SYSTEMATIC RISK FACTORS IN AUSTRALIAN
SECURITY PRICING
A Thesis
Submitted in fulfilment of requirements for the Degree of
Doctor of Philosophy in Economics and Finance
BY
MAZHARUL HAQUE KAZI B.Sc. (Hons) Economics, M.Sc. Economics, M.Com (Hons) Economics and Finance
School of Economics & Finance
College of Law & Business
University of Western Sydney
New South Wales
Australia
JULY 2004
Mazharul Kazi Ph.D Thesis
ii
CERTIFICATE OF ORIGINALITY
1. I certify that the substance of this thesis has not already been submitted to any other degree
and this is an original work of the undersigned for the proposed degree.
2. I certify that the substance of this thesis is not currently being submitted for any other degree
except this one; and
3. I certify that any assistance received in writing this thesis, and sources used, have duly been
acknowledged.
Sd/
Mazharul Haque Kazi
University of Western Sydney
Mazharul Kazi Ph.D Thesis
iii
DEDICATED TO
MY
PARENTS
Mazharul Kazi Ph.D Thesis
iv
ABSTRACT
In the economic environment of the information age, the performance of the stock market is
considered an important indicator of the health of a nations economy. Typically, the
performance of any stock market is reflected through stock market prices. When the stock
market tumbles, investors and others become nervous about the weakness of the economy.
When the stock market is strong and steady, everyone senses economic prosperity. It would
not be over emphasizing to state that, now the stock market is shedding value, it is having a
tremendous influence in shaping the overall economies of most developed nations around
the globe.
Investors come from various segments of the population having in common the capacity
and willingness to invest in stocks. The success of investors depends on their knowledge
and effectiveness in decision making. Some are effective enough to make a profit from
investing in the stock market, while others lose. There are many theories and schools of
thought operating in the field of stock investment. An abundance of literature, training and
analytical tools is available to assist investors in making their investment decisions.
Investors need especially to understand the stock pricing mechanism. Despite the
availability of assistance, the whole process seems to be very difficult for most investors.
The common investor understands that stock market investment is not completely risk-free.
Investment risk is quite different from other types of gambling risk, because investment
decision making processes deal with risk management rather than just taking a blind risk.
Risks are usually to be understood as systematic assertions on which an investment decision
Mazharul Kazi Ph.D Thesis
v
is based. In the relevant literature, the risks of investing in stocks are of two types. One type
of risk can be diversified away, while the other cannot. The risks that cannot be diversified
away are called systematic risks. The risk factors of the systematic risk category are
identified first in an investment decision making process. Usually, the systematic risk
factors vary from market to market and country to country. A review of the literature on
finance theory shows that only a few systematic risk factors affect the stock pricing
mechanism. If a systematic risk factor is responsive to any stock market price or industry
index, investors should closely watch the movement of that factor to manage their
investment risks effectively. A further consideration is that, under the prevailing worldwide
economic integration, stock markets around the globe are highly interrelated and are
strongly influenced by each other.
Against this backdrop, we have developed for empirical examination two research
questions from the perspective of the Australian stock market. The questions are: (i) what
systematic risk factors are influential for the Australian stock market returns in both the
long-and short-runs; and (ii) is the Australian stock market linked to developed stock
markets under the influence of globalization?
Through an extensive literature survey, we have identified a priori variables by considering
them as proxies for systematic risk factors in the stock market. We have also closely
investigated the methodological approaches suitable for empirical analyses to reveal the
precise characteristics of the long-run stock market pricing process. We have developed our
model for testing through employing econometrics software such as EViews4, SPSS 11,
Microfit 4.1 and SHAZAM 9. We have performed empirical tests to ascertain whether the
Mazharul Kazi Ph.D Thesis
vi
Australian stock market is responsive to the a priori variables, and if so, which ones and to
what extent. We have applied cointegration techniques to help answer both our research
questions. To answer the second research question, we have performed an analysis that
examined six overseas developed stock markets and asked whether the Australian stock
market is cointegrated with those markets in the long-run. We have, in other words,
conducted two studies to find scientific answers of our research questions
The results of our first study show that only a few systematic risk factors are responsible for
Australian stock market price movements in the long-run while short-run dynamics are in
force. The results of our second study confirm that the Australian stock market is being
influenced by a small number of overseas markets and it is integrated with those markets
under the influence of globalization.
Mazharul Kazi Ph.D Thesis
vii
ACKNOWLEDGEMENTS
I am grateful to Professor Tom Valentine and Dr Maria Varua for their help and support as
members of the panel supervising this PhD thesis. I am indebted to all staff members, colleagues
and friends of the School of Economics and Finance under the College of Law and Business,
University of Western Sydney (UWS), Australia, for their continued support.
I am also obliged to convey thanks to all staff members of the Post-Graduate Research Office,
Student Administration and the Libraries of UWS for their services. I acknowledge receiving help
in gathering data and other information materials from numerous sources, including SIRCA, ABS,
RBA, ASX, Morgan Stanley, HSBC, CNNFN, UNISON and Harcourt College Publishers.
I am thankful to all my friends and well-wishers both in Australia and abroad. Amongst others, in
random order, they are: Prof. Russel Cooper (UWS), Prof. Anis Chowdhury (UWS), Associate Prof.
Brian Pinkstone (UWS), Dr. Kevin Daly (UWS), Dr. Girija Mallik (UWS), Prof. Raja Junanker
(UWS), Associate Prof. Habibullah Khan (Singapore), Prof. Mustahidur Rahman (Jahangirnagar),
Prof. Kamal Ahmed (Dhaka), Prof. Abu Wahid (Tennessee), Associate Prof. Mohammed Quaddus
(Cartin), Associate Prof. Fazlur Rahman (UNSW), Associate Prof. Baker Ahmed (Illinois),
Associate Prof. Abul Shamsuddin (Newcastle), Associate Prof. Rafiqul Islam (Macquarie), Dr.
Kamran Ahmed (Latrobe), Dr Ziaul Haque (Melbourne), Dr. Akther Hossain (Newcastle), Dr.
Khoorshid Chowdhury (Woollongong), Associate Prof. Shahidur Rahman (Nanyang), Dr.
Shamsuzzaman Miah (UWS), Dr Shudhir Load (UWS), Dr. Masudul Haque(UWS) and Dr Abdul
Haque (Sydney).
I am very lucky to have received an abundance of divine love from both of my parents Mr. Harun-
al-Rashid Kazi and the late Mrs. Aklima Khatoon. I wish my father good health and long life. I pray
for the sanctity of my late mother. Also, I am grateful to each of my three brothers and four sisters
for their encouragement and patronage.
Finally, I owe gratitude to my family my wife Shaheen Akhter Banu, daughter Labeeba Nudrat
Kazi and son Atef Shadman Kazi, for their sacrifice, understanding and support, without which it
would not have been possible for me to complete this PhD thesis.
Mazharul Haque Kazi
Mazharul Kazi Ph.D Thesis
viii
CONTENTS
Certificate of Originality ii
Dedication iii
Abstract iv-vi
Acknowledgement vii
CONTENTS viii-xi
PART I
RESEARCH OBJECTIVE
1
CHAPTER 1
INTRODUCTION
2
Research Aim 3
Research Domain 5
Empirical Approach 7
Research Progression 9
PART II
LITERATURE REVIEW
10
CHAPTER 2
THEORETICAL ASPECTS OF SECURITY PRICING
11
Remembrance 15
Security Pricing 17
-Portfolio Selection 17
-Capital Asset Pricing Model 20
-Arbitrage Pricing 22
-APT of Ross 26
-Roll and Ross APT 28
Attributes of APT 36
Mazharul Kazi Ph.D Thesis
ix
Derived APT 47
-Factor Pricing Restrictions in APT 47
-Exact Pricing in a Noiseless APT 48
-Approximate Nonarbitrage Pricing Theory 49
-Compititive Equilibrium Derivations in APT 51
-Mean-Variance Efficiency and Exact Factor Pricing 52
-Dynamics in Asset Pricing 53
Summary 55
CHAPTER 3
EMPIRICAL FACADE OF ASSET PRICING THEORY
56
Empirical Episodes of APT 57
Akin Method 75
Summary 81
CHAPTER 4
FUNCTIONAL SECURITY PRICING FROM A MARKET VIEWPOINT
83
Summary 96
PART III
EMPIRICAL APPROACHES TO SECURITY MARKET PRICING
98
CHAPTER 5
METHODOLOGICAL OUTLINE
99
Empirical Methods in Asset Pricing 102
-General Linear Regression 102
-Maximum Likelihood 103
-Generalised Method of Moments 104
Mazharul Kazi Ph.D Thesis
x
-Seemingly Unrelated Regression 104
Factor Analysis 105
-Strict Factor Model 108
-Approximate Factor Model 109
-Conditional Factor Model 110
-Multi-Factor Models 111
Time Series Analysis 113
Cointegration 113
-Applications of Cointegration Analysis 116
Summary 134
CHAPTER 6
MODELLING
136
Pool of a priori Variables 136
-Variables for Modelling 140
Modelling 142
Summary 149
PART IV
METHODOLOGY
151
CHAPTER 7
(EMPIRICAL ANALYSIS 1)
ANALYSIS OF SYSTEMATIC RISKS FOR THE AUSTRALIAN STOCK
MARKET
152
Data and Variables 153
Hypothesized Relationships of Variables 157
-Unit Root and Breakpoint Tests 161
-Lag Length Selection 164
Cointegration Analysis 165
Mazharul Kazi Ph.D Thesis
xi
Caution 176
Summary 177
CHAPTER 8
(EMPIRICAL ANALYSIS 2)
INTERDEPENDANCE OF STOCK MARKETS UNDER GLOBALIZATION
178
Background 179
Data and Variables 182
Preliminary Tests 183
Analysis 1 184
-Cointegration Analysis 186
-Suplementary Tests 193
Analysis 2 194
-Structural Approach 195
-Autoregressive Approach 198
-Supplementary Tests 199
Summary 203
PART V
CONCLUSION
204
CHAPTER 9
CONCLUSION AND RECOMMENDATION
205
Outcomes 205
Recommendations 210
Future research 210
BIBLIOGRAPHY 211-226
ANNEXURE 1 227-232
Mazharul Kazi Ph.D Thesis 1
PART 1
RESEARCH OBJECTIVE
This part provides an entry into the thesis. It consists solely of one chapter, which
introduces the basic problem areas, objectives, scope, and processes of the overall
research program.
Mazharul Kazi Ph.D Thesis 2
CHAPTER 1
INTRODUCTION
Since the start of human civilization, we have been trying to make our lives evenly
structured in every sphere of activity. However, civilization often fails to act in an
orderly fashion. Nature is not orderly. Institutions are created as a response to
disorderly nature to minimize uncertainty. Still, human created institutions are not
perfect. World economies and global capital markets also lack orderliness. The capital
market in particular is our own creation, operating in a controlled environment, but we
do not fully understand how it works. We have been trying to understand how capital
flows from one investor to another; what causes price movements; how capital markets
behave; and why they behave in this fashion. We often attempt to explain the lack of
market orderliness and try to make capital markets neater. We develop models and
theories based on simplifying assumptions under a controlled environment to help us
understand markets [Peters (1991)].
One may legitimately ask: why should we bother with theories and models? Why do
we not go to the real world and study it? Why do we not get the facts? To answer these
questions, Watson (1963) indicated two difficulties with facts. First, it is not always
easy to determine exactly what a fact is. Secondly, there are simply too many facts
about economic life. Clearly, it is impossible to get all of the facts. Therefore, we must
depend on theory, which is the systematic description of reality. Theory selects
essential features and shows the connections between them. It consists of
generalizations and pronouncements of causal relationships. All theories of economics
Mazharul Kazi Ph.D Thesis 3
and finance consist of models. Models are sets of interconnected economic
relationships.
The models of any theory are clear and exact. Their foundations are carefully specified,
and so are the relationships among the variables. However, as simplifications of reality,
models have their limitations which theory recognizes. As there are always some
aspects of reality left out of the models, there must be caution in using them. Similarly,
theories may not completely explain the real world, yet they reveal some of the
structures of reality and guide us to understand the facts of our economic and financial
environment.
Research Aim
Both the national and international role of any security market is to provide a facility in
which investors and enterprises can come together with confidence to create prosperity
through sharing of risks and rewards. The security market helps facilitate the flow of
funds from investors to productive enterprises; this eventually stimulates economic
growth, creates national wealth, and generates employment and stability within society.
An effective security market is therefore a necessary condition for corporate vitality in
any national economy. It provides three principal opportunities: trading equities, debt
securities, and equity and index derivatives. Additionally, the security market is an
important conduit for the overseas flow of equity investments in any nation.
Mazharul Kazi Ph.D Thesis 4
Capital markets around the globe have an impact on the performance of national
economies. Economic activities are interrelated with capital market movements. Capital
markets are very volatile. When the market is bullish, it is generally a sign of a strong
economy. On the other hand, a bearish market indicates a weak economy. The volatility
of security prices has become a real phenomenon. A capital market can crash and shake
the whole economy. Whether the market crash is a signal for necessary correction or
refers to a downturn in economic activities has been the subject of ongoing research in
every country. How are securities being priced? Should the pricing mechanism rely on
fundamental, technical or behavioural variables? Although there are a good number of
theories and models available to explain some of these problems, shortcomings are still
evident. We aim here to identify and assess factors that contribute to changes in
security market prices. We first attempt to understand the prevailing theories and
empirical approaches taken to investigate the long-term relationships among the
variables, including their dynamic co-movement in the adjustment process to long-term
equilibrium in the Australian stock market.
This research investigates two issues. First, we determine whether there exists, based
on finance theory and empirical evidence, any long-term relationship between
systematic risk factors and Australian stock market returns, and if so, which ones.
Secondly, we study selected cross-country stock market relationships to determine
whether the Australian stock market is cointegrated with other developed stock markets
under prevailing global economic and financial settings.
More specifically, our research questions are:
Mazharul Kazi Ph.D Thesis 5
(i) Whether our a priori variables (derived from literature reviews,
intuition, and related checks and balances to proxy systematic risk
factors) influence the long-term return generating process in Australia,
and if so, whether these risk factors are being priced in the market.
(ii) Whether the Australian stock market is integrated to developed markets
(namely, the US, UK, Canada, Germany, France, and Japan) under the
influence of globalization, and if so, whether it is dynamic in nature.
Research Domain
There are many competing and complementary models and theories in asset pricing.
There are several models operating even within stock market pricing theory. Of the
prevailing models, arbitrage pricing theory (APT) is the most recent. APT has two
forms: the basic form of Ross, and an equilibrium form. APT can accommodate both
for latent factors and pre-specified factors in its empirical analyses, in line with its two
schools of thought. Additionally, linkages of APT to other models in financial theories
are inevitable. APT is an outcome of the empirical test on the capital asset pricing
model (CAPM) of Sharpe and Lintner (1960s) performed by Ross in the mid 1970s.
CAPM is often understood as a special case of APT. CAPM was also heavily
influenced by Markowitzs (1952) portfolio theory. Furthermore, there is evidence that
APT is linked with market equilibrium, efficient market hypothesis (EMH), market
microstructure, volatility, fundamental asset valuation theories, economic development,
Mazharul Kazi Ph.D Thesis 6
growth, and so on. Technological innovation, globalization, international market
characteristics, financial liberalization and speculative bubbles are also, to some extent,
linked with the development of APT models. There is confusion about the implications
of a multi-factor APT and multi-beta CAPM [Shanken (1995b), and others]. Recent
works in the area of asset pricing have even attempted to explore its linkages to an
intertemporal, dynamic, nonlinear, and chaotic world environment.
APT has not been without controversy. Shanken (1992, 1995a) and others emphasized
that for individual securities, the approximation implied by Rosss APT has been so
imprecise that it makes it impossible to test whether the APT is true or false. Shanken
argued that as the expected return for any security or portfolio is related only
approximately to its factor sensitivities, to find an exact pricing relationship further
assumptions are required. He argued that researchers who attempted to test APT were
actually testing an equilibrium form of APT rather than Rosss APT. Hence, they
would have confronted the inherent difficulties in identifying appropriate risk-free
assets and market portfolio in one-factor modelling that arise when testing the CAPM.
Dybvig and Ross (1985) later responded to these criticisms of APT. However, it is
understood that, like the CAPM, there are fundamental limitations to any empirical
testing of APT, especially in explaining the priced risk factors in a security market.
