0% found this document useful (0 votes)
380 views126 pagesExpert Trading Systems - Modeling Financial Markets With Kernel Regression
Authored by John R. Wolberg
Pages: 235
Uploaded by
arkanisgarth7209Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
380 views126 pagesExpert Trading Systems - Modeling Financial Markets With Kernel Regression
Authored by John R. Wolberg
Pages: 235
Uploaded by
arkanisgarth7209Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 126
WILEY TRADING ADVANTAGE
‘ding witout Fer Ricacd W. Ar
Neat Neto ine Sree Practing of Pinas Merkel Mac Ath
Option Marke Mating Aisa Bard
Mena Sirs Rte Doers Nase Bre
‘Technical Mar rind Tchad 3. er, J aod de BD
osha
pt of Chart Pate Tas Bao
ane Petey Can
Manage
Bi Ne ark rsd Sage
Bee Gd atonbeny Reba Anais / Sack Sdoeager
Schone Rts ah Scag
Geming the
“Expert Trading Syotere | Job B. Walberg
EXPERT TRADING
SYSTEMS
Modeling Financial
Markets with
Kernel Regression
JOHN R. WOLBERG
JOHN WILEY & SONS, INC.
New York * Chichester * Weinheim » Brisbane * Singapore * Toronto This bookie printed on acid free paper. @
Copyright © 2000 by John R. Wolberg. All rights reserved,
Published by John Wiley & Sons, Ine.
Published simultaneously in Canada,
[No part ofthis publication may be reproduce, stored in retrieval system oF
‘transmitted in any form or by aay means electronic, mechanic, photocopying,
cording, eanning or otherwise, exept as permitted under Sections 107 ot 108
ofthe 1976 United States Copyright Act, without either the priae written permission
ofthe Publisher, or authorization through payment of the appropriate per-opy fee to
the Copyright Clearance Center, 22 Rosewood Drive, Darers, MA 01928,
(878) 760.8400, fox (978) 750-<744, Request tothe Publier for permission
should be addressed to the Permissions Department, John Wiley & Sons, Inc
606 Third Avenue, New York, NY 10158-0012, 212) 850-6011, fx (212) 850.6008,
E-Mail: PERMREQ @ WILEY.COM,
‘This palication is designed to provide accurate and authoritative information in
‘regard to th subject matter covered. Ite eo withthe understanding that the
Dubliher it net engaged in rendering professional services, If professional advice cr
bother expert eslatance is require, the services ofa competent profesional person
‘Should be sought
Library of Congress Cataloging-in-Publication Data:
Wolberg, John R.
pert trading gystems: modeling Snancil markets with kernel
rogression abn R. Walberg
p._em—Wiey trading advantage)
Includes bibliographical references end index.
