Advanced Econometric Methods A Guide To

Estimation And Inference For Nonlinear Dynamic

Models Francisco Blasques download

https://ebookbell.com/product/advanced-econometric-methods-a-
guide-to-estimation-and-inference-for-nonlinear-dynamic-models-
francisco-blasques-35165190

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Nonparametric Econometric Methods Advances In Econometrics Qi Li

https://ebookbell.com/product/nonparametric-econometric-methods-
advances-in-econometrics-qi-li-1374378

Dynamic Model Analysis Advanced Matrix Methods And Unitroot

Econometrics Representation Theorems Ition 2nd Ed Mario Faliva

https://ebookbell.com/product/dynamic-model-analysis-advanced-matrix-
methods-and-unitroot-econometrics-representation-theorems-ition-2nd-
ed-mario-faliva-1223698

Topics In Dynamic Model Analysis Advanced Matrix Methods And Unitroot

Econometrics Representation Theorems Prof Mario Faliva

https://ebookbell.com/product/topics-in-dynamic-model-analysis-
advanced-matrix-methods-and-unitroot-econometrics-representation-
theorems-prof-mario-faliva-4211444

Advanced Econometric Theory John Somerset Chipman

https://ebookbell.com/product/advanced-econometric-theory-john-
somerset-chipman-50269850
Advanced Econometric Marketing Models Dennis Fox

https://ebookbell.com/product/advanced-econometric-marketing-models-
dennis-fox-2120488

Advanced Econometric Theory John Chipman

https://ebookbell.com/product/advanced-econometric-theory-john-
chipman-51228330

Financial Economics And Econometrics Routledge Advanced Texts In

Economics And Finance 1st Edition Nikiforos T Laopodis

https://ebookbell.com/product/financial-economics-and-econometrics-
routledge-advanced-texts-in-economics-and-finance-1st-edition-
nikiforos-t-laopodis-36137880

Readings In Unobserved Components Models Advanced Texts In

Econometrics Andrew C Harvey

https://ebookbell.com/product/readings-in-unobserved-components-
models-advanced-texts-in-econometrics-andrew-c-harvey-1795156

Applied Econometrics A Practical Guide Routledge Advanced Texts In

Economics And Finance 1st Edition Chungki Min

https://ebookbell.com/product/applied-econometrics-a-practical-guide-
routledge-advanced-texts-in-economics-and-finance-1st-edition-chungki-
min-49061048
ADVANCED ECONOMETRIC METHODS
A GUIDE TO ESTIMATION AND INFERENCE
FOR COMPLEX DYNAMIC MODELS

WITH APPLICATIONS IN BUSINESS, ECONOMICS, FINANCE,

ENGINEERING, MACHINE LEARNING AND DATA SCIENCE

DETAILED SOLUTIONS AND CODE SCRIPTS

AVAILABLE IN MATLAB, R AND PYTHON

Francisco Blasques

2019

ADVANCED TEXTS IN ECONOMETRICS

MACHINE LEARNING AND DATA SCIENCE A.
2
c Francisco Blasques, 2019

All rights reserved. No part of this publication may be reproduced,

stored in a retrieval system, or transmitted in any form, or by any means,
electronic, mechanical, photocopying, recording or otherwise, without the
prior permission in writing from the author.

This book was typeset by the author using LATEX.

Published by A. Publications’ advanced academic textbook series.

Advanced Texts in Econometrics, Machine Learning and Data Science.
4

In memory of my dear friend

Jean-Pierre Urbain
 Contents

Preface i

I Introduction 1

1 What you should already know! 3

1.1 Linear regression model . . . . . . . . . . . . . . . 3
1.2 Least squares estimator . . . . . . . . . . . . . . . 5
1.3 Least squares: consistency . . . . . . . . . . . . . . 6
1.4 Least squares: normality . . . . . . . . . . . . . . . 11
1.5 Linear dynamic model . . . . . . . . . . . . . . . . 16
1.6 Maximum likelihood estimator . . . . . . . . . . . 19
1.7 Maximum likelihood: consistency . . . . . . . . . . 21
1.8 Maximum likelihood: normality . . . . . . . . . . . 23

2 The challenges ahead... 27

2.1 Intractable estimators . . . . . . . . . . . . . . . . 27
2.2 Nonlinear regression . . . . . . . . . . . . . . . . . 28
2.3 Nonlinear dynamic models . . . . . . . . . . . . . . 29
2.4 Nonlinear dynamics and stationarity . . . . . . . . 32
2.5 Incorrect specification . . . . . . . . . . . . . . . . 34
2.6 Questions we would like to answer . . . . . . . . . 37

II Dynamic Models 39

3 Probability models 41
3.1 What is a probability model? . . . . . . . . . . . . 42
3.2 Why use probability models? . . . . . . . . . . . . 45
3.3 What is a DGP? . . . . . . . . . . . . . . . . . . . 47
3.4 Correct Specification . . . . . . . . . . . . . . . . . 48
3.5 Generality of linear models . . . . . . . . . . . . . 49
3.6 Nonlinear dynamic models . . . . . . . . . . . . . . 54
3.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . 74

5
6 CONTENTS

4 Stochastic properties 77
4.1 Stationarity and Ergodicity (SE) . . . . . . . . . . 78
4.2 Fading memory and limited dependence . . . . . . 99
4.3 Bounded unconditional moments . . . . . . . . . . 113
4.4 Models with exogenous variables . . . . . . . . . . 118
4.5 Time-varying parameter models . . . . . . . . . . . 121
4.6 Multivariate Contractions . . . . . . . . . . . . . . 125
4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . 127

5 Filter invertibility 133

5.1 Filters and DGPs . . . . . . . . . . . . . . . . . . . 133
5.2 Contraction and invertibility . . . . . . . . . . . . 136
5.3 Filter stationarity and moments . . . . . . . . . . . 137
5.4 Multivariate filters . . . . . . . . . . . . . . . . . . 141
5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . 146

III Estimation and Inference 147

6 Extremum estimators 149

6.1 M and Z estimators . . . . . . . . . . . . . . . . . 150
6.2 Existence and Measurability . . . . . . . . . . . . . 155
6.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . 160

7 Consistent Estimators 165

7.1 A general consistency theorem . . . . . . . . . . . 167
7.2 Uniform convergence . . . . . . . . . . . . . . . . . 172
7.3 Stochastic Equicontinuity . . . . . . . . . . . . . . 185
7.4 Identifiable uniqueness . . . . . . . . . . . . . . . . 190
7.5 Time-varying parameter models . . . . . . . . . . . 195
7.6 Estimation for misspecified models . . . . . . . . . 200
7.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . 202

8 Normal Estimators 209

8.1 An asymptotic normality theorem . . . . . . . . . 212
8.2 Approximate statistical inference . . . . . . . . . . 224
8.3 Estimation of asymptotic variance . . . . . . . . . 226
8.4 Moments and equicontinuity . . . . . . . . . . . . . 229
8.5 The CLT under misspecification . . . . . . . . . . . 232
8.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . 234
CONTENTS 7

IV Econometric Analysis 237

9 Analysis of complex models 239

9.1 Probabilistic analysis . . . . . . . . . . . . . . . . . 240
9.2 Value-at-Risk . . . . . . . . . . . . . . . . . . . . . 243
9.3 Forecasting . . . . . . . . . . . . . . . . . . . . . . 246
9.4 Impulse response functions . . . . . . . . . . . . . 253
9.5 Dynamic portfolio optimization . . . . . . . . . . . 257
9.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . 260

10 Method selection 267

10.1 Pseudo-True Parameters . . . . . . . . . . . . . . . 268
10.2 Estimator selection . . . . . . . . . . . . . . . . . . 275
10.3 Model selection . . . . . . . . . . . . . . . . . . . . 282
10.4 Ensemble Methods . . . . . . . . . . . . . . . . . . 299
10.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . 301

11 Structural Analysis 307

11.1 Structural vs predictive models . . . . . . . . . . . 307
11.2 Endogeneity on complex models . . . . . . . . . . . 311
11.3 Instrumental variables and AB testing . . . . . . . 317
11.4 Other Examples . . . . . . . . . . . . . . . . . . . 326
11.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . 333

V Appendix 335

A Random variables i

B Modes of Convergence ix

C Metrics and Norms xiii

D Well behaved functions xvii

8 CONTENTS
Preface

This book developed from the lecture notes of the graduate mas-
ters course ‘Advanced Econometrics’, that I have the pleasure of
teaching since 2014 at the VU University, in Amsterdam. The
text retains the informal classroom tone with which the lecture
notes were originally written. All chapters were written, first and
foremost, with my students in mind, so the text is addressed to
them. Let us begin...

Dear student, it is a pleasure to welcome you to the course

Advanced Econometrics! These lecture notes offer a review of
estimation and inference for complex dynamic models.

Let us start by clarifying one point immediately: what exactly

is a complex model ? Well, we shall say that a model is complex,
if we cannot obtain an analytic expression (i.e. a mathematical
formula) for the estimator of the parameters of the model.1
Until now, the models and estimators that you encountered on
your undergraduate studies were such that one could always
derive a closed-form analytic expression for the estimator, and
ultimately use that formula to study the properties of the
estimator. In contrast, we shall now focus on the theory and
practice of econometrics in settings where robust estimators and
model nonlinearities preclude the availability of simple formulas
and closed-form analytic expressions.

1
Note that, according to this definition, a model being ‘complex’ depends
not only on the model itself, but also, on the type of estimator being used.
So it may sometimes be more convenient to think of complexity as referring
to a specific pairing of model and estimator.

i
ii PREFACE

Another point worth mentioning is that the word advanced in

the title of this book should be suggestive of at least two things:
(a) that you are already familiar with a considerable range of
statistical and econometric methods;

(b) that you have enough mathematical background to under-

stand statistical and econometric theory at a considerable
level of generality.
Of course, there are plenty of books in econometrics that
offer considerably more challenging and technical reviews of
econometric techniques, provide substantially more detailed
and advanced discussions of the methods, and most books will
definitely be broader in scope and more comprehensive than
this one. However, those books are typically written for an
audience of researchers and academics. In contrast, with this
book, I try to make the theory and practice of complex dynamic
models accessible to graduate (master) students, or a readership
with heterogenous background but a reasonable prior knowledge
of probability, statistics and quantitative methods. This is a
‘classroom’ textbook in both the way it is written and the way
it is organized.

Whenever possible, I will try to start from the basics and

briefly recap the material that you learned in your introductory
courses of probability, statistics, econometrics, machine learning,
statistical engineering, quantitative finance or business analytics.
In any case, I will assume that you are already familiar with a
few simple models, e.g. the linear regression model or the linear
dynamic autoregressive model. I will also take for granted that
you have derived the properties of some estimators, like the
consistency and asymptotic normality of ordinary least squares
(OLS) and maximum likelihood (ML) estimators.

As we shall see however, the theory that you learned in your

introductory courses is quite limited, for at least three reasons:
1. it works only in the simplistic setting of linear models;

2. it does not extend to other estimators;

 iii

3. it requires the very restrictive assumption that the model

provides an exact description of reality (i.e. that the model
is correctly specified).

Let us make these limitations more concrete by looking at the

case of the least squares estimator for a simple linear dynamic
autoregressive model.

Example: Consider the linear autoregressive (AR) model

xt = βxt−1 + εt .

You surely recall that, given the simple linear structure of this
model, it is easy to obtain an expression for the least squares
estimator of the parameter β. Indeed, you might even recall that
it takes the form PT
xt xt−1
β̂T = Pt=2
T
.
2
t=2 xt−1
Analyzing the properties of this estimator is easy when we assume
that the linear model is the true model (i.e. when we impose a
correct specification assumption). Indeed, if the model is well
specified, then we can substitute xt for βxt−1 + εt and thus re-
write the estimator β̂T as follows
PT
εt xt−1
β̂T = β + Pt=2
T
. (1)
2
t=2 xt−1
|{z}
true parameter | {z }
remainder term

Finally, we exploit the linearity of the model once more! In par-

ticular, we recall that, in the linear setting, the condition |β| < 1
ensures that the time series {xt }t∈Z is stationary. The station-
arity property is important as it allows us to apply laws of large
numbers and central limit theorems to the remainder term in
equation (1). As a result, we are able to arrive at the conclusion
that:2
2
Read Section 2 for a detailed description of all the steps used in analyzing
the properties of the OLS estimator.
iv PREFACE

1. the estimator β̂T is consistent because the remainder term

vanishes to zero when the sample size T grows to infinity
PT
εt xt−1 p
Pt=2
T 2
→ 0 as T → ∞,
t=2 xt−1

2. the estimator β̂T is asymptotically normal because the stan-

dardized remainder term is asymptotically normally dis-
tributed
PT
√ εt xt−1 d
T Pt=2
T
→ N (0, σ 2 ) as T → ∞.
2
t=2 xt−1

The results described above were interesting at an introductory

level. However, they were applicable only in this very limited
setting of simple linear regression models and under the very
restrictive assumption that the model is correctly specified.

The linear nature of the models used in your introductory statis-

tics and econometrics courses was often imposed for simplicity.
Many models currently used in econometrics, statistics, machine
learning, statistical engineering, and big data analytics, are non-
linear in nature. From regime-switching models like the thresh-
old autoregressive (TAR) model of Tong and lim (1980), or the
smooth transition autoregressive (STAR) model of Terasvirta and
Anderson (1992), to models of time-varying volatility like the
generalized conditional heteroeskedasticity (GARCH) model of
Engle (1982) and Boreslev (1986), nonlinear models are present
in every corner of econometrics. Nonparametric and semi-
nonparametric dynamic econometric methods also call for a the-
ory that extends to nonlinear models. In machine learning and
big data analytics, nonlinear methods are pervasive and range
from nonlinear classifiers and nonlinear support vector machines
to artificial neural networks used in deep learning problems. In
all these settings, the estimator or training algorithm does not
have a closed form analytical expression (a ‘formula’) as in the
linear example above. Instead, the parameters are estimated by
numerical methods. In particular, the sum of squares are min-
imized numerically to obtain least squares estimators, and the
 v

likelihood function is maximized numerically to obtain maximum

likelihood estimators.

It seems to be generally accepted that the economy is non-

linear, in that major economic variables have nonlinear re-
lationships. Economic theory suggests models with floors
and ceilings, buffer stocks, and switching regimes. Invest-
ment functions, production functions and Phillips curves
are usually specified in nonlinear forms.
Granger and Terasvirta (1994)

The assumption of correct specification used in your introduc-

tory statistics and econometrics courses was also imposed essen-
tially to simplify the analysis. In practice, no one really believes
that such simple models are an exact description of the relation
between economic variables. Economic variables are most likely
generated by an immensely complex process involving potentially
infinitely many variables. Econometric models are simplifications
of reality that help us approximate a very complex economic
world and uncover basic relations between variables.