APT is essentially a multi-factor or k-factor model of asset pricing, where the factors
are not explicitly described. This matter of non described, unidentified factors is
perhaps an advantage that enables researchers to accommodate currently influential
variables in their factor modelling. Hence, researchers have an opportunity to progress
Mazharul Kazi Ph.D Thesis 7
with APTs inherent and ongoing and challenges for empirical testing. APT, like
CAPM, belongs to positive rather than normative theory. Thus, there exists an
opportunity in developing an APT model in such a way that represents positive as well
as normative theoretical views, or a positive approach that reflects normative
viewpoints. This amalgamation could be pursued in a market, which is not complete,
and in which equilibrium bubbles persist due to limited information or otherwise and
all investors are not equally informed. These are the incentives for pursuing research to
find out if the systematic or common risk factors are correctly priced in any stock
market.
Empirical Approach
The APT approach essentially seeks to measure the risk premia attached to various risk
factors and attempts to assess whether they are significant and priced into stock market
returns. There are quite a few approaches to studying security pricing behaviour,
namely, standard factor analysis, cross-sectional regression analysis, principal
component analysis, maximum likelihood analysis, multivariate analysis, and
generalized method of moments. Since APT analysis that considers factors as
inherently latent has shortcomings, the alternative approach that pre-specifies a set of
variables to act as common factors often performs well. This requires determining a
priori variables. These observable variables are pre-specified in the context of the
contemporary environment and economics and finance theory. The APT model
developed by Ross (1976) has been the most common framework in assessing the
influence of different factors.
Mazharul Kazi Ph.D Thesis 8
An alternative approach is cointegration analysis. The cointegration approach is very
useful in examining the relationships between economic variables and stock markets
from an empirical viewpoint. Chen, et al. (1986) provided the basis for the view that a
long-term equilibrium relationship exists between stock prices and macroeconomic
variables, and Granger (1986) verified this notion through cointegration analysis. A
simple statement of the cointegration approach is that a set of time series variables is
said to be cointegrated if they are integrated of the same order and a linear combination
of them is stationary [Maysami and Koh (2000)]. Such linear combinations would then
point to the existence of a long-term relationship between the variables [Johansen and
Juselius (1990)]. The advantage of cointegration analysis is that through building an
error correction model, the dynamic co-movement among variables and the adjustment
process towards long-term equilibrium can be examined.
To address both of our research questions we have predominantly used the
cointegration approach as this is the most recent methodology trend in empirical
testing. In our analysis of how asset pricing in Australia responds to systematic risk
factors, the a priori variables to proxy for systematic risk factors in asset pricing were
determined from empirical evidence and the unique conditions of the Australian social,
economic and market environment or from simple intuition. We first considered 15 a
priori variables and conducted some basic tests on them, which allowed us to keep only
six for ultimate consideration. For our second analysis, we considered seven developed
national stock markets indices as proxies and essentially applied cointegration as our
method of analysis from the perspective of the Australian stock market.
Mazharul Kazi Ph.D Thesis 9
Research Progression
We aim to pursue our study into empirical evidence for the Australian stock market
within the boundaries of asset pricing models and theories. In the following chapters
we have reviewed the existing literature in relation to security pricing theories and
empirical research. Chapter 2 reviews the literature from a theoretical perspective,
whereas Chapter 3 does the same from an empirical viewpoint. Chapter 4 surveys
various studies relevant to the security market pricing processes. Chapter 5 investigates
the empirical methodology appropriate to our study and Chapter 6 sets out our models
for the necessary analyses. Chapters 7 and 8 conduct the cointegration and related tests
and provide results. Chapter 9 concludes the thesis with its recommendations for future
research.
Mazharul Kazi Ph.D Thesis 10
PART II
LITERATURE REVIEW
This part of the thesis reviews the literature on asset pricing from theoretical, empirical
and stock market perspectives. It contains three chapters that investigate the whole
gamut of prevailing theories, empirical aspects and intuitions relating to stock markets
and its pricing mechanisms. Chapter 2 reviews the literature on asset pricing and other
theories. Chapter 3 investigates security pricing theories from an empirical perspective.
Chapter 4 examines recent trends in empirical studies on security pricing phenomena.
Mazharul Kazi Ph.D Thesis 11
CHAPTER 2
THEORETICAL ASPECTS OF SECURITY PRICING
Modern financial analyses are concerned with sources of common risk factors that
contribute to changes in security prices or values. By identifying such factors, investors
may be able to control their investment risks more efficiently and even improve their
investment returns [Kritzman (1995)]. Since risks and returns together play a vital role
in the process of security pricing, it is important to understand security market
behaviour.
There are many competing and conflicting theories of the determinants about the price
of a stock. These theories accord with their corresponding schools of thought
[Sivalingam (1990)]. There are a number of schools of thought in finance, including the
behavioural finance school. The behavioural or primeval school believes that the price
of a stock depends on what the next person in the market is willing to pay for it. In
other words, the value of a stock depends entirely on the psychology of investors in the
market. This school of thought considers that investors behave as a mob and tend to
build castles-in-the-air when stock prices move up. In other words, irrational investors
tend to push prices up because of optimistic expectations and tend to panic-sell when
they are pessimistic. An intelligent investor is advised by the advocates of this school to
read the mind of the crowd or the market and make buying or selling decisions just
before the crowd starts doing so. If the investor is able to outsmart the crowd, he/she
will be able to make abnormal or extraordinary profits out of such investment. Another
school, known as technical analysts or chartists, believes that there are repetitive
Mazharul Kazi Ph.D Thesis 12
patterns in movements in stock prices. Accordingly, to some extent, chartists agree with
the castle-in-the-air paradigm that the past market pricing behaviour of a stock contains
useful information that can be used to predict future prices [Sivalingam (1990)].
Fundamental analysts disagree with the castle-in-the-air concept and the chartists. The
fundamentalists claim that the price of a stock is determined by its intrinsic value or
market fundamentals. Accordingly, if the stock is selling at a price lower than its
intrinsic value (discounted present value of future cash flow), investing in that stock
would generate extraordinary or abnormal profits for investors. In essence, the stocks
intrinsic value and the market price would determine the buying or selling decision for
investors. Thus, if the stock is under-valued, investors should buy and if the stock is
over-valued, investors should sell. Fundamental analysts claim that they have a
methodology for determining whether a security is currently over-valued or under-
valued [Sivalingam (1990)].
Neo-classical theorists do not agree with the castle-in-the-air paradigm, the chartists or
the fundamentalists. They argue that the information contained in the past price or
market behaviour of a stock is already reflected in todays price. The neo-classicists
believe that prices follow a random walk and are unpredictable. Accordingly, they do
not recommend that investors look for bargains in the stock market [Sivalingam
(1990)].
In other words, the neo-classical school argues that, because the market is efficient,
there are no under-priced securities; hence no opportunity for a bargain exists in the
Mazharul Kazi Ph.D Thesis 13
market. Their advice for investors is to pick stocks based on the stocks risk and return
characteristics. This recommendation is based on their basic assumption that the return
on a stock is dependent on its risk class. The riskier the stock, the higher is its return
and vice versa. Modern portfolio theory, capital asset pricing theory, and arbitrage
pricing theory follow the neo-classical school of thought.
Finance theory is a branch of economics that customarily focuses on capital markets.
In the early life of the subject, emphasis was placed on describing the market
environment and valuing individual securities. In more recent years, modern financial
theories lay emphasis on broader aspects of the valuation of assets whose
characteristics extend across time, and which impose composite and complex risk on
investors. The development of asset pricing theory includes the valuation of a wide
range of financial assets and derivatives. One can present this development as a
coherent body of theory, rather than disjointed components. In particular, modern
financial theory is founded on three central assumptions, namely, markets are highly
efficient; investors exploit potential arbitrage opportunities; and investors are rational
[Dimson and Mussavian (1999)].
Within the neo-classical school, a common taxonomy usually divides modern financial
theories between normative and positive approaches. Normative theories are
prescriptive, and positive theories are descriptive. When a theory embraces some norms
or standards, mixing them with cause-effect analyses, it is normative or prescriptive
theory. When the theory confines itself to statements about cause-effect and the
fundamental relations that prevail, it is positive or descriptive theory [Watson (1963)].
Mazharul Kazi Ph.D Thesis 14
Markowitzs path-breaking mean-variance portfolio theory falls into the former
category, dealing as it does with rules for optimal portfolio choice by an individual
investor. Asset pricing theory can be classified as belonging to the latter, since it is
concerned with the determination of the prices of capital assets in a competitive market
environment [Sharpe (1990)].
The domain of positive financial theory is often further divided into a set of models that
are described as utility-based and a complementary set that are described as arbitrage-
based. Models that fall in the latter category derive implications from the assumption
that capital asset prices will adjust until it is impossible to find a profitable strategy that
requires no initial investment. Models that fall in the former category typically conform
to the conditions required for the latter, but produce stronger implications due to added
assumptions about the utility functions that investors are assumed to maximize [Sharpe
(1990)].
According to Popper (1934), any scientific theory should lead to propositions that are
potentially verifiable or replicable by experimental or empirical observation. Asset
pricing theory is no exception. The large volume of capital investment using these
models around the globe has created further pressure to continue research into the
development of appropriate techniques for testing these theories empirically. The
literature on asset pricing is crucial to advance our understanding about the systematic
or common risk factors that affect investment decisions as well as to develop
appropriate models that explain the microstructure of financial markets.
Mazharul Kazi Ph.D Thesis 15
Remembrance
Many issues addressed in modern finance are traced back to a remarkable paper
presented to the Imperial Academy of Sciences in St Petersburg by the Swiss
mathematician Bernoulli (1738). Bernoulli suggested that increases in wealth would
result in an increase in utility inversely related to the quantity of goods in an
individuals possession. This enabled him to demonstrate the trade-off between the
expected change in wealth and the risk associated with such an opportunity. Bernoullis
idea of decreasing marginal utility became central to economics by the late nineteenth
century, notably in price theory and the associated theoretical developments of the
notable late nineteenth century economists Walras (1834-1910) and Marshall (1842-
1924). In the literature of price theory, the concept of Bernoullis utility theory is
known as the St Petersburg Paradox [Watson (1963)]. Bernoulli also introduced the
concept of the maximization of expected utility. Despite endorsement by the scholar
Laplace, Bernoullis approach had little impact on the economics of decision making
under risk until the development of expected utility theory by Von Neumann and
Morgenstern during the mid 1940s and Savage in 1954 [Dimson and Mussavin (1999)].
In earlier ages, any economic problem was the topic of price theory that explained the
composition or allocation of total resources including capital. Price theory was the
centre of attention for economists during the eighteenth and nineteenth centuries, when
it was known as the theory of value. The theory of value played a leading role as one of
the intellectual foundations built after 1776. The theory of value was also at the heart of
the old controversies over capitalism and socialism. Karl Marxs critique of capitalism
Mazharul Kazi Ph.D Thesis 16
was based upon his special version of the theory of the value of commodities. Price
theory or the theory of value, therefore, has for a long time been the most important
issue in economics and finance [Watson (1963)].
The concept of risk is now endorsed in economics, and especially in modern finance.
However, this was not the case until 1921 when Knight made a distinction between risk
and uncertainty. According to Knight, when the randomness of an event can be
expressed in terms of numerical probabilities, the situation is said to involve risk. When
probabilities cannot be assigned to alternative outcomes, the situation is called
uncertainty. In a setting that embraces both risk and uncertainty, the model of general
equilibrium developed by Arrow and Debreu has been fundamental to economics and
finance since the early 1950s [Debreu (1959)]. Arrows theory of equilibrium operating
with incomplete asset markets is seminal. Arrow showed that by using the temporal
structure of the economy, equilibrium can be attained with a more limited number of
markets. He explained how markets that are almost complete through a series of
contingent claims resolve uncertainty. Arrows work has provided a conceptual
framework for the theory of asset pricing [Dimson and Mussavin (1999)].
Arrows concept of a complete market provided an opportunity to secure an individual
investor from any loss. In an economy where it is possible to insure against the range of
possible future outcomes, individuals are more likely to be willing to bear risk. It
provides encouragement to investors to hold diversified portfolios, rather than putting
all their eggs in one basket. Arrows framework set the direction for Markowitz (1952)
to take a more structured analysis of investors portfolio selection decisions.
Mazharul Kazi Ph.D Thesis 17
Security Pricing
Portfolio Selection
With the seminal work of Markowitz (1952), finance as a discipline achieved its
creditable place in the modern era. Ever since the day of Bernoulli, it was clear that
individuals would prefer to increase their wealth, and also to minimize the risk
associated with any potential return. Markowitz rejected the previous idea that there
might be a portfolio which gives investors both maximum expected return and
minimum risk. He showed that the portfolio with maximum expected return is not
necessarily the one with minimum variance. There is a rate at which the investor can
gain expected return by taking on variance, or reduce variance by giving up expected
return [Reilly and Norton (1999)].
Markowitzs (1990) work on portfolio theory primarily considers how an optimizing
investor would behave in a market environment. Yet, portfolio theory differs in three
major ways from marketing theories, that is, from theory of the firm and theory of the
consumer behaviour. First, it is concerned with investors rather than consumers or
firms. Secondly, it is concerned with economic agents who act under uncertainty.
Finally, it is a theory that is supposed to direct prices rather than quantity of products
supplied in the market or economy.
Mazharul Kazi Ph.D Thesis 18
The important contribution of Markowitz is his distinction between the variability of
returns from an individual security from a risky portfolio. He indicated that to make the
variance small it is not enough to invest in many securities. Rather, it is necessary to
avoid investing in securities with high covariances among themselves. Markowitz
demonstrated that one could identify a set of portfolios that provide the highest possible
expected return for a given level of risk or the lowest level of risk for each level of
expected return. These portfolios form an efficient frontier. He showed that the trade-
off between returns and risks would be economically efficient for the portfolios that fall
on this frontier. Markowitz (1959) also described a method of modelling the consumer
choice problem. If the set of portfolios that satisfy the mean-variance criterion is
identified and investors preferences can be modelled, then the investors chosen
portfolio would be the combination of risky assets that maximize expected utility
[Brailsford and Heaney (1998)].
The safety-first approach of Roy (1952) has many similarities to Markowitzs portfolio
selection framework. Roys model addressed the question of how an investor can
ensure a suitably small probability that wealth falls below some disaster level, which is
essentially the same problem that Markowitz addressed in a different fashion. Roys
approach was presented in a linear form that considered risk as an independent variable
and return as a dependent variable, and was adopted by finance professionals for quite
some time. Still, Markowitz is regarded as the pioneer of portfolio theory [Dimson and
Mussavin (1999)].
Mazharul Kazi Ph.D Thesis 19
Tobin (1958) extended Markowitzs analysis by showing how to identify, out of a set
of efficient portfolios, which one should be held by an individual investor. He showed
how investors should divide their funds between safe liquid assets and risky assets and
proposed a framework for asset allocation that is intuitively appealing. Tobin suggested
breaking down the portfolio selection problem into stages at different levels of
aggregation-allocation, first among and then within, asset categories. The asset mix
should reflect the degree of risk tolerance of the investor. The optimal portfolio of risky
assets, however, should be independent of the risk preferences of the investor. This
proposition, known as the separation theorem, provides a basis for identifying the
efficient portfolio and clarifies the task of portfolio selection. Yet, Tobins theorem
requires the uses of Markowitzs full covariance model to embrace the analysis of
individual securities as well as individual investors. However, the data and
computational requirements of this approach are enormous unless the universe of
securities is limited to a manageable level. Sharpe (1963) addressed these difficulties
when he devised his simplified model for portfolio analysis [Dimson and Mussavin
(1999)].
Sharpe (1963) drew on an insight of Markowitz (1959) that stocks are likely to co-
move with the market. His model assumes that security returns linearly relate to
fluctuations in a market index with a known degree of sensitivity. Security-specific
returns are generated with a known mean-variance. With only three parameters per
security, the tasks of risk measurement and portfolio optimization are greatly
simplified. Markowitz (1990) pointed to the fact that his portfolio theory considers how
an optimizing investor would behave, whereas the Sharpe-Lintner capital asset pricing
Mazharul Kazi Ph.D Thesis 20
model is concerned with economic equilibrium assuming all investors optimize in a
particular manner. However, Sharpes model has been extended to embrace more
complex issues concerning asset pricing in capital markets.
Capital Asset Pricing Model
Soon after the Markowitzs book on portfolio selection was published in 1959, Treynor
(1961) started to explore the possibility of developing a theory of asset pricing. Around
the same time, Sharpe set out to determine the relationship between the prices of assets
and their risk attributes. Sharpe aimed to use the theory of portfolio selection for
constructing a market equilibrium theory of asset prices under conditions of risk.
Sharpe noted that his model shed considerable light on the relationship between the
price of an asset and the various components of its overall risk [Dimson and Mussavin
(1999)].
In his Ph.D dissertation, Sharpe explored a number of aspects of Markowitzs portfolio
analysis. At that time, he called his approach the single-index model, which is now
known as the one-factor model. His key assumption was that security returns relate to
each other solely through responses to one common factor. In his dissertation he
addressed both normative and positive results. However, under the heading A Positive
Theory of Security Market Behaviour, his dissertation covered a result similar to the
capital asset pricing model (CAPM) [Sharpe (1990)].