ISBN 0-471-845085 (alk. paper)
1, Capital market--Mathematical models. 2. Expert systems
8, Regression analysis. L Tide. Series
4523.65 2000
sin’ o414-016118—de21 99.5708
Printed in the United States of America
wos76 54221
This book is dedicated to the new generation:
Yoni and Maya Sassoon, Noa Wolberg, Abigail Kimche,
Shelly Wolberg, Aviv Kimche, and those yet to come. CONTENTS
Preface
‘Acknowledgments
Chapter 1 Introduction
1a
12
13
14
15
16
Data Modeling
‘The Hills of the Galilee Problem
Modeling Financial Markets
Evaluating a Model
Nonparametric Methods
Fundamental versus Technical Analysis
Chapter 2 Data Modeling of Time Series
21
22
23
24
25
26
‘The Time Series Problem
Classical Methods of ‘Time Series Modeling
‘The Curse of Dimensionality
Candidate Predictors
‘The Equity Curve
‘Measuring the Efficiency of a Modeling Method
Chapter 3 Kernel Regression
Ba
B2
33
34
35
36
37
‘The Basic Concept
Higher Order Algorithms
‘The Bandwidth Concept,
Error Estimates
Applying Kernel Regression to Time Series Data
Searching for a Model
‘Timing Considerations
A
13
18
20
25
29
31
37
42
45
49
5B
56
62
65
69 x CONTENTS
Chapter 4 High-Performance Kernel Regression
4.1, Software Considerations
42 The p-Tree
4.3 Partitioning the Learning Data Set
44 Using the p-Tree
4.5. Time Weighting the Data
4.6 Multistage Modeling
4.7 The Search Engine
4.8 Computational Complexity
4.9 Parallel Processing
Chapter 5 Kernel Regression Software Performance
5.1 Software Evaluation
5.2 Searching Parameters
5.3 The Effect of Tree Height
5.4 Number of Nearest Neighbors
5.5 Processing Time per Space
5.6 Data Weighting
5.7 Comparing the Three Algorithms
Chapter 6 Modeling Strategies
"The Modeling Plan
Out-of Sample Testing
Modeling Dynamic Systems
Cross-Sectional Modeling
Combining Models
‘Measures of Performance
Using Kernel Regression for Classification
Problems
6.8. Fine Tuning a Medel
Chapter 7 Creating Trading Systems
7.1. Trading Systems
7.2 Generating Signals
7.3 Creating a Filter
7.4 Cross-Sectional Trading Systems
7.5 Updating Models
7.6 Stock Selection: A Case Study
73
73
4
78
85
91
92
94
96
101
105
105
109
115
120
124
132
138
143
143,
146
150
155
167
159
162
167
a7.
am
174
176
179
181
183
conTENTS
‘Appendix A Linear Least Squares With Data Weighting
Appendix B A Test for Significance of Variance
Reduction
Appendix C Comparing KR and Parametric Regression
Appendix D Comparing KR and Neural Networks
Appendix EA Test for Significance of
Fraction_Same Sign
Bibliography
Index
195
199
205
2a1
221
225
231 PREFACE
‘The use of computers for modeling financial markets has been
growing exponentially. Huge sums of money are being managed
based purely upon computer models. Dats is fed into the models
and directives are issued as output. The directives advise the
user regarding suggested changes in the positions of financial
instruments (e.g., stocks, bonds and commodities). The direc-
tives are based upon the results of models that predict future
price or volatility changes.
‘This book is directed towards people in the financial com-
munity who are involved in evaluating, developing, or using
market models and trading systems. Kernel regression (KR) is
a modeling technique that is particularly attractive for finan-
cial market applications. Although the primary emphasis of
this book is on financial modeling, the methodology can easily
bbe applied to any high dimensional modeling problem. KR has
‘many attractive features that make it a modeling technique
‘that should be in the arsenal of any data mining software
suite.
Attempts to model time series date back to the 1920's when
Yale developed the autoregressive technique for predicting the
annual number of sunspots. Since then the general subject of
time series analysis has become well established. There are a
number of techniques available for modeling time series in gen-
eral and financial markets in particular. Financial market mod-
eling is typically associated with large amounts of high
dimensional multivariate data. Furthermore, the data typically
has a low signal to noise ratio and the signals are usually non-
ai xiv PREFACE
linear. These problems make financial market modeling partic-
ularly challenging.
Another major problem associated with financial market
modeling is that one really doesn't know which if any related
time series are relevant. For example, assume we are trying to
develop a model to predict changes in the S&P price index. One
could list a variety of series that might affect the S&P index
(eg., short and long term interest rates, commodity prices, ete.).
For each related series a number of different indicators ean be
proposed as candidate predictors for the S&P model (e.g., the
one-day fractional changes in the short term interest rate).
Other candidate predictors can be suggested based upon several
related series (e.¢., changes in the ratio of long term to short
term interest rates). One can easily develop a set of hundreds of
candidate predictors that might or might not be included in the
resulting model (or models).
