018, 66, 455–465.
4. Tejado, I.; Pérez, E.; Valério, D. Fractional calculus in economic growth modelling of the Group of Seven.
Fract. Calc. Appl. Anal. 2019, 22, 139–157. [CrossRef]
5. Ming, H.; Wang, J.R.; Feckan, M. The Application of Fractional Calculus in Chinese Economic Growth
Models. Mathematics 2019, 7, 665. [CrossRef]
6. Tarasov, V.E. On history of mathematical economics: Application of fractional calculus. Mathematics 2019,
7, 509. [CrossRef]
7. Tarasova, V.V.; Tarasov, V.E. Economic interpretation of fractional derivatives. Progr. Fract. Differ. Appl. 2017,
3, 1–6. [CrossRef]
8. Hu, Z.; Tu, X. A new discrete economic model involving generalized fractal derivative. Adv. Differ. Equ.
2015, 65. [CrossRef]
9. Petrás, I.; Podlubny, I. State space description of national economies: The V4 countries. Comput. Stat.
Data Anal. 2007, 52, 1223–1233. [CrossRef]
10. Skovranek, T.; Podlubny, I.; Petrás, I. Modeling of the national economies in state-space: A fractional calculus
approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]
11. Yue, Y.; He, L.; Liu, G. Modeling and application of a new nonlinear fractional financial model. J. Appl. Math.
2013, 325050. [CrossRef]
12. Xu, Y.; He, Z. Synchronization of variable-order fractional financial system via active control method.
Cent. Eur. J. Phys. 2013, 11, 824–835. [CrossRef]
13. Tenreiro Machado, J.A.; Mata, M.E. Pseudo phase plane and fractional calculus modeling of western global
economic downturn. Commun. Nonlinear Sci. Numer. Simul. 2015, 22, 396–406. [CrossRef]
14. Tenreiro Machado, J.A.; Mata, M.E.; Lopes, A.M. Fractional state space analysis of economic systems.
Entropy 2015, 17, 5402–5421. [CrossRef]
15. Tarasov, V.E. Long and short memory in economics: Fractional-order difference and differentiation. IRA-Int. J.
Manag. Soc. Sci. 2016, 5, 327–334. [CrossRef]
16. Tarasova, V.V.; Tarasov, V.E. Exact discretization of an economic accelerator and multiplier with memory.
Fractal Fract. 2017, 1, 6. [CrossRef]
17. Caputo, M.; Carcione, J.M.; Botelho, M.A.B. Modeling extreme-event precursors with the fractional diffusion
equation. Fract. Calc. Appl. Anal. 2015, 18, 208–222. [CrossRef]
18. Scalas, E.; Gorenflo, R.; Mainardi, F. Fractional calculus and continuous-time finance. Physica A 2000,
284, 376–384.

242
Mathematics 2020, 8, 50

19. Aguilar, J.-P.; Korbel, J.; Luchko, Y. Applications of the fractional diffusion equation to option pricing and
risk calculations. Mathematics 2019, 7, 796.
20. Denison, E.F. Why Growth Rates Differ: Postwar Experience in Nine Western Countries; Brookings Institution:
Washington, DC, USA, 1967; ISBN 9780815718055.
21. Lucas, R.E. On the mechanics of economic development. J. Monet. Econ. 1988, 22, 3–42. [CrossRef]
22. Maddison, A. Explaining the economic performance of nations, 1820–1989. In Convergence of Productivity:
Cross-National Studies and Historical Evidence; Baumol, W.J., Nelson, R.R., Wolff, E.N., Eds.; Oxford University
Press, Inc.: Oxford, UK, 1994; pp. 20–61.
23. Van den Berg, H. Economic Growth and Development; World Scientific Publishing Co. Pte. Ltd.: Singapore,
2017; ISBN 978-981-4733-33-5.
24. Valério, D.; Sá da Costa, J. Introduction to single-input, single-output fractional control. IET Control
Theory Appl. 2011, 5, 1033–1057. [CrossRef]
25. Wang, S.; He, S.; Yousefpour, A.; Jahanshahi, H.; Repnik, R.; Perc, M. Chaos and complexity in a
fractional-order financial system with time delays. Chaos Solitons Fractals 2019, 109521. [CrossRef]
26. Tejado, I.; Pérez, E.; Valério, D. Results for Predictions of the Future Evolution of the GDP for the G20 Group.
2019. Available online: https://github.com/UExtremadura/Economic/blob/master/G20Results_Tejado_
et_al2019.rar (accessed on 16 December 2019).
27. The World Bank. World Bank Database. 2019. Available online: https://databank.worldbank.org/source/
world-development-indicators (accessed on 19 August 2019).
28. Tejado, I.; Pérez, E.; Valério, D. Economic Data for the G20 Group. 2019. Available online: https://
github.com/UExtremadura/Economic/blob/master/G20Data_Tejado_et_al_Mathematics19.xls (accessed
on 16 December 2019).
29. West Germany. 2018. Available online: https://en.wikipedia.org/wiki/West_Germany (accessed on
29 August 2018).
30. Lee, J.W.; Lee, H. Lee and Lee Long-run Education Dataset, Lee–Lee Database Version 2.2. 2018.
Available online: http://www.barrolee.com/Lee_Lee_LRdata_dn.htm (accessed on 29 August 2019).
31. Wittgenstein Centre for Demography and Global Human Capital. Wittgenstein Centre Data Explorer
Version 2.0. 2018. Available online: http://dataexplorer.wittgensteincentre.org/wcde-v2/ (accessed on
19 August 2019).
32. OECD. OECDiLibrary. 2019. Available online: http://dx.doi.org/10.1787/1036a2cf-en (accessed on
10 September 2019).
33. Federal Reserve Bank of St. Louis. Federal Reserve Economic Data. 2019. Available online: https://fred.
stlouisfed.org/ (accessed on 10 September 2019).
34. The Global Economy. Economic Indicators for Over 200 Countries: Germany. 2019. Available online:
https://www.theglobaleconomy.com/Germany/money_supply/ (accessed on 12 September 2019).
35. The Global Economy. Economic Indicators for Over 200 Countries: France. 2019. Available online: https:
//www.theglobaleconomy.com/France/money_supply/ (accessed on 12 September 2019).
36. The Global Economy. Economic Indicators for Over 200 Countries: Italy. 2019. Available online: https:
//www.theglobaleconomy.com/Italy/money_supply/ (accessed on 12 September 2019).

c 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

243
 mathematics
Article
The Application of Fractional Calculus in Chinese
Economic Growth Models
Hao Ming 1 , JinRong Wang 1,2 and Michal Fečkan 3,4, *
1 School of Mathematics and Statistics, Guizhou University, Guiyang 550025, China
2 School of Mathematical Sciences, Qufu Normal University, Qufu 273165, China
3 Department of Mathematical Analysis and Numerical Mathematics, Faculty of Mathematics,
Physics and Informatics, Comenius University in Bratislava, Mlynská Dolina, 842 48 Bratislava, Slovakia
4 Mathematical Institute, Slovak Academy of Sciences, Štefánikova 49, 814 73 Bratislava, Slovakia
* Correspondence: [email protected]

Received: 2 June 2019; Accepted: 23 July 2019; Published: 25 July 2019 

Abstract: In this paper, we apply Caputo-type fractional order calculus to simulate China’s gross
domestic product (GDP) growth based on R software, which is a free software environment for
statistical computing and graphics. Moreover, we compare the results for the fractional model with
the integer order model. In addition, we show the importance of variables according to the BIC
criterion. The study shows that Caputo fractional order calculus can produce a better model and
perform more accurately in predicting the GDP values from 2012–2016.

Keywords: Caputo fractional derivative; economic growth model; least squares method

MSC: 26A33

1. Introduction
As one of the most important macroeconomic statistics indicators, GDP is an effective tool for
people to understand and grasp the macroeconomic operation of a country; it is also an important basis
for formulating economic policies. However, the calculation of GDP is very complicated, so a good
economic growth model (EGM) can effectively form the economic progress problem, and it can reduce
the loss of human and material resources.
Derivatives and integrals are often used to describe the process of economic development.
However, there are still some shortcomings in using classical calculus to model real data. In recent
years, the existence of solutions to fractional order differential equations have been studied in [1–3].
In addition, fractional calculus is widely used to construct economic models; it incorporates the effects
of memory in evolutionary processes; experimental results show that the fractional order model is
superior to integer order model, such as [4–13].
Recently, Luo et al. [14] improved the fractional EGM model in [5] and adopted different
computational methods to simulate GDP via MATLAB, SPSS, and R software. The simulation
results showed that the newly-established fractional hybrid model had better performance than
the classical model.
In this paper, we adopt the idea in [14] to apply Caputo fractional order EGM and integer order
to study China’s GDP growth, as well as the minimum mean-squared-error (MSE) to estimate the
parameters in the model. In order to compare the fitting effect between the integer order and the
fractional order model, we establish the minimum absolute error coefficient, determination, and the
BIC index. Finally, we use the prediction effect of the absolute relative error evaluation model.
Summarizing, based on fractional calculus, this paper conducts modeling of China’s economic
growth. Through a case study, it shows that fractional calculus has a better effect than integral calculus

Mathematics 2019, 7, 665; doi:10.3390/math7080665 245 www.mdpi.com/journal/mathematics

Mathematics 2019, 7, 665

in modeling. It would be possible to use Monte Carlo simulation to generate sample data (see [15,16])
and then conduct modeling comparison. In this paper, real data are used for modeling and then
showing the advantage of fractional calculus. The purpose of the two methods is the same, but the case
analysis is often more complex and difficult than simulation, so a simulation is not used in this paper.