The assumption of correct specification is often the most

difficult condition of all to accept in a non-experimental
science such as economics. When formulating tractable
empirical models, we freely admit that we abstract from
reality, so that a model constitutes only an approximation
to a complex process. It is important to know what happens
to commonly used estimator and test statistics when the
model is mis-specified.
Domowitz and White (1982)

In this course we will move away from the simple linear setting
and abandon the restrictive assumption of correct specification.
We will venture together into unknown territory! It will require
some effort, but it will pay off! Together, we will learn how
to analyze the properties of the many nonlinear models that
econometricians, statisticians and data scientists of all fields use
on a daily basis. We will see that we can analyze the properties
of estimators even if they do not have a tractable analytic
vi PREFACE

expression. We will become familiar with a theory that applies

to many more estimators than just least squares, maximum
likelihood or method of moments. Finally, we will discover what
happens to estimators when the model is misspecified. In par-
ticular, we will see that the estimators converge to ‘pseudo-true’
parameters that have a very clear interpretation. We will show
that our mis-specified econometric models approximate the true
unknown data generating process, and how this approximation
depends crucially on the estimation method that we use.

Given the importance of these topics in econometrics and other

similar fields, it should not come as a surprise to anyone that
there exists a large number of research articles and textbooks
addressing these matters. Unfortunately, this literature is
typically directed only at an audience of academic researchers
interested in econometric theory. As a result, most authors
approach these topics at a level of technicality that is far too
high for the typical first-year graduate student. Furthermore,
the link between theory and practice is typically sacrificed. It is
up to the reader to understand how the theory may potentially
be applied to practical problems and state-of-the-art models.
With few exceptions, there seems to exist no middle ground
in the literature. Either these topics are ignored, or they are
directed at an advanced audience of researchers.

The aim of this book is twofold. First, it attempts to “digest”

and simplify excellent advanced textbooks and key research
articles, and hopefully make their content more easily accessible
to the typical graduate student. These include the work of
Gallant and White (1986), Nelson (1990, 1992), Bougerol (1993),
White (1994) and Potscher and Prucha (1997) and Straumann
and Mikosch (2006), among others. Second, the book tries to
link theory and practice at every step of the way. In particular,
through examples and exercises. In essence, this book attempts
to apply the relevant theory directly to models, methods and
data used by econometricians, statisticians, engineers, as well as
machine learning and big data scientists.

The ultimate objective of this book is to introduce students to

 vii

advanced methods and models in statistics and econometrics.

Hopefully, after reading it, you will be able to analyze the prop-
erties of complex nonlinear dynamic models and understand the
behavior of general estimators in advanced and complex settings.

In order to bring these advanced topics to the classroom, we

will sometimes have to sacrifice some generality in theorems and
assumptions in favor of clarity and simplicity. We will however
not sacrifice those things that truly matter! The fundamental
findings and insights of the advanced econometric and statistical
theory literature will not be compromised! Together, we shall
walk the road towards understanding the properties of advanced
nonlinear dynamic models and econometric methods.

SCRIPTS
This book offers a library of scripts in MATLAB, Python and R. These
scripts are used in examples and practical exercises throughout the book.
The reader is encouraged to download the relevant scripts from the au-
thor’s webpage https://personal.vu.nl/f.blasques/.

EXERCISE SOLUTIONS
This book is accompanied by a solutions manual for a number of selected
exercises. The reader is encouraged to download the solutions manual
from the author’s webpage https://personal.vu.nl/f.blasques/.

I would like to express my sincere gratitude to my past stu-

dents Emmely de Horde, Ruben Konijn, Sina Zolnoor and Tamer
Dilaver for their careful reading of these lecture notes and the
many helpful suggestions. A very special thanks goes to Tamer
Dilaver for pursuing this task with such persistence and diligence
for several months after the course was already over. I am also
thankful for Artem Duplinskiy for helping with the preparation
of the library of scripts.
viii PREFACE
 Part I

Introduction

1
Chapter 1

What you should

already know!

1.1 Linear regression model

Some additional reading material:

1. Wooldrige (2003), “A Modern approach to economet-
rics”
• Chapter 2.1, 2.2, 5.1, 5.2, 10.1, 10.2, 11.1 and 11.2
2. Heij et al. (2004), “Econometric methods with applica-
tions ...”
• Chapter 1.3.1, 2.1, 4.1.3, 4.1.4, 4.3.1 – 4.3.3, 7.1.1
– 7.1.3 and 7.2.2
3. Stock and Watson. (2007), “Introduction to Economet-
rics”
• Chapter 4.1, 4.2, 14.1–14.3, 17.2 and 17.3
4. Verbeek. (2004), “A guide to modern econometrics”
• Chapter 2.1, 2.2, 2.6.1, 2.6.2, 6.1, 8.1, 8.2 and 8.6.2

In your introductory econometrics courses, you have most likely

learned about regression. In particular, by now, you should be
familiar with the linear regression model

yt = α + βxt + εt

where yt is the dependent or endogenous variable, xt is the

independent or explanatory variable, εt is the error term or

3
4 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

innovation, and α and β are the fixed unknown parameters

typically called intercept and slope respectively.

You should also remember that when the error term εt satis-
fies the assumption that E(εt |xt ) = 0, then the linear regression
model is a model of the conditional expectation of yt given xt .

Assumption 1.1 (Exogeneity) E(εt |xt ) = 0 ∀ t ∈ N.

In many ways, Assumption 1.1 gives meaning to the regression.

Without it, the regression equation cannot be interpreted as the
conditional expectation of yt given xt . Indeed, recall that when
E(εt |xt ) = 0 holds, we have,
E(yt |xt ) = E(α + βxt + εt |xt )
= α + β E(xt |xt ) + E(εt |xt )
| {z } | {z }
=xt =0
= α + βxt .

So under Assumption 1.1, the errors εt account for the fact that
the relation between yt and xt holds only “on average”. Under
Assumption 1.1, the linear regression model states essentially
that, on average, the dependent variable yt is linearly related
to the explanatory variable xt . Additionally, the exogeneity
assumption ensures that the parameter β measures the expected
change in yt given a unit change in xt , and the parameter α
measures the expected value of yt when xt = 0.

Let us highlight at this point that the conditional expectation

E(yt |xt ) describes the predictive relation between yt and xt which
emerges from the data. As we shall see, when conducting struc-
tural or causal econometric analysis, we will sometimes be con-
fronted with regressions that do not satisfy this exogeneity as-
sumption. Additional effort will then be made to ensure that we
can still uncover the structural relation between yt and xt . Often
this requires that we build models for different conditional ex-
pectations, such as E(yt |xt , zt ), where zt are additional relevant
regressors correlated with xt , or E(yt |x̂t ) where x̂t is a prediction
1.2. LEAST SQUARES ESTIMATOR 5

for xt obtained from an instrumental regression. We shall dis-

cuss the fundamental difference between predictive modeling and
structural modeling in Chapter 11.

1.2 Least squares estimator

From your introductory econometrics courses you should be
familiar with the famous method of ordinary least squares
(OLS), first published by Legendre in 1806 but discovered by
Gauss in 1795 when he was only 18 years old!

As you may recall, the OLS estimator minimizes the sum of

squared residuals. In other words, the OLS estimator of the un-
known parameter vector (α, β), is defined as the random variable
(α̂T , β̂T ) that satisfies
T
X
(α̂T , β̂T ) = arg min (yt − α − βxt )2 .
(α,β) t=1

The idea of the OLS estimator is thus to estimate the unknown

parameters α and β, by selecting the values α̂T and β̂T that
minimize the sum of the squared residuals. In essence, as shown
in Figure 1.1 we select the line that best fits all the observations.

Figure 1.1: The ‘fitted line’.

6 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

Suppose for a moment that the intercept parameter α is known

to be zero and that we focus only on estimating the unknown
slope parameter β. As you may recall, finding the expression of
the OLS estimator β̂T is very simple! In particular, we can easily
find the value β̂T that minimizes the sum of squared residuals
T
X
β̂T = arg min (yt − βxt )2
β t=1

by taking the derivative of the least squares criterion function

T
X
(yt − βxt )2
t=1

with respect to β and setting it to zero! This is just high-school

mathematics: at the minimum, the derivative must be zero!

Since the derivative takes the form

T
∂ Tt=1 (yt − βxt )2
P X
= −2 (yt − βxt )xt
∂β
t=1

we conclude that the OLS estimator β̂T has the following expres-
sion
T PT
X yt xt
−2 (yt − β̂T xt )xt = 0 ⇔ β̂T = Pt=1
T
. (1.1)
2
t=1 t=1 xt

As we shall see now, the fact that we can derive analytically

expression of the OLS estimator in this simple regression setting
is a big help for understanding the properties of the estimator.

1.3 Least squares: consistency

As you surely remember, the OLS estimator β̂T derived in
equation (1.1) is a random variable because it is a function of
the random data y1 , y2 , ..., yT and x1 , x2 , ..., xT . In particular, as
shown in Figure 1.2, every new sample of data gives us a new
estimate β̂T for the unknown slope parameter β.
1.3. LEAST SQUARES: CONSISTENCY 7

Figure 1.2: Different OLS estimates (β̂, α̂) of the unknown parameters α and
β, obtained from different samples of data generated by the same true model with
α = 0 and β = 1.

You may remember from your introductory econometrics courses

that one of the most fundamental properties that an estimator
should satisfy is the property of consistency. An estimator β̂T is
said to be consistent for a parameter β if it converges in probabil-
ity to the parameter β as the sample size T increases to infinity.
In the definition below, we denote the convergence in probability
p
by the symbol →; see also Appendix B for a review of convergence
modes.

Definition 1.1 ( Consistent Estimator) An estimator β̂T is

p
said to be consistent for the parameter β if and only if β̂T → β
as T → ∞.

Later in this book we will give a precise definition of convergence

in probability and also study other modes of convergence.
However, for now, it suffices to recall that the estimator β̂T
converges in probability to β if the probability that β̂T is close
to β increases to 1 as the sample size T goes to infinity.

In your introductory econometrics courses, you most certainly

learned how to prove that the OLS estimator is consistent when
8 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

the linear regression model is correctly specified. The crucial thing

to remember is that,

If the linear regression model is correctly specified. Then you

know that each yt is exactly given by yt = α + βxt + εt , for some
unknown (α, β) ∈ R2 .

Assumption 1.2 (Correct specification) The linear regres-

sion model yt = α + βxt + εt is correctly specified.

Consider again the case where α = 0. Then, under Assumption

1.2, we can substitute yt by βxt + εt in the expression of the OLS
estimator and obtain
PT PT
t=1 yt xt t=1 (βxt + εt )xt
β̂T = PT = PT
2 2
t=1 xt t=1 xt

Re-arranging this expression gives

PT 2
PT PT
t=1 βxt t=1 εt xt x t εt
β̂T = PT + PT = β + Pt=1
T
2 2 2
t=1 xt t=1 xt t=1 xt

Re-writing the OLS estimator in this way

PT
t=1 xt εt
β̂T = β + T
x2
|{z} P
true parameter | t=1{z t }
remainder term

is very useful because now we can obtain the desired consistency

p
β̂T → β as long as we can show that the remainder term vanishes
to zero as the sample size grows,
PT
x t εt p
Pt=1
T 2
→ 0 as T → ∞.
t=1 xt

Showing that the remainder term vanishes to zero is very easy!

We just have to apply a Law of Large Numbers. For example,
if the data is independent and identically distributed (iid), then
we can use the following law of large numbers proven by Ja-
cob Bernoulli in his masterpiece “Ars Conjectandi” published in
1713.
1.3. LEAST SQUARES: CONSISTENCY 9

Theorem 1.1 (Bernoulli’s Law of Large Numbers) Let

z1 , ..., zT be iid random variables with finite first moment
E|z1 | < ∞. Then we have
T
1X p
zt → E(zt ) as T → ∞.
T
t=1

The law of large numbers tells us that regardless of the distribu-

tion of thePvariables z1 , ..., zT , we know that the sample average
z̄ = (1/T ) Tt=1 zt is a consistent estimator of the mean E(zt ) as
long as the random variables are iid and have bounded moments.
The law of large numbers is one of the most important results in
statistics, and a remarkable piece of mathematics! This much cele-
brated theorem was first stated without proof by the italian mathe-
matician Gerolamo Cardano in the 1500’s. The first proof of the law
of large numbers appeared in Jacob Bernoulli’s master piece “Ars
Conjectandi”, published in 1713, eight years after his death. Jacob
Bernoulli is famous for having taken more than 20 years to com-
plete the proof of the law of large numbers. Bernoulli himself was
very proud of being the first to give a rigorous proof of the law of
large numbers and named this theorem the Golden Theorem, but
today it is known as Bernoulli’s Theorem.1 The term ‘Law of Large
Numbers’ was coined in 1837 by the French mathematician Simeon
Poisson. A simpler proof was obtained by the great russian mathe-
matician Pafnuty Chebyshev in 1874 using an unproved inequality.
Finally, his student Andrey Markov wrote a complete proof in his
doctoral thesis in 1884.

In order for the law of large numbers to be useful in our problem,

we must first multiply the numerator and the denominator of the
remainder term by 1/T to obtain
PT 1 PT
x t εt t=1 xt εt
Pt=1
T 2
= T
1 PT 2
t=1 xt T t=1 xt

By Theorem 1.1, we can apply the law of large numbers to both

the numerator and the denominator separately, as long as the
sequence {(xt , et )} is iid with finite first moments.
1
Another theorem known as Bernoulli’s Theorem in physics is actually named
after Jacob Bernoulli’s nephew, Daniel Bernoulli, a great mathematician himself.
10 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

Assumption 1.3 (Random samples and bounded moments)

The random sequence {(xt , et )} is iid with bounded moments
E|xt εt | < ∞ and 0 < Ex2t < ∞.

Note that the assumption that the vector sequence {(xt , et )} is

iid implies that both {xt et } and {x2t } are also iid sequences. Un-
der Assumption 1.3 we can thus apply Bernoulli’s law of large
numbers to both sample averages and obtain,
T T
1X p 1X 2 p
xt εt → E(xt εt ) and xt → E(x2t ) as T → ∞.
T T
t=1 t=1

Additionally, note also that Assumption 1.1 ensures that

E(xt εt ) = 0 since, by the law of total expectation, we have



E(xt εt ) = E E(xt εt |xt ) = E xt E(εt |xt ) = E(xt · 0) = 0.