Mazharul Kazi Ph.D Thesis 21
Sharpe (1964) indicated that one of the problems which plagued him in attempting to
predict the behaviour of capital markets was the absence of a body of positive
microeconomic theory dealing with conditions of risk. The pervasive influence of risk
in financial markets forced him to adopt models of price behaviour, models which are
little more than assertions. In assessing the desirability of a particular investment,
Sharpe (1964) noted that an investor should make decisions based on only two
parameters: the expected return and the standard deviation of returns. In other words,
he considered the mean-variance approach under certain conditions in order to ascertain
the investors preference function. Sharpes paper was supplemented by contributions
from Lintner (1965) and Mossin (1966).
A number of researchers attempted to relax the strong assumptions that underpin the
original CAPM of Sharpe, Lintner and Mossin. The most frequently cited modification
was by Black (1972). Black showed that the model needed to be modified when
borrowing at a risk-free rate was not available. His version is known as the zero-beta
CAPM. Brennan (1970) found that when taxes were introduced into the equilibrium,
the structure of the original CAPM held. Mayer (1972) showed that when the market
portfolio included non traded assets, CAPM remained identical in structure to the
original version. Solnik (1974) and Black (1974) observed that CAPM was extendable
to encompass international investments. Williams (1977) noted that the theoretical
validity of the model was relatively robust if the assumption of homogenous return
expectations was relaxed. Treynor and Black (1973) showed how best to construct
managed portfolios (as a blend of a passive portfolio and an active portfolio) by linking
the CAPM with the index model of Sharpe (1963). Research aimed at extending the
Mazharul Kazi Ph.D Thesis 22
classical one-period setting of CAPM to a continuous-time environment is generally
known as the intertemporal capital asset pricing model (ICAPM).
Until the early 1980s, CAPM was the standard and popular tool used to describe
common stock returns in an efficient market. A large number of empirical studies
published during the 1970s concluded that the CAPM was a good descriptor and
predictor of the behaviour of stock returns in the US, Europe and Japan [Hawawini
(1993)]. In the late 1970s, however, the validity of CAPM was seriously questioned, by
a number of researchers including Roll (1977), both on conceptual and empirical
grounds [Roll and Ross (1980)]. During the mid 1970s, Ross (1976) developed an
alternative pricing model called arbitrage pricing theory.
Arbitrage Pricing Theory
The core idea of Rosss arbitrage pricing theory (APT) is that only a small number of
systematic influences affect the long-term average returns on securities. Hence, APT is
a factor model. Unlike Sharpes (1963) single-index model, APT includes multiple
factors that represent the fundamental risks in asset returns and thus the prices of
securities. Multi-factor models allow an asset to have not just one, but many, measures
of systematic risk. Each measure captures the essential sensitivity of the asset to the
corresponding pervasive factor. Thus, APT is also a multi-factor equilibrium pricing
model that is more general than the CAPM. On both theoretical and empirical grounds,
APT is an attractive alternative to CAPM. It is argued that APT requires less stringent
and presumably more plausible assumptions and is more readily testable since it does
Mazharul Kazi Ph.D Thesis 23
not require the measurement of market portfolios. Often, APT explains the anomalies
found in the application of CAPM to asset returns [Dhrymes, et al. (1984, 1985)].
APT conventionally assumes that the returns on securities are linearly related to a small
number, k, of common or systematic factors rather than a single factor, β. The model
applies to any set of securities as long as their number, n, is much larger than the
number k of common factors [Hawawini (1993)]. APT does not specify what the k-
factors are; rather it has kept this open for consideration by researchers [Brailsford and
Heaney (1998), Reilly and Norton (1999)]. Moreover, the model does not require that
investors hold all outstanding securities; hence the market, which is central to CAPM,
plays no role in APT [Dimson and Mussavian (1999)].
To understand the basic formulations of both APT and CAPM, let us consider three
common factors that influence the stock market. According to APT, with Rf being the
risk-free return the expected return on asset i is
E(Ri) = Rf + λ1 b1i + λ2 b2i + λ3 b3i (2.01)
where λ1, λ2 and λ3 are three risk-premia corresponding to the three sensitivity
coefficients of asset i, that is, b1i, b2i and b3i. These sensitivity coefficients are similar to
the β coefficient in CAPM.
E(Ri) = Rf + [E(Rm)−Rf] βI (2.02)
Mazharul Kazi Ph.D Thesis 24
where the β of a stock is a measure of the sensitivity of that stocks returns to the return
of the market as a whole. Likewise, b1i is a measure of the sensitivity of the returns of
stock i to the returns of the first common factor, and so on. There is a structural
similarity between the APT model and CAPM. Assuming there is only one common or
systematic factor affecting all outstanding securities and that this factor is the return on
the market as a whole, then the APT pricing relationship collapses to the CAPM pricing
relationship. In this case the sensitivity coefficient to the unique market factor is the
beta (β) coefficient and the risk premium of APT (λ1) is the CAPM risk premium
[E(Rm)−Rf].
APT may appear to be similar to the multiple betas in the intertemporal capital asset
pricing model (ICAPM) of Metro (1973), but this is not the case. Although the CAPM
of Sharpe (1964) and Lintner (1965) identifies the factor as the return on the market
portfolio, Roll (1977) provides convincing arguments for the unobservability of market
returns. APT essentially is an arbitrage-based model, while CAPM is utility-based.
Again, Ross viewed APT as an arbitrage relation rather than an equilibrium condition
[Dimson and Mussavian (1999)].
APT has two basic forms: the first is the nonarbitrage theorem, derived by Ross (1976);
the second is an equilibrium model [Brailsford and Heaney (1998)]. Factor pricing
theories, such as the APT of Ross and the equilibrium APT, occupy a central place in
the theory of asset pricing. The APT of Ross and the equilibrium APT extend the basic
intuition of CAPM to large diversified asset markets without imposing strong
Mazharul Kazi Ph.D Thesis 25
assumptions on the nature of investors preferences and asset return distributions
[Nawalkha (1997)].
Given perfectly competitive and frictionless capital markets where investors believe in
a k-factor return generating model, APT assumes that random returns for n assets,
denoted by R(i), for i= 1, 2, n, equal the expected return for each asset E(i), plus k
risk premiums B(i,k)D(i). Each risk premium is determined by the k market risk
components D(i), and the sensitivity of the asset to that risk component B(i), plus an
unexplained residual term e(i), as follows:
R(i) = E(i) + k∑ B(i, k) D(i) + e(i) (2.03)
The arbitrage portfolio has a proportion of wealth x(i) invested in each asset and has
returns equal to x(i)R(i, k):
x(i)R(i,k) =x(i)E(i) + x(i)B(i,1)D(1) +
+ x(i)B(i,k)D(k) + x(i)e(i) (2.04)
Purchases of new assets will exactly equal the proceeds of short sales of other assets, so
that the portfolio has no net effect on investment. The x(i) are chosen in such a way that
the sum of the x(i)B(i,k) equals zero. This implies that the portfolio has no systematic
risk. As the number of securities in the portfolio increases, the unsystematic risk
approaches zero:
Mazharul Kazi Ph.D Thesis 26
i
∑ x(i) e(i) = 0
Thus, an arbitrage portfolio has no net investment and no risk (either systematic or
unsystematic) and earns no return. For other portfolios, with both risk and investment,
returns on assets are a linear combination of a risk-free rate λ(0), plus a risk premium
for each risk component to which the asset is sensitive, x(i)B(i, k)λ(k):
E(i) = λ(0) + k
∑ x(i) B(i,k) λ(k) (2.05)
A risk-free or zero beta portfolio, for which all B(i,k) = 0, would yield the risk-free rate
of return [McGowan and Francis (1991)]: E(i) = λ(0) = E(0).
APT of Ross
Ross (1976) started with the following return generating process for securities to
construct his form of the APT model for testing purposes:
rt = Et + ft B + ut (2.06)
where rt is an m-element row vector containing the observed rates of return at time t for
the m securities. Et is an m-element row vector containing the expected (mean) returns,
while ft is a k-element vector of common (but unobservable) factors affecting security
returns, both at time t. B is a k×m matrix of parameters indicating the sensitivity of
Mazharul Kazi Ph.D Thesis 27
securities to the common factors, and ut is the idiosyncratic component of the error
term. Ross showed that if the number of securities, m, is sufficiently large, there exists
a (k+1)-element row vector ct such that
Et = ct B*, t = 1, 2, T (2.07)
where B*′ = [e:B′] and e is an m-element column of ones. The empirical tests of the
APT model, therefore, are based upon a two-stage factor analytic approach [Dhrymes,
et al. (1985)].
Not all studies supported APT. However, most studies reported positive relationships
between the APT factors and asset cross-sectional mean returns. In the early 1980s, at
least three studies directly addressed the problem of factor identification for testing
APT. Fogler, et al. (1981) attempted to define the relationship between estimated
factors and a market index, the US Treasury bond rate, and AA (Double A
class/quality) utility bond rates. They indicated that there was a significant relationship
between security returns and these three independent variables. Anckonie (1983)
employed a more systematic approach to factor identification. He used principal
component analysis to decompose the variation of quarterly returns for two industries:
large commercial banks and regulated utilities. Seven significant factors were found
through this study. These seven factors were then regressed against 13 macroeconomic
variables. Significant relationships were found for each APT factor. Anckonies results
also indicated that industrial production, inflation, and interest rates are the three most
important macroeconomic factors. Chen, et al. (1983) found a statistically significant
Mazharul Kazi Ph.D Thesis 28
relationship between factors using security returns and unanticipated inflation, long-
term interest rates, industrial production, and risk premium. Chen, et al. (1983) did not
find statistically significant relationships with a market index, consumption, or oil
prices. Their results were similar to those of Anckonie (1983).
Later, Leudecke (1984) and Trzcinka (1986) both explored the Chamberlain and
Rothschild (1983) concept of an approximate factor structure using US data. Leudecke
(1984) examined a sample of 392 assets and found that the first eigenvalue is dominant.
Even for his large sample, the approximate size of the k-factors was not obvious, as his
analysis did not decisively distinguish between the approximate one-factor structure
and the many-factor structure. Trzcinka (1986) conducted another analysis on 865 firms
over 1,069 weeks. He found one large factor and no obvious way to choose more than
one. He also suggested that the number was very unlikely to be zero, which supported
the crucially fundamental APT assumption of the existence of an underlying factor
structure.
Roll and Ross APT
Most APT tests employed the methodology suggested by Roll and Ross (1980),
commonly known as the RR method [Hawawini (1993)]. Roll and Ross (1980)
followed a two step procedure in their empirical tests of APT. In the first step, they
estimated the expected returns and the factor coefficients on individual asset returns
from time series data. In the second step, they used the estimates from the first step to
test the basic cross-sectional pricing conclusion of APT.
Mazharul Kazi Ph.D Thesis 29
Roll and Ross (1980) used daily return data, from July 1962 to December 1972, for a
sample of 1,260 NYSE and AMEX listed companies. They divided their sample of
stocks into 42 groups of 30 securities each. From the return data they established each
groups variance-covariance matrix. Due to the lack of high-powered computing
facilities at that time, they analyzed each group separately. By applying maximum
likelihood factor analysis, they determined simultaneously the number of common
factors, Fj, and their corresponding sensitivity coefficient, bij, called the factor loading.
Finally, they performed a cross-sectional test of the APT model by running a linear
regression in which the average stock return, E(Ri), was the dependent variable and the
set of factor loadings, bij, were the independent variables. Their model allows them to
estimate the risk premium associated with each factor and determine the number of
factors that are priced according to APT. Roll and Ross (1980) found that for 75% of
the groups, there is a 50% chance that five factors are significant; and that three to four
factors explain the cross-sectional variation in average stock returns. They concluded
that the APT model performs well in empirical tests on securities.
During the mid 1980s, the RR method was applied to European data by Dumontier
(1986), Winkelmann (1984) and Beenstock and Chan (1983, 1984) for France,
Germany and the UK respectively. The European evidence indicated that equity pricing
was generally consistent with the RR model. However, the number of significant
common factors from the French, German and UK data varied. In the case of France,
seven common factors were significant in explaining the variance of common stock
returns. For the German data, the number of factors retained was dependent on the
Mazharul Kazi Ph.D Thesis 30
criterion employed to select those that are significant. The smallest number of factors
identified was one, while the largest number was 43. The British study suggested 15 to
20 relevant factors.
The disparity in the number of significant common factors reported might have been
due to differences in: (a) the size of groups analyzed; (b) the statistical technique
employed to extract the factors from the data; and (c) the type of criterion adopted to
select those factors. On the other hand, the cross-sectional evidence provided three to
five factors that seemed to explain the cross-sectional variation of the average returns
on the French stocks. The number was eight for the German stocks and one to eight for
the UK stocks.
A major weakness of these tests was their inability to identify the nature of common
factors since they treated them as inherently latent. An alternative approach that pre-
specifies a set of economic (or financial) variables to act as common factors was likely
to perform well. Upon determination of a priori variables, this approach of testing APT
examines whether the sensitivity coefficients of stock returns to these factors explain
the cross-sectional variation of average stock returns. Chen, et al. (1986) and Hamao
(1986) applied this approach to US and Japanese stocks respectively.
Chen, et al. (1986) tested whether innovations in macroeconomic variables are risks
and warrant reward in the stock market. They pre-specified their variables in
accordance with the theory that a few macroeconomic variables should systematically
affect stock market returns. These variables were: (i) the spread between long and short
Mazharul Kazi Ph.D Thesis 31
interest rates; (ii) expected and unexpected inflation; (iii) industrial production; and (iv)
the spread between high and low grade bonds. They found these sources of risk were
significantly priced in the US stock market.
Chen, et al. (1986) considered the common belief that asset prices are sensitive to
external forces. Consistent with the ability of investors to diversify unsystematic
influences, pervasive or systematic external influences are the likely sources of
investment risk. They understood that an additional component of long-run returns was
required whenever a particular asset was influenced by systematic economic news and
no extra reward for bearing diversifiable risk. Since APT has been silent about which
events are likely to influence all assets, a gap exists between the theoretically exclusive
importance of systematic state variables and complete ignorance about their identity.
The co-movements of asset prices suggest the presence of underlying exogenous
influences, yet it is difficult to determine the responsible variables. As an example,
Chen, et al. (1986) argued that a variable should be one that has no direct influence on
current cash flows but does describe the changing investment opportunity set.
Accordingly, stock prices of Chen et al. (1986) were determined by expected
discounted dividends:
( )E c
pk
= (2.08)
where c is the dividend stream and k is the discount rate. This implies that actual
returns in any period are:
Mazharul Kazi Ph.D Thesis 32
[ ]p
c
k
dk
cE
cEd
p
c
p
dp +−=+)(
)(
(2.09)
At least trivially, it follows that the systematic forces that influence returns are those
that change discount factors, k, and expected cash flows, E(c).
On the other hand, Hamao (1986) selected a set of eight economic variables as likely
common factors that would affect the returns of Japanese stocks. His pre-determined
variables were: (i) the monthly growth rate in an index of industrial production; (ii)
changes in expected inflation; (iii) unexpected inflation; (iv) unexpected changes in the
risk premium with the risk premium defined as the spread between government and
corporate bonds; (v) unexpected changes in the slope of the term structure of interest
rates with the slope defined as the spread between the rate on long-term government
bonds and the short-term risk-free rate; (vi) unexpected changes in the forward yen-
dollar exchange; (vii) the growth rate in oil prices; and (viii) the return of a market
index of TSE stocks. Only three factors were found to be significant for the Japanese
common stock returns. These were: (a) changes in expected inflation; (b) unexpected
change in the risk premium; and (c) unexpected changes in the slope of the term
structure of interest rates.
McGowan and Francis (1991) studied asset pricing using a multivariate statistical
analysis technique. Like factor analysis and principal component analysis, the ability of
the multivariate technique to discern interesting empirical insights depends upon the
size and heterogeneity of the sample and the way data are analyzed. Their approach had
Mazharul Kazi Ph.D Thesis 33
three stages. In stage one principal component analysis was used to calculate factor
scores for 86 different industry indices. They extracted 10 orthogonal principal
components of stock price volatility. In stage two principal component analysis was
again used to calculate factor scores for 95 macroeconomic time series from the data
set. It detected nine principal components of economic activity. In stage three each of
10 principal component scores (derived from stage one) was regressed one-at-a-time on
all nine macroeconomic variables (identified from stage two) to test statistically
significant relationships between the APT factors and macroeconomic factors. They
observed that at least one principal component of economic activity had a statistically
significant impact on each of the 10 principal components of stock price variation.