‘To develop models for financial markets, methodologies are
required whieh allow rapid analysis even though the number of
candidate predictors is large. If we start with several hundred
candidate predictors we will certainly eliminate most of these
prior to completing the modeling task. The methodology should
be geared towards finding a subset or subsets of the candidate
predictor space that have an acceptable level of predictive
power. KR, if applied properly, is an excellent method for devel-
oping such models. It-can be extremely fast, a basic requirement
when one considers the number of subsets of the candidate pre-
dictor space which might be examined as part of the modeling
process. KR can be used as a stand-alone modeling technique or
as a preprocessor for slower techniques such as Neural Net-
works. (A comparison of KR and Neural Networks is included in
Appendix D. The results illustrate the complementary nature of
these two modeling methodologies.)
‘This book is geared to three types of readers. The first group
includes those who are interested in modeling in general and
desire an overview of the KR technique. The second group
includes those involved in the development and/or usage of KR.
software. The third group includes readers primarily interested
in the development of computerized trading systems. ‘The first
two chapters include an introduction to the subject and an
PREFACE ~
overview of the modeling process. The next three chapters
include the technical details of the KR method. In Chapter 3 the
mathematical basis of the KR method is developed. Chapter 4
introduces a data structure that permits an efficient implemen-
tation of KR. Chapter 5 provides information regarding the
effect of the various parameters upon performance. This chap-
ter is particularly useful for users of KR software. The method-
ology used in Chapter 5 utilizes artificial data and introduces a
general approach to evaluating modeling software that can also
be applied to other modeling techniques. Chapter 6 discusses
modeling strategies and is relevant to analysts and managers
interested in planning a modeling project. Some of the ideas
introduced in Chapter 6 can also be applied to other modeling
techniques. The final chapter discusses the application of pro-
dictions from the computer models to the development of trad-
ing systems.
The technical chapters are written for readers who are
familiar with college level mathematics but are not necessarily
mathematical statisticians. Many of the books on time series
are directed towards the statistical community and as a result,
the mathematies and notation schemes are difficult to follow for
the non-statistician. The emphasis in this book is on applica-
tion, evaluation, and implementation rather than topics of con-
cern primarily to statisticians. Numerical examples rather than
theorems and lemmas are used to help the reader understand
the details
John R. Wolberg
‘Technion - Israel Institute of Technology
Haifa, areal
‘wolbergshitech.technion.aeil
October, 1999 ACKNOWLEDGMENTS
am most indebted to David Aronson for introducing me to the
field of financial market forecasting. What started out as a sum-
mer project at Raden Research in the early 1980's has turned
out to be a long-term relationship. Through the many projects
that we've worked on together I've developed an understanding
of what is required to develop models and how one goes about.
testing and applying them.
A number of Raden customers have given me the opportu-
nity to develop software and to learn what the real world is all
about. In particular I would like to thank Alan Bechter, Peter
Borish, Ed Bosarge, John Deuss, David Hirschfeld, John Huth,
Barry Honig, Paul Tudor Jones, Sandor Strauss, and Tom
Wright.
Some of my colleagues at the ‘Technion have been particu-
larly helpful. On a number of occasions I've received statistical
advice from Paul Feigin. Miriam Zacksenhouse helped me with
the neural network research that forms the basis of Appendix D.
Eyal Zussman was also helpful in this area. Alon Itai introduced
‘me to the literature on the K Nearest Neighbor problem. I would
also like to acknowledge the help that I have received from the
many students who have taken my graduate course in The
Design and Analysis of Experiments. For many years part of the
course has been devoted to nonparametric statistical modeling
(including kernel regression) and the feedback I've received from
the students has helped me gain insight into the problem.