As a result, if Ex2t > 0, the continuous mapping theorem ensures

that,2
PT
xt εt p E(xt εt ) 0
Pt=1 → 2 = = 0 as T → ∞.
T 2
t=1 xt
Ext Ex2t

We can thus conclude that the OLS estimator β̂T is consistent

for β because
PT
x t εt p
β̂T = β + Pt=1
T
→ β + 0 = β.
2
t=1 xt

We have thus proven the following theorem.

2
The condition Ex2t > 0 is needed to ensure that the fraction is continuous
p
and we obtain 1 PT1 x2 → Ex1 2 .
T t=1 t t
1.4. LEAST SQUARES: NORMALITY 11

Theorem 1.2 (Consistency of OLS) Let the observed data

{yt } and {xt } be obtained from the correctly specified
(Assumption 1.2) linear regression model

yt = βxt + εt ∀ t ∈ N.

Assume that the regressor is exogenous (Assumption 1.1) and

that {xt et } and {x2t } are iid sequences with finite first moment
E|xt εt | < ∞ and 0 < Ex2t < ∞ ( implied by Assumption 1.3).
Then the OLS estimator β̂T is consistent for β
p
β̂ → β as T → ∞.

1.4 Least squares: normality

As you may remember from your introductory courses, econo-
metricians are often interested in conducting statistical inference
on the unknown parameter β.

When the distribution of the estimator β̂T is known under

some null hypothesis (for example H0 : β = 4), then we can
calculate the probability of observing a certain estimate (for
example β̂T = 3.5) under the null hypothesis. If the probability
of observing the estimate β̂T = 3.5 under the null hypothesis
that β = 4 is very low, then we reject H0 . If the probability of
observing β̂T = 3.5 under the null hypothesis H0 is high, then
we do not reject H0 . In other words, we use the p-value to
conduct inference.

This way of conducting inference is much more intuitive that may

seem at first. Actually, it is the way we make sense of the world in
our every day lives! If from 100 coin tosses we obtained 98 heads and
2 tails, then most of us would find it extremely hard to believe that
the coin is fair. The reason behind this is simple: if the coin is fair,
then the probability of observing 98 heads and 2 tails is incredibly
small! So observing the outcome of 98 heads and 2 tails should
lead us to reject the hypothesis that the coin is fair. In contrast,
an outcome of 46 heads and 54 tails has much higher probability
and should not lead us to reject the hypothesis that the coin is fair.
Clearly, the main ingredient needed for this type of inference, is to
know the distribution of events under the null hypothesis that we are
trying to test.
12 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

In most cases, the distribution of the estimator is not known

exactly, but we can derive an ‘approximate’ distribution. For
example, when data is iid, then we can use a central limit theorem
to show that the estimator is approximately normally distributed
d
in large samples. Below, the symbol → denotes convergence in
distribution.3

Definition 1.2 (Asymptotically normal estimator) An esti-

mator β̂T is said to be asymptotically normal for the parameter
β if and only if
√
d
T β̂T − β → N (0, σ 2 ) as T → ∞.

The variance σ 2 of the asymptotic normal distribution is

called the asymptotic variance of the estimator β̂T .

In this simple well specified linear regression model, it is very

easy to show that the OLS estimator is asymptotically normal.
We just have to apply a Central Limit Theorem!

Theorem 1.3 (Lindeberg-Levys Central Limit Theorem)

Let z1 , ..., zT be iid random variables with E(z1 ) = µ and
Var(z1 ) = σ 2 < ∞. Then
√ 1 X T 
d
T zt − µ → N (0, σ 2 ) as T → ∞.
T
t=1

The central limit theorem tells us that, regardless of the distribu-

tion of the data, the appropriately standardized sample average
converges in distribution to a normal random variable as the sam-
ple size grows to infinity. All we need is for the random variable
of interest to be iid and have two bounded moments.
The central limit theorem is an absolutely remarkable result! It is,
together with the law of large numbers, one of the most important
results in probability and statistics! The french mathematician Abra-
ham de Moivre provided a remarkable first proof of the central limit
3
We will define this mode of convergence appropriately at a later stage.
1.4. LEAST SQUARES: NORMALITY 13

theorem in 1733. Unfortunately, the result passed unnoticed to ev-

eryone else. Pierre-Simon Laplace rediscovered the central limit the-
orem in 1812 in his Theorie Analytique des Probabilites. Finally,
Aleksandr Lyapunov brought the theorem to fame in 1901! In 1934,
the great Alan Turing, the founder of computer science, dedicated his
doctoral dissertation to finding a new proof of the central limit the-
orem, only to discover (too late) that Lindeberg had already written
a similar proof in 1922.

We can now use the central limit theorem to obtain the asymp-
totic normality of our OLS estimator β̂T when the linear regres-
sion model is correctly specified. First
√ we use the correct speci-
fication Assumption 1.2 to rewrite T (β̂ − β) as
PT PT
√ √  √
t=1 xt εt x t εt

T (β̂T − β) = T β + PT − β = T Pt=1 T
2 2
t=1 xt t=1 xt

Second, we multiply the numerator and denominator of the re-

mainder term by 1/T to obtain

√ √ 1 Tt=1 xt εt
P
T
T (β̂T − β) = T 1 PT
2
T t=1 xt
√  
T T1 Tt=1 xt εt − E(xt εt )
P
= 1 PT
.
2
T t=1 xt

Note that we were allowed to subtract the term E(xt εt ) in the last
equality above, because we have already seen that Assumption
1.1 implies that E(xt εt ) = 0. Finally, we can apply a central
limit theorem to the numerator and a law of large numbers to
the denominator as long as E|xt εt | < ∞.

Assumption 1.4 (Additional moment) E|xt εt |2 < ∞.

Armed with our new moment assumption we obtain,

T
!
√ 1X d
 
T xt εt − E(xt εt ) → N 0, σε2 E(x2t ) as T → ∞,
T
t=1

T
1X 2 p
xt → E(x2t ) as T → ∞.
T
t=1
14 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

Finally, we can combine these two limit results by appealing to

Slutsky’s Theorem to obtain4
PT
√1
√ t=1 xt εt


T N 0, Var(xt εt )
d
T (β̂T − β) = →
Ex2t
1 PT 2
T t=1 xt


N 0, Var(xt εt ) d   1 2 
where naturally = N 0, Var(x t ε t ) .
Ex2t E(x2t )

The expression for the asymptotic variance can be further simpli-

fied when the error term εt and the regressor xt are independent.

Assumption 1.5 (Stricter form of exogeneity) εt is inde-

pendent of xt for every t.

Recall that the product xt εt has mean zero, and hence that
Var(xt εt ) = E(xt εt )2 . Now, under Assumption 1.5 we have that
Var(xt εt ) = Ex2t Eε2t = σε2 Ex2t , and hence, we obtain the following
simplification
PT
√1
√ t=1 xt εt
 −1 
T d
→ N 0 , σε2 E(x2t )

T (β̂T − β) = 1 PT .
2
T t=1 xt

We have thus proven the following theorem.

4
The theorem proven by the famous econometrician Eugene Slutsky in 1925
d p d
states that if zt → z and wt → w, then zt · wt → z · w.
1.4. LEAST SQUARES: NORMALITY 15

Theorem 1.4 (Asymptotic Normality of OLS) Let the ob-

served data {yt } and {xt } be obtained from the correctly spec-
ified ( Assumption 1.2) linear regression model

yt = βxt + εt ∀ t ∈ N.

Assume that εt is independent of xt for all t (Assumption 1.5),

and that {xt et } and {x2t } are iid sequences with finite
first moment E|xt εt | < ∞ and 0 < E|x2t | < ∞
( implied by Assumption 1.3). Suppose further that {xt et } has fi-
nite variance Var(xt et ) = σ 2 < ∞ (implied by Assumption 1.4).
Then the OLS estimator β̂T is asymptotically normal for β
√ d
 −1 
T (β̂T − β) → N 0, σε2 E(x2t )

as T → ∞.

In practice, since the true expectation E(x2t ) isPunknown, we must

approximate it by its sample counterpart T1 Tt=1 x2t . Hence we
√
obtain the following approximate distribution for T (β̂T − β)

√ T
approx
 1 X 2
−1 
T (β̂T − β) ∼ N 0 , σε2 xt .
T
t=1

Re-arranging the expression gives us the approximate distribu-

tion of the OLS estimator β̂T
 T
approx X
−1 
β̂T ∼ N β, σε2 x2t .
t=1

This approximate distribution is extremely useful as it allows us

to conduct inference on the true unknown parameter! In par-
ticular, it is useful because now we can calculate the probability
of observing a certain estimate β̂T under a given null hypothesis
H0 for the parameter β. Suppose for example that you obtain
an estimate for the parameter of β̂T = 3.5. Should we believe
the hypothesis H0 : β = 0 that the true unknown parameter β is
equal to zero? Well, if the probability of observing β̂T = 3.5 when
β = 0 is very low, then we will reject H0 because it seems un-
reasonable. If the probability is quite high, then the hypothesis
seems reasonable and we do not reject it!
16 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

1.5 Linear dynamic model

In your introductory econometrics courses you surely learned
about linear time-series models. Most probably, you studied
linear time-series models like the autoregressive model of order
1, also called the AR(1) model. Recall that a random sequence
{xt }t∈Z is said to be generated by a linear Gaussian AR(1) model
if it is given by,

xt = φxt−1 + εt ∀t∈Z

where the innovation sequence {εt }t∈Z is an iid normally dis-

tributed sequence with εt ∼ N (0, σε2 ) for every t.

Note that data generated by this AR(1) model satisfies a very

strict form of exogeneity. Namely, that the error term εt is
independent of the regressor xt−1 at every time t. This is true
in particular, for every so-called causal solution of the AR(1)
model. Indeed, since xt−1 is a function of past error terms
xt−1 = f (εt−2 , εt−3 , εt−4 , ...), and εt is independent of εt−j for
all j > 0, we obtain naturally that xt−1 is itself independent
of εt . From an estimation standpoint, independence is not
needed, but we require at least an exogeneity assumption such
as E(εt |xt−1 ) = 0 ∀ t ∈ Z, if we wish to uncover the causal
relation between xt−1 and xt .

Linear autoregressive models like the AR(1) are very useful in

modeling the temporal dependence that we usually observe in
economic and financial time-series. In the AR(1) model, this de-
pendence can be well understood by noting that the distribution
of xt depends on the value of xt−1 . In particular, if we have
a sequence of T random variables x1 , ..., xT generated from an
AR(1). Then it is clear that the distribution of x2 conditional on
x1 is given by5

x2 |x1 ∼ N (φx1 , σε2 ).

5
Indeed, remember that if we condition on the value of xt−1 , then xt−1 is just
a constant. So xt is just a constant φxt−1 plus a normal random variable εt ∼
N (0, σε2 ). Hence, the distribution of xt conditional on xt−1 is simply N (φxt−1 , σε2 ).
1.5. LINEAR DYNAMIC MODEL 17

Similarly, the distribution of x3 conditional on x2 is given by

x3 |x2 ∼ N (φx2 , σε2 ). In fact, we have that the distribution of any
xt conditional on xt−1 is given by

xt |xt−1 ∼ N (φxt−1 , σε2 ).

You may also remember from your introductory courses that a

time series {xt }t∈Z is stationary if its unconditional distribution is
invariant in time. Figure 1.3 shows a typical path of a time-series
generated by an AR(1) model with time-varying conditional dis-
tribution. Observing this single path could lead us to think that

Figure 1.3: Single path [above] shows time-varying conditional mean. Multiple
paths [below] show invariance of the distribution (mean and variance are clearly
constant over time).

the unconditional mean is slowly changing over time, with long

periods where it is persistently below zero, and other periods of
time when it remains above zero. However, the plot with multiple
paths reveals that unconditional mean, unconditional variance
and other unconditional features of the distribution are actually
time-invariant. In other words, it suggests a stationary sequence
with time-invariant unconditional distribution. The word ‘uncon-
ditional’ is crucial here! The conditional distribution of {xt }t∈Z
18 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

may obviously change! In fact, we have already seen that the

distribution of xt conditional on xt−1 is changing over time since

xt |xt−1 ∼ N (φxt−1 , σε2 ).

In particular, note that the conditional mean of xt is φxt−1

whereas the conditional mean of xt+1 is φxt . Figure 1.3 shows
however that the unconditional properties of the data are
time-invariant! Just like we must toss a coin multiple times
to evaluate if its distribution is fair, we must also draw multi-
ple paths of a time-series to evaluate its distributional properties.

With this in mind, we now recall the definition of strict sta-

tionarity that you may have encountered in your undergraduate
d
studies. In the definition below, the symbol = denotes equality
in distribution.

Definition 1.3 (Strict Stationarity) A time-series {xt }t∈Z is

said to be strictly stationary if the distribution of every finite
sub-vector is invariant in time
d
(x1 , ..., xh ) = (xt+1 , ..., xt+h ), ∀ (t, h) ∈ N × N.

You also learned that the linear Gaussian AR(1) model is strictly
stationary as long as |φ| < 1. In other words, the Gaussian AR(1)
is stationary as long as it does not exhibit ‘too much’ temporal
dependence.

Theorem 1.5 Let {xt }t∈Z be a time-series generated by the

linear Gaussian AR(1) model

xt = φxt−1 + εt ∀t∈Z

with |φ| < 1 and innovations {εt }t∈Z that are NID(0, σε2 ).
Then {xt }t∈Z is strictly stationary.

This stationarity property of the time-series is important for un-

derstanding the properties of estimators because it allows us to
make use of laws of large numbers and central limit theorems.
1.6. MAXIMUM LIKELIHOOD ESTIMATOR 19

1.6 Maximum likelihood estimator

As you surely remember, the famous method of maximum
likelihood first introduced by Gauss in “Theoria motus corporum
coelestium in sectionibus conicis solem ambientium” in 1809, is
one of the most important statistical discoveries of all times!