Thus, the results of McGowan and Francis (1991) verified the relationships between
stock price movements and the real sector of the economy that are suggested by
economic theory.
The seminal work of Fama and French (1993) to ascertain common risk factors in the
returns on stocks and bonds based on pre-specified financial variables has been widely
discussed. Fama and French (1993) identified five common risk factors. Out of the five,
three were significant stock market factors and two bond market factors. The stock
market factors were factors related to the size and book-to-market equity, and an
overall market factor. Bond market factors were related to maturity and default risks.
Fama and French (1993) extended their earlier asset-pricing tests [Fama and French
(1992a)] in three ways. These were: (i) the extension of the set of asset returns to be
explained; (ii) the expansion of the set of variables used to explain returns; and (iii) a
Mazharul Kazi Ph.D Thesis 34
different approach from those of Fama and French (1992a) and Fama and MacBeth
(1973) in testing asset pricing models.
Fama and French (1993) used the time series regression approach of Black, Jensen and
Scholes (1972). Monthly returns on stocks and bonds were regressed against the returns
on a market portfolio of stocks and mimicking portfolios for the size, book-to-market
equity (BE/ME), and term structure risk factors in returns. The time series regression
slopes were treated as factor loadings that, unlike size or BE/ME, had a clear
interpretation as risk factor sensitivities for the securities. Time series regressions were
used to study two additional issues of asset pricing. The first was whether variables
related to average returns (for example, size and book-to-market equity) act as proxies
for sensitivity to common risk factors in returns for rationally priced assets. The time
series regressions gave direct evidence on this issue. Their regression slopes and R2
values showed that mimicking portfolios for risk factors related to size and BE/ME
captured shared variation in stock and bond returns not explained by other factors. To
study the second issue, Fama and French (1993) used returns (monthly stock/bond
returns minus the one-month Treasury bill rate) as dependent variables in their time
series regressions, and excess returns or returns on zero-investment portfolios as
explanatory variables. In such regressions, their well-specified asset-pricing model
produced intercepts that were indistinguishable from zero [Merton (1973), Fama and
French (1993)].
Fama and French (1993) also checked the robustness of their inference in the following
manner. First, they used the residuals from the five-factor time series regressions to
Mazharul Kazi Ph.D Thesis 35
check that the regressions captured the variation through time in the cross-section of
expected returns. Secondly, they examined whether their five risk factors captured the
January seasonal in stock and bond returns. Thirdly, they contemplated split-sample
regressions that used one set of stocks in the explanatory returns and another disjoint
set in the dependent returns. Finally, they examined whether the stock-market factors
that captured the average returns on portfolios were informative enough.
In observing the predictability of the regression residuals, Fama and French (1993)
used (i) excess returns or returns on zero-investment portfolios; (ii) dividend yields
(D/P); (iii) spreads of low-grade over high-grade bond yields (that is, default spreads or
DFS); (iv) spreads of long-term over short-term bond yields (that is, term spreads or
TS), and (v) short-term interest rates (that is, one-month bill rate or RF). They
hypothesized that if these five factors captured the cross-section of expected returns,
the predictability of stock and bond returns should be embodied in the explanatory
returns in the regressions. To test their hypothesis, they estimated 32 time series
regressions:
ep (t+1) = k0 + k1( )
( )
D t
P t+ k2DFS(t) + k3TS(t) + k4RF(t) + ηp(t+1) (2.10)
The ep (t+1) were the time series of residuals for 25 stock and seven bond portfolios
from the regressions. Their estimates did not produce any evidence that the residuals
from the five-factor time series regressions were predictable. In 32 regressions, 15
produced negative values of 2
R . Only four R2 exceeded 0.1; the largest was 0.3. Fama
Mazharul Kazi Ph.D Thesis 36
and French (1996) concluded that, in various cases, average returns on common stocks
are related to firm-characteristics and, except for the continuation of short-term returns,
the anomalies largely disappear in a three-factor model. Their results were in harmony
with the rationales of both capital asset pricing model and APT.
Attributes of APT
Most applications of security pricing aimed to find factors that explain security returns
and the associated risk premiums using multi-factor models. Multi-factor models of
security returns are normally divided into three types: macroeconomic, fundamental,
and statistical factor models [Connor (1995)]. Macroeconomic factor models use
observable economic time series, such as inflation and interest rates, as measures of the
pervasive shocks to security returns. Fundamental factor models use the returns to
portfolios associated with observed security attributes such as dividend yields, the
book-to-market ratio, and industry identifiers. Statistical factor models derive their
pervasive factors from factor analysis of the panel data set of security returns.
Connor (1995) investigated the explanatory power of the three types of factor models
for security returns in the US capital market. He observed that statistical and
fundamental factor models substantially outperformed the macroeconomic factor
model. The fundamental factor model slightly outperformed the statistical factor model.
From the point of view of theoretical consistency and intuitive appeal, macroeconomic
factor models performed better. Although interesting and useful, his results did not
Mazharul Kazi Ph.D Thesis 37
allow him to make final recommendations about selecting a specific model out of the
three.
It is also conventional to divide factor models into two schools of thought [Zhou
(1999)]. The first school of thought takes the stand that the factors are inherently latent
and unobservable directly from market data. Rosss (1976) basic APT provides
theoretical justification for the use of latent factors. The second school of thought takes
a more pragmatic stand. Rather than identifying either observable or unobservable
factors by using any asset pricing theory, this school treats factors as pre-specified
economic or financial variables that seem related to asset returns by simple reasoning
or plain intuition.
Consistent with the latent factor school, there are two notable methods that can be used
to extract factors from asset return data. These are Connor and Korajczyks (1986)
asymptotic principal components approach, and the standard factor analysis approach
[Seber (1984)]. There are two weaknesses with these approaches. First, they rely on
information on security returns alone, making it difficult to link the estimated factors to
economic fundamentals. Secondly, there is an error-in-variables problem when the
estimated factors are used to test APT. Although the one-step procedures, namely, the
maximum likelihood and Bayesian approaches, will not be subject to an error-in-
variables problem as suggested by Geweke and Zhou (1996), they will not be as robust
as the approaches proposed by Zhou (1999).
Mazharul Kazi Ph.D Thesis 38
The advantage of approaches consistent with the pre-specified factor school of thought
is their flexibility in including useful economic, financial or perhaps any other form of
variables in the analysis. Examples of such studies include Chen, et al. (1986) and
Fama and French (1993). Yet, one may run into the danger of including too many
variables that are highly correlated with one another and hence redundant, resulting in
an overstatement of the number of factors and inaccurate estimation of parameters.
Conversely, the exclusion of important variables from consideration might also give
estimation problems. Furthermore, as highly correlated variables are included, say, in a
linear regression system, it is often difficult to interpret the associated factor risk
premiums, as they might simply reflect the same source of economic risk factor [Zhou
(1999)]. To solve the problems associated with the testing of APT, Zhou (1999)
suggested a new framework that applies to both the latent and pre-specified factor
schools.
For the latent factor school, which treats the risk factor as unobservable, Zhou (1999)
modelled the latent factors explicitly as a linear function of economic variables plus a
noise. Then he proposed to use those linear combinations of the economic variables that
best forecast the latent factors. There are three interesting aspects of this new approach.
First, it estimates the latent factors by using the best forecasts from the economic
variables, naturally linking the latent factors to economic fundamentals. Secondly, the
estimation of the factors is carried out jointly with the estimation of the factor load
parameters or sensitivity of variables. Thus, the previous error-in-variables problem is
no longer found in the test of APT. Finally, the new approach has the flexibility of
Mazharul Kazi Ph.D Thesis 39
using many economic variables to analyze a factor model with only a few factors,
which is in contrast to Chen, Roll and Ross (1986).
For the pre-specified factor school, for the purposes of estimating the minimum number
of factors needed to explain asset returns, Zhous (1999) procedure based on the
generalized method of moments of Hansen (1982) is proposed. This minimum number
is not necessarily the number of pre-specified variables. The procedure also yields the
best linear combinations (that is, minimum number) of pre-specified economic
variables to serve the non redundant factors. An interesting feature of this method is
that the factors so obtained are orthogonal to one another. Consequently, it allows the
use of potentially many pre-specified economic or other form of variables to generate a
few orthogonal factors.
The APT of Ross (1976) hypothesized that the cross-sectional distribution of expected
returns of financial assets can be approximately measured by their sensitivities, called
factor loadings or κ, to k unknown economic factors. Chamberlain and Rothschild
(1983) extended the APT of Ross (1976) and Ross (1977) which postulated that with a
factor structure the mean returns are approximately linear functions of the factor
loadings, and provided a unique approach often known as principal components
analysis. This extended approach indicates that when κ eigenvalues (the weights in the
principal components) of the population covariance matrix increase without bound as
the number of securities in the population increases, elements of the corresponding κ
eigenvectors of the covariance matrix can be treated as the factor sensitivities.
Mazharul Kazi Ph.D Thesis 40
Connor and Korajczyk (1988) have shown that Chamberlain and Rothschilds (1983)
findings hold good even for the sample covariance matrix. They have shown that the
eigenvectors can be used instead of statistical factor loadings in the return generating
model for a large (infinitely many assets) economy. This does not necessarily imply
that the eigenvectors can be used in a cross-sectional model of security pricing in a
small (finitely fewer assets) economy.
Beggs (1986) provided an insight into APT operating in a finite economy by using
quadratic preferences and standard regression analysis. He observed that the standard
APT result of a linear relationship between expected returns and the covariances of
returns is approximately true, although it does not hold exactly in a finite economy. He
provided an expression for the error of the approximation and suggested some
computational procedures to keep the error small. However, principal component
analysis is found to be equivalent to factor analysis only when the idiosyncratic risks of
all firms are equal; otherwise the use of factor analysis is less complicated [Shukla and
Trzcinka (1990)].
It is customary to contemplate two approaches that identify factors: factor analysis and
cross-sectional regression analysis. The factor analysis approach isolates factors by
observing common variables in the returns of different securities. These factors are
statistical constructions that represent some underlying sources of risk; that source may
or may not be observable. Cross-sectional regression analysis, on the other hand,
requires that we define a set of security attributes that measure exposure to an
Mazharul Kazi Ph.D Thesis 41
underlying factor and determine whether differences across security returns correspond
to differences in these security attributes [Kritzman (1995)].
As a common guide, factor analysis is used as a method of analyzing APT that begins
by calculating the daily returns of a representative sample of stocks during some period
and then isolates the factors that underpin stock market performance. To isolate the
factors, it computes the correlations between the (daily) returns of each stock and the
returns on every other stock. Then it seeks to identify groups consisting of stocks that
are highly correlated with each other but not with the stocks outside the group.
Similarly, it proceeds to isolate other groups of stocks whose returns are highly
correlated with each other within their own group, until it isolates all the groups that
seem to respond to a common source of risk. Then the factor analysis pursues the task
of identifying the underlying sources of risk for each group. Initially, it relies on
intuition to identify the factor that underlies the common variation in returns among
stocks. Then it tests the intuition. In doing so, this approach defines a variable that
serves as a proxy for the unanticipated change in the factor value [Kritzman (1995)].
Factor analysis looks at covariation in returns and attempts to identify the sources of
this covariation, whereas cross-sectional regression analysis requires specification of
the sources of return covariation and attempts to affirm that these sources do indeed
correspond to differences in return. Based upon intuition and prior research, researchers
infer attributes, rather than factors, that correspond to differences in stock returns. Once
a set of attributes is specified, a measure of sensitivity to the common sources of risk
can be determined. The process regresses the returns across a large sample of stocks
Mazharul Kazi Ph.D Thesis 42
during a given period, say a month, on the attribute values for each stock as of the
beginning of the month. Then it repeats this regression over many different periods. If
the coefficients of the attribute values are not zero and are significant in a sufficiently
high number of regressions, this approach concludes that differences in returns across
the stocks relate to differences in their attribute values [Kritzman (1995)]. According to
this approach, a securitys return in a particular period equals:
Ri = α + λ1 bi1 + λ2 bi2 + ……… + λn bin + εI (2.11)
where Ri is the return of security i; α is a constant; λ1, λ2, . λn are marginal returns to
attributes 1, 2, .n respectively; bi1 is the attribute 1 of security i; bi2 is the attribute 2
of security i,; bin is the attribute n of security i; and εi is the unexplained component
of security is return. It is not necessary for the coefficients in the above formula to be
significantly positive or negative on average over all regressions. In some periods, there
may be positive returns associated with an attribute, and in some other period there may
be negative returns associated with an attribute. Consequently, the average value for a
coefficient over many regressions may be zero. Nonetheless, the attribute would still be
important if the coefficient were not zero in a large number of regressions. One can
measure the extent to which a coefficient is significant in a particular regression by its
t-statistic.
Kim and Lee (1995) presented the APT with the assumption that k-linear factor models
generate random returns on any set of assets:
Mazharul Kazi Ph.D Thesis 43
Rit = Ei + 1
k
k=∑ bik fkt + εit, i = 1,N, t = 1,T (2.12)
where Ei is the expected return on the ith
asset; bik is the sensitivity of the return on asset
i to factor fkt; fkt is a mean zero common factor to the return of all assets under
consideration; and εit is the nonsystematic idiosyncratic error to the ith
asset with
E(ε1tƒ1t,…ƒkt) = 0, Cov (ε1t, ƒkt) = 0 (2.13)
and
Ψ = Cov (ε1t,…εkt )′ = diag ( )1
2 2,.....
Nε εσ σ (2.14)
In vector notation,
Rt = E + Bft + εt (2.15)
where Rt = (R1t,RNt)′; E = (E1,EN)′; ft = (f1t,fkt)′; εt = (ε1t, εNt)′; B = (b1,bk);
and bk = (b1k,bNk)′. The idiosyncratic error terms are assumed to all equal zero (that
is, nonexistent) in the principal component analysis, whereas they exist in the factor
analysis [Kim and Lee (1995)].
More recently, Zhou (1999) has suggested models, in line with both the latent and pre-
specified factor schools, for consideration when using the generalized method of
Mazharul Kazi Ph.D Thesis 44
moments for testing APT. When factors are pre-specified variables, he assumed M pre-
specified observable factors X1, X2, XM and allowed returns to be generated by
factors in the following linear form:
rit = αi + βi1 Χ1t + … + βiM ΧMt + εit i = 1,…N (2.16)
where εit is the model disturbance with a zero mean conditional on available
information. It can have conditional heteroskedasticity as shown by Hansen (1982).
Since the factor selection procedure may include more variables than necessary, some
of the variables could be correlated and hence redundant.
Obtaining the minimum number of orthogonal factors that are necessary depends on the
rank of the regression coefficient matrix B = (βij), N × M formed by the betas. If rank
B=(βij) = K< M, there exists an N×K rank matrix A and K×M matrix C such that B is their
product. This means H0: B = AC; A: N × K; and C: K × M and
rit = αi + Ai1 ƒ1t + … + AiK ƒKt + εit i = 1,…N (2.17)
where ƒ1t ƒKt are new factors linear combinations of the pre-specified ones:
ƒkt = Ck1 X1t + … + CkM XMt, k = 1, … K (2.18)
Mazharul Kazi Ph.D Thesis 45
Thus, the K-factors are orthogonal and out of the M individual factors, K-linear
combinations are sufficient to explain returns.
In the case of a model where factors are inherently latent, Zhou (1999) begins with a
one-factor model, which is extendable to multi-factor or K-factor models.
Rit = αi + βi ƒt + εit, i = 1, … N (2.19)
where Rit is the return of asset i in excess of the risk-free rate, ƒt is the latent factor and
εit is the model residual with zero mean. Without a loss of generality, ƒt is assumed to
have zero mean and unit variance. Its realizations are unobservable directly from
market data. As the factor is unknown, it may be extracted through the applications of
Connor and Korajczyks (1986) asymptotic principal component approach and the
standard factor analysis see, for example, Seber (1984). It is usually difficult to
relate the estimated factor to X1,XM (a set of M given observable economic or
financial variables) which have zero expected values. To overcome this, Zhou (1999)
projected ƒt onto X1, XM to obtain
ƒt = C1 X1t + … + CM XMt + υt (2.20)
where υt is the projection error which is uncorrelated with X1, XM, that is, Ε[Χm εit] =
0. This is not restrictive because εit measures the idiosyncratic risk, whereas both ƒt and
the observable factors are pervasive risks that are common to all assets. The advantage
of the above projection formula is that latent factors are modelled explicitly as a linear
Mazharul Kazi Ph.D Thesis 46
function of observable factors plus white noise. By estimating the parameters, C1CM,
the contributions of the known variables to the latent factor can be estimated.
Combining the two equations, we get:
Rit = αi + βi (C1 X1t + … + CM XMt) + uit, where uit = βi υt + εit (2.21)
This can be written in vector form as:
Rt = α + BXt + ut (2.22)
where Rt is an N-vector of the returns; α , Xt and ut are defined similarly; and B is an
N×M matrix of regression coefficients, B = βC′ with β = (β1,…βN)′, N×1; and C =
(C1,…CM)′, M×1.