For software support I have relied upon Victor Leikehman
and Ronen Kimche from Insightware Ltd. They are both highly
wi wit ACKNOWLEDGMENTS
talented software specialists and their help has been most
appreciated. Whenever I ran into a particularly nasty bug I
could always count upon either Victor or Ronen finding the
source of the trouble,
‘To write this book I had to learn Microsoft Word and my
daughter Tamar Kimche got me over the initial problems. There
are still a few thousand features that I haven't learned yet but,
at this point T ean manage!
A few people do not fall into any of the above categories but
nevertheless deserve acknowledgement. I would like to thank
George Butler and Richard Waldstein for their help in my early
days in the modeling business. Most recently Tim Masters has
been extremely helpful in answering my many questions
regarding neural networks. He has published a number of books
on the subject and is a real font of knowledge.
‘The BARRA Corporation provided the historical data used
in the stock selection case study discussed in Chapter 7, I would
like to thank them for being so cooperative and for their per-
mission to include the results of the study in this book.
would also like to acknowledge the help I received from the
‘Technion—Israel Institute of Technology. The Technion Fund
for Promotion of Research supported some aspects of the
research upon which this book is based.
Finally, since I've already mentioned my daughter Tammy,
Iwould also like to acknowledge the other members of my fam-
ily just for being there and making life so pleasant: my wife
Laurie, my sons Dave and Danny, my other daughter Beth, all
their spouses and the grandchildren.
EXPERT TRADING
SYSTEMS INTRODUCTION
1.1 DATA MODELING
‘The concept of a mathematical model has been with us for thou-
sands of years, going back to the ancient Egyptians and Greeks
who were known for their mathematical ability. They used
mathematical expressions to quantify physical ideas. For exam-
ple, we are all familiar with Pythagorus’s theorem:
c= Nats an
which is used to compute the length of the hypotenuse of a right
triangle once the lengths of the other two sides are known. For
‘this example, C is called the dependent variable and A and B
are the independent variables. To generalize the concept of a
mathematical model, let us use the notation:
Y=/%) az)
where Y is the dependent variable and X is the independent
variable. For the more general case, both X and Y might be vee-
tors. The function f might be expressed as a mathematical equa-
tion, or it might simply represent a surface that describes the
relationship between the dependent and independent variables.
Data modeling is a process in which data is used to deter-
mine a mathematical model. Although there are many data
1 INTRODUCTION
‘modeling techniques, all of them can be classified into two very
broad categories: parametric and nonparametric methods.
Parametric methods are those techniques that start from a
known functional form for f(X). Probably the most well-known
parametric method is the method of least squares. Most books
on numerical analysis include a discussion of least squares, but
usually the discussion is limited to linear least squares. The
more general nonlinear theory is extremely powerful and is an
excellent modeling technique if (X) has a known functional
form. For many problems in science and engineering, X) is
known or can be postulated based on theoretical considerations.
For such cases, the task of the data modeling process is to deter-
mine the unknown parameters of fX) and perhaps some mea-
sure of their uncertainties.
‘As an example of this process, consider the following exper-
iment: the count rate of a radioactive isotope is measured as a
function of time. Equation (1.2) for this experiment can be
expressed as
YeAct)+B as)
where the dependent variable ¥ is the count rate (i.e., counts
per unit time), A is the amplitude of the count rate originating
from the isotope (i.e., counts per unit time at x equal to zero),
‘kis the unknown decay constant, x is the independent variable
(which for this experiment is time), and B is the background
count rate. This is a very straightforward experiment, and
nonlinear least squares can be used to determine the values of
A, k, and B that best fit the experimental data. As a further
bonus of this modeling technique, uncertainty estimates of
the unknown parameters are also determined as part of the
analysis.
In some problems, however, (X) is not known. For exam-
ple, let's assume we wish to develop a mathematical model
that gives the probability of rain tomorrow. We can propose @
list of potential predictors (ie., elements of a vector X) that
can be used in the model, but there is really no known fune-
tional form for (X). Currently, a tremendous amount of inter-
est has been generated in weather forecasting, but the general
approach is to develop computer models based on data that
THE HILLS OF THE GALILEE PROBLEM 3
yield predictions but are not based on simple analytical fune-
tional forms for (UX).