Given T observations x1 , ..., xT from the time-series {xt }t∈Z ,

the likelihood function f (x1 , ..., xT ; θ) is simply the joint density
function of the vector (x1 , ..., xT ) under the parameter vector
θ = (φ, σε2 ) that defines the distribution of the data. The only
difference between the joint density and the likelihood is that the
joint density is seen as a function of the data x1 , ..., xT , whereas
the likelihood function is seen as a function of the parameter
vector θ. The maximum likelihood (ML) estimator θ̂ T of the un-
known parameter θ is defined as the maximizer of the likelihood
function
θ̂ T = arg max f (x1 , ..., xT ; θ).
θ

Since a joint density function can always be factorized into the

product of conditional and marginal densities,

f (x1 , x2 ; θ) = f (x1 ; θ) × f (x2 |x1 ; θ),

f (x1 , x2 , x3 ; θ) = f (x2 , x1 ; θ) × f (x3 |x2 , x1 ; θ)
= f (x1 ; θ) × f (x2 |x1 ; θ) × f (x3 |x2 , x1 ; θ),
T
Y
f (x1 , ..., xT ; θ) = f (x1 ; θ) × f (xt |xt−1 , ...x1 ; θ).
t=2

it follows that we can re-write the likelihood function as

T
Y
f (x1 , ..., xT ; θ) = f (x1 ; θ) × f (xt |xt−1 , ...x1 ; θ).
t=2

Re-writing the likelihood as a product of conditional densities is

very useful because we often know the distribution of xt condi-
tional on xt−1 . Indeed, as we saw above, in the linear Gaussian
AR(1) model we have that

xt |xt−1 ∼ N (φxt−1 , σε2 ).

20 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

This means that the density f (xt |xt−1 , ...; θ) is given by the well
known formula
f (xt |xt−1 , ...x1 ; θ) = f (xt |xt−1 ; θ)
1 h (x − φx )2 i
t t−1
= p exp − .
2πσε2 2σε2

As a result, taking the logarithm of the likelihood, we have that

the ML estimator θ̂ T satisfies6
T
X p (xt − φxt−1 )2
θ̂ T = arg max − log 2πσε2 − .
θ 2σε2
t=2

Finding the maximum of the likelihood function is easy! We

simply take the derivative and set it to zero!

Suppose for simplicity that σε2 = 1. Then θ = φ and the deriva-

tive of the likelihood takes the form
T
∂ log f (x1 , ..., xT ; φ) X
= (xt − φxt−1 )xt−1
∂φ
t=2

and since, by construction, φ̂T satisfies,

T
∂ log f (x1 , ..., xT ; φ̂T ) X
=0 ⇔ (xt − φ̂T xt−1 )xt−1 = 0
∂φ
t=2

it follows that
PT
xt xt−1
φ̂T = Pt=2
T
.
x 2
t=2 t−1

The ML estimator φ̂T is the same as the OLS estimator of φ.

You can verify this yourself! Having the analytical expression of
the ML estimator φ̂T is very useful as it becomes very easy to
study the properties of the estimator.
6
Note that we start at t = 2 and simply treat x1 as a fixed value because x0 is
unknown. We call this conditional ML by error decomposition. If we assume a dis-
tribution for x1 then we call it ML by error decomposition with exact initialization.
1.7. MAXIMUM LIKELIHOOD: CONSISTENCY 21

1.7 Maximum likelihood: consistency

Similarly to the consistency of the OLS estimator, we can now
obtain the consistency of the ML estimator by appealing to the
correct specification of the model and a law of large numbers.

Assumption 1.6 (Correct specification) The Gaussian lin-

ear AR(1) model defined as xt = φxt−1 + εt , where {εt } is an
iid sequence of Gaussian random variables xt ∼ N (0, σε2 ) with
0 < σ 2 < 1, is correctly specified.

If the linear AR(1) model is correctly specified, then we can sub-

stitute xt by φxt−1 + εt in the expression of the ML estimator to
obtain
PT PT
t=2 xt xt−1 t=2 (φxt−1 + εt )xt−1
φ̂T = PT = PT
2 2
t=2 xt−1 t=2 xt−1
PT
εt xt−1
= φ + Pt=2 T
.
2
t=2 xt−1

Multiplying both the numerator and denominator of the remain-

der term by 1/T we obtain
PT 1 PT
εt xt−1 t=2 εt xt−1
Pt=2
T 2
= T
1 PT 2
.
t=2 xt−1 T t=2 xt−1

Finally, we can show that

1 PT
T t=2 εt xt−1 p E(εt xt−1 ) 0
→ 2 = = 0
E(x2t )
1 PT 2
T t=2 xt−1
E(xt )

by applying a law of large numbers to both the numerator and

the denominator. This time however, the data {xt }t∈Z is not iid.
Hence, we cannot use Bernoulli’s law of large numbers stated
in Theorem 1.1. Instead, since {xt }t∈Z is strictly stationary for
|φ| < 1 we will make us of the following law of large numbers for
strictly stationary data proved by George Birkhoff in 1931.
22 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

Theorem 1.6 (Birkhoff - Khinchin Theorem) Let {zt }t∈Z be

a strictly stationary and ergodic random sequence with finite
first moment E|z1 | < ∞. Then we have
T
1X p
zt → E(zt ) as T → ∞.
T
t=2

Using this law of large numbers we obtain the consistency of the

ML estimator in the strictly stationary (ensured by |φ| < 1),
correctly-specified, linear Gaussian AR(1) model since
PT
εt xt−1 p
φ̂T = φ + Pt=2 T
→ φ + 0 = φ as T → ∞.
2
t=2 xt−1

Note that Eεt xt−1 = 0 holds since εt is independent of xt−1 has

mean zero. Indeed, recall the assumption of correct specification
of the AR(1) model stated above ensures that the error term εt
is independent of the regressor xt−1 at every time t, since xt−1 is
a function of past error terms xt−1 = f (εt−2 , εt−3 , εt−4 , ...), and
{εt } is iid.

We have thus proven the following theorem.

Theorem 1.7 (Consistency of ML Estimator) Let {xt }t∈Z be

generated by a strictly stationary (|φ| < 1), correctly-specified
(Assumption 1.7), linear Gaussian AR(1) model

xt = φxt−1 + εt ∀t∈Z

with innovations that satisfy {εt }t∈Z ∼ NID(0, σε2 ) with 0 <
p
σε2 < ∞ Then, φ̂T → φ as T → ∞.

Note that the consistency theoremPabove imposes P sufficient con-

1 1
ditions for the sample averages T xt−1 εt and T x2t−1 to sat-
isfy the law of large numbers in Theorem 1.6. In particular, it is
easy to show that when {xt }t∈Z is strictly stationary and ergodic,
then the sequences {xt−1 εt }t∈Z and {x2t }t∈Z are also strictly sta-
tionary and ergodic. Furthermore, it is also easy to show that
1.8. MAXIMUM LIKELIHOOD: NORMALITY 23

E|xt−1 εt | < ∞ and E|xt |2 < ∞ as long as σε2 < ∞. Hence the
conditions required by Theorem 1.6Pto apply the law P of large
numbers to the sample averages T1 xt−1 εt and T1 x2t−1 are
easily satisfied!

1.8 Maximum likelihood: normality

Proving the asymptotic normality of the ML estimator φ̂T is easy
in the setting of a strictly-stationary, correctly-specified, linear
Gaussian AR(1) model. First, we use the correct-specification
√
assumption to substitute xt by φxt−1 +εt and re-write T (φ̂T −φ)
as
PT PT
√ √  √
t=2 xt−1 εt xt−1 εt

T (φ̂T − φ) = T φ + PT − φ = T Pt=2 T
.
2 2
t=2 xt−1 t=2 xt−1

Second, we multiply both the numerator and denominator of the

remainder term by 1/T to obtain
PT
√ √ 1
t=2 xt−1 εt
T
T (φ̂T − φ) = T 1 PT 2
T t=2 xt−1
√  1 PT 
T T t=2 xt−1 εt − E(xt−1 εt )
= 1 PT
.
x 2
T t=2 t−1

Finally, we apply a central limit theorem to the numerator and

a law of large numbers to the denominator. Since the data is
no longer iid, we make use of the following central limit theorem
for strictly stationary and ergodic martingale difference sequence
proved by Billingsley in 1961.

Definition 1.4 (Martingale difference sequence) A sequence

{zt }t∈Z is said to be a martingale difference sequence if and
only if E(zt |zt−1 , zt−2 , ...) = 0 ∀ t ∈ Z.
24 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!

Theorem 1.8 (Billingsley’s Central Limit Theorem) Let

{zt }t∈Z be a strictly stationary and ergodic martingale differ-
ence sequence of random variables with E(z1 ) = µ, Var(z1 ) =
σ 2 < ∞. Then
√ 1 X T 
d
T zt − µ → N (0, σ 2 ) as T → ∞.
T
t=2

Application of the central limit theorem to the numerator and a

law of large numbers to the denominator yields7
T
!
√ 1X d
 
T xt−1 εt − E(xt−1 εt ) → N 0, σε2 E(x2t−1 ) ,
T
t=2

T
1X 2 p
and xt−1 → E(x2t−1 ) as T → ∞.
T
t=2

Hence, by Slutsky’s theorem we obtain

PT
√1
√ t=2 xt−1 εt N 0, σε2 E(x2t−1 )


T d
T (φ̂T − φ) = →
E(x2t−1 )
1 PT 2
T t=2 xt−1
0, σε2 E(x2t−1 )


N  −1 
N 0 , σε2 E(x2t−1 )

and ∼ .
E(x2t−1 )

Recall that the correct specification assumption ensures that εt

and xt−1 are independent, so we have that Var(xt εt ) = σ 2 Ex2t .
Furthermore, the assumption of stationarity |φ| < 1 and Gaus-
sian errors ensures all the necessary bounded moments for εt and
xt . We have thus proven that the following theorem holds.
7
To see that xt−1 εt is a martingale difference sequence just define Ft−1 :=
(xt−2 εt−1 , xt−3 εt−2 , ...) and apply the law of total expectation to obtain



E xt−1 εt |Ft−1 = E E xt−1 εt |xt−1 , Ft−1 Ft−1


= E xt−1 E εt |xt−1 , Ft−1 Ft−1



= E xt−1 E(εt ) Ft−1 = E xt−1 · 0|Ft−1 = 0.
1.8. MAXIMUM LIKELIHOOD: NORMALITY 25

Theorem 1.9 (Asymptotic Normality of ML Estimator) Let

{xt }t∈Z be generated by a strictly stationary (|φ| < 1),
correctly-specified (Assumption 1.7), linear Gaussian AR(1)
model
xt = φxt−1 + εt ∀ t ∈ Z
with innovations that are iid Gaussian {εt }t∈Z ∼ N ID(0, σε2 )
with σε2 < ∞ Then,
√ d
 −1 
T (φ̂T − φ) → N 0 , σε2 E(x2t−1 )

as T → ∞.

With this, we conclude Chapter 1. Having completed our recap

of the material of introductory econometrics, we are now ready
to set sail into the unknown! In Chapter 2, we shall discuss the
limitations of the introductory econometrics material covered in
this first chapter, and highlight the challenges that lie ahead.
26 CHAPTER 1. WHAT YOU SHOULD ALREADY KNOW!
Chapter 2

The challenges ahead...

The results obtained in Chapter 1 for the least squares and max-
imum likelihood estimators were only possible because the mod-
els we considered were sufficiently simple and assumed to be cor-
rectly specified. If we consider more general nonlinear models and
allow for the possibility of model mis-specification, then things
become more complicated. In particular, it is important to note
that:

1. the linearity of the models was crucial for being able to

obtain an analytical expression for the estimator;

2. the linearity of the model was important in establishing the

strict stationarity of the time-series model;

3. the correct specification assumption was essential for re-

writing the estimator in deviations from the true parame-
ter.

2.1 Intractable estimators

As we shall now see, nonlinear models typically lead to estimators
that are analytically intractable. This essentially means that its
is difficult or impossible to find a ‘mathematical formula’ for the
estimator. From a practical point of view, the estimation can be
easily carried out using numerical methods. For example, you
can use a simple gradient-based algorithm to find the minimum

27
28 CHAPTER 2. THE CHALLENGES AHEAD...

of the least squares function or the maximum of the likelihood

function. However, from a theoretical perspective, the absence
of an expression for the estimator means that we cannot analyze
the properties of the estimator using the same methods as we did
in Chapter 1.

2.2 Nonlinear regression

Economic theory often tells us that economic variables are nonlin-
early related between them. A simple example is the AK produc-
tion function with additive errors where output yt is a function
of capital xt according to

yt = αxβt + εt

where α > 0 and β ∈ (0, 1) are unknown parameters, and εt

is an additive error term.1 Suppose for simplicity that α = 1.
Then, the least squares estimator β̂T of the unknown parameter
β that defines the curvature of the production function (hence
the marginal productivity of capital) is defined as
T

2
yt − xβt .
X
β̂T = arg min
β t=1

Until now everything is fine. However, we will run into trouble if

we try to derive the analytic expression of the estimator. Indeed,
setting the derivative of the least-squares function to zero yields
T
− xβ̂t T )2
PT
∂ t=1 (yt
(yt − xβ̂t T )xβ̂t T log(xt ) = 0.
X
=0 ⇔ −2
∂ β̂T t=1

Can you solve this equation for β̂T ? It will be very difficult, and
many times even impossible! In the linear regression setting that
1
Note that the AK production function with multiplicative errors, which is given
by
yt = αxβ
t εt
can be re-written in linear form by taking logs since one obtains

log yt = a + b log xt + ut

where a = log α, b = β and ut = log εt . However, this linear transformation does

not work for an AK production function with additive errors.
2.3. NONLINEAR DYNAMIC MODELS 29

we covered in Chapter 1 it was easy to solve for the estimator

β̂T . But now it is not!

It is often very difficult or impossible to derive the

analytic expression of the estimator in nonlinear re-
gression models!

Note that from a practical point of view, the absence of an

expression for the estimator does not prevent us from obtaining
point estimates for the parameters. Indeed, we can always find
the point that minimizes the least squares function numerically,
and we will have our estimate β̂T . But how reliable is this
point estimate? We can only answer this question if we know
the properties of the estimator. But this raises a fundamen-
tal question: How can we study the stochastic properties of
our estimator without knowing its analytical expression? With-
out a formula for the estimator, how can we study its properties?

Finally, please keep in mind that the AK production function is

obviously not the only example of a nonlinear relation between
economic variables. On the contrary, nonlinear relations are per-
vasive in economics in both theory and in practice.