When applying this method to analyze APT, Zhou (1999) assumes that the returns on a
vector of N assets relate to K latent factors by a K-factor model. Thus, the equation for
APT becomes:
rit = Ε[rit] + βi1 ƒ1t + … + βiK ƒKt + εit i = 1,…N, t=1,T (2.23)
where rit is the return on asset i at time t; ƒKt the kth
factor at time t; εit the idiosyncratic
factor of asset i at time t; βiK the factor loading of the kth
factor for asset i; N the number
of assets; and T the time periods. This equation is equivalent to the following:
Mazharul Kazi Ph.D Thesis 47
rt = Ε[ rt ] + βƒt + εt (2.24)
where rt is an N×1 vector of returns, Ε[rt] is the expected returns conditional on ƒt; and
β, N×K, ƒt and εt are defined accordingly. The standard assumptions for the factor
model are: Ε[ƒt] = 0, Ε[ƒt ƒ′] = Ι, Ε[εt|ƒt] = 0, Ε[εt ε′ t|ƒt] = Σ.
With the factor model, the classic APT is derived under the assumption that the residual
covariance matrix Σ is diagonal [Shanken (1992)], although this assumption has been
relaxed in the study of Zhou (1999) by a much weaker one where Σ does not need to be
a diagonal matrix as long as the matrix is positive definite.
Derived APT
Factor Pricing Restrictions in APT
The APT pricing process is often derived from the factor models of equity returns as
under:
λλι Β+≈ 0][ nrE (2.25)
where λ0 is a constant; λ is a k-vector of factor risk premia; and ιn is an n-vector of
ones. The approximate equality sign ≈ reflects the fact that APT holds only
approximately. For APT to become an accurate pricing model, on average, it requires
that the economy possess a large number of traded assets [Connor and Korajczyk,
Mazharul Kazi Ph.D Thesis 48
(1995)]. The derivation of APT follows various approaches, namely, exact pricing in a
noiseless factor model; approximate nonarbitrage, competitive equilibrium derivation;
mean-variance efficiency and exact factor pricing; and pricing dynamics.
Exact Pricing in a Noiseless APT
A noiseless factor model is one with no idiosyncratic risk, where [ ]r E r f= + Β . This is
of course a strong restriction on asset returns. Yet it is useful for the intuition that APT
provides. Here, an exact arbitrage argument is sufficient. It does not require a large
number of assets, and there is no approximation error in the APT pricing restriction to
satisfy Ross (1977). To derive APT, simply project E[r] on ιn and B to obtain
projection coefficients λ0 and λ, and a projection residual vector η:
ηλιλ +Β+= nrE 0][ (2.26)
The projection residual properties provide 0η ′Β = and 0nη ι′ = . Viewing the n-vector,
η, as a portfolio of asset purchases and sales, the portfolio will have zero cost structure.
If this portfolio possessed positive expected returns, it would represent an arbitrage
opportunity. This condition is inconsistent with even the weakest type of pricing
equilibrium. The expected payoff of the portfolio here is [ ]E rη η η′ ′= and this can be
zero if η = 0. Thus, the pricing model of Ross, 0[ ] nE r t λ λ≈ +Β , holds with equality in
the absence of arbitrage opportunities.
Mazharul Kazi Ph.D Thesis 49
Connor and Korajczyk (1995) combined the noiseless factor model, [ ]r E r f= +Β , with
the Ross pricing result, 0[ ] nE r ι λ λ= +Β , to show that
)(0 λλι +Β+= fr n
(2.27)
where a unit-cost portfolio is any collection of assets such that 1=′ nιω . The payoff to
a unit-cost portfolio is a portfolio return. A unit-cost portfolio with 0=Β′ω (that is, no
factor risk) has a risk-free return equal to λ0. As long as the (k+1)×n matrix [ι n, B] has
the rank k+1, one can construct such a portfolio, and identify λ0 as the risk-free return.
A unit-cost portfolio with a unit sensitivity to factor j and zero sensitivity to the other
factors will have expected returns λ0+λ j. Hence, the k-vector λ measures the risk
premium per beta-unit of each factor risk. This risk premium depends on factor
rotation, which affects the scales of the betas [Connor and Korajczyk (1995)].
Approximate Nonarbitrage Pricing Theory
The argument of the noiseless APT is extended to a strict or approximate factor model
case, where the pricing relationship holds approximately in an economy with many
assets. Connor and Korajczyk (1995) combined the original formulation of Rosss
(1976) APT with some refinements of Huberman (1982) and considered the orthogonal
price, η, as in the noiseless model above. Accordingly, the nth
portfolio consists of
holdings of the first n assets in proportion to their price deviations, scaled by the sum of
squares of these deviations, as under:
Mazharul Kazi Ph.D Thesis 50
( )
nn
n n
ηωη η
= ′
(2.28)
Using the same steps as in the noiseless case, one can show that the cost of each of
these portfolios is zero, the expected payoff of each is one, and the variance is
2( )n n n n nVη η η η−′ ′ . Using the property of maximum eigenvalues, we can have
( )n n n n n nV Vη η η η′ ′≤ , where n
V denotes the maximum eigenvalue of V n. Thus, the
portfolio variance is less than or equal to 1( )
n n nVη η −′ . This is the case in an economy
with a large, but finite, number of assets.
For an economy with an infinite number of assets, Chamberlain (1983) extended the
model by way of expanding the space of portfolio returns to include infinite-
dimensional linear combinations of asset returns, as under:
nn
nrr′
=→∝
ωω lim (2.29)
where rn denotes the n-vector of the first n of the infinite set of assets, and rw defines a
limit portfolio return as the limit of the returns to n-asset portfolios as n tends to
infinity. The limit takes the second-moment norm 2[ ]r E rω ω= . The simple form of
convergent sequence of portfolios is (1/ ,1/ ,...1/ )n n n nω = . This sequence of portfolio
weights converges to a zero vector. Yet, the limit portfolio of this sequence is a well-
defined non zero return in some cases. However, limit portfolio returns can be perfectly
Mazharul Kazi Ph.D Thesis 51
diversified, that is, they have idiosyncratic variance of exactly zero [Connor and
Korajczyk (1995)].
Ross (1978b) and Kreps (1981) developed an exact nonarbitrage pricing theory, which
is not same as APT. Their approach suggested that in the absence of exact arbitrage
opportunities there must exist a positive, linear pricing operator over state-contingent
payoffs. Chamberlain and Rothschild (1983) showed that in an infinite-asset model the
approximate APT is an extension of Ross-Krepss exact nonarbitrage pricing theory.
Reisman (1988) extended the Chamberlain and Rothschild (1983) result to general
normal linear spaces. He showed that APT can be reduced to an approximation.
Competitive Equilibrium Derivations in APT
Imposing competitive equilibrium, an alternative to the approximation approach can
also derive APT. This gives a stronger pricing approximation, and links the APT with
other equilibrium-based pricing models. Considering a risk-averse investors utility
function, u(.), for end-of-period wealth, and the approximate factor model of returns
that includes the additional assumption that idiosyncratic risks are conditionally mean-
zero, gives the following factors:
0]|[ =fE ε (2.30)
If W0 is the investors wealth at time 0, then in a competitive equilibrium, the following
first order portfolio optimization condition would hold for every investor:
Mazharul Kazi Ph.D Thesis 52
γιω nrrWuE =′′ ])([ 0 , (2.31)
for some positive scalar γ. For simplicity, letting W0 = 1, and inserting
[ ]r E r f ε= +Β + into 0[ ( ) ] nE u W r rω ι γ′ ′ = as well as separating three additive terms
and bringing constants outside, the expectations operator gives:
)]([
])([][ 0
ruE
ruErE
n
ωεωλλι
′′′′
+Β+=′ (2.32)
where λ0= / [ ( )]E u rγ ω′ ′ and [ ( ) ] / [ ( )]E u r f E u rλ ω ω′ ′ ′ ′= − . The competitive equilibrium
derivations of the APT assume a sufficient set of conditions so that the last term of the
above equation,[ ( ) ]
[ ( )]
E u r
E u r
ω εω
′ ′′ ′
, is approximately a vector of zeros.
Mean-Variance Efficiency and Exact Factor Pricing
The mathematics of mean-variance efficiency are often employed to reiterate the APT
pricing restriction. This is useful for economic modelling purposes. Considering a set
of unit-cost portfolios as linear combinations of n-vectors of asset returns, r, for any
portfolio return, rω, one can define the one-factor projection equation as:
ε++= bfrEr ][ (2.33)
Mazharul Kazi Ph.D Thesis 53
where [ ]f r E rω ω= − . A single-beta-pricing model holds with respect to the above
equation if:
λλι brEn += 0][ (2.34)
for some scalars λ0 and λ. Defining a mean-variance efficient portfolio as a unit-cost
portfolio that minimizes variance subject to [ ]E r cω ′ = for some c, one can show
that 0
[ ] nE r bι λ λ= + is both the necessary and sufficient condition for the mean-
variance efficiency of ω. Thus 0[ ] nE r bι λ λ= + is equivalent to providing that ω is
mean-variance efficient.
This relationship between mean-variance efficiency and beta pricing can be extended to
a multi-beta model with a given n-vector of returns r. For any set of k portfolio returns,
rω1, rω2,…, rωk, with the projection equation [ ]r E r f ε= +Β + (where f j = rω j E[rω j] for
j =1,2,,k), linear pricing with respect to these factor-mimicking portfolios takes the form
0[ ]n
E r ι λ λ= + Β .
Dynamics in Asset Pricing
Dynamic versions of APT usually specify an exogenous factor model for the cash flows
(dividends) paid by firms while deriving the factor model for the prices of securities
endogenously. Often, discrete time dynamic models are useful to explain asset pricing
[Jagannathan and Viswanathan (1988); Connor and Korajczyk (1989, 1995)].
Mazharul Kazi Ph.D Thesis 54
Even if factor loadings (betas) for the cash flow process are time invariant, the betas of
asset returns (relevant to factor-mimicking portfolios) will be functions of the current
information set. In the static APT one can replicate the priced payoff from a security
with the risk-less asset and factor-mimicking portfolios. Jagannathan and Viswanathan
(1988) showed that in a multi-period economy there exists a different set of riskless
assets for every maturity, that is, a discount bond with that maturity. Thus, although the
risky components of assets payoff are derived by say the one-factor model, asset
returns may follow an infinite factor structure corresponding to a portfolio mimicking
the single factor, plus the returns on discount bonds for every maturity [Connor and
Korajczyk (1989)].
Connor and Korajczyk (1989) developed a multi-period model in which they assumed
that per-share dividends, rather than asset returns, could be used in an approximate
factor model. They observed that expected returns follow the exact APT pricing
restriction at each date. Yet, the general format of their model could not be identified
since the beta coefficients and factor risk premium varied through time. Thus, they had
to add new conditions that give a statistically identified model. Among other works in
this field, extensions of dynamics into the continuous-time economy, intertemporal
equilibrium asset pricing and intertemporal capital asset pricing are notable
contributions to the literature [Reisman (1992a); Ohlson and Garman (1980);
Chamberlain (1988); Merton (1973); and Connor and Korajczyk (1989, 1995)].
In brief it is understood that factor analysis does not actually observe underlying
factors; rather it identifies the common sources of attributes or aptitudes (treating them
Mazharul Kazi Ph.D Thesis 55
as factors) that help in explaining the performances of variables under study. Thus,
factors derived through factor analysis are not always amenable to appropriate
interpretation [Kritzman (1995)]. Parallel to factor analysis, principal component
analysis has been used by some researchers to identify the factors that influence stock
returns in finance. In this regard eigenvalues are calculated to construct relevant
eigenvectors which are similar to factors in factor analysis.
Summary
This chapter has reviewed finance theories that are related to security market behaviour
and pricing mechanisms. Both the popular theories of the capital asset pricing model
(CAPM) and APT are found useful for security market analyses or to describe stock
returns in an efficient market. Still, APT is popular as a superior alternative to the
CAPM. There are two forms of APT: the nonarbitrage pricing theory of Ross, and the
equilibrium APT. Rosss APT is more popular. Traditionally, the testing of APT is
done through factor analysis, as most researchers represented APT as a k-linear factor
model, where the number of common factors, k, is unknown. The Roll and Ross (RR)
method of testing APT has been popular within traditional applications. Our
contribution in this chapter has been to comprehensively review the asset pricing
literature from a theoretical perspective.
Mazharul Kazi Ph.D Thesis 56
CHAPTER 3
EMPIRICAL FACADE OF ASSET PRICING THEORY
Generally, Arbitrage Pricing Theory (APT) indicates that an assets expected return is a
linear function of a set of sensitivity coefficients to systematic or common factors. The
number of common factors, k, and their nature, are not known. Hence, tests of the APT
model are often difficult to design. To avoid the confusion that arose in the literature
about the relationship between the theoretical underpinnings of APT and its empirical
tests, Dybvig and Ross (1985) noted that the empirical APT differs from the theoretical
APT. Yet, in deriving the empirical APT, one has to understand the underlying
theoretical premise of APT. The generalized theoretical APT states that asset returnsiR%
are generated by a factor model, 1
k
i i ij j i
j
R E β ε=
= + ∂ +∑ %% %, where
j∂% is a vector of systematic
risk factors that influence asset returns; iβ is a vector of factor loadings on asset i; Ei is
the expected return on asset i; and iε is the idiosyncratic noise term associated with
asset i. When assumed that iε% is independent across assets,
iε% is independent of the
factors with a mean zero, and that the factorsj∂% have also mean zero and that variances
exist, the empirical APT can be derived from the theoretical APT as suggested by
Dybvig and Ross (1985). Thus, the theoretical APT simply asserts that the new assets
expected returns are linear in factor loadings in the following form:
∑
=
+=k
j
ijjirE
1
βλ (3.01)
Mazharul Kazi Ph.D Thesis 57
where r is a constant that equals the riskless rate of return if there is a riskless asset, and
λ is a vector of risk premia. The arbitrage in APT comes from the simple proof that
the theoretical version 1
k
i i ij j i
j
R E β ε=
= + ∂ +∑ %% % implies the empirical version
1
k
i j i j
j
E r λ β=
= + ∑ in the absence of arbitrage alone, for the case in which all the
iε% s
are zero. Customarily, the derivation of APT involves establishing that theiε% s can be
neglected when pricing assets. Most derivations are based on the intuition that the
idiosyncratic error terms represent diversifiable risks and therefore should not affect
asset pricing.
A good number of empirical studies, including Gehr (1978), tested the efficacy of APT
models without attempting to identify the risk factors. In these studies, portfolio returns
were decomposed using factor analysis. To test the ability of the factors to explain
mean returns, the factor loadings derived from factor analysis were used as independent
variables in a cross-sectional regression with mean portfolio rates of return.
Empirical Episodes of APT
The original APT of Ross (1976) hypothesized that the cross-sectional distribution of
expected returns of financial assets could be approximately measured by their
sensitivities, usually called factor loadings,κ, to k unknown economic factors.
Chamberlain and Rothschild (1983) extended Rosss (1976, 1977) APT, which
postulated that with a factor structure the mean returns are approximately linear
Mazharul Kazi Ph.D Thesis 58
functions of the factor loadings. Rothschild (1983) accordingly provided a unique
approach, known as the principal components analysis.
Connor and Korajczyk (1988) have shown that Chamberlain and Rothschilds (1983)
findings hold good even for the sample covariance matrix. They have shown that
eigenvectors instead of statistical factor loadings can be used in the return generating
model for a large (infinitely many assets) economy.
Beggs (1986) provided an insight into APT operating in a finite economy by using
familiar quadratic preferences and standard regression analysis. He observed that the
standard APT result of a linear relationship between expected returns and the
covariance of returns is approximately true, although it does not hold exactly in a finite
economy. Beggs (1986) provided an expression for the error of the approximation,
suggesting some calculation procedures to keep the error small. However, principal
component analysis is found to be equivalent to factor analysis only when the
idiosyncratic risks of all firms are equal; otherwise the use of factor analysis is less
complicated [Shukla and Trzcinka (1990)].
McGowan and Francis (1991) used both multivariate analysis techniques and principal
component analyses in their study. They exploited the ability of multivariate statistical
analysis techniques (like factor analysis and principal component analysis) to discover
interesting empirical insights dependent upon the size and heterogeneity of the sample
and the approach of data analysis. Their research aimed at ascertaining the principal
components of stock price fluctuations as well as the principal components of
Mazharul Kazi Ph.D Thesis 59
macroeconomic fluctuations during simultaneous sample periods. Then the
macroeconomic principal components were compared to principal components of stock
prices in an effort to determine primary relationships. To analyze the relationships, the
authors selected a highly diverse sample of macroeconomic data (consisting of 95
variables) in conjunction with the highly diverse sample of stock prices.