Another area in which known functional forms for (X) are
not practical are for financial markets. It would be lovely to dis-
cover a simple equation to predict the price of gold tomorrow or
next week, but so far no one has ever successfully accomplished
this task (or if they have, they are not talking about it). Never-
theless, billions of dollars are invested daily based on computer
models for predicting movement in the financial markets. The
modeling techniques for such problems are typically nonpara-
metric and are sometimes referred to as data-driven methods.
Within this broad class, two subclasses can be identified that
cover most of the nonparametric analyses being performed
today: neural networks and nonparametric regression.
‘The emphasis in this book is on nonparametric methods and
in particular on the nonparametric kernel regression method,
One problem with this class of methods is that they can be very
computer intensive. Emphasis will therefore be placed on devel-
oping very efficient algorithms for data modeling using kernel
regression. We live in a complicated nonlinear world, and it is
often necessary to use multiple dimensions to develop a model
with a reasonable degree of predictive power. Thus our discus-
sion must include the development of multidimensional models,
‘We must also consider how one goes about evaluating a model,
Can itbe used for predicting values of ¥? How good are the pre
dictions? These are the sorts of questions considered in the fol-
lowing chapters.
1.2 THE HILLS OF THE GALILEE PROBLEM
From ny office here in Haifa, I can look out the window and see
the hills of the Galilee. I've used these hills to pose a problem to
students: design a program that estimates height as a function
of position for any point in the Galilee. Assume that we are lim-
ited to 10,000 sample data points. Let’s say we limit the prob-
Jem to a square that starts from a point in Haifa port and
extends 30 kilometers east and 30 kilometers north. Anyone
who has seen this part of the globe knows that the Galilee is 4 INTRODUCTION
quite irregular. It includes valleys, small hills, and some larger
hills that might even be called mountains,
The first simple-minded approach is to fit a grid over the
entire area and distribute the data points evenly throughout.
this area. Spreading 10,000 points over 900 square kilometers
implies a separation distance of 300 meters. This might be rea-
sonable if we were trying to develop a model for elevation in the
middle of Kansas, and it might also be reasonable for some of
the valleys of the Galilee, but it is certainly not a reasonable
mesh size for some of the more mountainous areas. Ideally, we
would like to concentrate our points in the hillier regions and
use fewer points in the flatter regions. But by doing this we
introduce a new level of complexity: how do we estimate height
‘as a function of position for a nonuniform grid?
The uniform grid suffers from a lack of resolution but allows
the user an incredibly simple data structure: a two-dimensional
matrix of heights, Thus, to find the height at point (X, Y), all we
need to do is calculate where this point falls in the matrix. For
example, consider the point X = 12342 and Y = 18492 where X
is the distance in meters going east from our 0,0 point and Y is
the distance in the northward direction. If we denote a point in
the matrix as (I, J) and the matrix as H, then (X, ¥) is located
northeast of point J = 41, J = 61. We can then use simple two-
dimensional linear interpolation to determine the height at (X,
Y) using the four surrounding points:
a
HEIGHTXY)=¥ Ys s+) as
ma io
where
fin, m) = Hin, m) * WOY.n,m) as)
and the weight terms W(X,Y,n,m) are calculated as follows:
‘WOK ¥, 1.1) = (800 ~X + 800 1) * (300 - ¥ + 300 .1¥(300*300) (1.6)
WOGY.L + 1d) = (X— 300 1) * (800 — ¥ + 300 J(800"800) (1.7)
WOY.L,J-+1) = (800 ~ X + 300 1) * CY - 300 J/{300°300) (1.8)
THE HILLS OF THE GALILEE PROBLEM 5
WX YL+1,J + 1) = (X—300 1 * (¥ ~ 800 (300"300) (1.9)
The problem with this simple-minded approach is illus-
trated in Figure 1.1. The estimate of the height of the hilltop
just northeast of point (I,J) would be clearly underestimated. If
the points are chosen such that we concentrate points in the
hilliest regions, we introduce a number of problems:
1. If we use N points to estimate HEIGHT(X,Y), how do we
choose the N points?