2.3 Nonlinear dynamic models

The problem with nonlinear time-series models is similar to that
of the nonlinear regression. In particular, it will become difficult
(or even impossible) to obtain nonlinear expressions for the esti-
mator. Consider the self-excited smooth-transition autoregressive
(SESTAR) model of Terasvirta and Anderson (1992)
γ
xt = xt−1 + εt ∀t∈Z
1 + exp(α + βxt−1 )

where γ, α and β are unknown parameters and {εt }t∈Z is Gaus-

sian iid sequence with εt ∼ N (0, σε2 ) for every t. Note that this
nonlinear autoregressive model can be re-written as
γ
xt = φt−1 xt−1 +εt with φt−1 = ∀ t ∈ Z.
1 + exp(α + βxt−1 )
30 CHAPTER 2. THE CHALLENGES AHEAD...

Unrelated Random Variables Linear Relationship

4 20

2 10

0 0

−2 −10

−4 −20
−10 0 10 −10 0 10

Quadratic Relationship Exponential Relationship

40 40

20 20

0 0

−20 −20
−10 0 10 −10 0 10

Figure 2.1: Scatter plots with linear and nonlinear relations between variables

The SESTAR model is useful to describe for example the changes

in the dependence of economic macroeconomic aggregates (like
GDP, aggregate consumption, or aggregate investment) that
occur over the business cycle. In particular, it is useful to
capture the fact that temporal dependence in these variables
tends to be higher in recessions than expansions. Hence, the
autoregressive coefficient φt−1 tends to be larger (closer to 1) in
recessions than expansions.

This model is more general than the linear AR(1) model consid-
ered in Chapter 2 because it ‘contains’ the linear AR(1) model
as a special case. In particular, for α = 0 and β = 0 we obtain

xt = φxt−1 + εt with φ = γ/2 ∀ t ∈ Z.

In econometric terminology we say that the SESTAR model

nests the AR(1) model.
2.3. NONLINEAR DYNAMIC MODELS 31

In practice, estimation of the parameters of the SESTAR model is

not difficult since we know that the distribution of xt conditional
on xt−1 is given by
!
γ
xt |xt−1 ∼ N xt−1 , σε2 .
1 + exp(α + βxt−1 )

and hence we can use numerical methods to maximize the log

likelihood function which takes the form
T
X p 1  γ 2
− log 2πσε2 − 2 xt − xt−1 ,
2σ 1 + exp(α + βxt−1 )
t=2

However, as before we are faced with the problem of finding an

analytic expression for the ML estimator. Taking the derivative
of the likelihood function with respect to the parameters of inter-
est and equating to zero (as we did in the AR(1) model) leads to
complicated expressions that are difficult to solve for the estima-
tor. We are thus left again with the following question: without
an analytic expression for the estimator how can we analyze its
properties?
A wide class of other models in econometrics, machine
learning, engineering and big data analytics, typically lead to
analytically intractable estimators. The examples range from
dynamic nonparametric and semi-nonparametric models in
econometrics, to nonlinear classifiers and recurrent artificial
neural networks in machine learning and big data analytics.
Time-varying parameter models like the local level model in
Durbin and Koopman (2012), or the autoregressive conditional
heteroeskedasticity model (GARCH) of Engle (1982) and Boller-
slev (1986) also lead to intractable estimators. They are simple
to estimate by maximum likelihood, since it is easy to maximize
the likelihood using numerical methods. However, the estimators
do not have a closed form analytical expression.

As you probably know already, the GARCH model is used to

filter a time-varying volatility σt2 . This model is very useful in
empirical finance to keep track of changes in the volatility of
32 CHAPTER 2. THE CHALLENGES AHEAD...

financial markets. Specifically, the GARCH models a series of

returns {yt }t∈Z as
yt = σt εt ∀ t ∈ Z
where {εt }t∈Z is an iid standard Gaussian sequence εt ∼ N (0, 1),
and {σt }t∈Z is a time-varying volatility that is filtered using the
update equation,
2
σt+1 = ω + αyt2 + βσt2 .
The parameters ω, α and β are unknown and must be estimated
from the sample of observed data. In practice, estimation is easy
since we can easily maximize the likelihood function numerically.
However, there exists no analytical expression for the estimators
ω̂T , α̂T and β̂T .

2.4 Nonlinear dynamics and stationarity

Another important difficulty that econometricians are confronted
with when using nonlinear models is that of understanding the
stochastic properties of the time-series.

Consider again the linear AR(1) model

xt = φxt−1 + εt ∀ t ∈ Z.

In your introductory econometrics courses you surely learned that

the AR(1) model is stable if |φ| < 1, and hence {xt }t∈Z is sta-
tionary when {εt }t∈Z is a white noise sequence or a mean-zero
iid sequence. At the time, the stationarity results were obtained
by unfolding the process in time to obtain the following repre-
sentation, which holds for |φ| < 1,
∞
X
xt = φj εt−j .
j=0

With this expression at hand, it is easy to calculate uncondi-

tional means, variances and covariances and, hence, easy show
that the sequence is weakly stationary if |β| < 1 and {εt } is
white noise. This expression also allowed us to conclude that
{xt } is strictly stationary as xt is a measurable function of an
2.4. NONLINEAR DYNAMICS AND STATIONARITY 33

iid sequence {εt }.

You probably also learned in introductory econometrics that

when φ = 1, then {xt }t∈Z follows a non-stationary random walk
process. Figure 2.2 shows that the random walk is non-stationary
because its unconditional variance grows over time.

Figure 2.2: Simulated random walks. This stochastic process is non-stationary

because the unconditional variance grows over time.

While in linear models it is easy to find conditions for stationarity.

In nonlinear models a number of challenges arise. As an example,
consider the nonlinear autoregressive model

xt = g(xt−1 ; β) + εt ∀t∈Z

where g is some nonlinear function with parameter β. Under

what conditions is {xt }t∈Z stationary? Answering this question
is not easy, because we cannot easily unfold the process as we did
in the linear case and find an expression for xt as a function of
all past innovations,

xt = G(εt , εt−1 , εt−2 , ...).

34 CHAPTER 2. THE CHALLENGES AHEAD...

This inability to unfold the process and derive any conclusions

about its stationarity is very problematic. Indeed, without
knowing the stochastic properties of {xt }t∈Z we cannot make
use of laws of large numbers and central limit theorems for
establishing the properties of the estimator!

As a second example, consider the GARCH model for time-

varying conditional volatilities,

yt = σt εt , εt ∼ N (0, 1) ,
2
σt+1 = ω + αyt2 + βσt2 ∀ t ∈ Z.
For which values of ω, α and β is {yt }t∈Z a strictly stationary
time-series? We know that {yt }t∈Z has a time-varying conditional
variance. However, its unconditional variance may be invariant
over time, and hence, {yt }t∈Z may be strictly stationary for ap-
propriate values of ω, α and β. Again, unfolding this process and
obtaining an expression for xt as a function of past innovations
is not easy, and this spells trouble for us. Figure 2.3 suggests
that the GARCH model might indeed generate strictly station-
ary time-series.

2.5 Incorrect specification

We have already seen that the lack of an analytic expression
for the estimator is a problem that occurs when we consider
more general nonlinear models. As we will now see, even if
the estimator is analytically tractable, we cannot analyze its
properties in the usual way if the model is mis-specified.

The assumption of correct specification used in your introductory

econometrics courses (and earlier in Chapter 1) is very restrictive.
It was used essentially to simplify things. In practice, no one re-
ally believes that such simple models are an exact description of
the relation between economic variables. Economic variables are
most likely generated by an immensely complex process involving
potentially infinitely many variables. Both economic and econo-
metric models are simplifications of reality that help us uncover
basic relations between variables. For this reason, it is crucial
Discovering Diverse Content Through
Random Scribd Documents
 It was a pictorial Temperance drama—so essentially dramatic
indeed, that on its first appearance it immediately found its way to
the stage. *

* It was published by the late David Bogue, of Fleet Street.

The story of The Bottle is unfolded in eight designs executed in

glyphography—a process by which it was possible to execute the
immense number of copies which the publisher anticipated, and with
good reason, would be required by the public, but which is
ungrateful and unfaithful to the touch of the artist.
In the first plate we have a cosy family party. The open cupboard
is well supplied. The children are playing by the hearth; a kitten is
toying with the cat’s tail upon the rug; the mantelpiece is loaded
with pretty ornaments; there is a picture of a village church against
the wall; at the table the husband and wife are seated at dinner, and
he is handing her a glass, which she coyly refuses. Under the plate
we read: “The bottle is brought out for the first time: the husband
induces his wife ‘just to take a drop.’” The interest deepens apace.
The effect of the first drop is seen in Plate 2. The sottish husband,
with a pipe hanging from his mouth, his kerchief awry, his clothes in
disorder, sits drowsy with drink, his children looking fearingly at him,
while the wife is giving a bundle of clothes to the servant girl, to
pawn, “to supply the bottle.” The starved cat is licking an empty
platter upon the table; the cupboard door ajar discloses empty
shelves. In the next plate “an execution sweeps off the greater part
of the furniture,” but the drunken man and wife huddle themselves
before the fire, and “comfort themselves with the bottle.” There are
Hogarthian touches, developing the story throughout the series. In
this plate the china cottage upon the mantelpiece is broken, and the
husband’s battered hat upon a peg is the only ornament to the bare
walls. From the empty house the family repair to the streets to beg,
“and by this means they still supply the bottle.” In the fifth plate,
“cold, want, and misery” have destroyed their youngest child, and
still “they console themselves with the bottle.” A little open coffin is
in the room, and while the eldest girl weeps over it, the father and
mother drink, and weep also. A broken toy dog is upon the
mantelpiece near a candle, with a bottle for a candlestick. An old
shawl is fastened before the window with a fork. There are only a
few sticks in the fire. In the next scene the husband has his wife by
the throat; and his children and neighbours intervene.
“Fearful quarrels and brutal violence,” says the artist-preacher,
“are the natural consequences of the frequent use of the bottle.”
Murder is the next scene. The wife lies dead, with the doctor leaning
over her, and all the horrible commères who gather round death in
the dark, byways of great cities, are staring and talking. The
murderer is in the clutches of the police; the boy looks on aghast,
holding his chin, and trembling in his rags; the bottle, which has
done the deed, is shivered upon the floor and the fragments lie near
a broken pipe, a ragged slipper and a battered hat. The final scene
is a mad-house. “The bottle has done its work; it has destroyed the
infant and the mother, the boy and the girl left destitute and thrown
on the streets, and has left the father a hopeless maniac.” The figure
of the madman before the caged fire is a very powerful bit of
realism.
The moral of “The Bottle” was enforced by the poetic genius,
Charles Mackay. His “Gin-Fiend” sang to the scratching of
Cruikshank’s needle—

“There watch’d another by the hearth,

With sullen face and thin;
She utter’d words of scorn and hate,
To one that stagger’d in.
Long had she watch’d; and when he came,
His thoughts were bent on blood;
He could not brook her taunting look,
And he slew her where she stood.
‘And it’s hip I’ said the Gin-Fiend, ‘hip, hurrah!
My right good friend is he;
He hath slain his wife, he hath given his life,
And all for the love of me!’”

Regarded as a sample of Cruikshank’s art power, these plates are

far below the level of his best. We do not perceive here the master-
craftsman. His dramatic force is evident in every plate. He tells his
story with the fulness and intensity which are in all his pictorial
narratives; but the drawing is without grace, and the types, with the
exception of the husband, are wanting in that strong individuality he
generally realized.
In a letter to Mr. Forster (September 2nd, 1847), Mr. Dickens
describes the impression “The Bottle” made on him:—
“At Canterbury yesterday, I bought George Cruikshank’s ‘Bottle.’ I
think it very powerful indeed: the two last plates most admirable,
except that the boy and girl in the very last are too young, and the
girl more like a circus-phenomenon than that no-phenomenon she is
intended to represent. I question, however, whether anybody else
living could have done it so well. There is a woman in the last plate
but one, garrulous about the murder, with a child in her arms, that is
as good as Hogarth. Also the man who is stooping down, looking at
the body. The philosophy of the thing, as a great lesson, I think all
wrong; because, to be striking, and original too, the drinking should
have begun in sorrow, or poverty, or ignorance—the three things in
which, in its awful aspect, it does begin. The design would thus have
been a double-handed sword—but too ‘radical’ for good old George,
I suppose.”
And yet such calamities as that which “old George” has drawn
happen every day; beginning not in sorrow, or poverty, or ignorance,
but in little yieldings to temptation, in apparently trivial and
accidental excesses. What constitutes intemperance? According to
Dr. Alfred Carpenter, any consumption of alcohol sufficient to furnish
the blood with one part of alcohol in five hundred of blood, is
dangerous to health, and therefore is an act of intemperance. A
more moderate indulgence, he says, is not yet proved to be
deleterious. The late Dr. Anstie put temperance in a different way. An
average man or woman cannot, according to him, take more than a
couple of glasses of sherry daily without injury. Dr. Carpenter has
denounced the habitual use of stimulants, even in a very diluted
form, to enable the drinker to do more work than he could get
through without them, as unquestionably injurious—and therefore
an act of intemperance. There is not a middle-aged man of
education who has not come across the wrecks of lives where the
ruin was a gradual giving way to-the temptation of stimulants.
The police courts unfold daily stories of clerks and others, holding
positions of honour and of trust, who have first staggered out of the
straight path under the influence of drink. Cruikshank’s beginning of
his drama is only too true to life; and I think he would have made a
mistake, that he would have weakened the tremendous force of his
moral, if he had put the excuse of sorrow, or poverty, or ignorance
into his opening scene. As his story, stands, it teaches humble and
happy households, in a rough text which all who run may read, to
have a care whenever the bottle appears on the scene; and to lose
no opportunity of impressing, upon the children the danger, of
putting; the enemy near, their mouths, who may steal away, not
their brains only, but their heart and soul.
“Coarsely designed and coarsely executed, yet very suggestive,
very full of that story-teller’s power which was so much Hogarth’s
and his own,” as Mr. Frederick Wedmore remarks, “Cruikshank’s
‘Bottle,’ and the ‘Drunkard’s Children,’ which immediately followed it,
albeit executed when the finer qualities of his genius were suffering
decay, must always be welcomed as admirable contributions to the
matériel of Temperance advocacy.” Cruikshank used to relate how,
when his “Bottle” was finished, and he was anxious to secure for this
first Temperance sermon the widest possible publicity, he carried the
plates to Mr. William Cash, then chairman of the National
Temperance Society, for his approval, and the support of his
powerful Association. Mr. Cash, although a Quaker, was a gentleman
with a very sharp, humorous manner. Having attentively examined
the series, he turned upon the artist, and asked him how he himself
could ever have anything to do with using “The Bottle,” which, by his
own showing, was the means of such dreadful evil? Cruikshank, in
his own forcible way, described how he was “completely staggered”
by this point-blank question. He said, when he had left Mr. Cash, he
could not rid himself of the impression that had been made upon
him. After a struggle, he did not get rid of it, but acted upon it, by
resolving to give his example as well as his art to the total
abstainers.
He was immediately rewarded by the extraordinary success which
“The Bottle” achieved. It was sold by tens of thousands, and was the
talk of the day. If it has not directly led to a tangible result, as
Hogarth’s “Harlot’s Progress” is said to have led to the foundation of
the Magdalen Hospital, it and the “Drunkard’s Children,” a poor
sequel (but then sequels are always poor), have had the effect of
powerful, popular, and permanent sermons against the monster evil
of our time.
Not the least of the artist’s rewards was the tribute to his genius it
inspired in Mr. Matthew Arnold, who wrote:—

TO G. CRUIKSHANK,

ON SEEING, IN THE COUNTRY, HIS PICTURE OF “THE BOTTLE.”