2. Models Description
We select six explanatory variables in this paper, and they are land area (LA) (km2 ), cultivated
area (CL) (km2 ), total population (TP) (million), total capital formation (TCF) (billion), exports of goods
and services (EGS) (billion), and general government final consumer spending (GGFCS) (billion), and
the explained variable is GDP (billion). The data used in this paper were all Chinese data from the
world bank from 1961–2016.
In order to simplify the expression, we define the following symbols:

x1 x2 x3 x4 x5 x6 y n k t
LA CL TP TCF EGS GGFCS GDP NVM NPM year

The general expression of the EGM is y = f ( x1 , x2 , · · · ), where f is the given function. Thus,
the integer order model (IOM) and Caputo fractional order model (CFOM) are considered as:

• IOM:  t
dx7 (t)
y(t) = ∑ c j x j ( t ) + c4
t0
x4 (t)dt + c7
dt
,
j=1,2,3,5,6

• CFOM:
7
α
y(t) = ∑ ck ( Dt0k,t xk )(t),
j =1

where t0 and αk represent the starting year and order respectively; in addition, the Caputo
α
derivative Dt0k,t xk for a given function xk is defined as (see [1]):

 t dxk (s)
α 1
Dt0k,t xk (t) = ds
ds, t > t0 , 0 < αk ≤ 1.
Γ (1 − α k ) t0 ( t − s )αk

In order to facilitate the comparison of GDP between different years, the GDP, TCF, EGS,
and GGFCE used here were converted into unchangeable local currency. The data from 1961–2011
were selected as the training sample, and data from 2012–2016 were used as the test sample. Moreover,
we used the average absolute deviation (MAD) and the coefficient of determination (R2 ) to evaluate
the model, and the absolute relative error criterion was used to compare the prediction effect of the
model. Recall the following definitions:
n
∑ |yi − ŷi |
i =1
MAD = ,
n
and:
 
 y − ŷi 
AREi =  i , i = 1, 2, · · ·, n,
y  i

and:
n
∑ (yi − ŷi )2
i =1
R2 = 1− n .
∑ (yi − ȳ)2
i =1

246
Mathematics 2019, 7, 665

We often used the Akaike information criterion (AIC) and Bayesian information criterion (BIC)
for the selection of variables in the model. Compared with the BIC criterion, the AIC criterion has the
phenomenon of over-fitting. Therefore, we adopted the following BIC criterion:
 
n
1 p log n
BIC = log
n ∑ (yi − ŷi ) 2
+
n
,
i =1

and:
 ( BIC − BIC ) 
exp − j
2
min

ωj = p  ( BIC − BIC )  .
∑ exp −
j min
2
j =1

3. Main Results

3.1. Economic Data for China

By using the Chinese economic data from 1961–2016 in unchangeable local currency, we apply R
software to get the following figure (see Figure 1).

x1 x2
9388250

1200000
Arable area (km2)
Land area (km2)

9388230

1100000
1000000
9388210

1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010

year year

x3 x4
130000

250000
Population (Ten thousand)

GCF (Hundred million)

110000

150000
90000

50000
70000

1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010

year year

x5 x6
1e+05
140000

8e+04
GGFCE (Hundred million)
100000
EGS (Hundred million)

6e+04
60000

4e+04
2e+04
0 20000

0e+00

1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010

year year

Figure 1. Data for China from 1961–2016. EGS, exports of goods and services.

247
Mathematics 2019, 7, 665

3.2. Parameter Estimation

In this paper, we used R software and the least squares method to obtain the coefficient estimation
in the integer order and Caputo fractional order models. Moreover, according to the MSE criteria,
we gave the order of the Caputo fractional order model, and the following data were obtained (see
Table 1).Table 2 shows the significance test results of the IOM and CFOM coefficients.

Table 1. The coefficients and orders of the integer order model (IOM) and the Caputo fractional order
model (CFOM).

IOM CFOM IOM CFOM

α1 0 −0.5389 c1 −0.0051 0.0027
α2 0 −1.3704 c2 0.0286 0.0029
α3 0 −0.6873 c3 0.3220 −0.3619
α4 −1 0.0960 c4 0.1147 0.6058
α5 0 −0.7777 c5 0.4229 0.2329
α6 0 0.0331 c6 3.5943 1.3208
α7 1 3.5251 c7 0.7978 0.1087

Table 2. Significance level of the Caputo model.

IOM CFOM
Variable t-Value p-Value t-Value p-Value
x1 −5.066 7.76 × 10−6 10.750 6.86 × 10−14
x2 2.853 6.58 × 10−3 12.048 1.58 × 10−15
x3 4.127 1.61 × 10−4 −10.697 8.04 × 10−14
x4 6.601 4.41 × 10−8 16.408 2.00 × 10−16
x5 5.498 1.83 × 10−6 19.996 2.00 × 10−16
x6 10.692 8.14 × 10−14 6.645 3.80 × 10−8
x7 2.128 3.90 × 10−2 2.304 2.60 × 10−2

The results in Table 2 show that when the significance level was 0.05, the coefficients of IOM and
CFOM passed the significance test.

3.3. Model Evaluation

In order to compare the performance of limited samples between IOM and CFOM, we present the
values of MAD, R2 , and BIC index in the training sample set (see Table 3).

Table 3. Sample performance of IOM and CFOM.

Index MSE MAD R2 BIC

IOM 15,497,849 2,430.793 0.9991 17.0959
CFOM 1,906,429 1070.643 0.9999 15.0004

We adopted the BIC criterion to select variables in the model, and the importance of each variable
was obtained, represented by ω (see Table 4).

248
Mathematics 2019, 7, 665

Table 4. The importance of variables based on BIC.

Variable IOM CFOM

BIC without x1 17.47825 15.81924
one variable x2 17.18849 16.06651
x3 17.34602 15.91584
x4 17.70702 15.02607
x5 17.54180 15.73556
x6 18.29923 16.06451
x7 17.11674 15.02607
ω found from the x1 14.40% 12.92%
BIC without x2 16.64% 11.42%
one variable x3 15.38% 12.31%
x4 12.84% 19.22%
x5 13.95% 13.48%
x6 9.55% 11.43%
x7 17.25% 19.22%

3.4. Fitting Results

Now, we give the fitting results of IOM and CFOM based on R software (see Figure 2).
5e+05

Original Data
4e+05

IOM (R2 = 0.9991)

FOM (R2 = 0.9999)

GDP(Hundred million)

3e+05
2e+05
1e+05
0e+00

1960 1970 1980 1990 2000 2010

year

Figure 2. Data fitting.

3.5. Predicted Results

Finally, we present the forecast results of the IOM and CFOM models for China’s GDP data from
2012–2016, and we calculate ARE index values, as shown in Table 5.

Table 5. Our results.

IOM CFOM
Year Real Value
Predict Value AREi Predict Value AREi
2012 557,487.6 554,010.4 0.6237% 558,937.6 0.2601%
2013 600,735.4 611,769.9 1.8368% 601,514.2 0.1296%
2014 644,575.1 665,424.7 3.2346% 638,231.2 0.9842%
2015 689,052.1 741,602.5 7.6265% 675,637.3 1.9468%
2016 735,218.6 810,156.9 10.1927% 714,153.2 2.8652%

4. Conclusions
We selected six economic indicators in this paper and used IOM and CFOM to model China’s
GDP growth from 1961–2011. The fitting results showed that CFOM was significantly better than

249
Mathematics 2019, 7, 665

IOM. To further illustrate the forecasting effect of the CFOM model, we presented the GDP forecast
for China from 2012–2016 and compared it with the real value. It was found that the CFOM model
not only had an advantage in fitting China’s GDP growth, but also predicted it better. Finally, since
all data were discrete, we intend to extend our study by applying the Caputo differences to create a
fractional discrete time EGM.

Author Contributions: The contributions of all authors (H.M., J.W. and M.F.) are equal. All the main results were
developed together.
Funding: This work is partially supported by the National Natural Science Foundation of China (11661016),
the Major Research Project of Innovative Group in Guizhou Education Department ([2018]012), the Slovak
Research and Development Agency under Contract No. APVV-14-0378, and the Slovak Grant Agency VEGANo.
2/0153/16 and No. 1/0078/17.
Acknowledgments: The authors thank the referees for their careful reading of the article and insightful comments.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Kilbas, A.A.; Srivastava, H.M.; Trujillo, J.J. Theory and Applications of Fractional Differential Equations; Elsevier:
Amsterdam, The Netherlands, 2006.
2. Diethelm, K. The Analysis of Fractional Differential Equations; Lecture Notes in Mathematics; Springer:
New York, NY, USA, 2010.
3. Zhou, Y.; Wang, J.; Zhang, L. Basic Theory of Fractional Differential Equations, 2nd ed.; World Scientifc:
Singapore, 2016.
4. Škovránek, T.; Podlubny, I.; Petráš, I. Modeling of the national economies in state-space: A fractional calculus
approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]
5. Tejado, I.; Valério, D.; Pérez, E.; Valério, N. Fractional calculus in economic growth modelling: The Spanish
and Portuguese cases. Int. J. Dyn. Control 2017, 5, 208–222. [CrossRef]
6. Machado, J.A.T.; Mata, M.E. A fractional perspective to the bond graph modelling of world economies.
Nonlinear Dyn. 2015, 80, 1839–1852. [CrossRef]
7. Machado, J.A.T.; Mata, M.E.; Lopes, A.M. Fractional state space analysis of economic systems. Entropy
2015, 17, 5402–5421. [CrossRef]
8. Machado, J.A.T.; Mata, M.E. Pseudo phase plane and fractional calculus modeling of western global economic
downturn. Commun. Nonlinear Sci. Numer. Simulat. 2015, 22, 396–406. [CrossRef]
9. Tarasova, V.V.; Tarasov, V.E. Elasticity for economic processes with memory: Fractional differential calculus
approach. Fract. Differ. Calc. 2016, 6, 219–232. [CrossRef]
10. Tarasov, V.E.; Tarasova, V.V. Harrod-Domar growth model with memory and distributed lag. Axioms
2019, 8, 9. [CrossRef]
11. Tarasov, V.E.; Tarasova, V.V. Dynamic Keynesian model of economic growth with memory and lag.
Mathematics 2019, 7, 178. [CrossRef]
12. Tarasov, V.E.; Tarasova, V.V. Macroeconomic models with long dynamic memory: Fractional calculus
approach. Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
13. Tarasov, V.E. Generalized memory: Fractional calculus approach. Fractal Fract. 2018, 2, 23. [CrossRef]
14. Luo, D.; Wang, J.; Fečkan, M.F. Applying fractional calculus to analyze economic growth modelling. J. Appl.
Math. Stat. Inform. 2018, 14, 25–36. [CrossRef]
15. Fulger, D.; Scalas, E.; Germano, G. Monte Carlo simulation of uncoupled continuous-time random walks
yielding a stochastic solution of the space-time fractional diffusion equation. Phys. Rev. E 2008, 77, 021122.
[CrossRef] [PubMed]
16. Gerardo-Giorda, L.; Germano, G.; Scalas, E. Large-scale simulations of synthetic markets. Commun. Appl.
Ind. Math. 2014, 6, 535.