2. How many points should we use (i.e., what should be the
value of N)?
3. Once we have found N points, what can we do to estimate
the height at X,¥?
The simple data structure used for a uniform grid is no longer
applicable. One approach is simply to list data in an array of
Tength 10,000 and width 3 (ie., column 1 is X(), column 2 is Yl),
and column 3 is H(). If our choice of N points are the N nearest
neighbors to (X,Y), then we will have to run through the entire
ue? —
Leon |\
7 im cy) :
a ||
Howto
Figure 1.1 Map with lines of constant elevation. Note hilltop surrounded by
four grid points. 6 INTRODUCTION
list to find these nearest points. This would require 10,000 ealeu-
lations of distance to identify the N nearest neighbors.
We could improve our search using the same 10,000 by 3
matrix by just sorting the data on one of the columns (either the
X or ¥ column), Then to find the N nearest neighbors we could
use a binary search to find the region of the matrix in which to
concentrate our search, For example, let's say that the value X =
12342 falls between rows 4173 and 4174 of the H matrix (after
sorting in the X direction). In mathematical terms, the following
inequalities are satisfied:
A(A173, 1) < =X and A(4174, 1) >X ato)
where column 1 of the A matrix is the X values of each point.
Let's say that we use the four closest points for our estimation
of HEIGHTYX,Y) (ie., N = 4); then we need some rule to decide
which rows of the matrix should be considered. One simple
heuristic is to consider the 10 points below X and the 10 points
above X (i.e., rows 4164 to 4183). ‘The distance squared (Dsqr)
from row i to the point X,Y is simply:
Dsgr =
XAG, DP + ~AG2y" aay
We use Dsgr rather than distance to avoid the need to take
unnecessary square roots. The 20 values of Dsgr are compared,
and the points represented by the smallest four are chosen. It is
possible that some or all of the four nearest points to (X,Y) fall
outside this range of rows. However, more than likely we will
find some, if not all, of the four nearest neighbors using this
method. By sorting the data, we reduce the number of values of
Dsgr that must be determined from 10,000 to 20, which ropre-
sents a tremendous saving in eompute time (if we plan to uae
the program to determine elevation at many points).
Compared to the uniform grid, this choice of data structure
results in a factor of 3 increase in the size of the matrix (i.e.,
30,000 numbers instead of only 10,000). We also require an
increase in computational time because we must now initiate a
search to find the N nearest neighbors. This is the price we
must pay to improve the accuracy of the computed values of ele-
vation while limiting the size of the data array to 10,000 points.
MODELING FINANCIAL MARKETS 7
Once the N nearest points have been located, the next prob-
Jem to be addressed is the task of converting the heights of
these points into an estimate of HEIGHT (X,Y). A number of
approaches can be used to solve this problem, and they are dis-
cussed in later chapters. The main purpose of this discussion is
to introduce the concept of estimating the value of a dependent.
variable (in this case HEIGHT) as a function of one or several
independent variables (in this case X and ¥) when the relation-
ship is complex. Trying to determine one global equation that
relates HEIGHT as a function of X and Y is useless. The best
‘one could do is to find separate equations for many subregions
in the 30 by 30 kilometer space. Using a parametric technique
such as the method of least squares, an equation for HEIGHT
as a function of X and Y within each region could be deter-
mined. An alternative approach is to use a nonparametric
method such as kernel regression. This method is developed
and discussed in the following chapters.
1.3 MODELING FINANCIAL MARKETS
‘The purpose of developing models for financial markets is to end
up with a means for making market predictions. One typically
attempts to develop models for predicting price changes or mar-
ket volatility. The hope associated with such efforts is to use the
‘model (or models) as the basis for a computerized trading system.