“Artist, whose hand, with horror wing’d hath torn

From the rank life of towns this leaf! and flung
The prodigy of full-blown crime among
Valleys and men to middle fortune born,
Not innocent, indeed, yet not forlorn—
Say, what shall calm us when such guests intrude
Like comets on the heavenly solitude?
Shall breathless glades, cheer’d by shy Dian’s horn,
Cold-bubbling springs, or caves?—Not so!
The soul Breasts her own griefs, and, urged too fiercely, says—
‘Why tremble? True, the nobleness of man
May be effaced; man can control
To pain, to death, the bent of his own days.
Know thou the worst!
So much, not more, he can. ‘”
Original Size -- Medium-Size
 CHAPTER III. GEORGE
CRUIKSHANK AS A TEETOTALER.

G
eorge Cruikshank was an enthusiast in all things to which he
gave his mind. He did nothing in a halfhearted way. Whether
preparing to address a great Exeter Hall audience on the evils
of drunkenness, or marching at the head of his riflemen, or
arraying himself in a table-cover to enact the part of Lord Bateman;
in small things as in great, he was ever at fever-heat. He would have
made a good actor, had he not been incapable of a moment’s
repose; he would have been an admirable Temperance advocate,
had it been in him to give himself pause in order to think over the
heads of his discourse; he would have been a good volunteer officer,
had it been possible for him to sit quiet in his saddle. But he seemed
to be troubled with an excess of life. Life at fever-heat is the
dominant characteristic of all his work. The “quiet spaces” in his
etchings are rare.
Having been converted by his own “Bottle” to total abstinence
from fermented liquors, he could be nothing less than an earnest
and a vehement worker in the cause. He threw himself heart and
soul into it; and during the thirty remaining years of his life his zeal
never slackened, and he had never made sacrifices enough in it. His
impulsive advocacy often took ludicrous forms. He sometimes
offended people by his denunciations of even the most moderate
drinkers, but he never made an enemy by his gaucherie or his
downright phrases imported into quiet circles, because the parity of
his motive and the well-known impetuosity of his nature excused
him. I can remember, in the first year of his total abstinence,
meeting him at a ball given in Fitzroy Square, by Mr. Joshua
Mayhew, the father of Horace and the Brothers Mayhew. He danced
and was light-hearted with the youngest; but when at supper the
wine began to circulate, he stole round to the head of the table,
and, laying his hand upon the shoulder of the venerable host (who
was a very haughty and quick-tempered old gentleman), said, in a
deep, warning voice, “Sir, you are a dangerous man.” Mr. Mayhew
had a glass of wine in his hand, and was about to drink a toast to
the health of one of his sons, when Cruikshank’s hand fell upon his
shoulder. “I look upon every wine-drinker,” Cruikshank added firmly,
“as a dangerous man, sir.” The company, knowing the hot temper of
their host, expected an explosion of rage; but it was staunched by
Horace Mayhew, who burst into a hearty laugh, and told his father to
go on, for “it was only dear old George.”
In the same way, when dining at the Mansion House, Cruikshank,
at the passing of the loving-cup, would go through an extraordinary
pantomime before all the company, expressive of his horror of strong
drinks. He would shake his hand angrily at the Lord Mayor, and raise
his arms with horror while his neighbour quaffed of the cup. The
company humoured the eccentric old gentleman; for, in their hearts,
they could not but respect his downright earnestness. He lost no
opportunity. Returning home at the head of his volunteer corps, he
showed his jaded officers, who had freely taken beer, how fresh he
was—on two oranges.
“Ah! you may laugh,” he would say, when his friends bantered him
about his aggressive protests in society; “you may laugh, but I can
tell you this——the presence of the old jackdaw checked the
drinking, if didn’t stop it, and I am very grateful to feel sure of that.”
* As Mr. Sala has observed, “the veteran sticks bravely to his text.”
And well he might, for his temperance renewed his youth. “He
neither smoked tobacco nor drank fermented liquors in his old age;
but he was a hearty eater, an early riser, and a vigorous walker and
his reward was that which, according to Gray, is only felt by boys at
school—a perpetual ‘sunshine of the breast.’” He was fond of
showing this vigour renewed by temperance, at every possible
opportunity; for he very wisely regarded it as his most forcible
argument. It enabled him, in his old age, to capture a burglar on his
own premises. The story runs that when he was following the
burglar to the station, with the police, he drew him under a lamp,
and told him that he could see drink had brought him to this—
adding that he himself drank nothing but water. “I wish I’d ha’
known that,” said the ruffian, “I’d ha’ broken your head for you.”
Cruikshank delighted to show an audience how he could hold a
tumbler full of water steady upon the palm of his outstretched hand.
At eighty, he was seen in costume at a fancy dress ball at Willis’s
Rooms, joining heartily in the dance, and letting everybody know
that it was “water that did it.”

* Grace Stebbing’s article on Cruikshank in the Graphic.

It was very difficult to obtain from him the toleration of tobacco

smoke in his company; for, after he had given up alcoholic
stimulants, he threw away his pipe. He would say to a man of letters
whom he favoured, laying his hand upon his arm, and turning those
fierce eyes of his full upon him, “I want you to give up drinking and
smoking, and you tell me that if you don’t smoke you can’t write.
Now, I’ll meet you half-way. Give up the drink, and you may smoke
—-just a little.” But, as a rule, he was as stern in the matter of
tobacco as in that of beer or gin. One evening M. Legros, the
distinguished French artist, lighted a cigarette in his hall as he was
leaving Mornington Place. “To that vice,” said “the inimitable George”
in his deepest tone, “I was a slave for many years, but now I am a
free man.”
To it also, it must be added, he owed one of his most imaginative
and delightful etchings,—“The Triumph, of Cupid,”—published in his
“Table Book.”
His earnestness was extravagantly expressed in all things. As a
furious anti-Papist, he would draw aside and shake his coat when
Sisters of Charity or a Catholic priest passed him. “Do you see that
fellow in front?” he suddenly asked a friend with whom he was
walking. It was a workman quietly enjoying his pipe. “Do you know
what I would do to him if I were a man of fortune? I’d kick him! To
think that any man should be fool enough to place a tube between
his lips, and go puff, puff, puff!” This was his “counter blast.” And he
glared at the workman as he passed on. He had himself been an
inveterate smoker for more than forty years!
On another occasion he drew sharply up before the windows of
his old wine merchant, and called out, “Give me back my thousand
pounds!”
When the Crystal Palace was opened at Sydenham, Cruikshank, in
his rage that it had not been made a Temperance palace, drew some
extravagant drawings of the opening ceremony for Messrs. Cassell,
one of which represented the Archbishop of Canterbury bestowing
his blessing upon a public-house.
Dining one day at Grampian Lodge, Forest Sill, with his friend Dr.
Rogers, he suddenly began to tell the company that he had had a
vision the night before. Then he related it with much gesticulation,
and with dramatic effect.
He had seen two devils in council. One had said, “England is
moral, prosperous, happy—this will never do. How can we put an
end to it? Her crops are splendid; look, for instance, at her barley,
her-” The second devil interrupted: “I have an idea. Her barley,
which makes such splendid food, let us teach them to soak it, to
sour it, to make it ferment; in short, to turn it into a tempting
poison.”
“Agreed!” cried the first devil.
“Why,” the second devil continued, “we will actually make them
drink it of their own accord; they shall lift the poison to their own
lips with their own hands.”
“Ha! ha!” shouted the first devil; “and then of course, there will be
murder, robbery, violence, and misery all over the land.”
“The devils have had their way,” the old man added his keen eyes
glancing round the table to mark the effect of his vision.
He was indeed, as a writer called him, a “muscular teetotaler.”
“In his time,” a Temperance writer * records to his honour, “he
must have attended thousands of temperance meetings, and at
these few men were more welcome.
* The Temperance Record, February 7th, 1878.