c 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

250
 mathematics
Article
Deep Assessment Methodology Using Fractional
Calculus on Mathematical Modeling and Prediction
of Gross Domestic Product per Capita of Countries
Ertuğrul Karaçuha, Vasil Tabatadze, Kamil Karaçuha *, Nisa Özge Önal and Esra Ergün
Informatics Institute, Istanbul Technical University, 34467 Istanbul, Turkey; [email protected] (E.K.);
[email protected] (V.T.); [email protected] (N.Ö.Ö.); [email protected] (E.E.)
* Correspondence: [email protected]

Received: 19 February 2020; Accepted: 9 April 2020; Published: 20 April 2020 

Abstract: In this study, a new approach for time series modeling and prediction, “deep assessment
methodology,” is proposed and the performance is reported on modeling and prediction for upcoming
years of Gross Domestic Product (GDP) per capita. The proposed methodology expresses a function
with the finite summation of its previous values and derivatives combining fractional calculus and
the Least Square Method to find unknown coefficients. The dataset of GDP per capita used in this
study includes nine countries (Brazil, China, India, Italy, Japan, the UK, the USA, Spain and Turkey)
and the European Union. The modeling performance of the proposed model is compared with the
Polynomial model and the Fractional model and prediction performance is compared to a special
type of neural network, Long Short-Term Memory (LSTM), that used for time series. Results show
that using Deep Assessment Methodology yields promising modeling and prediction results for GDP
per capita. The proposed method is outperforming Polynomial model and Fractional model by 1.538%
and by 1.899% average error rates, respectively. We also show that Deep Assessment Method (DAM)
is superior to plain LSTM on prediction for upcoming GDP per capita values by 1.21% average error.

Keywords: deep assessment; fractional calculus; least squares; modeling; GDP per capita;
prediction; LSTM

1. Introduction
In the last quarter of the century, the data exchange with not only person to person but also, machine
to machine has increased tremendously. Developments in technology and informatics in parallel
with the development of data science lead the companies, institutions, universities and especially,
the countries to give priority to evaluating produced data and predicting what can be forthcoming.
The modeling of all technical, economic, social events and data has been the interest of scientists
for many years [1–4]. Many authors have been investigating the modeling and predicting events,
options, choices and data. Especially, there is a huge research interest in finding any relation between
telecommunication, economic growth and financial development [5–12]. One of the approaches to
model a physical phenomenon or a mathematical study is to model the dependent variable satisfying
differential equation with respect to the independent variable. However, the differential equations with
an integer-order proposed for mathematical economics or data modeling cannot describe processes
with memory and non-locality because the integer-order derivatives have the property of the locality.
On the other hand, the fractional-order differential equation is a branch of mathematics that focuses on
fractional-order differential and integral operators and can be used to address the limitations of integer
order differential models. Using the fractional calculus or converting the integer-order differential
equation into the non-integer order differential equations lead to a very essential advantage which is

Mathematics 2020, 8, 633; doi:10.3390/math8040633 251 www.mdpi.com/journal/mathematics

Mathematics 2020, 8, 633

memory property of the fractional-order derivative. This is very crucial for models related to economics
which in general, deal with the past and the effect of the past and now on future [12,13]. The memory
capability of the fractional differential approach is the foundation of our motivation.
Fractional calculus (FC) as a question to Wilhelm Leibniz (1646–1716) first arose in 1695 from
French Mathematician Marquis de L’Hopital (1661–1704) [11]. The main question of interest was what
if the order of derivative were a real number instead of an integer. After that, the FC idea has been
developed by many mathematicians and researchers throughout the eighteenth and nineteenth centuries.
Now, there exist several definitions of the fractional-order derivative, including Grünwald-Letnikov,
Riemann-Liouville, Weyl, Riesz and the Caputo representation. The fractional approach is used in
many studies because the fractional derivative represents the intermediate states between two known
states. For example, zero order-derivative of the function means the function itself while the first-order
derivative represents the first derivative of the function. Between these known states, there are infinite
intermediate states [11]. The use of semi-derivatives and integrals in the mass and heat transfer become
an important instant in the field of fractional calculus due to employing the mathematical definitions
into physical phenomena [12,13]. In the last decade, using fractional operators which explain the
events, situations or modes between two different stages or the phenomena with memory provide
more accurate models in many branches of science and engineering including chemistry, biology,
biomedical devices, nanotechnology, diffusion, diffraction and economics [12–31]. In References [25–31],
the modeling and comparison of the countries and trends in the sense of economics and its parameters
are implemented. In References [25,26], economic processes with memory are discussed and modeling
is obtained by using the fractional calculus. The studies with similar purposes as we aim such as
modeling or prediction exist. In these studies, the fractional calculus is employed to model the
given dataset and to predict for the forthcoming. In Reference [28], the orthogonal distance fitting
method is used. The study is trying to minimize the sum of the orthogonal distance of data points
in order to obtain an optimized continuous curve representing the data points. In Reference [32],
the one-parameter fractional linear prediction is studied using the memory of two, three or four
samples, without increasing the number of predictor coefficients defined in the study. In Reference [33],
the generalized formulation of the optimal low-order linear prediction by using the fractional calculus
approach is developed with restricted memory. All these studies focus on modeling or prediction for a
phenomenon with fractional calculus. Also, in our previous studies, methods based on FC that works
for modeling were introduced. In these studies, the children’s physical growth, subscriber’s numbers
of operators, GDP per capita were modeled and compared with other modeling approaches such as
Fractional Model-1 and Polynomial Models [34–36]. According to the results, proposed fractional
models had better results compared to the results obtained from Linear and Polynomial Models [34–36].
Our previous works do not take into account the previous values of the dataset for any time instant.
Their purpose is to model the dataset with minimum error and faster way compared with classical
methods such as Polynomial and Linear Regression.
In this study, we extend our prior works by predicting the next incoming values as well as
modeling the data itself. We introduce a new mathematical model, namely “Deep Assessment,”
based on the fractional differential equation for modeling and prediction by using the properties of
fractional calculus. Different to the literature and our early studies mentioned above, this model can
be used for prediction as well as modeling. The proposed approach is built on the fractional-order
differential equation and corresponding Laplace transform properties are utilized. Here, the modeling
is implemented with mathematical tools similar to those developed in the previous study [4] with
a different approach in which the finite numbers of previous values and the derivatives are taken
into account. Then, the prediction is obtained by assuming a value in a specific time can be expressed
as the summation of the previous values weighted by unknown coefficients and the function to be
modeled is continuous and differentiable. In this way, the proposed method takes previous values and
variation rates between different time samples (derivative) of the dataset into account while modeling

252
Mathematics 2020, 8, 633

the data itself and predicting upcoming values. Combining the previous values with the variations
weighted by the unknown coefficients lead to calling the method “deep assessment.”
In this study, we assessed the proposed method by the modeling, testing and predicting GDP
per capita of the following countries and the European Union: Brazil, China, European Union, India,
Italy, Japan, the UK, the USA, Spain and Turkey. GDP per capita is a measure of a country’s total
economic output divided by the number of the population of the country. In general, it is a reasonable
and good measurement of a country’s living quality and standard [37]. Therefore, the modeling of
GDP per capita is crucial and predicting GDP per capita is very essential not only for researchers but
also for companies, investors, manufacturers and institutions. To assess the performance of Deep
Assessment in modeling, we compare the proposed model with Polynomial Regression and Fractional
Model-1 [34]. Besides, in the same way, for the prediction, we compared the model with Long-Short
Term Memory (LSTM), a special type of neural networks used in time series problems.
The structure of the study is the following. Section 2 explains the formulation of the problem.
After that, Section 3, namely Our Approach, is devoted to explaining how to obtain modeling, simulation,
testing and prediction. Then, in Section 4, the results are presented. Lastly, Section 5 highlights the
conclusion of the study.