McGowan and Francis (1991) grouped each stock market index (consisting of 86
variables) under 10-factor classifications for which the index had the highest factor
loading. Their factor loading value (provided in the parentheses in Table 1) indicates
which industry indices were the most significant in explaining the stock price returns
for a factor. Thus, their principal component analysis grouped together those industries
that are most responsive to the same factor, that is, the risk component. They defined
their 10 industry factors into three broad groups, namely, primary, secondary, and
tertiary. The primary factor is an extracted factor composed of stock market indices that
had their highest factor loading on that factor. The secondary factor is a factor for
which no industry had its highest factor loading. That is, it was composed of indices
that were more highly related to a primary factor yet had a sufficient additional
common relationship with each other to be included in another grouping. The
remainder of the factors are tertiary factors, which had only one large industry factor
loading. Factors one to six, and nine were primary factor; factors seven and 10 were
secondary factors; and factor 8 was a tertiary factor. Details are presented in Table-3.1:
Mazharul Kazi Ph.D Thesis 60
TABLE 3.1
Factor 1
Foods (0.52); Textile products (0.59); Textile-apparel manufacturers (0.79); Home furnishings (0.69); Containers-
paper (0.62); Publishing (0.70; Building materials-composite (0.73); Building materials-roofing & wallboard (0.62);
Tires & rubber goods (0.64); Shoes (0.71); Containers-metal & glass (0.52); Building materials-cement (0.61); Metal
fabricating (0.67); Building materials- GTG & PLMBG (0.54); Building materials-air conditioning (0.73); Machine
tools (0.63); Pollution control (0.61); Auto merch mach (0.61); Electric household appliances (0.53); Automobiles
(0.62); Automobile- except (0.65); Automotive-trucks & parts (0.73); Aerospace (0.62); Mobile homes (0.54); Toys
(0.70); Leisure time (0.78); Air transport (0.51); Air freight (0.56); Radio-TV broadcasters (0.72); Retail stores-
composite (0.56); Retail stores-discount (0.75); Retail stores-food chains (0.54); Investment companies (0.56);
Homebuilding (0.73); Hotels-Motels (0.80); Entertainment(0.62);Conglomerates (0.68)
Factor 2
Beverages-brewers (0.46); Beverages-distillers (0.48); Beverages-soft drinks (0.72); Tobacco (0.52); Drugs (0.83);
Hospital supplies (0.70); Soaps (0.75); Cosmetics (0.78); Office & business equipment (0.73); Office & business
equipment IBM (0.73); Electric & electronic-major (0.52); Electronic equipment (0.63); Restaurants (0.67)
Factor 3
Metals-miscellaneous (0.63); Coal-bituminous (0.63); Forest products (0.55); Paper (0.71); Chemicals (0.64); Steel-
except USS (0.76); Steel (0.77); Copper (0.62); Aluminum (0.75); Machinery-agriculture (0.56); Machinery-
construction & materials handling (0.55); Machinery-industrial & specialty (0.61); Rail road equipment (0.44); Rails
(0.52)
Factor 4
Telephone (0.71);Telephone-except ATT (0.64); Electric power (0.69); Natural gas- distribution (0.63); Natural gas-
pipe lines (0.48); Personal loans (0.54); S7L holding companies (0.60); Life insurance (0.58); Multiline insurance
(0.52); Property & casualty insurance (0.54)
Factor 5
Oil-composite (0.81); Oil-crude producers (0.62); Offshore drilling (0.50); Oil-intec domestic (0.75); Oil-intec
international (0.73); Oil well equipment & service (0.54)
Factor 6
Sugar refiners (0.68)
Factor 7
Offshore drilling (-0.39); Tires & rubber goods (0.34); Pollution control (-0.37); Automobile (0.43); Automobiles-
except (0.36)
Factor 8
Banks-NYS (0.46); Banks-outside NYS (0.40)
Factor 9
Gold mining (0.86)
Factor 10
Truckers (0.40); Air transport (0.37); Air freight (0.40)
Source: Adapted from McGowan and Francis (1991).
Mazharul Kazi Ph.D Thesis 61
Connor (1995) presented an overview of the empirical procedures for the basic three
types of factor models, namely, macroeconomic, statistical, and fundamental models.
Input requirements, outputs and relevant estimation techniques corresponding to each
type of model are presented in the Table 3.2:
TABLE 3.2
Overview of the Empirical Procedures of Factor Models Model Type Inputs Estimation Technique Outputs
Macroeconomic Security Returns
Economic Variables
Time series regression Security factor betas
Statistical Security Returns Iterated time-series or Cross-
sectional regression
Statistical factors and
Security factor betas
Fundamental Security Returns
Security Characteristics
Cross-sectional Regression Fundamental factors
Source: Connor (1995).
Roll and Ross (1980) conducted their empirical tests of APT using data from individual
equities during the period 1962 to 1972. Their research design followed a two-step
procedure. In the first step, the expected returns and factor coefficients were estimated
from time series data on individual asset returns. The second step used the first stage
estimates to test the basic cross-sectional pricing conclusion. Since APT applies to
subsets of the universe of assets, it does not require a particular choice of a substitute
for the market portfolio, which is required for the capital asset pricing model. They
used data as described in Table 3.3.
Mazharul Kazi Ph.D Thesis 62
TABLE 3.3
Data Description Source Daily Return File: Center for Research in Security Prices, Graduate School of
Business, University of Chicago, US.
Selection Criterion
By alphabetical order into groups of 30 individual securities from those listed on NY
or American Exchanges on both 3 July 1962 and 31 December 1972. The alphabetical
last 24 securities were not used since complete groups of 30 were required.
Basic Data Unit Return adjusted for all capital changes and dividends, if any, between adjacent trading
days, i.e., [ ], , , 1( )/ 1j t j t j tdρ ρ −+ − , where ρ =price, d = dividend, j = security
index, t = trading day index.
Maximum Sample Size
per Security
2619 daily returns.
Number of Selected
Securities
1260 (42 groups of 30 each)
Source: Roll and Ross (1980).
Brown and Weinstein (1983) proposed a new approach for estimating as well as testing
asset pricing models this was in the context of the bilinear paradigm of Kruskal
(1978). The bilinear paradigm suggests that testing of APT or asset pricing models does
not require that the factors common across securities have a multivariate normal
distribution. The authors found the approach relatively simple and general compared to
others. They described a bilinear model as the matrix of return observations, R = λ0 e ′ +
Ak Bk + U, where e is a column vector of ones, which can be re-written as R = AB + U.
This model generates observed returns in the context of an arbitrage pricing model with
a pre-specified k-number of factors. Thus, a simple test of the theory would be to check
whether the factor structure, factor prices, and implied risk-free rate given by the matrix
A are constant across groups of securities at each point in time.
Brown and Weinstein (1983) used the same set of data as considered by RR (Roll and
Ross, 1980). They started with 42 groups of 30 stocks each, arranged alphabetically,
Mazharul Kazi Ph.D Thesis 63
and considered the daily returns (including dividends) for the period from 3 July 1962
to 31 December 1972. To be eligible for inclusion in a group, the stocks had to be listed
on one of the two major US stock exchanges on both the starting and ending dates for
the period. They combined two consecutive RR groups to form each of their 60 security
groups. They thus obtained 21 groups of 60 securities each. The results obtained
through the application of their approach of factor analyses for 30 security groups were
not identical with those of RR, for two reasons. First, their requirements were different
in the sense that they performed an analysis for 60 security groups to describe common
observations across 60 securities as opposed to the 30 securities of RR. Thus, each of
their runs used fewer observations than RR. Secondly, the results differed because they
used different data sources to RR.
Chen, et al. (1986) indicated that the systematic forces that influence US equity returns
are those that change discount factors, k, and expected cash flows, E(c). Since we are
only concerned with business institutions, stock prices can be defined as expected
discounted dividends p = ( )E c
k. The discount rate is an average of rates over time, and it
changes with both the level of rates and the term structure spreads across different
maturities. Unanticipated changes in the riskless interest rate will therefore influence
pricing. Through their influence on the time value of future cash flows, they will
influence returns. Unanticipated changes in the risk premium will also influence
security returns, because the risk premium is to an extent determined by the discount
rate. On the demand side, real consumption changes owing to the changes in the
indirect marginal utility of real wealth will influence pricing, and often such effects
show up as unanticipated changes in the risk premium.
Mazharul Kazi Ph.D Thesis 64
Chen, et al. (1986) also inferred that expected cash flows change because of both real
and nominal forces. Changes in the expected rate of inflation influence nominal
expected cash flows as well as the nominal rate of interest. The unanticipated price
level change will have a systematic effect. Relative prices usually change with the
general inflation level. There can also be a change in asset valuation associated with
changes in the average inflation rate. Finally, changes in the expected level of real
production affect the current real value of cash flows. Thus, innovations in the rate of
productive activity have an influence on stock returns through their impact on cash
flows. Accordingly, Chen, et al. (1986) constructed two series of variables: a basic
series, and a derived series. Their basic series of variables along with their symbols,
definition/sources are as presented in Table-3.4:
TABLE 3.4
Glossary and Definitions of Basic Series Variables Symbol Variable Definition/Source
I Inflation Log relative of US Consumer Price Index
TB Treasury-bill rate End-of-period return on 1-month bills
LGB Long-term government bonds Return on long-term government bonds
IP Industrial Production Industrial production during month
Baa Low-grade bonds Return on bonds rated Baa and under
EWNY Equally weighted equities Return on equally weighted portfolio of NYSE listed
stocks
VWNY Value weighted equities Return on a value weighted portfolio of NYSE listed
stocks
CG Consumption Growth rate in real per capita consumption
OG Oil prices Long relative of Producer Price Index / Crude Petroleum
series
Source: Chen, et al. (1986).
Mazharul Kazi Ph.D Thesis 65
The derived series of variables considered by Chen, et al. (1986) are as in Table 3.5:
TABLE 3.5
Glossary and Definitions of Derived Series Variables Symbol Variable Definition/Source
MP(t) Monthly growth, industrial production loge[IP(t)/IP(t-1)]
YP(t) Annual growth, industrial production loge[IP(t)/IP(t-12)]
E[I(t)] Expected inflation Fama and Gibbon (1984)
UI(t) Unexpected inflation I(t) E[I(t)|t-1]
RHO(t) Real interest (ex post) TB(t-1) I(t)
DEI(t) Change in expected inflation E[I(t + 1)|t] E[I(t)|t-1]
URP(t) Risk premium Baa(t) LGB(t)
UTS(t) Term structure LGB(t) TB(t-1)
Source: Chen, et al. (1986).
Faff (1988) examined empirically the issues concerning APT using Australian equity
data and employing the approach of Chamberlain and Rothschild (1983). He considered
a sample of 140 equities over 12 years from 1974 to 1985. He encountered problems
associated with missing data. A total number of 157 firms were detected having no
more than four missing observations; 94 firms did not have any missing information;
25 firms had four; 10 firms had three; 23 firms had two; and five firms had one missing
observation. The sample of 140 was achieved by a random selection of eight from 25 in
the group of firms having exactly four missing data points. Values of all missing
observations assigned the appropriate monthly average of available observations for
each affected firm. In his sample, Faff (1988) distributed the firms across industries in
the manner shown in Table 3.6 below:
Mazharul Kazi Ph.D Thesis 66
TABLE 3.6
Distribution of Sample Firms Across Industries ASE Industry Group Number of Firms
1. Metals
2. Solid Fuels
3. Oil & Gas
4. Developers and Contractors
5. Building Materials
6. Electrical, Household Durable
7. Alcohol and Tobacco
8. Food, Household Goods
9. Textiles and Clothing
10. Automotive
11. Chemicals
12. Light Engineering
13. Heavy Engineering
14. Paper and Packaging
15. Banks and Finance
16. Insurance
17. Retail
18. Merchants and Agents
19. Transport
20. Media
21. Property Trust
22. Other Services
23. Miscellaneous, Diversified and Industrials
24. Diversified Resources
18
7
5
7
14
6
3
10
1
3
5
5
7
4
4
4
6
2
5
5
1
9
5
4
Source: Faff (1988).
Faff (1992) undertook a multivariate test of the equilibrium version of APT for the
Australian equity market applying the asymptotic principal component technique of
Connor and Korajczyk (1988). Faffs (1992) study aimed at extending and improving
his 1988 empirical examination of APT on Australian data. He observed that the
Connor and Korajczyk (1988) method performs well in relatively small sub-periods.
It uses a multivariate approach, permits time-varying factor risk premiums, and
incorporates monthly (seasonal) effects. The iterative process of the asymptotic
principal component technique followed by Faff (1992) can be understood from Table 3.7:
Mazharul Kazi Ph.D Thesis 67
TABLE 3.7
The Iterative Process in Asymptotic Principal Component Technique 1. Form the cross-product matrix (Ω) of excess returns.
2. Calculate eigenvectors for the cross-product matrix.
[The first k-eigenvectors represent proxies for the independent variables of
1 1 2 2 .....it i t i t ki kt itr B F B F B F ε= + + + + , where rit = (Rit Rft), the excess return for asset i in period t;
and Fjt = ( )jt jtfγ + = 1,.,k, the realized risk premium plus the factor realization for factor j in period t].
3. For each individual asset in the sample run a regression of excess returns on the first k eigenvectors obtained
and calculate the standard deviation of residuals.
4. Scale the excess returns of each asset by its associated residual standard deviation obtained and form a new
scaled cross-product matrix of excess returns.
5. Repeat steps 2 to 4 until convergence achieved.
Source: Summarised from Faff (1992).
Faff (1992) gathered data from the price relatives file of the Centre for Research in
Finance (CRIF) at the Australian Graduate School of Management. His data set
covered the 165-month period from January 1974 to September 1987, which was
divided into three non overlapping 55-month sub-periods for analysis. Securities were
included in a sub-period sample if they had a complete price relative history in that
period. This produced sample sizes of 303, 158 and 340 for the three sub-periods
respectively. Faff (1992) used value weighted market index as supplied by CRIF, while
he estimated the risk-free rate from a series of monthly observations on 13-week
Treasury Notes. Likewise, he used portfolios rather than individual assets for analytical
convenience.
Hamao (1988) examined APT empirically using Japanese data. He considered
macroeconomic state variables similar to those used in Chen, et al. (1986). He utilized
Mazharul Kazi Ph.D Thesis 68
neither the usual factor analysis nor mimicking portfolios analysis in deriving factors
that influence asset pricing in APT. Rather, he derived variables from the basic
economic theory of asset pricing that is applicable regardless of the location of the
markets and that reflects different magnitudes of parallel but different economy specific
economic conditions. Since the Japanese economy is heavily dependent on trade,
Hamao (1988) considered factors related to international trade in addition to basic
economic variables. He used the monthly return data on TSE Section I for the period
January 1975 to December 1984. The data were gathered from the Nomura Research
Institute. Out of 1066 companies, some were excluded because of missing
observations. The number of stocks excluded from the data series varied from month to
month. They varied from 53 to 188. For convenience, stocks were grouped into 20
equally weighted portfolios, with an approximately equal number of securities sorted
by size (that is, price to stock times outstanding volume) of a company at the end of the
previous month.
To determine the macroeconomic state variables, Hamao (1988) looked for exogenous
macro-variables that affect the future cash flows or the risk adjusted discount rate of
companies. He did not include equity related variables since the study aimed to identify
macroeconomic forces that only influence the stock market. Accordingly, Hamao
(1988) considered seven variables: industrial production, inflation, risk premium, term
structure, foreign exchange, market indices, and oil prices.
The structural specifications and/or definition of a priori variables considered by
Hamao (1988) are provided in Table 3.8.
Mazharul Kazi Ph.D Thesis 69
TABLE 3.8
Structurally Specified Priori Variables A Priori Variable Structural Specification
Industrial Production
The monthly index of production, IP(t), obtained from Ministry of International
Trade and Industry statistics. The growth rate of industrial production:
( ) ln ( ) ln ( 1)MP t IP t IP t= − −
It was moved forward by one month since MP(t) actually reflects the production
change in month t-1 (at least in part). The average MP over the previous five years
for the same month was subtracted from the MP series to yield the variable MPSA(t).
Inflation
Used the Fama and Gibbons (1984) methodology as in Chen, Roll and Ross to derive
the expected inflation variable. Following the Fisher theorem, the one month interest
rate GR(t-1) observed at the end of month t-1 applicable to month t was broken down
into: an expected real return for month t and expected inflation rate:
( 1) [ ( ) | 1] [ ( ) | 1]GRt E p t t E I t t− = − + −
With some modification and adjustments in the process of estimation, monthly
inflation variables were constructed as:
( ) ( ) [ ( ) | 1]UI t I t E I t t= − − and
( ) [ ( 1) | ] [ ( ) | 1]DEI t E I t t E I t t= − − −
where UI is unanticipated inflation and DEI is the change in expected inflation. If the
inflation forecast is based on information other than the previous forecast errors, DEI
would have different information from UI.