The style of his advocacy was peculiar, he passed from grave to

gay with facility, but he never lost sight of the great object he had in
view. He seemed for years, to be deeply impressed at the numerous
murders that were taking place, all of them, or nearly so, being in
the last instance, if not in the first, attributable to drink. He used to
exclaim, with deep fervour, ‘Can nothing be done to stop these
dreadful murders?’ The clear remedy of total abstinence from that
drink which was their inciting cause then came naturally from his
lips; but though individuals responded to his appeal, the general
mass of the public remained unmoved. Sometimes he would suggest
a deputation to the House of Lords. But though this idea was not
acted upon, yet he lived to see that august assembly collect
evidence well fitted to be of service to them, and also to the public
at large. Mr. Cruikshank’s powers of mimicry were also very great,
and often has he convulsed his audience with his inimitable acting;
but, at the same time, there was no mistaking his deep earnestness,
and the sincerity with which he expressed the convictions of his
heart.
He did his utmost, when the teetotalers had failed at the Crystal
Palace, to establish a teetotal palace in the old Surrey Zoological
Gardens; and he was drawn in state from the Hampstead Road to
Walworth, in a carriage and four, to open a bazaar in aid of the
scheme. He even prepared a design for the building. But although
many went to cheer the honest, earnest old man, few remained to
invest, and the design fell to the ground. It may have been some
consolation to him and to his Temperance friends to mark,
afterwards, the services which the Crystal Palace was destined to
render to the cause of Temperance, for a drunkard has hardly ever
been seen under its shining roof.
Cruikshank could never convert his mother to his views. She lived
with him during the latter years of her life, and died under his roof,
in the care of a most reverent and attentive son.. She had always
been a careful, sober body, and would not be coerced, because her
son could not take his beer or toddy without committing excesses.
She had been a handsome woman in her days, a grandson records,
and it was picturesque to see the lame old lady, leaning upon her
crutch, and wrapped in a plaid,—with her shrivelled features and
wild grey hair,—raise her withered arm, and with the old fire declare
that she would not surrender her principles. A glass of beer with
dinner, and a little toddy at bedtime, she had always taken, and she
took them to the end, and George had to submit.
Addressing, on one occasion, a Temperance oration to a Bristol
audience, he appealed to his female hearers not to believe that
“nourishing stout” was necessary to nursing-mothers; and he
pointed to himself as a melancholy example, saying, “My mother first
lifted the poisoned chalice to my lips.” His aged mother read this in
the morning paper. Her wrath was violent. “What!” she cried, “am I
to be told publicly, at eighty years of age, that I, who always begged
and prayed him to be sober, taught him to drink?” Her son did not
return home for several days; but he heard of his speech in no
uncertain tones when he presented himself to the old lady, who had,
in his youth, often physically chastised him for his excesses.
Perhaps the best specimen of his manner of laying his subject
before an audience is the speech which he delivered at the Grand
Demonstration of the National Temperance League, in the Guildhall,
on the 19th of November, 1864. It wants his by-play, his dramatic
delivery, his grotesque movements, and then the solemn sounds of
his voice, to be completely understood; but it is sufficiently original
and suggestive as reported:—
“My Lord Mayor, Ladies and Gentlemen,—-My worthy friend the
doctor has given you a very excellent discourse upon his own
profession. It So happens that as I was coming to this meeting I met
with a gentleman who had just been to consult his medical man;
and finding I was coming to this meeting, he laughed at all idea of
abstaining from intoxicating drinks. He told me he had been to see a
very eminent member of the medical profession. I asked him what
was the result.
 “He said the physician told him he wanted a stimulant and
prescribed one. I said, ‘What did you give him:’
“He replied, ‘Of course, I gave him the usual fee—a guinea.’ I said,
‘I can show you how to save that guinea in future. If you will give
me half of it. I will give it to some good charity, and the other half
you may keep in your pocket.’ He said, ‘How is that?’ I said, ‘Instead
of going to the physician, go to the publican, and tell him what is the
matter with you, and he will prescribe the same thing; and if the
landlord is not in, say the same to the potboy, and he will do as well.
Rely upon it, they will prescribe exactly the same thing as the doctor,
and the effects will be the same. Now, I must say one word, if you
please, to defend a very eminent prince who has been mentioned
here to-night. I am sorry to say it happened to be my fate to hold up
to ridicule the Prince Regent—very often indeed; but he was not
such a bad man as he is represented to be. It must be recollected
that if he committed excess in the way of drinking, it was then the
fashion for all the eminent persons to get drunk. No man was
considered a liberal man—no man was considered a gentleman, in
fact, unless he made his companions drunk; and therefore, with all
due respect to my friend Mr. Scott, who mentioned the circumstance,
it must be recollected that about half a century back it was the
fashion—it is a fearful thing, but it was the fashion—of gentry to get
drunk; therefore we ought to make allowances. But now, my Lord
Mayor, to come to this very serious question. This hall is the place
where the great City feasts are held, and the question is, is it
possible that there can be any grand entertainment given without
mixing up with it the intoxicating cup? What will be said? It is very
well for you, my lord, who are almost an abstainer yourself—very
well for you—but what will be said of another Lord Mayor who
comes here and gives a dinner without wine and beer? What will be
said of him? He will be called a shabby fellow; and the question is,
whether the guests will not all be melancholy. It will, perhaps, be
somewhat in this style: ‘Have a little more soup?’ ‘No, thank you.’
‘More fish?’ ‘No, thank you.’ ‘Bit of fowl?’ ‘No.’ ‘Venison?’ ‘No.’ ‘What,
can’t you eat any more?’ ‘No, I don’t like it: I want something to
drink.’ There is the serious thing: what is to be done? There is one
way of settling that question. It is supposed that there can be no
sociality, no comfort, no enjoyment, without intoxicating drinks.
Now, I recommend the next Lord Mayor who may succeed our
honoured chairman, if he be in favour of the moderate use of these
delightful drinks, to be so good as to ask the present company to
come to dinner. Wouldn’t you enjoy yourselves? And then, when we
have had enough to eat, and want something to drink, here you are
(holding up a glass of water)—Mr. Chairman, your very good health!
Ladies and gentlemen, your good health! (drinking the toast.) We
should have a jolly time of it. (Loud and long-continued cheering.)
Mr. Morley says we will take the sherbet without the punch. That is
the way in which these things are looked at; but supposing that it is
impossible that any social enjoyment can be had without the use of
these stimulants, let us take another view of the question. I have
had the honour of dining here, and I have enjoyed myself very
much, not only in the time when I used to take wine myself—
because I recollect there was such a time as that—but when I have
been a teetotaler I have been, here, and enjoyed my dinner very
much indeed, without any of these drinks. But supposing we had
this hall upon the occasion of the Lord Mayor’s feast with the most
elegant people in the world (for I believe of all the people in the
world the British people are the best looking and the best dressed):
imagine the scene! The tables are set out in the most splendid
manner; everything looks grand and happy; but what is going on
outside? Ay! my friends, the most splendid monument in the world
where this drink is used in moderation as it is in this country, may in
the inside be a splendid monument of good order, taste, and
sobriety, but at the outside there is filth and dirt and crime through
drink. I say, suppose these social meetings cannot be enjoyed
without these drinks, let us look at the outside. Now, there are a
certain number of circumstances or acts committed in society, which
are always injurious, not only to the individual himself, but also to
society at large. Now, I do not mean to say that every teetotaler is
an honest man. There may be some dishonest fellows amongst
them. I have heard of two in the last thirty years. This reminds me,
by-the-bye, of a teetotal turnkey at Coldbath Fields. There were two
youths brought into the prison, who were teetotalers, and the other
turnkeys jeered the teetotal turnkey upon it. He said, ‘It is true that
there are two teetotalers here, but they are here only for begging,
whereas you have about fifteen hundred brought in who drink, and
they are most of them committed for stealing.’ There are a number
of besetting sins connected with drinking, such as robberies, brutal
assaults, garotting, house-breaking, suicide, and murder. By-the-bye,
speaking of murder, there has been a very strong feeling existing for
many years, and still increasing, against the punishment of death. I
think it is a very horrible thing indeed to hang anybody; but, my
friends, do not forget that it is a still more horrible thing for one to
be murdered. Do not let us forget that. There was a young man in
the country a little while ago hanged for murder—quite a young
man. It was a sad thing indeed, no doubt, to see this poor fellow
gibbeted; but what was he hanged for? He had been drinking on the
Saturday night, and he murdered a young woman as she was going
to church on the Sunday morning. Do not forget that these horrible,
detestable, damnable crimes are committed under the influence of
drink. We will talk about doing away with the punishment of death
after we have stopped murders. I had the honour of speaking in the
Mansion House when Mr. Charles Pearson, the City solicitor, brought
on the question about the convicts; and I told the Lord Mayor then,
that if we could do away with intoxicating liquors altogether, we
might wheel out that dreadful instrument the gibbet into the Old
Bailey, and make a bonfire of it. I believe you will find, if you go into
the question, that there is hardly a murder committed in this country
out of a hundred—I may say out of a thousand—not ten out of ten
thousand—but drink has something to do with it. Remove the drink,
and you will stop murder. But there is a gentleman who ought to
have been speaking instead of myself, and therefore I will not detain
you much longer; but I will say this, my friends, and call your
attention to it especially, that the teetotal question has now been
before the world for about thirty years, and during that short time I
challenge any one to point out any teetotaler who has been
committed for a brutal assault upon his wife, or for garotting, or
picking pockets, or house robbery, or murder. I challenge the world
to produce one single case wherein any real teetotaler has been
convicted of one of those crimes. Then, if this be so, what have we
to do but to spread this Temperance movement throughout the
length and breadth of the land? and then we should stop, if not all
crimes, if not all offences, still the great majority of them; and that is
what we are aiming at. And recollect this, my friends, that we are
not a society formed merely for the purpose of reclaiming the
drunkard. It is a very good thing to do so, and I am sorry to say that
my time is so occupied that I am almost in despair. I have six most
dreadful cases in hand at the present moment There is nobody to
assist them. I could not go to the brewer or distiller, and ask them to
give me funds for the support of these people whom they have
rained; and why not? Because there is blood upon the money. I
would not have it. But I had to-day a letter imploring me for help
from the nephew of an old friend of my father. What am I to do? I
have a lady in the country at this moment, the wife of a barrister,
who is starving for want of help, and whose husband has been
ruined by drinking. My time is occupied, and my friends are gone,
and I am called upon for all I can afford. But, my friends, if you do
away with these drinks, you do away with these cases. But it is
utterly impossible to go into the evils arising out of these drinks in
the time I have to speak—they are so extensive; all I have to say is,
‘Go on and prosper!’ and prosper we shall. I cannot sit down without
saying that I look upon this meeting to-night to be one of the
grandest movements that this cause has ever had. I say it from my
heart, and think that those gentlemen who have assisted in getting
up this meeting deserve our best thanks.”
The idea of a temperance Lord Mayor’s Banquet suggests, no
doubt, many vastly amusing incidents and episodes to the mind of
the comic writer, but honest-hearted George Cruikshank could not,
and would not, in his latter dav, see any element of fun in
drunkenness, and he was quite in earnest in recommending the next
Lord Mayor to fill his loving-cup with pump water. *
* Since Cruikshank delivered the above speech, a Temperance
banquet has been held at the Mansion House.

The account he gave, moreover, of his trouble about the many

people who were seeking his assistance, was true of his experience
year after year. His doors were besieged. He was waylaid by
petitioners for his known bounty (the recklessness of which, as we
have seen, Dickens reproved) whenever he went abroad. A poor
man himself, for ever in money troubles, even to the end of his
laborious life, his heart lay always open to a tale of distress. He was
never without “cases” on hand.
It has been remarked of his Temperance days, by one of his
friendliest critics, that his style suffered from the contraction of his
ideas and sympathies, “and it cannot be questioned that with the
general public his reputation declined in proportion to the increase of
his popularity among the teetotalers.” He lost heavily, in a pecuniary
sense, by his Temperance advocacy. Publishers ceased to employ
him. He remarked that, for the last ten years of his life, he was
without commissions. He had refused none, he would say. He was
willing to work, and he held that his powers were unimpaired.
Temperance preacher; to them the inimitable George, the
illustrator of Boz, the kindly satirist, the creator of “Points of
Humour,” the illustrator of Grimm, was dead.
And, firmly believing this, the brave old man held on in the rigid
course of duty he had laid down for himself. He had seen all the
horrors which lie behind drunkenness; in his early time he had
himself been a tavern hero; and he had dedicated the remainder of
his life to the work of warning the rising generation out of the path
in which he himself had stumbled.
“I come forth,” he said, in one of his earliest temperance
harangues, “to set by my humble example the opinion of this
unthinking world at defiance.”
But the public had come to regard him simply as a keen a sense of
the ridiculous as most men. I can see clearly what is ridiculous in
others. I am so sensitive myself, that I am quite alive to every
situation, and would not willingly place myself in a ridiculous one;
and, I must confess, that if to be a teetotaler was to be a milksop, if
it was to be a namby-pamby fellow, or a man making a fool of
himself or of others, then indeed I would not be one—certainly not;
but if, on the contrary, to be a teetotaler is to be a man that values
himself, and tries by every means in his power to benefit others; if to
be a teetotaler is to be a man who tries to save the thoughtless from
destruction; if to be a teetotaler is to be a man who does battle with
false theories and bad customs, then I am one. I have been a
convert but a short time, not much over twelve months. I only wish
that I could say, with Dr. Gourley, that I had never taken a glass of
spirits in my life; I wish that I had acted upon the principles of total
abstinence only thirty years ago; for if I had, I am convinced that at
this time I should have been much better, both in body and mind. I
have experienced much benefit already, both physically and
mentally. I never did sneer at or scorn the question of Temperance,
yet I never thought that I should stand up as a teetotal advocate;
but I am proud that I have been put into the position in which I am
now placed.
Later on, still conscious of the disadvantage at which he was
placing himself as an artist, he said to another audience—.
“When I left off drinking wine altogether, and became a total
abstainer, I became a healthier and stronger man, more capable of
meeting the heavy responsibilities that were upon me, and for the
following two years I had my life renewed, and all the elasticity of
my schoolboy days came back to me. Domestic afflictions then came
upon me, ending in death, and my spirits and health were crushed
down. In this extremity I applied to my medical adviser. He said,
‘Medicine is of no use to you; you must drink wine again.’ I refused,
and my medical friend called in some others of his profession; he
told me they had had a consultation, the result being that all of
them agreed it was necessary I should drink wine to restore my
sinking constitution. I replied,
“‘Doctor, I’ll take your physic, but not your wine. Let me try
everything else first, and only when there is no other chance give
me wine, because I feel there is a great principle at stake in this
matter.’ I have said, and I believe, wine is unnecessary, even as a
medicine, and I do not wish to do a single act which would tend to
weaken or destroy the weight and force of that conviction. And here
I stand. I have not tasted the vile and destroying enemy, and I am
almost restored to health, without having risked the violation of my
principles. I call this a triumph; and I stand here as an evidence that
wine is totally unnecessary, even as a medicine.”
Much later, we find the preacher an octogenarian—albeit rudely
buffeted by the world, and well-nigh forgotten as a living artist—still
true to his noble text. “Alcoholic liquors,” he exclaimed to an
audience, little more than two years before his death, “were
recommended to keep up strength! But what kept up his strength?
He had not taken a drop of wine, beer, or any alcoholic drink, for
twenty-seven years, and he would be eighty-three next September, if
he lived till then. What was it, then, that kept up his strength? Since
he had given up drinking beer and smoking, he had had a higher
enjoyment of life, because all his nervous system was in proper
tone.”
Cuthbert Bede, who knew Cruikshank intimately in his teetotal
days, has drawn this graphic picture of the Temperance advocate at
home:—
“Though I had interchanged letters with Mr. George Cruikshank for
several years, it was not until early in the autumn of 1853 that I
made his personal acquaintance. He had asked me to write a serial
story for a projected publication to be illustrated by himself ; and, as
it would simplify matters if we could talk over the subject together, I
went up from the country to London to call upon him. He was then
living in Mornington Crescent, near to Regent’s Park. Numerous
portraits had made me familiar with his personal appearance, so that
I needed not to be told who was the gentleman who so courteously
received me downstairs, and then took me upstairs to his
comfortable studio, where he introduced me to his wife. Some of our
first conversation, indeed, was on the subject of his portrait; for,
among the pictures on his walls, I had noticed the original of the
portrait by Frank Stone, which was engraved on steel for the
Omnibus, and was certainly a far more flattering representation of
George Cruikshank than the caricaturist’s sketches of himself. I told
him that I considered the best portrait of himself was to be found in
his own etching, ‘The Reverie,’ published in his Table- Book, and in
every respect a wonderfully fine specimen of his art and genius. I
also referred to his own account of ‘My Portrait,’ in the Omnibus, in
which, with his own pen and pencil, he portrayed himself, and made
comments on a curious description of himself that had been given in
a publication called ‘Portraits of Public Characters’; how he was said
to be of the medium height, with a forehead of a prominently
receding shape, with a handsome pair of whiskers, and hair
partaking of a lightish hue; and, moreover, how the ludicrous and
extraordinary fancies with which his mind was constantly teeming
often imparted a sort of wildness to his look frighten from his
presence those unacquainted with him.
“He read these and similar passages to me, and was immensely
tickled at their egregious absurdity. In truth, his manner at once
impressed me as being peculiarly gentle, and kind, and genial.
Instead of assuming any airs of superiority, I found him possessed of
all the humble modesty and chivalrous courtesy of the truly great
artist and thorough gentleman; and although I was quite young, and
he was in his sixty-first year, he treated me as though I had been his
equal, if not superior, in ability. We had so much to talk about, and
he had so much to show me, that my first interview with George
Cruikshank had been prolonged to nearly four hours before I
became aware how quickly the time had flown. The time had then
arrived for their luncheon, or early dinner; and as both Mr. and Mrs.
Cruikshank pressed me to stay, and I had by this time overlapped
the hour at which I had made another engagement, I readily and
peculiarity to his manner, which would suffice to consented to
remain, and we went downstairs to dinner. ‘There will be nothing
else than a leg of mutton,’ said Cruikshank. ‘I happen to know that,
for I came in with it,’ I replied; ‘for as I knocked at the hall door the
butcher’s boy was down in the area, delivering the leg of mutton to
the cook.’ Cruikshank seemed to be greatly amused at this, for he
laughed heartily, and said to his wife, ‘My dear, Mr. ———— came in
with the mutton.’ Something in the occurrence seemed to mightily
tickle his fancy, for more than once he repeated the words to his
wife, ‘My dear, Mr. ———— came in with the mutton!’ It was while I
was eating it that I terribly forgot myself. The day was very sultry; it
was five hours since I had breakfasted; we had been busy talking,
and I felt thirsty. So, while the parlourmaid was handing something
to me. I asked her to give me a glass of beer. She replied, ‘We have
no beer, sir.’ ‘Then,’ I said, ‘please to bring me the sherry.’ ‘There is
no sherry, sir.’ Whereupon my host interposed, and laughingly
explained that he could not allow the introduction of any alcoholic
liquor into his house; and that, while I was his guest, I must content
myself with drinking water. Then I suddenly remembered that which
I ought not to have forgotten, even for a moment, that my host had
devoted himself to teetotal, who, six years before, had drawn the
eight scenes of ‘The Bottle,’ and had thereby struck a powerful blow
at one of the greatest vices of the age.
“I duly apologized for my forgetfulness; and the incident naturally
led Cruikshank to dilate on that important theme, in furtherance of
which he so steadily devoted his great powers to the very end of his
career, with a persistent courage and devoted zeal that won for him
the genuine respect and admiration of those even who could not
wholly agree with him in details. I was one of those. I could travel
with him, very willingly, up to a certain point, after which our paths
parted, and we ‘agreed to differ.’ I could accompany him to
temperance, but not to total abstinence. During the remainder of the
time that we occupied over dinner, we scarce spoke on any other
subject than that which gave ism, and that I was sitting at the table
of the artist rise to the scenes of ‘The Bottle,’ ‘The Drunkard’s
Children,’ ‘The Gin Trap,’ ‘The Gin Juggernaut’—and, at a later
period, his large oil-painting, ‘The Worship of Bacchus.’
“Our discussion on the subject was preserved with perfect good
humour; so much so, that I ventured to remind him that only a year
or so before he had been converted to teetotalism he had
caricatured Father Mathew, in an etching for the Comic Almanac for
1844, representing him as an old pump. I reminded my host that
these were his sentiments for more than fifty years of his life, and
that he had never during that period objected to the moderate use
of alcoholic liquors, although he had always vigorously lashed their
gross abuse; and I pleaded that I had not lived for half those years
that I had named, and that I might be pardoned for my
forgetfulness in asking his servant for beer and wine.
“Then he told me how the crying sin of the age had sunk deep
down into his heart, especially when he had seen it flourishing, like
an upas tree, in all its foul deformity, in those courts and alleys into
which he was so often led in search of subjects for his pencil; and
how the design for ‘The Bottle’ had gradually grown upon him, and
the necessity for practising what he preached, which he found he
could do only by cutting himself adrift from all alcoholic drinks. He
also explained how his plans to disseminate the scenes of his
‘Drunkard’s Progress,’ in such a form and at such a low price that
they should reach those masses for whom he specially designed
them, were hampered and well-nigh frustrated, chiefly by the cost of
engraving such large drawings on wood; and how the new art of
glyphography had come to his assistance, and enabled him to draw
the eight designs, and to sell them (with Dr. Charles Mackay’s
explanatory poem) for a shilling—which in the year 1847 was an
extraordinarily low price for such a production. ‘You will remember,’
he said, ‘how Maclise represented me seated on a beer-barrel,
getting my inspiration from pothouse scenes, and pencilling them on
the crown of my hat?’ ‘Yes, I remember: it was in the Fraser gallery
of portraits.
And you have amply proved to the world since then that you can
turn to the best account, and for the public good, the people and
incidents that you saw in those places.’ I told him that of ‘The Bottle’
and ‘Drunkard’s Children’ series I preferred the one where the poor
girl commits suicide from Waterloo Bridge—the idea of the body
falling from a height being so vividly conveyed to the eye, as to
impress one with the conviction that we can really see the swift
descent of that ‘one more unfortunate.’”
An instance of Cruikshank’s earnestness in the Temperance cause
happened in May 1854. He had been invited to preside over a
meeting of total abstainers, to be held in Sadler’s Wells Theatre, a
place associated in his mind with the glories of his friend Joe
Grimaldi, the clown, and the days when he was a frequenter of the
clown’s club, “The Crib,” hard by. The great Temperance advocate, J.
B. Gough, was to address the audience. Cruikshank introduced him
in his own original way, delivering, as the papers remarked, a speech
full of piquant and incontrovertible truth. But it was at the close of
the orator’s speech that the chairman proved himself equal to the
occasion. Seeing that the audience were under the spell of Mr.
Gough’s eloquence, he rose and exhorted them at once to come
forward and sign the pledge. With this he advanced to the footlights,
bridged the orchestra with a few planks, and stood by to receive the
ladies who came forward in crowds, many of them leading their
children. So delighted was the artist with the number of converts he
led to the table to sign the pledge, that he drew the scene for the
Illustrated London News, with himself for central figure.
I remember attending another meeting in George Cruikshank’s
company. It was a gathering of London pickpockets, called by Mr.
Henry Mayhew, when he was engaged upon his London Labour and
London Poor inquiry. The solemn, but still somewhat grotesque
impressiveness of the Temperance preacher, as he rose, while that
dreadful company of keen-eyed vicious lads were eating the plain
Temperance supper which had been provided for them, to bid them
renounce the evils of their way, and as a beginning, to shun the
bottle and the beer-pot, dwelt long in my memory. “Man,” said Lord
Lytton, “has no majesty like earnestness.” That night, honest, whole-
hearted Cruikshank, as with wild gesticulation he talked to “the dear
lads”—for the forlornest and wickedest waif was dear to him—was
clothed in majesty; and it cowed a man at hand, who acknowledged,
within his hearing, that he had smuggled something stronger than
water into the room.
Original Size -- Medium-Size
 CHAPTER IV. THE TRIUMPH OF
BACCHUS.