2. Formulation of the Problem

In this section, the mathematical foundation of the proposed method is given. Before going
into the mathematical manipulations, it is better to explain the approach and the main steps for the
formulation. The study aims to model and then, to predict GDP per capita data at any time t by using
the previous GDP per capita values of the countries. Here, we assume that countries’ historical data
and the change of these data over time create an eco-genetics for the forthcoming. In other words,
mathematically, GDP per capita at a time t is assumed to be the summation of both its previous
values and the changes in time with unknown constant coefficients. In the second stage, we express
a function for the GDP per capita as a series expansion by using Taylor expansion of a continuous
and bounded function. Then, the differential equation obtained from this series expansion is defined.
After that, the unknown constant coefficients are found by the least-squares method. The method aims
to minimize the error between the proposed GDP per capita function and the dataset.
First, it is a reasonable idea to approximate a function g(x) as the finite summation of the
previous values of the same function weighted with unknown coefficients αk and the summation of the
derivatives of the previous values of the same function weighted with unknown coefficients βk because,
intuitively, the recent value of data, in general, is related to and correlated with its previous values
and the change rates. The purpose is to find the upcoming values of any dataset with a minimum
error by employing the previously inherited features of the dataset. As a starting point, an arbitrary
function is assumed to be approximately the finite summation of the previous values and the change
rates weighted with some constant coefficients. To use the heritability of fractional calculus, this
presupposition for modeling of the function itself and predicting future values is done [6,28,34].
l l

g(x)  αk g(x − k) + β g (x − k ). (1)
k =1 k =1 k

Here, g is the first derivative of g(x − k) with respect to x. After assuming Equation (1), the function
g(x) can be expanded as the summation of polynomials with unknown constant coefficients, an as
given in Equation (2). Here, g(x) is assumed to be a continuous and differentiable function.
∞
g(x) = an xn . (2)
n=0

253
Mathematics 2020, 8, 633

Then, g(x − k) becomes as Equation (3)


∞
g(x − k ) = an (x − k)n (3)
n=0

The final form of g(x) is given as Equation (4).

l ∞ l ∞
g(x)  αk an (x − k)n + β an n(x − k)n−1 . (4)
k =1 n=0 k =1 k n=0

After combining αk an as akn , βk an as bkn and approximating Equation (4), Equation (5) is obtained.
Here, truncation of ∞ to M is performed. After truncation, the first derivative of g(x) is taken and
given in Equation (6).


l 
M 
l 
M
g(x)  akn (x − k)n + bkn n(x − k)n−1 (5)
k =1 n=0 k =1 n=0

dg(x) l M l  M
 akn n(x − k)n−1 + b n(n − 1)(x − k)n−2 . (6)
dx k =1 n=1 k =1 n=1 kn

The expression given in Equation (7) is the definition of Caputo’s fractional derivative [11].
Throughout the study, Caputo’s description of the fractional derivative is employed.

γ dγ g ( x ) 1 x
g(n) (k)dk
Dx g(x) = = , (n − 1 < γ < n). (7)
dxγ Γ (n − γ) 0 (x − k)γ−n+1

In Equation (7), Γ(1 − γ) is the Gamma function, the fractional derivative is taken with respect to x
in the order of γ and g(n) corresponds to the nth derivative again, with respect to x. In our study, n = 1
is assumed and the fractional-order spans between 0 and 1. Here, two expansions are done to express
g(x), approximately. The first one is to express the function as the finite summation of the previous
values of the function. Second, expressing the function g(x) as the summation of polynomials known
as Taylor Expansion assuming that g(x) is a continuous and differentiable function.
Finally, the mathematical background is enough to go further in the proposed methodology.
As a summary, above, we mentioned three important tools. First, a function is expressed as the
summation of its previous samples. Second, Taylor expansion for a continuous and differentiable
function is defined. After that, the Caputo definition of the fractional derivative is given. Now, it is time
to express Deep Assessment Methodology by using fractional calculus for the modeling and prediction.
Apart from above, there is an assumption that the fractional derivative f (x) in the order of γ is equal to
Equation (8). After this assumption, it is required to find unknown f (x) which satisfies the fractional
differential equation below and models the discrete dataset.

dγ f (x) l ∞ l ∞
γ  akn n(x − k)n−1 + b n(n − 1)(x − k)n−2 , (8)
dx k = 1 n = 1 k =1 n=1 kn

where, f (x) stands for the GDP per capita of the countries and x corresponds to the time.
Note that, in (6), allowing the order of the derivation in the left-hand side of Equation (6) to be
non-integer gives a more general model [28]. This generalization is employed in Deep Assessment
Methodology for f (x) which stands for the GDP per capita.
Here, the motivation is to find akn and bkn given in Equation (8). To find the unknowns, the
differential equation needs to be solved. The strategy is as follows—first, it is required to take the
Laplace transform which leads to having an algebraic equation instead of a differential equation.
In other words, the Laplace transform is taken for Equation (8) to reduce the differential equation

254
Mathematics 2020, 8, 633

to algebraic equation, then, by using inverse Laplace transform properties, the final form of f (x) is
obtained as Equation (9) [11].
l ∞ l ∞
f (x, γ)  f (0) + k =1 n=1 akn Ckn (x, γ) + k =1 n=1 bkn Dkn (x, γ),
where,
Γ (n+1)
Ckn (x, γ)  Γ (n+γ)
(x − k)n+γ−1 (9)
Γ (n+1)
Dkn (x, γ)  Γ(n+γ−1)
(x − k)n+γ−2 .

To obtain the numerical calculation, the infinite summation of polynomials is approximated as a

finite summation given in Equation (10).
l M l M
f (x, γ)  f (0) + akn Ckn (x, γ) + b D (x, γ). (10)
k =1 n=1 k =1 n=1 kn kn

Here, f (0), akn and bkn are unknown coefficients that need to be determined. Note that, below,
properties of the Laplace transform (L) are given to find Equations (9) and (10) [11].
 γ 
 Γ (n) d f (x)
L (x − k)n−1 = n e−ks and L γ = sγ F(s) − sγ−1 f (0) for 0 < γ < 1.
s dx

where, L stands for the Laplace transform and L[ f (x)] = F(s).

For the numerical calculation, the infinite summation is converted into a finite summation, as
given in Equation (10).

3. Our Approach

3.1. Modeling with Deep Assessment

In this part, the methodology for the modeling of the problem is given in detail. To predict the
upcoming years, the problem has four regions as given in Figure 1. Dataset spans in Region 1, 2 and 3.
Note that, there is no data for Region 4 where the prediction is aimed. Region 1 is called “before
modeling region” which consists of historical data. Each of the coefficients (x − k)n+γ−1 and derivative
coming from previous values of GDP per capita for different values of k and multiplication by different
weights as given in Equation (10) will add the contribution to the recent data. For modeling, the
historical data is employed directly for the modeling of the data located in Region 2. Region 2 and 3 are
named as modeling and testing, respectively. In the modeling region, the GDP per capita is tried to be
modeled and the unknown coefficients are found. Note that, the approach uses the previous l values
(Pi−1 , Pi−2 , . . . , Pi−l and corresponding f (i − 1), f (i − 2), . . . f (i − l)) for arbitrary Pi located in Region 2.
The third region consists of the data used to test for upcoming predictions. Finally, Region 4 is called
the “prediction region” where the aim is to find the GDP per capita values for the time that the actual
values have not known yet and implement prediction. The region division is required because there
are parameters given in the previous section (Equation (10)) such as M, l, γ which need to be found
before the prediction. In Region 2, the modeling is done to find the optimum values of coefficients
akn , M, l, γ in Equation (10) for modeling. To model the data, Least Squares Method is employed,
which is explained later in this section. After that, one of the purposes of the study is achieved. This is
the modeling of the data using the fractional approach. Then, the second purpose comes which is to
predict the values of GDP per capita for the upcoming unknown years. In order to find optimum
M, l, γ values for the prediction, Region 3, namely testing is needed. In the region, there is an iterative
solution where the real discrete data is again known. For instance, in Region 3, it is required to find
f (m1 + 1). Then, by using the proposed method employing the fractional calculus and Least Squares
Method, f (m1 + 1) is obtained with a minimum error by optimizing M, l, γ values for f (m1 + 1) itself.
Then, f (m1 + 1) is included the dataset for the next test which is done for f (m1 + 2). This continues
up to f (m). Then, with optimized M, l, γ, the predicted f (mx ) is found in Region 4.

255
Mathematics 2020, 8, 633

To model the known data, f (x) representing the data optimally should be obtained. In other
words, the unknowns akn , bkn and f (0) in Equation (10) or Equation (11) should be determined. For this,
the Least Squares Method is employed.
l M l M
f (i, γ) = f (0) + a C (i, γ) + b D (i, γ). (11)
k =1 n=1 kn kn k =1 n=1 kn kn

In Equation (12), the squares of total error T 2 is given. The main purpose of the modeling region
is to minimize T 2 by a gradient-based approach which requires minimization of the square of the total
error as the following.

m1
T 2 = (Pi − f (i, γ))2 (12)
i=l

∂T 2
= 0,
∂ f (0)
∂T 2
= 0,
∂art
and
∂T 2
= 0.
∂brt
where, r = 1, 2, 3, . . . l and t = 1, 2, 3, . . . M.
∂T 2 ∂T 2
It is better to give an example of how to obtain ∂art
= 0 and ∂brt
= 0.

∂ 
m1
∂T 2
=0→ (Pi − f (i, γ))2 = 0
∂art ∂art
i=l

Then,

m1
2 [Pi − f (i, γ)]Crt (i, γ) = 0
i=l
⎧
m1 ⎪
⎫

m1 
m1  ⎪ l 
M ⎪
⎪
⎨ ⎬
Crt (i, γ)Pi = f (0) Crt (i, γ) + ⎪
⎪ a C ( i, γ ) ⎪
⎪ C (i, γ)
⎩ kn kn
⎭ rt
i=l i=l i=l k =1 n=1

∂T 2
The same procedure is followed for ∂brt
= 0.

∂ 
m1
∂T 2
=0→ (Pi − f (i, γ))2 = 0
∂brt ∂brt
i=l

Then,

m1
[Pi − f (i, γ)]Drt (i, γ) = 0
i=l
⎧
m1 ⎪
⎫

m1 
m1  ⎪ l 
M ⎪
⎪
⎨ ⎬
Drt (i, γ)Pi = f (0) Drt (i, γ) + ⎪
⎪ a D ( i, γ ) ⎪
⎪ D (i, γ)
⎩ kn kn
⎭ rt
i=l i=l i=l k =1 n=1

This leads to having a system of linear algebraic equations (SLAE) as given in (13).