Risk Premia
Changes in the risk premium in the financial market affects the value of an asset
through changes in discount rates. Through a process of treating the spread between
government and corporate bonds yields of long-term and short-term maturity or high-
grade and low-grade bonds yield differentials; and well-diversified managed
portfolio, the constructed risk premia variables are:
( ) ( ) ( )UPREL t EL t LGB t= − and
( ) ( ) ( )UPRMF t MF t LGB t= −
where EL(t), LGB(t), and MF(t) are the returns on electricity bonds, long-term
government bonds, and the managed funds respectively.
Mazharul Kazi Ph.D Thesis 70
Term Structure
To capture the effects of changes in interest rates, the return difference between long
bonds and Gensaki considered. The constructed term structure variable is:
( ) ( ) ( 1).UTS t LGB t GRt= − −
The interpretation here is that if UTS is positive, then long bonds have risen implying
that their yield has fallen which lowers the slope of the yield curve.
Foreign Exchange
To capture the economys heavy orientation towards international trade, the volatility
in foreign exchange rates were considered as having substantial systematic effects
upon Japanese equities. With appropriate assumptions concerning risk-neutrality,
efficiency of the forward exchange market, and forecasting of spot rate, the exchange
rate change variable is:
1( ) ln lnt t tUYEN t F S−= −
Additionally, the terms of trade have some direct effect on the economy that
influences its exchange rate. Considering terms of trade (TT) data, and the definition
of TT = (Price index of export goods)/(Price index of import goods), the change in TT
is:
( ) ln ( ) ln ( 1).DTT t TT t TT t= − −
Market Indices
To capture unexpected shocks to the economy, market indices were included. The two
market indices included in the analysis are: return on the value weighted TSE-I index
including dividends, VW(t); and return on the equally weighted TSE-I index including
dividends, EW(t).
Oil Prices
Since the Japanese economy is heavily dependent on foreign oil, the price of oil were
included as a systematic factor influencing the Japanese stock market. The oil price
(growth) variable, OG, constructed as the monthly log difference of the Arabian Light
Spot price translated into Japanese yen using month-end spot exchange rates.
Source: Derived from Hamao (1988).
Fama and French (1992a) studied the joint roles of the market beta, firm size, E/P,
leverage, and book-to-market equity in the cross-section of average stock returns. Their
results suggested two empirically determined variables, the size and book-to-market
equity, which explained satisfactorily the cross-section of average returns on the
NYSE, AMEX, and NASDAQ stocks for the period 1963 to 1990.
Mazharul Kazi Ph.D Thesis 71
Later, Fama and French (1993) extended Fama and French (1992a) in three ways. First,
they expanded the set of asset returns for explanation. Secondly, they expanded the set
of variables to explain returns. Finally, they employed a different approach to testing
asset pricing models. Monthly returns on stocks and bonds were regressed against the
returns on a market portfolio of stocks and mimicking portfolios for size, book-to-
market equity, and term structure risk factors in returns. They used the time series
regression approach of Black, Jensen, and Scholes (1972). In their explanatory
variables of the time series regression, they included the returns on a market portfolio
of stocks and mimicking portfolios. They sought explanations for the returns on
government bond portfolios in two maturity ranges, corporate bond portfolios in five
rating groups, and 25 stock portfolios formed based on size and book-to-market equity.
Hughes (1985) tested APT using monthly security returns for Canadian data. He found
the existence of multiple factors in generating Canadian security returns for the period
1971 to 1980. His empirical procedures include classical factor analysis. His
formulated minimum variance portfolios attempt to derive the time series of returns on
general economic factors, sensitivities of returns for each of 220 firms to that of each
factor, and expected returns on the factors. At least 12 common factors were found to
be those which generate Canadian security returns. Out of the 12 factors, three to four
factors had statistically significant risk premia. Hughes generated an APT functional
relationship based on the notion that all individuals believe that security returns are
generated by the same small number of common factors.
Mazharul Kazi Ph.D Thesis 72
Dynamic versions of APT usually specify an exogenous factor model for the cash flows
(dividends) paid by firms while deriving the factor model for the prices of securities
endogenously. Often, discrete time dynamic models are useful to explain asset pricing
[Jagannathan and Viswanathan (1988); Connor and Korajczyk (1989, 1995)]. Even if
factor loadings (betas) for the cash flow process are time invariant, the betas of asset
returns (relevant to factor-mimicking portfolios) will be functions of the current
information set. In the static APT one can replicate the priced payoff from a security
with the risk-less asset and factor-mimicking portfolios. Jagannathan and Viswanathan
(1988) showed that in a multi-period economy there exists a different set of riskless
assets for every maturity, that is, a discount bond with that maturity. Thus, although the
risky components of an assets payoff are derived by, say, a one-factor model, asset
returns may follow an infinite factor structure corresponding to a portfolio mimicking
the single factor, plus the returns on discount bonds for every maturity [Connor and
Korajczyk (1989)].
Connor and Korajczyk (1989) developed a multi-period model in which they assumed
that per-share dividends, rather than asset returns, could be used in an approximate
factor model. They observed that expected returns follow the exact APT pricing
restriction at each date. Yet, the general format of their model could not be identified
since the beta coefficients and factor risk premium varied through time. Thus, they had
to add new conditions on preferences and the stochastic process for dividends that give
a statistically identified model. Among other works in this field, extensions of
dynamics into the continuous-time economy, intertemporal equilibrium asset pricing
and intertemporal capital asset pricing are notable contributions to the literature
Mazharul Kazi Ph.D Thesis 73
[Reisman (1992a), Ohlson and Garman (1980), Chamberlain (1988), Merton (1973),
and Connor and Korajczyk (1989, 1995)].
Aitken, et al. (1996) studied the Australian trading system. They found price clustering
pervasive, tending to follow an overall pattern somewhat similar to that found in US
security markets. They noticed that clustering resulted from imprecise beliefs or
haziness about firm value and firm conventions. Thus they found that the clustering
increases with the price of stock (reflecting imprecision in beliefs about firm value) and
higher market-wide volatility, own stock volatility, trade size, and the size of bid-ask
spread. Clustering is lower when price discovery is likely to be more efficient.
Brailsford and Easton (1991) examined the impact of the seasonality factor on
Australian equity returns for the period 1936 to 1957. Their major finding was that
equity returns in January are consistently higher than returns in any other month, and
that returns in February and June are consistently lower than returns in other months.
Also, the returns in July were higher, and in March were lower. However, January,
February and June were significant in their study.
Easton and Faff (1994) conducted a study of the robustness of the day-of-the-week
effect in Australia. Their results showed that sample size could distort the interpretation
of classical test statistics unless the significance level is adjusted downward. They
noticed that the Australian day-of-the-week effect was independent of that of US.
Mazharul Kazi Ph.D Thesis 74
Faff and Heaney (1999) analyzed the relationship between inflation and equity returns
in Australia over the period January 1974 to March 1996. Their analysis was based on
monthly and quarterly data and they used value weighted equity indices at both the
aggregate market and industry level. Three price indices, the quarterly consumer price
index, the manufactured materials used index, and the manufacturing articles produced
index, both as monthly and quarterly values, were used to measure inflation. The result
showed the impact of expected inflation on industry returns varied considerably over
three sub-period observations. Changes in government inflation policy appeared to
have the greatest impact on expected company inflation in Australia.
Faff and Brailsford (1999) investigated the sensitivity of Australian industry equity
returns to an oil price factor over the period 1983 to 1996. They employed an
augmented market model to establish the sensitivity. The important findings were as
follows. First, beyond the influence of the market, they detected a degree of
pervasiveness of an oil price factor across some Australian industries. Secondly, oil
price sensitivity was found to be significantly positive in the oil, gas and other
diversified resources industries. Similarly, they found significant negative oil price
sensitivity in the paper and packaging, and transport industries. They observed that
long-term effects persist in the market, although some firms passed on oil price changes
to its customers or hedged the risk.
Mazharul Kazi Ph.D Thesis 75
Akin Methods
Analyses of the factor structure of systematic risks in security markets returns actually
predate the APT. In addition to the pricing implications, the need for describing the
covariance structure of asset returns, Σ, becomes significant in the literature [Connor
and Korajczyk (1995)]. The covariance matrix of asset returns is a major component of
portfolio optimization problem. Estimation of the unrestricted covariance matrix of n
securities requires the estimation of n×(n+1)/2 distinct elements. The single-index or
diagonal model of Sharpe (1963) postulated that all of the common elements of returns
were due to assets relationship with the index. Thus, only 3×n parameters needed to be
estimated: n betas relative to the index, n unique variances, and n intercept terms. This
approach reduced much of the noise in estimating Σ. The single-index model can be
viewed as a strict one-factor model with a pre-specified factor. Specifically, the single-
index model does not describe all of the common movements across assets, which
means the residual matrix is not diagonal. On the other hand, with k factors there are
only n×(k+2) parameters to estimate: n× k betas, n intercepts or means, and n unique
variances.
If one becomes primarily interested in the evidence regarding the pricing implications
of APT, one main implication is that expected returns on assets are approximately
linear in their sensitivities to the factors, that is, 0[ ] nE r ι λ λ≈ +Β . With some additional
restrictions, it is possible to obtain the pricing relation of APT as an equality. Since
standard statistical methods are not amenable to testing approximations, most empirical
tests evaluate whether 0[ ] nE r ι λ λ≈ +Β holds as an equality. Often, joint tests of APT
Mazharul Kazi Ph.D Thesis 76
with any ancillary assumptions are performed for obtaining the exact pricing relation,
that is, 0[ ] nE r ι λ λ= +Β [Shanken (1985)].
Once the relevant factors are identified or estimated, approaches to analyzing and
testing APT proceed in most cases by analyzing and testing other competitive asset
pricing models, for example, the capital asset pricing model, and by investigating
various aspects of 0[ ] nE r ι λ λ= +Β . Some researchers have focused on evidence
regarding the size and significance of the factor risk premium vector, λ. One testable
restriction of the model is that the implied risk premiums are the same across subsets of
assets. This implies that, if we partition the return vector, r, into components, r1, r
2,…,
rs, with B
i representing the same partitioning of B, the subset pricing relations become
0[ ]i n i i iE r ι λ λ= +Β , where i = 1, 2, , s. It follows that 0 0
iλ λ= and iλ λ= for all i. An
additional restriction implied by the pricing model is that variables in the agents
information set should not allow investors to predict expected returns which differ from
the relation in0[ ] nE r ι λ λ= +Β . These restrictions form the basis for testing APT.
Now, the exact pricing relation,0[ ] nE r ι λ λ= +Β , along with the factor model for the
return generating process, [ ]r E r f ε= +Β + , imply that the n-vector of returns at time t,
rt, is given by:
0, 1 1
( )n
t t t t tr fι λ λ ε− −= +Β + + (3.02)
Mazharul Kazi Ph.D Thesis 77
The riskless rate of return,0, 1tλ −
, and the risk premia, λ t-1, have a time t-1 subscript since
they are determined by expectations conditional on information at time t-1. If one
observes the return on the risk-free asset, 0, 1tλ −
, one would obtain an equivalent relation
between returns in excess of the risk-free rate 0, 1
n
t t tR r ι λ −= − , B, and the factor returns,
1t tfλ − + . This relationship takes the following form:
1
( )t t t t
R fλ ε−= Β + + (3.03)
Almost every empirical study of APT involves analyses of a panel of asset returns in
which we observe a time series of returns (t = 1,2,,T) on a cross-sectional sample of
assets or portfolios (the n individual assets in rt or Rt). Although all empirical studies
combine cross-sectional and time series data, it is common to classify them as cross-
sectional or time series studies on the basis of the approach used in the final testing
stage of the analysis. A study conditional on B, 0, 1 1( )
n
t t t t tr fι λ λ ε− −= +Β + + , and
1( )t t t tR fλ ε−=Β + + , can be thought of as a cross-sectional regression in which the
parameters being estimated are 0, 1t
λ − and (
1t tfλ − + ). Conversely, conditional on
0, 1tλ −
,
the formulae (1t t
fλ − + ), 0, 1 1
( )n
t t t t tr fι λ λ ε− −= +Β + + , and
1( )
t t t tR fλ ε−= Β + + can be
thought of as a time series regression in which the parameters being estimated are
estimates of B only. Some studies, however, jointly estimate the model in one step.
A recent study by Merville, Hayer-Yelken and Xu (2001) agreed that stock returns can
be better understood using several dimensions. Accordingly, they investigated two
Mazharul Kazi Ph.D Thesis 78
matters. First, they studied the number of fundamental factors [Fama and French
(1992)] that drive the returns. Secondly, they tried to bridge the gap between the
statistical factors and economic factors by interpreting the former using the latter. They
used both principal components analysis as well as the mispricing approach of Connor
and Korajcyzk (1993) to determine the number of pricing factors. They relied on
principal components analysis because it does not require pre-specifying the number of
explanatory factors or variables. The principal components analysis helped them
identify the meaning of the factors found in the portfolio analysis because they relied
on the correlation structure of the data inputs. Accordingly, through principal
components analysis they were able to study both the number and the nature of the
factors driving returns over time without pre-specifying them. Once factors are
constructed, regression analysis is used to determine the consistency of the factor
model with the corresponding APT model. This technique also allowed them to
calculate a relative measure of the importance of each factor found through principal
components analysis. Their results indicate that there are most likely three significant
fundamental factors that explain common stock prices: (i) market return, which also
includes idiosyncratic return; (ii) market capitalisation; and (iii) the investment
opportunity set. Moreover, they found that higher-order factors are uniquely identified
with macroeconomic variables.
The research work of Bai and Ng (2002) attempted to determine the number of factors
in approximate factor models. Their study proposes econometric theory for factor
models of large dimensions. Their idea is that variations in a large number of economic
variables can be modelled by a small number of reference variables and this can be
Mazharul Kazi Ph.D Thesis 79
analyzed. Their focus was to determine the number of factors, r, which is an unresolved
issue in the rapidly growing literature on multi-factor models. They first established the
convergence rate for the factor estimates that will allow consistent estimation of r. They
then proposed some panel criteria and showed that the number of factors can be
consistently estimated using the criteria. A theory has been developed under the
framework of large cross-sections, N, and large time dimensions, T. They later
conducted simulations that showed that the proposed criteria have good finite sample
properties in many configurations of the panel data encountered in practice.
They considered the following model, which treats Xit as observed data for the ith
cross-
section unit at time t, for i =1,,N, and t =1,,T:
i t i t i tX F eλ ′= + (3.04)
Ft is a vector of common factors; t
λ is a vector of factor loadings associated with Ft,
and eit is the idiosyncratic component of Xit. The product i tFλ ′ is called the common
component of Xit. The above equation then becomes the factor representation of data.
As the factors, their loadings, and the idiosyncratic errors are not observable, this type
of factor analysis allows for dimension reduction and is in that sense a useful statistical
tool. Usually, many economic and financial analyses including APT fit naturally into
this framework.
Mazharul Kazi Ph.D Thesis 80
However, as security prices are responsive to external factors, the relationship between
equity markets and exogenous variables is not necessarily unidirectional. Various
researchers considered diverse a priori (pre-specified or choice) variables. Chen, et al.
(1986) considered nine a priori macroeconomic variables. They were inflation, the
treasury-bill rate, long-term government bonds rate, industrial production, low-grade
bonds, equally weighted equities, value weighted equities, consumption, and oil prices.
On the other hand, Fama and French (1993) considered five variables that can explain
average returns, but do not have any special standing in conventional asset pricing
theory. Those specific financial variables were size, leverage ratio, earnings/price ratio,
book-value/market-value ratio, and overall market beta.
McGowan and Francis (1991) considered 25 representative macroeconomic factors
(chosen from between 86 and 95 variables) in their research design to identify
relationships between APT factors and macroeconomic variables in a broader fashion.
McGowan and Francis (1991) factors were: overall production; inflation; money
supply; personal consumption; interest rates; weekly wages; wholesale sales; lagging
indicators; labour force; building construction; total federal debt; federal budget
financing; total construction; new one-family houses; business formation; exports and
imports; population; inventories; leading indicators; non market interest bearing debt;
average weekly hours of service workers; free reserves of all banks; budget deficit and
current liabilities; consumer credit outstanding; commercial banks asset value; and net
budget receipt plus money stock.