‘To minimize equity drawdowns, most computerized trading sys.
tems use separate models for different markets and perhaps for
different trading frequencies. By trading several models simulta-
neously, the equity decreases due to vue mudel ure hopefully bal-
anced by equity increases from other models. Thus one would
‘expect a smoother portfolio equity curve than what one might get
by trading a single model. This well-known concept is called
diversification and is discussed in most books on finance. Data is
fed into the models, and the system issues trading directives. The
directives are usually in the form of buy and sell signals.
Using past history, one can simulate the performance of a
system based on the issued signals. Thus for each model an 8 INTRODUCTION
equity curve can be generated for the simulated time period.
One can look at the equity curves of each model separately and
develop strategies for combining the various models into a mul-
timodel trading system with a single combined equity curve
Whether one looks at the individual equity curves or the com-
bined equity curve, measures of performance are required.
Obviously, one must look at profitability (for example, annual
rate of return), but it should be emphasized that profitability is
not the sole measure of the value of a trading system. Typically,
one also computes the risk associated with the system and then
combines profitability and risk into some measure of perfor-
mance. There are many definitions of risk. In his Nobel Prize-
winning work on Modern Portfolio Theory, Markowitz used the
standard deviations of equity changes as the measure of risk
associated with an equity curve.* Many other definitions of risk
are in use, some of which are included in a discussion of mea-
sures of performance in Section 1.4.
One danger associated with the modeling process described
here is that if enough combinations of models and markets are
tried, we will end up with a “successful” combination that only
works for the modeling data but is not based on models with
real predictive power. For such cases we can expect failure
when the models are applied to unseen data. ‘To protect against
this possibility, one should test the entire system using unseen
data (e., data not used in the modeling process). Only if it per-
forms well on this data should one actually start using it to
trade real funds.
‘The task of developing a model for a financial market is
guite different from the Hills of the Galilee problem discussed
in the previous section, That problem exhibited the following
characteristics:
1. The number of independent variables (.e., 2) was known.
2. For every combination of the independent variables (ie,
X and Y), there was one correct value of the dependent
variable (.e., H).
3. The dependent variable H could be modeled as a function
of X and Y, and with enough data points the model could
be made to be as accurate as desired.
MODELING FINANCIAL MARKETS 9
‘The modeling of financial markets is quite different in all of
these respects.
1. The number of independent variables required to develop
‘a model with a reasonable degree of predictive power is
unknown. Indeed, one does not even know if it is possible
to obtain a decent model with the available data
2, Fora given set of independent variables, there is no guar-
antee that if the values of the independent variables are
the same for two data points, the values of the dependent
variable will also be the same. In other words, for each
combination of the independent variables there is a range
of possible values of the dependent variable.
8. Regardless of the number of available data points, there
is no hope of converging to a model free of error. (In other
words, we assume that our final model will contain some
noise. Our hope is to obtain a model in which the signal is
strong enough that the predictive power of the model will
be of some value.)
Financial markets can be characterized as having a low sig-
nal-to-noise ratio. In other words, a large fraction of the change
in price from one time period to the next appears to be a random
shock. In addition, the small signal typically varies in a highly
nonlinear manner over the modeling space. Often, however, the
random shock is not totally random if one considers other
related time series. By bringing in more relevant information
(c., related time series), a greater fraction of the price changes
can be explained. The interesting aspect of financial market
modeling is that one does not need to obtain a high degree of
predictability to develop a successful trading system. For exam-
ple, if models could be developed for a dozen different: markets
and each model could consistently explain 5 percent of the vari-
ance in the price changes (see Section 1.4 for a definition of
‘Variance Reduction), a highly profitable trading system could be
developed based on these models!