[A].[B] = [C] (13)

where, [A], [B] and [C] is shown in Equations (14), (15) and (16), respectively.

256
Mathematics 2020, 8, 633

To find f (x) the continuous curve modeling with a minimum error, the optimum fractional-order
γ is inquired between (0, 1). Then, with optimum fractional-order γ, the unknown coefficients are
determined. In the study, the GDP per capita of Brazil, China, the European Union, India, Italy, Japan,
the UK, the USA, Spain and Turkey were used from 1960 until 2018 [38]. The dataset is shown in
Tables A1 and A2.
Among them, the year 2018 is in Region 3 as testing to predict for the next years.
Here,
t (years): [1960, 1961, ..., 2018]
i (points): [1, 2, ..., 59]
Pi (value of i): [P1 , P2 , . . . , P59 ]
Pi : It shows the actual GDP per capita of each country in each ith year. For example, P2 is the GDP
per capita of the country in 1961.
i: It stands for the number for each year. For example, i = 1 for 1960, i = 3 for 1962 and i = 59
for 2018.  
A1,1 A1,2
A= (14)
A2,1 A2,2
A matrix consists of the matrix set below, where,

Ckn (x, γ) = Ckn and Dkn (x, γ) = Dkn

⎡ m 1 m 1 1
m 1
m 1
m 1
m ⎤
⎢⎢ m1 − l + 1 C11 ... C1M C21 ... C2M ... Cl1 ... ClM ⎥⎥
⎢⎢⎢ ⎥⎥
⎥⎥
⎢⎢ i=l i=l i=l i=l i=l i=l
⎥⎥
⎢⎢⎢ 1
m 1
m 1
m 1
m 1
m 1
m
⎥⎥
⎢⎢ C11 C11 C11 . . . C1M C11 C21 C11 ... C2M C11 ... Cl1 C11 ... ClM C11 ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ i=l i=l i=l i=l i=l i=l ⎥⎥
⎢⎢
m
m
m 1
m 1
m 1
m ⎥⎥
A1,1 = ⎢⎢⎢⎢ ⎥⎥
1 1 1
C12 C11 C12 . . . C1M C12 C21 C12 ... C2M C12 ... C11 C12 ... ClM C12 ⎥⎥
⎢⎢ i=l i=l i=l i=l i=l i=l ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ . . . . .. .. .. .. .. .. .. ⎥⎥
⎢⎢ . . .. .. ⎥⎥
⎢⎢ . . . . . . . . . ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ 1
m 1
m 1
m 1
m 1
m 1
m ⎥⎥
⎢⎣ Clm C11 ClM . . . C1M ClM C21 ClM ... C2M ClM ... Cl1 ClM ... ClM ClM ⎥⎦
⎡ i=l i=l i=l i=l i=l i=l ⎤
⎢⎢ D 1 1 1 1 1 1
m m m m m m
⎢ C11 D11 . . . C1M D11 C21 D11 ... C2M D11 ... Cl1 D11 ... ClM D11 ⎥⎥⎥⎥
⎢⎢⎢ 11
i=l i=l i=l i=l i=l i=l ⎥⎥
⎢⎢ m1 ⎥⎥
⎢⎢ 1
m m 1 1
m 1
m 1
m 1
m ⎥
⎢⎢ D12 C11 D12 . . . C1M D12 C21 D12 ... C2M D12 ... Cl1 D12 ... ClM D12 ⎥⎥⎥⎥
⎢⎢⎢ ⎥⎥
A2,1 = ⎢⎢⎢ i=l . i=l i=l i=l i=l i=l i=l ⎥⎥
⎥⎥
⎢⎢ . . . .. .. .. .. . . . ⎥⎥
⎢⎢ .. .. .. .. .. .. .. ⎥⎥
⎢⎢ . . . . ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ m
⎢⎢ 1 1
m 1
m m 1 1
m 1
m .
..
m 1
⎥⎥
⎣ DlM C11 DlM . . . C1M DlM C21 DlM ... C2M D1M ... Cl1 DlM ClM DlM ⎥⎦
i=l i=l i=l i=l i=l i=l i=l
⎡ m 1 1
m 1
m ⎤
⎢⎢ D11 ... D1M ... DlM ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ i=l i=l i=l ⎥⎥
⎢⎢ m ⎥⎥
⎢⎢ 1 1
m 1
m
⎥⎥
⎢⎢ ⎥⎥
⎢⎢ i=l D11 C11 . . . i=l D1M C11 ... DlM C11 ⎥⎥
⎢ i=l ⎥⎥
⎢⎢⎢ m 1 1
m 1
m ⎥⎥
A1,2 = ⎢⎢⎢⎢ D C ... D1M C12 ... DlM C12 ⎥⎥
⎥⎥
⎢⎢ i=l 11 12 i=l i=l ⎥⎥
⎢⎢ ⎥⎥
⎢⎢⎢ .. .. .. .. .. ⎥⎥
⎥⎥
⎢⎢ . . . . . ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ m ⎥⎥
⎢⎢⎣ 1 D C 1 1
m m
... D1M ClM ... DlM ClM ⎥⎦
11 lM
=l i=l i=l
⎡ im ⎤
⎢⎢ D D 1 1
1 m m
··· ··· ⎥⎥
⎢⎢ 11 11 D1M D11 DlM D11 ⎥⎥
⎢⎢ i=l i=l i=l ⎥⎥
⎢⎢ m ⎥⎥
⎢⎢ 1 1
m 1
m ⎥⎥
⎢⎢ D11 D12 · · · D1M D12 ··· DlM D12 ⎥⎥
⎢⎢ ⎥⎥
A2,2 = ⎢⎢⎢ i=l i=l i=l ⎥⎥
⎥⎥
⎢⎢ . . . . . ⎥⎥
⎢⎢ .. .. .. .. .. ⎥⎥
⎢⎢ ⎥⎥
⎢⎢ m ⎥⎥
⎢⎢ 1 1
m 1
m
⎥⎥
⎢⎣ D11 DlM · · · D1M DlM ··· DlM DlM ⎦
i=l i=l i=l

T
[B] = f (0) a11 a12 ... a1M a21 a22 ... a2M . . . al1 . . . alM b11 b12 ··· b1M b21 ... b2M ... bl1 bl2 ... blM (15)

257
Mathematics 2020, 8, 633

 T
1
m 1
m 1
m 1
m 1
m 1
m 1
m
[C] = Pi Pi C11 Pi C12 ... Pi ClM Pi D11 Pi D12 ... Pi DlM (16)
i=l i=l i=l i=l i=l i=l i=l

Figure 1. The regions of the dataset.

3.2. Prediction with Deep Assessment

To find the optimized values of the unknowns for the prediction, the testing region (3rd region) is
required. The predictions obtained in the test region (m1 < i < m) are also given in Table 1. For testing,
the data up to m1 = 58 have been taken into consideration in the operations. The f (m1 + 1) value was
found from the obtained modeling. Then, the value is kept, and the next step was started again for
( f (m1 + 2)). These operations are done until the last value of the test zone. In our case, m = 59.

Table 1. Comparison of modeling results (γ, M and MAPE values) of countries for l = 10.

γ Polynomial Fractional
γ Deep Assessment *
Country Deep Model * Model-1 * M
Fractional Model-1 (l<i<m)
Assessment (l<i<m) (l<i<m)
US 0.44 0.54 0.81% 1.01% 1.06% 15
UK 0.14 0.85 5.38% 7.03% 6.61% 15
Brazil 0.06 0.58 7.26% 7.13% 9.00% 17
China 0.03 0.95 2.84% 5.62% 5.67% 11
India 0.15 0.02 3.09% 2.51% 4.10% 16
Japan 0.26 0.69 4.45% 4.64% 5.82% 20
EU 0.06 0.89 4.02% 3.41% 5.71% 20
Italy 0.39 1 4.70% 8.81% 8.81% 9
Spain 0.22 0.58 4.44% 6.49% 6.36% 13
Turkey 0.71 0.01 6.09% 11.81% 8.93% 10
* MAPEModeling values.

The last region is called the “Prediction Region.” Here, using Region 1, 2 and 3, the prediction
for the upcoming years is obtained. After having modeled and tested regions, the unknowns in
Equation (11) have already found in an optimal manner. After testing, Region 4 is started. In the
region, the first prediction f (m + 1) is found by using the coefficients and unknowns found by the

258
Mathematics 2020, 8, 633

testing region. After that, the first predicted value ( f (m + 1)) is included in Region 3 (testing) for the
consecutive prediction f (m + 2). This procedure is reiterated and recycled up to f (mx ).
The prediction results for 2019 are given in Table 2. For example, as of the end of 2019 ( f (m + 2)),
Brazil, China, European Union, India, Italy, Japan, the UK, the USA, Spain and Turkey’s GDP per
capita values are expected as listed.

Table 2. Test (m1 < i < m) results (γ, l, M and MAPE) of GDP per capita for corresponding countries.

γ Deep
Country γ l M Deep Assessment *
Interpolation Learning *
Brazil 0.18 24 3 0.32 0.1303% 0.4728%
China 0.97 11 3 0.5 0.7147% 1.6365%
0.3379%
India 0.96 3 2 0.99 0.7203%
5
Italy 0.43 20 4 0.43 0.1048% 3.0796%
Japan 0.57 4 3 1 0.3499% 1.1091%
Spain 0.99 2 3 0.99 0.0560% 1.5683%
Turkey 0.39 17 4 0.39 0.1167% 2.3691%
EU 0.32 20 5 0.22 0.1044% 0.2522%
US 0.39 25 2 0.18 0.1081% 0.8424%
UK 0.18 18 7 0.05 0.9129% 3.0508%
* MAPEPrediction values.