M
r. Wedmore, in his critical sketch of Cruikshank, has
described in a few pregnant sentences, how in his later days
the public fell away from the great humourist and subtle
observer:—
“As time went on apace, neither the passage of time itself, nor the
hard work which crowded the days of his maturity in art, nor the
comparative neglect of the later years, when Cruikshank, no longer
quite in the movement of the day, was solaced by visits in the
Hampstead Road, chiefly of a very few who were collectors of his
work, or of some stray humourist still faithful and confident in the
achievement of so many years ago. As time went on, Cruikshank
wore well and slowly, so that it was truly said of him that he looked
as if he had once been very old and then had forgotten it. Employed
no longer in sketching and satirising the society of which he was
hardly any more a part, he betook himself, a good deal by choice, to
work more distinctly ambitious than any he had attempted when his
hand was really the strongest and his brain the most fertile. He
furnished the design for a monument to King Robert the Bruce. He
painted in oils, not only this or that moral lesson, but a tale of
heroism in humble life. No doubt the absence of the knowledge of
academical draughtsmanship told against him not less in 1871 than
it would have done half a century before, and no doubt the absence
of any capacity for the subtle modulations of colour—nay, the
absence even of sensitiveness to these—made his painting in oil a
failure when judged by the side even of quite every-day work by
every-day artists. Thus it was that no fresh honours came to him
when he was still eager for them. The popularity of the great days
was a little forgotten by the public in the presence of the failure of
the most recent. And then again, advertised poverty is never a
helpful thing. We worship merit a little, but success more, and
success must have its stamp. The public of Cruikshank narrowed. Of
course critics and journalists—the men whose business it is to keep
in memory some work that the chance public praises one day and
forgets the next—knew that Cruikshank was great, and how he had
been great, and having in more than one notable instance said so
faithfully during his old age, said so again last month, when he died.
And of course, again, so much of his work having become rare,
collectors of it had arisen—curious and anxious seekers, to whose
interest we shall owe the preservation of many of his early and many
even of his riper things. For them, when Cruikshank’s work was
pretty well accomplished, and ‘finis’ seemed about to be written to
that immense volume of production, Mr. G. W. Reid engaged on a
task of care—the great catalogue raisonné in which, with here and
there errors not easily avoided, he has chronicled well-nigh five
thousand designs: ‘the smiling offspring,’ as Thackeray so admirably
said of them, ‘the smiling offspring of painful labour.’ But in the main
Cruikshank was forgotten, and the weekly smiles—feint though now
and again they needs must be, and of indulgence rather than
commendation—which are given by the English public to the efforts
of our youngest English humour, a little trivial and slight, had ceased
to be bestowed on that larger and more massive humourist who
lingered from the past he was part of.”
This is very true, and is a very sad story skilfully told. Think what
would have become of the neglected or forgotten humourist, if,
when the mere laughing public had turned away from him to Leech
and Doyle, and Tenniel and Du Maurier, he had not been fired with
the ardour of an apostle in the cause he had taken up. His Almanac
had failed for lack of readers; and David Bogue had thrown up
Cruikshank’s Magazine, after the second number—convinced that the
artist had outlived his public. His ambition to become a painter was
mercifully renewed, with the renewal of his health and mind,
through temperance. Full of vigour he used to say, “A painter should
paint from his shoulder, sir.” He became almost wholly a serious man
in his work, and appealed to a public in a new capacity. He resolved,
stimulated by the success of “The Bottle,” to execute a great picture
that should remain behind him, a monument of his genius, and an
immortal Temperance lesson.
In the early ardour of his second youth he had braced himself to
supply, so far as he might, albeit he had reached his sixtieth year,
the deficiency in his art education, by working as a student at the
Royal Academy. He had, he believed, all his powers unimpaired; why
then should he not yet obtain the academical knowledge, of which
he had been deprived, as he had said bitterly, through the
improvident habits of his whisky-drinking father. Mr. Charles
Landseer says: “He entered as student at the Royal Academy, during
my keepership, April 22nd, 1853; but made very few drawings in the
Antique, and never got into the Life. He was placed upon the Turner
Fund in 1866—£50 per annum. I have heard that he made an
application to Fuseli for admission to the R.A., and was informed that
the school was too full, but that he might go and draw there if he
could find a place.” *

* Letter from Charles Landseer to B. J., Feb. 18, 1878.

This is the brief record of George Cruikshank’s relations with the

Academy. He was past the years when men learn. Time pressed too
heavily upon the elderly man to leave him patience for the slow
progress from the “Antique” to the “Life.” He had been at the “Life”
in his own keen way since he was a boy; and he must be content to
paint with the imperfect but original knowledge which had sufficed
for his etchings.
And so he turned to his easel, and painted in oils, with something
of his own inimitable power of concentration and dramatic story-
telling, such subjects as he had treated in earlier days with his
etching-needle. His “Tam o’ Shanter,” “Grimaldi the Clown Shaved by
a Girl,” “The Runaway Knock,” “The Fairy Ring,” “Titania and Bottom
the Weaver,” “Dressing for the Day,” “A New Situation,” and
“Disturbing the Congregation,” were exhibited at the Royal Academy
or at the British Institution; and were welcomed, for the fancy, the
life, the humour that were in them—although they were one and all
crude or violent in tone, and betrayed in every part a hand
unpractised with the brush, and an eye dead to the delicacies of
colour. They were, in truth, such bits of humour or fancy as the
master humourist was wont in the old time to throw off at the rate
of two or three in a week—only laboriously rendered in oils. The
Runaway Knock, for instance, might be a plate in the “Sketch-Book,”
or in “Points of Humour”—and the remark applies to Grimaldi being
Shaved by a Girl, and the Disturbing the Congregation—which latter,
to the artist’s great delight, the Prince Consort, who was one of
Cruikshank’s cordial admirers, bought. Some of these fetched high
prices. The Fairy Ring, the most imaginative, and as a composition
the best of Cruikshank’s oil-paintings, painted in 1855, was a
commission given to the artist by Mr. Henry Miller, of Preston—the
price being £800. * The fairy revel is full of exquisitely suggestive
bits. The canvas swarms with fairy life, and abounds with fanciful
episodes.

* It is now in the possession of Captain Douglass Kennedy,

of Summerfield, Kirkby-Lonsdale, Mr. Miller’s son-in-law.

The grace and spirit with which the artist could treat fairy or elfin
life may be seen in scores of his earlier works. Look at this “Fairy
Revenge,” from “Scott’s Demonology,” drawn in 1833.

Original Size -- Medium-Size

 “The Runaway Knock” is, as the reader will perceive, simply such a
bit of Cruikshankian humour as he had been wont to treat with his
etching-needle. It is full of life and excitement. The entire
household, to the pug puppy-dog, has been aroused; nor could the
painter refrain from throwing life into the carved stone head over the
street-door. Again, “Disturbing the Congregation” is an etching
subject, elaborated. A little boy, in church, has dropped his pegtop,
and the awful eye of the beadle (Cruikshank created the British
beadle as a humorous figure) is upon him. The Prince Consort,
whom a genuine bit of humour delighted, was glad to add this most
characteristic Cruikshank to his collection.
Cruikshank’s old friend, Clarkson Stanfield, first persuaded him to
trust himself to oils. In his tinted designs, he showed that sense of
colour which was everywhere manifest in the etchings of his best
time—in his designs to Ainsworth, for instance. The watercolour
drawings for his Walter Scott plates, again, are admirable. * But in
oil, it must be repeated, he failed utterly.

* They were for a long time the property of Mr. Lumley, of

Her Majesty’s Theatre, and on the sale of his effects
passed, fortunately, into the hands of Mr. Ellison, who
bequeathed his collection to the South Kensington Museum.

The touch of the etcher remained. He was hard and crude. The
first painting he exhibited was “Bruce attacked by Assassins”—the
Bruce upon a burlesque horse smothered in drapery! It was
exhibited at the Royal Academy. His next picture was “Moses
dressing for the Fair”—a subject more within his power; but it was
coarse, inharmonious, and sketchy. The wonder was that Cruikshank
could not perceive that he was on the wrong road. So far, however,
was he from suspecting this, that he was constantly meditating great
historical subjects; and actually “got in” upon a spacious canvas the
Battle of Agincourt. He even began a scriptural subject, “Christ riding
into Jerusalem.” But the genius that could realize a street or fairy
mob * upon a surface no broader than the palm of the hand, could
not paint a battle-piece. Without his outline he was all abroad. The
sacred subject remained in the studio, with many other canvases, to
the end. It was his “Battle of a Gin Court,” in his “Sunday in London”
that showed the master. He admitted, when it was suggested to him,
that the “etching-point feeling” was always in his fingers, giving a
“living” sensation to the brush, and that he never could get rid of it.
His Falstaff tormented by the Fairies, was, on the whole, the painting
he completed with most thorough satisfaction to himself.

* “Cruikshank’s crowds give one exactly the impression of

reality. They show a certain monotony, from the common
impulse of the mob, yet they are full of characteristic
figures, no two exactly alike. There is also all the due
sense of air, and motion, and fluctuation about them. They
are penetrable crowds, especially the Irish, which he
delights to draw,—true mobiles,—ready to break out into
new mischief, or disperse before the onslaught of the
Saxon.”—Francis Turner Palgrave.

Mr. Wedmore, dwelling on the shortcomings of biographers,

complains that where an artist is the subject they tell “not much of
the work he had planned but never executed; work, nevertheless, on
which perhaps he had set great store, and looked forward to
completing, and ‘purposes unsure.’ ‘That weighed not as his work,
yet swelled the man’s amount.’
“You should see the comedies I have not written,” said a pensive
playwright. Cruikshank was, throughout his life, disturbed by
unfulfilled dreams of great subjects, with which he felt his genius
could cope. He would have grappled with Milton, as we have seen,
but hard fate kept him tied to bread-and-cheese work, and to minor
themes. His “Pilgrim’s Progress” remains unfinished, and, even so far
as he executed it, unpublished. *

* The plates are in the possession of Mr. Truman.

“Ah! who shall lift that wand of magic power,

And the lost clue regain?
The unfinished window in Aladdin’s tower,
Unfinished must remain.”

It would have gone sadly then, when the publishers could no

longer find profit in his work, when the public had turned from his
old-fashioned humour and fancy, to the fresher and more peaceful,
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com