As an example of the modeling process, consider the prob-
lem of predicting a future price of gold. What are the indepen-
dent variables? The best that an analyst can do is to propose a 10 INTRODUCTION
set of candidate predictors. Each of these predictors must be
backward looking. In other words, their values must be known
at the times the predictions are made. How does one come up
with a decent set of candidate predictors? A massive body of
research has been devoted to this problem. Many articles and
books have been written on this subject for many different
financial markets. This book concentrates on how to select a
model once a set of candidate predictors has been proposed.
The focus of the book is on evaluating the candidate predictors
individually and together once they have been specified. How-
ever, some comments are made in Chapter 2 regarding the
specification of candidate predictors.
The number of candidate predictors available for developing
prediction models is limited only by the imagination of the ana-
lyst. ‘The primary source of candidate predictors is from the time
series being modeled. For the gold model, one would first use the
Gold price series as a source of candidate predictors. Often a
number of predictors might be variations on the same theme.
For the gold example, the most obvious choices of candidate pre.
dictors are past changes in gold prices. For example, the relative
change in the prices of gold over one, two, and three time periods
can be selected as the first three candidate predictors:
1~—LAG(GOLD,1/GOLD
= 1-LAG(GOLD,2VGOLD
1~LAG(GOLD,3YGOLD
‘The variable GOLD represents a time series of gold prices, and
the LAG operator returns the series being lagged by the num-
ber of records indicated as the second parameter. The next
theme might be ratios based on the current price of gold and
moving averages of gold prices, For example:
1. X4 = GOLD/MAGOLD,3) - 1
2. X5 = GOLD/MA(GOLD, 10) - 1
8. X6 = GOLD/MA(GOLD,50) - 1
‘The operator MA is a moving average over the number of time
periods indicated by the second parameter. 'The candidate pre-
EVALUATING A MODEL a
dictor X4 is the deviation from 1 of the ratio of the current price
of gold divided by the moving average of gold over the last three
time periods. The candidate predictor XS is similar to X4 but is
based on a longer time period, and X6 is based on the longest
time period. Negative values of these three candidate predictors
mean that the latest price of gold is less than the three moving
averages. Next, we might start considering data from other
markets: (e.g., XT = 1 - LAG(S&P, 1/S&P). The variable S&P
represents a time series of the S&P price index. When one
starts considering all the possible predictors that might influ-
cence the future change in the price of gold, the set of candidate
predictors can become huge.
A reasonable approach to modeling when the number of
candidate predictors is large is to consider subspaces. For
example, assume we have 100 candidate predictors. We might
first consider all 1D (one-dimensional) spaces. We try to find a
model based on X1 (i.e., ¥ = iX1), then AX2), up to X100).
After all 1D spaces have been considered, we then proceed to
2D spaces. If we examine all 2D combinations (i.e., AX1,X2) up
to (X99, X100), then we must consider 100*99/2 = 4950 differ-
ent 2D spaces. Proceeding to 3D spaces, the number of combi-
nations increases dramatically. For 100 candidate predictors,
the total number of 3D spaces is 100*99%98/6 = 161700.
Clearly, some sort of strategy must be selected that limits the
process to an examination of only the spaces that offer the
greatest probability of success.
One question that comes to mind is the upper limit for the
dimensionality of the model. How far should the process be con-
tinued? three dimensions? four dimensions? For a given number
of data points, as the dimensionality of the model increases, the
sparseness of the dala also increases. To discuss sparseness
from a more quantitative point of view, let’s first define a region
in a space as a portion of the space in which all the signs of the
values of the various X’s comprising the space do not chango.
For example, if we have N data points spread out in a 2D space,
then we have an average of N/4 points per region. Let's say we
are looking at the 2D region made up of candidate predictors X5
and X17, Assume further that both X5 and X17 have been nor-
malized so that their means are zero, The four regions are: R INTRODUCTION
1. X5>0 and X17>0
2. X60
3. XB > 0 and X17 <0
4. X5