In Figure 2, the algorithm for prediction with DAM is illustrated. The first step of the algorithm
is to initialize the parameters (l, M, x1 , x2 , . . . xm and P1 , P2 , . . . Pm ). Then, the counter variable N
is introduced, which counts the number of prediction steps. The total number of required predicted
steps is denoted as n0 . As an initial value, the fractional-order γ is assigned 0 and the increment is
0.01 for each loop to find the optimized value. For each value of γ between 0 and 1, matrix A given
as Equation (14) is created and then, the unknown coefficients given in Equation (10) are calculated.
After that, using the actual data in Region 1 and Region 2, the modeling of data between Pl and Pm is
actualized for Region 2. Then, the error defined in Equation (12) is calculated. The value of the error
is analyzed and compared to previously obtained values. If it is smaller than the previous one, the
corresponding fractional-order value is memorized. At the end of Loop II, the optimal value of the
fractional-order, which coincides with the optimal modeling is found and corresponding coefficients
given in Equation (10) is determined. Then, the prediction for the next forthcoming value is made
with Equation (10). After that, all the procedures starting from the increment of N is repeated so that
the previously predicted value is added to the initial data for the next step prediction. This process is
repeated up to the termination of Loop I. Finally, n0 the number of predictions is obtained. Keep in
mind that, for the parameters l and M, there exist two loops starting from 1 to L0 and 1 to M0 searching
the optimum values of the parameters in order to get the outcomes with a minimum error for the
testing region, respectively. Here, L0 and M0 are pre-defined some constant values.

259
Mathematics 2020, 8, 633

Figure 2. The algorithm for the prediction.

3.3. Long Short-Term Memory

In our study, we compare the modeling with the polynomial curve fitting method and in the
prediction, we compare Deep Assessment with the LSTM method. Conventional neural networks are
insufficient for modeling the content of temporal data. Recursive neural networks (RNN) model the
sequential structure of data by feeding itself with the output of the previous time step. LSTMs are
special types of RNNs that operate over sequences and are used in time series analysis [39]. An LSTM
cell has four gates: input, forget, output and gate. With these gates, LSTMs optionally inherit the
information from the previous time steps. Forget gate ( f ), input gate (i) and output gate (o) are
sigmoid functions (σ) and they take values between 0 and 1. Gate g has hyperbolic tangent (tanh)
activation and is between −1 and 1. The Gate and forward propagation equations are listed below
as Equations (17)–(22). Here clt and hlt refer to cell state and hidden state of layer l at time step t,
respectively. Each gate takes input from the previous time step (hlt−1 ) and previous layer (htl−1 ) and has
its own set of learnable parameters W’s and b’s.
 
ft = σ W f [hlt−1 , hl−1
t ] + bf (17)
 
it = σ Wi [hlt−1 , hl−1
t ] + bi (18)
 
ot = σ Wo [hlt−1 , hl−1
t ] + bo (19)
 
gt = tanh W g [hlt−1 , hl−1
t ] + bg (20)

clt = f
clt−1 + i
g (21)
 
hlt = o
tanh clt (22)

Here,
is the Hadamard product. Each LSTM neuron in a network may consist of one or more cells.
In every time step, every cell updates its own cell state, clt . Equation (22) describes how these cells get
updated with forget gate and input gate; f gate decides how much of previous cell state that cell should
remember while i gate decides how much it should consider the new input from the previous layer.
Then, LSTM neuron updates its internal hidden state by multiplying output and squashed version of

260
Mathematics 2020, 8, 633

clt . An LSTM neuron gives outputs only in hidden state information to another LSTM neuron. Gate o
and ct are used internally in the computation of forward time steps [40]. To forecast time series and
compare our proposed approach to neural networks, we employed a stacked LSTM model with 2
layers of LSTMs (each having 50 hidden units) and a linear prediction layer. LSTM model is trained
with the Adam optimizer [40].

4. Numerical Results
In this section, we report the modeling and prediction performance of the Deep Assessment
methodology. Further, we compare the proposed method to other modeling and prediction approaches
such as Polynomial Model, Fractional Model-1 [34,35] and LSTM. In this section, results are reported
with the Mean Average Precision Error (MAPE) metric and calculated as follows:
 ∼ 
1 k  v(i) − v(i) 
MAPE =   × 100, (23)
k i=1  v(i)
∼
where k is the total number of samples, v(i) is the actual value and v(i) is the predicted value for
ith sample.
Before presenting the results, it is important to highlight that for modeling, M0 and l0 are taken 20
and 10, respectively whereas for prediction, M0 and l0 are taken 8 and 25, respectively. The number of
prediction, n0 is equal to 1.

4.1. Modeling Results

In this part, we compare the modeling performance with Polynomial, Fractional Model-1 and
Deep Assessment models.
To achieve modeling, l value needs to be investigated. For the modeling of the GDP per capita of
each country, the required previous data l of past years used in the algorithm differs after optimization.
In order to make a fair evaluation, l value is fixed among all countries to 10. Modeling results for
Deep Assessment, Polynomial Model and Fractional Model-1 are shown in Table 1. Optimized M
values after processing can be seen in the last column. The Deep Assessment model has a %4.308
average MAPE and outperforms Polynomial and Fractional Model-1 by %1.538 and %1.899 average
error rates. All three methods model the US best with %0.81, %1.01 and %1.06 error. Further, in the
case of Italy, Fractional Model-1 uses the fractional-order value of 1 and produces %8.81 MAPE, equal
to the Polynomial method as expected because for the fractional-order value of 1 is the same with the
Polynomial Method. However, DAM yields fractional order of 0.39, decreasing the error to 4.70%,
justifying the advantage of employing fractional calculus and previous values of the data itself.
 
1 m  P(i) − f (i, γ) 
MAPEModeling =   × 100. (24)
m−l+1 i=l  P(i) 

GDP per capita data, Deep Assessment, Polynomial and Fractional Model-1 modeling results
are shown in Figure 1 for each country. One can conclude that when data points have high variance
all models produce high error rates, as in Turkey and Italy. For Japan and Brazil, DAM (Deep
Assessment Method) and Polynomial models produce similar results. Also, it can be seen from the
Figure 3, both Deep Assessment and Fractional Model-1 have a low bias when compared to the
Polynomial model and overfits to dataset less. This is possible because of the memory property of the
proposed approach. Except for Brazil, India and the EU, the proposed method yields superior results
compared to other models.

261
Mathematics 2020, 8, 633

(a): Modelling results of Brazil). (b): Modelling results of China.


(c): Modelling results of India.
G0RGHOOLQJUHVXOWVRI,WDO\


(e): Modelling results of Japan . (f): Modelling results of Spain.

Figure 3. Cont.

262
Mathematics 2020, 8, 633


(g): Modelling results of Turkey .
K0RGHOOLQJUHVXOWVRI(8


(i): Modelling results of US. (j): Modelling results of UK.

Figure 3. Modelling results of the countries (Brazil, China, India, Italy, Japan, the UK, the USA,
Spain and Turkey) and the European Union or Deep Assessment (Blue), Fractional Model 1 (Yellow),
Polynomial Model (Purple).

4.2. Prediction Results

In this section, we compare the accuracy rate of the prediction of Deep Assessment and Deep
Learning models. As in modeling, the GDP per capita dataset is used to assess the performance of the
proposed method. Table 2 illustrates optimized γ, l, M values and the corresponding performance of
DAM and LSTM. Here, column 6 reports the performance of DAM while column 7 represents LSTM.
Column 5 shows that the Deep Assessment methodology predicts GDP per capita with an average
0.29% error with predicting all countries with 1.< (less than 1 percent) of error. The best-predicted
country is Spain while UK’s prediction is the least accurate with 0.91% error. On the other hand,
LSTM yields 1.51% error on average. For both DAM and LSTM, UK yields the highest error. Table 2

263
Mathematics 2020, 8, 633

demonstrates that in the implemented setting, DAM outperforms LSTM by 1.21% average error and
produces fair results.
 
1 m  P(i) − f (i, γ) 
MAPEPrediction = 
  × 100. (25)
m1 − l + 1 i=m1  P(i) 

Table 3 reports the prediction of GDP per capita for the year 2019 is illustrated in Table 2 for both
DAM and LSTM methods. For countries Brazil, China, India, Turkey, the UK and the US, predictions
obtained by the two models are similar. On the other hand, Italy and Spain yield different results.

Table 3. GDP per Capita Prediction of Countries for 2019 (US dollars).

Country Deep Assessment Deep Learning

Brazil 7932 8013
China 10,312 10,273
India 2154 1967
Italy 39,028 35,141
Japan 34,421 37,994
Spain 30,385 35,372
Turkey 8260 8920
US 65,767 63,844
UK 44,897 44,702
EU 40,487 36,487

5. Conclusions
In this study, a model called “Deep Assessment” is introduced which employs Fractional Calculus
to model discrete data as the summation of previous values and derivatives. Different to the literature
and our previous work, the proposed approach also predicts the incoming values of the discrete data
in addition to modeling. The method is evaluated on modeling and predicting GDP per capita, using a
dataset including the period of 1960–2018 for nine countries (Brazil, China, European Union, India,
Italy, Japan, UK, the USA, Spain and Turkey) and the European Union. Using the fractional differential
equation and the summation of previous values for the modeling of GDP per capita at a specific time
instant bring non-locality, memory and generalization of the problem for different fractional order.
In experiments, first, GDP per capita is modeled. The Deep Assessment model has a 4.308% average
MAPE and outperforms Polynomial and Fractional Model-1 by 1.538% and 1.899% average error rates
for modeling. For prediction, LSTM, a special type of neural network is used to assess the performance
of the model. In the selected test region, it is shown that Deep Assessment is superior to LSTM by 1.51%
average error. Results illustrate that the proposed method yields promising results and demonstrates
the benefits of combining fractional calculus and differential equations. Evaluation of multivariable
and multifunctional problems, analyzing time windows, randomness, noise and error changes are left
to future work.

Author Contributions: The contribution of each author is listed as follows. E.K. has contributed to supervision,
conceptualization, investigation, methodology, and administration. V.T. plays an important role in resources,
supervision, and validation. K.K. supported conceptualization, writing, and editing. N.Ö.Ö. was the key person
about visualization, investigation, administration, validation, and writing. E.E. has contributed to validation,
visualization, writing, and editing. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Istanbul Technical University (ITU) Vodafone Future Lab with the grant
number ITUVF20180901P11.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses or interpretation of data; in the writing of the manuscript or in the decision to
publish the results.

264
Mathematics 2020, 8, 633

Appendix A

Table A1. GDP per capita (US dollars) values of the countries.

i Years Brazil China EU India Italy

1 1960 210.1099 89.52054 890.4056 82.1886 804.4926
2 1961 205.0408 75.80584 959.71 85.3543 887.3367
3 1962 260.4257 70.90941 1037.326 89.88176 990.2602
4 1963 292.2521 74.31364 1135.194 101.1264 1126.019
5 1964 261.6666 85.49856 1245.499 115.5375 1222.545
6 1965 261.3544 98.48678 1346.058 119.3189 1304.454
7 1966 315.7972 104.3246 1448.551 89.99731 1402.442
8 1967 347.4931 96.58953 1546.804 96.33914 1533.693
9 1968 374.7868 91.47272 1602.06 99.87596 1651.939
10 1969 403.8843 100.1299 1762.472 107.6223 1813.388
11 1970 445.0231 113.163 1950.732 112.4345 2106.864
12 1971 504.7495 118.6546 2195.145 118.6032 2305.61
13 1972 586.2144 131.8836 2611.729 122.9819 2671.137
14 1973 775.2733 157.0904 3296.935 143.7787 3205.252
15 1974 1004.105 160.1401 3685.596 163.4781 3621.146
16 1975 1153.831 178.3418 4274.046 158.0362 4106.994
17 1976 1390.625 165.4055 4406.238 161.0921 4033.099
18 1977 1567.006 185.4228 4968.988 186.2135 4603.6
19 1978 1744.257 156.3964 6064.883 205.6934 5610.498
20 1979 1908.488 183.9832 7377.165 224.001 6990.286
21 1980 1947.276 194.8047 8384.718 266.5778 8456.919
22 1981 2132.883 197.0715 7391.077 270.4706 7622.833
23 1982 2226.767 203.3349 7093.702 274.1113 7556.523
24 1983 1570.54 225.4319 6859.966 291.2381 7832.575
25 1984 1578.926 250.714 6572.019 276.668 7739.715
26 1985 1648.082 294.4588 6775.647 296.4352 7990.687
27 1986 1941.491 281.9281 9265.924 310.4659 11,315.02
28 1987 2087.308 251.812 11,432.23 340.4168 14,234.73
29 1988 2300.377 283.5377 12,711.96 354.1493 15,744.66
30 1989 2908.496 310.8819 12,936.46 346.1129 16,386.66
31 1990 3100.28 317.8847 15,989.22 367.5566 20,825.78
32 1991 3975.39 333.1421 16,496.51 303.0556 21,956.53
33 1992 2596.92 366.4607 17,919.02 316.9539 23,243.47
34 1993 2791.209 377.3898 16,256.42 301.159 18,738.76
35 1994 3500.611 473.4923 17,194.12 346.103 19,337.63
36 1995 4748.216 609.6567 19,898.44 373.7665 20,664.55
37 1996 5166.164 709.4138 20,295.17 399.9501 23,081.6
38 1997 5282.009 781.7442 19,121.21 415.4938 21,829.35
39 1998 5087.152 828.5805 19,763.51 413.2989 22,318.14
40 1999 3478.373 873.2871 19,698.89 441.9988 21,997.62
41 2000 3749.753 959.3725 18,261.97 443.3142 20,087.59
42 2001 3156.799 1053.108 18,457.89 451.573 20,483.22
43 2002 2829.283 1148.508 20,055.33 470.9868 22,270.14
44 2003 3070.91 1288.643 24,310.25 546.7266 27,465.68
45 2004 3637.462 1508.668 27,960.05 627.7742 31,259.72
46 2005 4790.437 1753.418 29,115.63 714.861 32,043.14
47 2006 5886.464 2099.229 30,960.56 806.7533 33,501.66
48 2007 7348.031 2693.97 35,630.94 1028.335 37,822.67
49 2008 8831.023 3468.304 38,185.62 998.5223 40,778.34
50 2009 8597.915 3832.236 34,019.28 1101.961 37,079.76
51 2010 11,286.24 4550.454 33,740.65 1357.564 36,000.52
52 2011 13,245.61 5618.132 36,506.64 1458.104 38,599.06
53 2012 12,370.02 6316.919 34,328.82 1443.88 35,053.53
54 2013 12,300.32 7050.646 35,683.86 1449.606 35,549.97
55 2014 12,112.59 7651.366 36,787.23 1573.881 35,518.42
56 2015 8814.001 8033.388 32,319.45 1605.605 30,230.23
57 2016 8712.887 8078.79 32,425.13 1729.268 30,936.13
58 2017 9880.947 8759.042 33,908 1981.269 32,326.84
59 2018 8920.762 9770.847 36,569.73 2009.979 34,483.2

265
Mathematics 2020, 8, 633

Table A2. GDP per capita (US dollars) values of the countries.

i Years Japan Spain UK US Turkey

1 1960 478.9953 396.3923 1397.595 3007.123 509.4239
2 1961 563.5868 450.0533 1472.386 3066.563 283.8283
3 1962 633.6403 520.2061 1525.776 3243.843 309.4467
4 1963 717.8669 609.4874 1613.457 3374.515 350.6629
5 1964 835.6573 675.2416 1748.288 3573.941 369.5834
6 1965 919.7767 774.7616 1873.568 3827.527 386.3581
7 1966 1058.504 889.6599 1986.747 4146.317 444.5494
8 1967 1228.909 968.3068 2058.782 4336.427 481.6937
9 1968 1450.62 950.5457 1951.759 4695.923 526.2135
10 1969 1669.098 1077.679 2100.668 5032.145 571.6178
11 1970 2037.56 1212.289 2347.544 5234.297 489.9303
12 1971 2272.078 1362.166 2649.802 5609.383 455.1049
13 1972 2967.042 1708.809 3030.433 6094.018 558.421
14 1973 3997.841 2247.553 3426.276 6726.359 686.4899
15 1974 4353.824 2749.925 3665.863 7225.691 927.7991
16 1975 4659.12 3209.837 4299.746 7801.457 1136.375
17 1976 5197.807 3279.313 4138.168 8592.254 1275.956
18 1977 6335.788 3627.591 4681.44 9452.577 1427.372
19 1978 8821.843 4356.439 5976.938 10,564.95 1549.644
20 1979 9105.136 5770.215 7804.762 11,674.19 2079.22
21 1980 9465.38 6208.578 10,032.06 12,574.79 1564.247
22 1981 10,361.32 5371.166 9599.306 13,976.11 1579.074
23 1982 9578.114 5159.709 9146.077 14,433.79 1402.406
24 1983 10,425.41 4478.5 8691.519 15,543.89 1310.256
25 1984 10,984.87 4489.989 8179.194 17,121.23 1246.825
26 1985 11,584.65 4699.656 8652.217 18,236.83 1368.401
27 1986 17,111.85 6513.503 10,611.11 19,071.23 1510.677
28 1987 20,745.25 8239.614 13,118.59 20,038.94 1705.895
29 1988 25,051.85 9703.124 15,987.17 21,417.01 1745.365
30 1989 24,813.3 10,681.97 16,239.28 22,857.15 2021.859
31 1990 25,359.35 13,804.88 19,095.47 23,888.6 2794.35
32 1991 28,925.04 14,811.9 19,900.73 24,342.26 2735.708
33 1992 31,464.55 16,112.19 20,487.17 25,418.99 2842.37
34 1993 35,765.91 13,339.91 18,389.02 26,387.29 3180.188
35 1994 39,268.57 13,415.29 19,709.24 27,694.85 2270.338
36 1995 43,440.37 15,471.96 23,123.18 28,690.88 2897.866
37 1996 38,436.93 16,109.08 24,332.7 29,967.71 3053.947
38 1997 35,021.72 14,730.8 26,734.56 31,459.14 3144.386
39 1998 31,902.77 15,394.35 28,214.27 32,853.68 4496.497
40 1999 36,026.56 15,715.33 28,669.54 34,513.56 4108.123
41 2000 38,532.04 14,713.07 28,149.87 36,334.91 4316.549
42 2001 33,846.47 15,355.7 27,744.51 37,133.24 3119.566
43 2002 32,289.35 17,025.53 30,056.59 38,023.16 3659.94
44 2003 34,808.39 21,463.44 34,419.15 39,496.49 4718.2
45 2004 37,688.72 24,861.28 40,290.31 41,712.8 6040.608
46 2005 37,217.65 26,419.3 42,030.29 44,114.75 7384.252
47 2006 35,433.99 28,365.31 44,599.7 46,298.73 8035.377
48 2007 35,275.23 32,549.97 50,566.83 47,975.97 9711.874
49 2008 39,339.3 35,366.26 47,287 48,382.56 10,854.17
50 2009 40,855.18 32,042.47 38,713.14 47,099.98 9038.52
51 2010 44,507.68 30,502.72 39,435.84 48,466.82 10,672.39
52 2011 48,168 31,636.45 42,038.5 49,883.11 11,335.51
53 2012 48,603.48 28,324.43 42,462.71 51,603.5 11,707.26
54 2013 40,454.45 29,059.55 43,444.56 53,106.91 12,519.39
55 2014 38,109.41 29,461.55 47,417.64 55,032.96 12,095.85
56 2015 34,524.47 25,732.02 44,966.1 56,803.47 10,948.72
57 2016 38,794.33 26,505.62 41,074.17 57,904.2 10,820.63
58 2017 38,331.98 28,100.85 40,361.42 59,927.93 10,513.65
59 2018 39,289.96 30,370.89 42,943.9 62,794.59 9370.176

266
Mathematics 2020, 8, 633

References
1. Beaudry, P.; Portier, F. Stock prices, news, and economic fluctuations. Am. Econ. Rev. 2006, 96, 1293–1307.
[CrossRef]
2. Simpson, D.M.; Rockaway, T.D.; Weigel, T.A.; Coomes, P.A.; Holloman, C.O. Framing a new approach to
critical infrastructure modelling and extreme events. Int. J. Crit. Infrastruct. 2005, 1, 125–143. [CrossRef]
3. Ho, T.Q.; Strong, N.; Walker, M. Modelling analysts’ target price revisions following good and bad news?
Account. Bus. Res. 2018, 48, 37–61. [CrossRef]
4. Karaçuha, E.; Tabatadze, V.; Önal, N.Ö.; Karaçuha, K.; Bodur, D. Deep Assessment Methodology Using
Fractional Calculus on Mathematical Modeling and Prediction of Population of Countries. Authorea 2020.
[CrossRef]
5. Forestier, E.; Grace, J.; Kenny, C. Can information and communication technologies be pro-poor?
Telecommun. Policy 2002, 26, 623–646. [CrossRef]
6. Kretschmer, T. Information and Communication Technologies and Productivity Growth: A Survey of the Literature;
OECD Digital Economy Papers; OECD Publishing:: Paris, France, 2012; Volume 195. [CrossRef]
7. Han, E.J.; Sohn, S.Y. Technological convergence in standards for information and communication technologies.
Technol. Forecast. Soc. Chang. 2016, 106, 1–10. [CrossRef]
8. Grover, V.; Purvis, R.L.; Segars, A.H. Exploring ambidextrous innovation tendencies in the adoption of
telecommunications technologies. IEEE Trans. Eng. Manag. 2007, 54, 268–285. [CrossRef]
9. Torero, M.; Von Braun, J. (Eds.) Information and Communication Technologies for Development and Poverty
Reduction: The Potential of Telecommunications; International Food Policy Research Institute: Washington, DC,
USA, 2006.
10. Pradhan, R.P.; Arvin, M.B.; Norman, N.R. The dynamics of information and communications technologies
infrastructure, economic growth, and financial development: Evidence from Asian countries. Technol. Soc.
2015, 42, 135–149. [CrossRef]
11. Podlubny, I. Fractional differential equations. Math. Sci. Eng. 1999, 198, 79.
12. Machado, J.T.; Kiryakova, V.; Mainardi, F. Recent history of fractional calculus. Commun. Nonlinear Sci.
Numer. Simul. 2019, 16, 1140–1153. [CrossRef]
13. Tarasov, V.E. On history of mathematical economics: Application of fractional calculus. Mathematics 2019, 7,
509. [CrossRef]
14. Baleanu, D.; Güvenç, Z.B.; Machado, J.T. (Eds.) New Trends in Nanotechnology and Fractional Calculus
Applications; Springer: New York, NY, USA, 2010.
15. Gómez, F.; Bernal, J.; Rosales, J.; Cordova, T. Modeling and simulation of equivalent circuits in description of
biological systems-a fractional calculus approach. J. Electr. Bioimpedance 2019, 3, 2–11. [CrossRef]
16. Machado, J.T.; Mata, M.E. Pseudo phase plane and fractional calculus modeling of western global
economic downturn. Commun. Nonlinear Sci. Numer. Simul. 2015, 22, 396–406. [CrossRef]
17. Moreles, M.A.; Lainez, R. Mathematical modelling of fractional order circuit elements and bioimpedance
applications. Commun. Nonlinear Sci. Numer. Simul. 2017, 46, 81–88. [CrossRef]
18. Tabatadze, V.; Karaçuha, K.; Veliev, E.I. The Fractional Derivative Approach for the Diffraction Problems:
Plane Wave Diffraction by Two Strips with the Fractional Boundary Conditions. Prog. Electromagn. Res. 2019,
95, 251–264. [CrossRef]
19. Sierociuk, D.; Dzieliński, A.; Sarwas, G.; Petras, I.; Podlubny, I.; Skovranek, T. Modelling heat transfer in
heterogeneous media using fractional calculus. Philosophical Trans. R. Soc. A 2013, 371, 20120146. [CrossRef]
20. Bogdan, P.; Jain, S.; Goyal, K.; Marculescu, R. Implantable pacemakers control and optimization via fractional
calculus approaches: A cyber-physical systems perspective. In Proceedings of the Third International
Conference on Cyber-Physical Systems, Beijing, China, 17–19 April 2012.
21. Magin, R.L. Fractional calculus models of complex dynamics in biological tissues. Comput. Math. Appl. 2010,
59, 1586–1593. [CrossRef]
22. Chen, Y.; Xue, D.; Dou, H. Fractional calculus and biomimetic control. In Proceedings of the 2004 IEEE
International Conference on Robotics and Biomimetics, Shenyang, China, 22–26 August 2004; pp. 901–906.
23. Dimeas, I.; Petras, I.; Psychalinos, C. New analog implementation technique for fractional-order controller:
A DC motor control. Aeu-Int. J. Electron. Commun. 2004, 78, 192–200. [CrossRef]

267
Mathematics 2020, 8, 633

24. Sierociuk, D.; Skovranek, T.; Macias, M.; Podlubny, I.; Petras, I.; Dzielinski, A.; Ziubinski, P. Diffusion process
modeling by using fractional-order models. Appl. Math. Comput. 2015, 257, 2–11. [CrossRef]
25. Tarasova, V.; Tarasov, V. Elasticity for economic processes with memory: Fractional differential calculus
approach. Fract. Differ. Calc. 2016, 6, 219–232. [CrossRef]
26. Tarasov, V.; Tarasova, V. Macroeconomic models with long dynamic memory: Fractional calculus approach.
Appl. Math. Comput. 2018, 338, 466–486. [CrossRef]
27. Tejado, I.; Pérez, E.; Valério, D. Fractional calculus in economic growth modelling of the group of seven.
Fract. Calc. Appl. Anal. 2019, 22, 139–157. [CrossRef]
28. Škovránek, T.; Podlubny, I.; Petráš, I. Modeling of the national economies in state-space: A fractional calculus
approach. Econ. Model. 2012, 29, 1322–1327. [CrossRef]
29. Inés, T.; Valério, D.; Pérez, E.; Valério, N. Fractional calculus in economic growth modelling: The Spanish
and Portuguese cases. Int. J. Dyn. Control 2017, 5, 208–222.
30. Ming, H.; Wang, J.; Fečkan, M. The application of fractional calculus in Chinese economic growth models.
Mathematics 2019, 7, 665. [CrossRef]
31. Blackledge, J.; Kearney, D.; Lamphiere, M.; Rani, R.; Walsh, P. Econophysics and Fractional Calculus:
Einstein’s Evolution Equation, the Fractal Market Hypothesis, Trend Analysis and Future Price Prediction.
Mathematics 2019, 7, 1057. [CrossRef]
32. Despotovic, V.; Skovranek, T.; Peric, Z. One-parameter fractional linear prediction. Comput. Electr. Eng. 2018,
69, 158–170. [CrossRef]
33. Skovranek, T.; Despotovic, V.; Peric, Z. Optimal fractional linear prediction with restricted memory. IEEE
Signal Process. Lett. 2019, 26, 760–764. [CrossRef]
34. Önal, N.Ö.; Karaçuha, K.; Erdinç, G.H.; Karaçuha, B.B.; Karaçuha, E. A Mathematical Approach with
Fractional Calculus for the Modelling of Children’s Physical Development. Comput. Math. Methods Med.
2019, 2019, 3081264. [CrossRef]
35. Önal, N.Ö.; Karaçuha, K.; Erdinç, G.H.; Karaçuha, B.B.; Karaçuha, E. A mathematical model approach
regarding the children’s height development with fractional calculus. Int. J. Biomed. Biol. Eng. 2019, 13,
252–260.
36. Önal, N.Ö.; Karaçuha, K.; Karaçuha, E. A Comparison of Fractional and Polynomial Models: Modelling on
Number of Subscribers in the Turkish Mobile Telecommunications Market. Int. J. Appl. Phys. Math. 2019, 10,
41. [CrossRef]
37. Maddison, A. A comparison of levels of GDP per capita in developed and developing countries, 1700–1980.
J. Econ. Hist. 1983, 43, 27–41. [CrossRef]
38. World Bank Databank World Development Indicators. Available online: https://databank.worldbank.org/
source/world-development-indicators# (accessed on 10 December 2019).
39. Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput. 1997, 9, 1735–1780. [CrossRef]
[PubMed]
40. Kingma, D.P.; Ba, J. Adam. A method for stochastic optimization. arXiv 2014, arXiv:1412.6980.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

268
 MDPI
St. Alban-Anlage 66
4052 Basel
Switzerland
Tel. +41 61 683 77 34
Fax +41 61 302 89 18
www.mdpi.com

Mathematics Editorial Office

E-mail: [email protected]
www.mdpi.com/journal/mathematics
MDPI
St. Alban-Anlage 66
4052 Basel
Switzerland
Tel: +41 61 683 77 34
Fax: +41 61 302 89 18
www.mdpi.com ISBN 978-3-03936-119-9