Mazharul Kazi Ph.D Thesis 81
Johansen (2000) pointed to the appropriateness of applying cointegration techniques
even for APT, as testing the arbitrage pricing theory often describes a one-period model
for asset prices and it derives a restriction on the mean return by assuming that there
should be a nonarbitrage opportunity by creating a portfolio with positive excess mean
return and no risk. Although the exact factor model does not allow this possibility, the
lack of the restriction opens an approximate arbitrage opportunity by diversification
over many assets. Johansen indicated that if we consider a multi-period model, and
instead of the asset returns consider the cumulated asset returns or the log prices, then
these variables are found to be nonstationary. Johansen and Lando (1999) suggested
that, if we fit a cointegration model with a linear term restricted to the cointegrating
relations, we will get a model for a rebalanced portfolio. The nonarbitrage condition
would be that any portfolio with linearly increasing mean and constant risk must have
the same mean return. Thus, the APT hypothesis is a restriction on the deterministic
terms and the cointegrating vectors [Johansen (2000)].
Summary
We have reviewed empirical APT in this chapter to ascertain whether the theoretical
underpinnings of APT and its empirical forms are the same. We observed that empirical
APT differs from theoretical APT in many ways. Still, theoretical APT implies an
empirical version of APT. We saw that traditional approach to test an APT model
follows statistical factor analysis, although many researchers used other methods,
including regression and principal component analyses. From a number of legitimate
versions of APT, the Roll and Ross two-step procedure is found popular in empirical
Mazharul Kazi Ph.D Thesis 82
analysis. And yet, many researchers tried various other testing methods, including
cointegration, on different versions of APT, including the dynamic version of APT.
Any empirical approach to APT always maintained harmony with the corresponding
school of thought on the factors (either latent or pre-specified). In this chapter we
reviewed the relevant empirical works of APT for the purposes of identifying an
appropriate approach to quantitative analysis.
Mazharul Kazi Ph.D Thesis 83
CHAPTER 4
FUNCTIONAL SECURITY PRICING FROM A MARKET
VIEWPOINT
A clear view of security market pricing procedures and related issues are important to
an astute investor. This chapter will briefly review those studies in asset pricing
analysis that are particularly relevant from a security market viewpoint.
Security market prices are represented by security market indices, which are often
measured by a stock price index, a way of measuring the performance of a market over
time. Indices are regarded as an important indicator by the finance industry and the
investing public at large. An index can be used as a benchmark by which an investor or
fund manager compares the returns of their own portfolio. As the share market index
represents the return of the market as a whole, one can easily evaluate market or
industry performance using this index. The market index can also be used to define the
universe from which investors or fund managers pick their stocks. Furthermore, indices
are useful to form indexed portfolios as the basis for estimating the risk of an investors
own portfolio. Fund managers portfolios can either be indexed (risk-neutral),
overweight or underweight, compared to the weights in their benchmark [Standard and
Poors (2002)].
We also acknowledge that significant risk factors need to be priced. Hence,
understanding systematic risk factors and determining whether these factors are being
appropriately priced for both an individual asset and stock indices are important to any
Mazharul Kazi Ph.D Thesis 83
CHAPTER 4
FUNCTIONAL SECURITY PRICING FROM A MARKET
VIEWPOINT
A clear view of security market pricing procedures and related issues are important to
an astute investor. This chapter will briefly review those studies in asset pricing
analysis that are particularly relevant from a security market viewpoint.
Security market prices are represented by security market indices, which are often
measured by a stock price index, a way of measuring the performance of a market over
time. Indices are regarded as an important indicator by the finance industry and the
investing public at large. An index can be used as a benchmark by which an investor or
fund manager compares the returns of their own portfolio. As the share market index
represents the return of the market as a whole, one can easily evaluate market or
industry performance using this index. The market index can also be used to define the
universe from which investors or fund managers pick their stocks. Furthermore, indices
are useful to form indexed portfolios as the basis for estimating the risk of an investors
own portfolio. Fund managers portfolios can either be indexed (risk-neutral),
overweight or underweight, compared to the weights in their benchmark [Standard and
Poors (2002)].
We also acknowledge that significant risk factors need to be priced. Hence,
understanding systematic risk factors and determining whether these factors are being
appropriately priced for both an individual asset and stock indices are important to any
Mazharul Kazi Ph.D Thesis 84
investor or fund manager. If systematic risk factors are cointegrated with security
market prices, this knowledge is useful to investors in their decision making processes.
This type of study is becoming a focal point in finance. The studies suggest that stock
market returns are predictable from a variety of financial and macroeconomic variables.
The results of various studies suggest that financial and macroeconomic variables
influence stock prices across a variety of markets and time horizons [Campbell (1987),
French, Schwert and Sambaugh (1987), Fama and French (1989), Balvers, Cosimano
and McDoanld (1990), Been, Glostone and Jagannathan (1990), Cochrane (1991),
Campbell and Hamao (1992), Ferson and Harvey (1993), Glosten, Jagannathan and
Runkle (1993), Pesaran and Timmerman (1995, 2000), and McMillan (2001)].
Both international investors and researchers have been focusing their attention on stock
markets and fundamental macroeconomic dynamic interactions in countries (including
ASEAN) that provide attractive investment opportunities to foreign investors
[Wongbangpo and Sharma (2002)]. A number of studies, for example, modelled
relationships between US share prices and real economic activity [Fama (1981), Geske
and Roll (1983), Huang and Kracaw (1984), Chen, Roll and Ross (1986), Chen (1991),
Dhakal, Kandil and Sharma (1993), Abdullah and Hayworth (1993)]. Poon and Taylor
(1991) and Cheng (1995) investigated the relationships between the UK stock market
and macroeconomic factors, while Hamao (1988), Brown and Otsuki (1990), and
Mukharjee and Naka (1995) studied the Japanese market. Moreover, Fung and Lie
(1990) considered the Taiwanese market and Kwon, Shin and Bacon (1997)
investigated the Korean market for any relationship with their respective macro-
variables. The outcomes of all these studies suggest that, with minor degrees of
Mazharul Kazi Ph.D Thesis 85
variation, fundamental macroeconomic dynamics are indeed influential factors for
stock market returns.
The cointegration approach has been used by many researchers to analyse pricing
factors and to observe the relationships between economic variables and stock markets.
Despite the fact that the cointegration approach is still evolving within the realm of
time series analysis, it has become popular in empirical work in both economics and
finance since its introduction in the 1980s by Granger (1981) and Engle and Granger
(1987).
There has been a noticeable tendency to shift from the classic testing approach to
testing APT with cointegration techniques as it is a more efficient method. The shift
can partly be explained by the fact that researchers are now assessing whether
systematic risk factors or a priori variables are priced in stock markets rather than just
condensing factors dimensions for analyzing the asset pricing of particular markets
[Ackert and Racine (1999), Nasseh and Strauss (2000), McMillan (2001), Mookerjee
and Yu (1997), Maysami and Koh (2000), Shamsuddin and Kim (2000, 2003)].
Researchers are aiming to investigate both the long-run and dynamic relationships
between a priori variables and stock market prices. The trend is towards the
cointegration technique. Empirical frameworks are being developed in line with the
cointegration techniques of Engle and Grager (1987), Granger (1988) and Johansen
(1988, 1991, 1995, and 2000).
Mazharul Kazi Ph.D Thesis 86
McMillan (2001), using US data, undertook to test whether a cointegrating vector
exists between stock market indices and industrial production, inflation, money supply
and the interest rate. The results provided positive evidence of cointegration between
both the US market index (S&P 500) and the Dow Jones Industrial Average index
(DJIA) and macroeconomic activity variables. McMillan (2001) focused his research
on both the long- and short-run relationships between stock markets rather than on
individual stock returns and financial and macroeconomic variables. He pointed to the
fact that research on long-run relationships between stock markets and the unit root
components of economic activity variables is yet to receive significant attention.
By using the concept of cointegration, McMillan (2001) investigated empirically long-
run equilibrium relationships between stock market indices and pre-selected economic
and financial variables. He also used variance decompositions to see which
macroeconomic factors explain a substantial part of the variation in stock prices over
the short and medium-run, namely, one, four and eight years. Variance decompositions
are constructed from a vector autoregression (VAR) with orthogonal residuals and can
directly address the contribution of macroeconomic variables in forecasting the
variance of stock prices [Sims (1980), Litternman and Weis (1985)].
McMillan (2001) used Johansens (1991, 1995) vector error correction model (VECM)
for ascertaining the cointegrating relationship between integrated time series through
representation by a vector autoregression. He asked whether each series was integrated
of the same order by considering both the Augmented Dickey-Fuller and Phillips-Peron
tests. He assumed that each series contained a single unit root, and thus was integrated
Mazharul Kazi Ph.D Thesis 87
of the same order. He found a potential for co-movement between these series exists,
meaning a linear combination of this co-movement is stationary. This suggests a long-
run relationship between these variables.
Researchers who have investigated long-run relationships between macroeconomic
variables and stock market indices focused their attention on determining the dynamic
relationships between a priori variables and a representative stock market index
[Maysami and Koh (2000), Mookerjee and Yu (1997), Shamsuddin and Holmes
(1997), Mallik, et al. (2001), Shamsuddin and Kim (2000, 2003), Nasseh and Strauss
(2000)]. The proxy variables chosen by these researchers varied from one stock market
to another. Also, the analytical methods varied noticeably. Yet, the basis of all these
studies in security pricing originated from the theoretical perspective of APT.
To assess the impact of economic forces on the stock market, Maysami and Koh (2000)
related monthly stock market levels and monthly stock returns using Johansens
VECM. Maysami and Koh (2000) adopted Mukharjee and Nakas (1995) lesson for
Singapore. Shamsuddin and Kim (2000, 2003) examined cross-country stock market
relationships by employing the cointegration technique of Johansen (1988). Chen
(1995) and Clare, Maras and Thomas (1995) used pair-wise cointegration in studying
the performances of a group of countries.
While examining the dynamic relationships between the Singapore stock market and
the two well-developed stock markets of the US and Japan, Maysami and Kohs (2000)
process of selecting macroeconomic variables for the Singapore stock market was
Mazharul Kazi Ph.D Thesis 88
based on the simple and intuitive financial theory suggested by Mukherjee and Naka
(1995), and Chen, et al. (1986). They recognized that although the Engle and Grager
(1987) two-step error correction model (ECM) was suitable for use in the multivariate
context, VECM yielded efficient estimators of cointegrating vectors. This was because
VECM is a full information maximum likelihood estimation method. It allows
cointegration to be tested in a whole system of equations in only one step without
requiring a specific normalized variable. VECM avoids carrying over the errors from
the first step into the second, and it does not require prior assumptions about the
variables (endogenous or exogenous).
In estimating a VECM, Maysami and Kho (2000) first checked for stationarity and unit
roots by performing the Augmented Dickey-Fuller and Phillips-Peron tests on the
variables in levels and first differences. Only variables integrated of the same order
seemed cointegrated, and the unit root tests helped them to determine which variables
were integrated of order one, or I(1). They chose lag lengths using Sims likelihood
ratio test. For simplicity, they used the multivariate forms of the Akaike Information
Criterion and Schwartz Bayesian Information Criterion. Their choice of the number of
maximum cointegrating relationships was based on the trace
λ tests. The max
λ test is used
to test specific alternative hypotheses.
Having determined the order of cointegration, Maysami and Kho (2000) analyzed the
relevant cointegrating vector and the speed of adjustment coefficients. They performed
tests on the parameters of the cointegrating vector using the likelihood ratio test. This
was required for understanding whether stock prices contributed to the cointegrating
Mazharul Kazi Ph.D Thesis 89
relation and also if the macroeconomic variables were significant in the cointegrating
relationship.
Maysami and Kho (2000) used monthly time series in natural logarithms of these
variables: share price index (LSESt), M2 money supply (LM2t), consumer price index
(LCPIt), industrial production index (LIPt), domestic export (LTDEt), bank rate
(LSTBt), and yield on government securities (LLTBt). They gathered time series data
from various sources including the Public Access Time Series System, an online
service of the Singapore Department of Statistics. Data on the exchange rate (LERt)
was obtained from the International Financial Statistics, published by the International
Monetary Fund. Both the US stock market price (LUSt) and the Japanese stock market
price (LJPNt) were obtained from the US Bureau of Labour Statistics. Definitions and
transformations of these variables are presented in Tables 4.1 and Table 4.2
respectively.
TABLE 4.1
Definitions of Variables Variables Definitions of Variables
LSESt Natural logarithm of the index of the market-value weighted average of month-
end closing prices for all shares listed on the Stock Exchange of Singapore
LERt Natural logarithm of the month-end exchange rate of the Singapore
LM2t Natural logarithm of the month-end M2 money supply in Singapore.
LCPIt Natural logarithm of the month-end Consumer Price Index.
LIPt Natural logarithm of the month-end Industrial Production Index.
LSTBt Natural logarithm of the month-end 3-month Interbank Offer Rate.
LLTBt Natural logarithm of the month-end yield on 5-year government securities.
LUSt Natural logarithm of the month-end stock-price index of the United States.
LJPNt Natural logarithm of the month-end stock price index of Japan.
LTDE Natural logarithm of the month-end Total Domestic Export from Singapore.
Source: Maysami and Kho (2000).
Mazharul Kazi Ph.D Thesis 90
TABLE 4.2
Time-Series Transformations Transformation Definitions of Transformations
∆SESt = LSESt LSESt-1 Monthly return on the Singapore stock market (ex-dividend).
∆ERt = LERt LERt-1 Monthly change in exchange rate.
∆M2t = LM2t LM2t-1 Monthly growth rate of money supply.
∆CPIt = LCPIt LCPI.t-1 Monthly realized inflation rate.
∆IPt = LIPt LIPt-1 Growth rate of industrial production.
∆STBt = LSTBt LSTBt-1 Monthly return on 3-month interbank market (short- term).
∆LTBt = LLTBt LLTBt-1 Monthly return on government bonds (long-term).
∆USt = LUSt LUSt-1 Monthly return on US stock market (ex-dividend).
∆JPNt =LJPNt LJPNt-1 Monthly return on Japanese stock market (ex-dividend).
∆LTDE = LTDEt LTDEt-1 Monthly change in Singapores total domestic exports
Source: Maysami and Kho (2000).
Maysami and Kho (2000) used seasonally adjusted month-end data for the period from
January 1988 to January 1995. The index they used was a value weighted index of all
the shares listed, which provided them with a broad measure of market levels. Based on
their test results, Maysami and Kho (2000) concluded that there were two cointegrating
relationships at the 5% level of significance.
Earlier, Mookerjee and Yu (1997) studied the Singapore stock market pricing
mechanism by investigating whether there were long-term relationships between
macroeconomic variables and stock market pricing. They chose the all-share price
index to broadly represent Singapore stock market prices. They found that three out of
four macroeconomic variables were cointegrated with stock market prices. To test for
informational efficiency they employed cointegration and causality techniques because
these techniques allowed for any potential linkages between variables in the long-run as
well as in the short-run.
Mazharul Kazi Ph.D Thesis 91
Mookerjee and Yu (1997) began by conducting tests of nonstationarity. They
considered four pre-selected macroeconomic variables that included the narrow money
supply, the broad money supply, nominal exchange rates, and foreign currency reserves
to ascertain whether these variables were related to Singapore stock market prices in
both the long- and short-runs. In their unit root testing process, Mookerjee and Yu
(1997) could not reject the hypothesis for all five variables at the 1% significance level.
But the hypothesis could be rejected in first differences at the 5% significance level.
The authors used the Engle and Granger (1987) method for identifying causality and
cointegration between stock market prices and macroeconomic variables. They also
employed the Johansen (1991) approach for further confirmation of their results. The
Engle-Granger tests of cointegration suggested that the Singapore stock market was
efficient in the long-run with respect to its exchange rate. Alternatively, they used the
causality test of Granger (1996) for short-run informational efficiency. As the test of
causality is sensitive to lag length selection as per Thornton and Batten (1985), they
conducted causality tests between stock price change and each macroeconomic variable
series up to the 12th lag. They observed that changes in Singapore stock prices had a
causal relationship only with the narrow money supply. To obtain concrete evidence
concerning the efficiency of the Singapore stock market, Mookerjee and Yu (1997)
conducted further tests that entailed calculating the forecasted or anticipated
components of macroeconomic variables.
Nasseh and Strauss (2000) also found support for the existence of significant long-run
relationships between stock market prices (represented by relevant share price indices)
and domestic and international economic activity in six countries that included France,
Mazharul Kazi Ph.D Thesis 92
Germany, Italy, Netherlands, Switzerland and the UK. In their study Johansens
cointegration tests demonstrated that stock price levels were significantly related to
industrial production, business surveys of manufacturing orders, short- and long-term
interest rates as well as foreign stock prices, short-term interest rates, and production.
Nasseh and Strauss (2000) also used variance decomposition methods that supported
the strong explanatory power of macroeconomic variables in contributing to the
forecast variance of stock market prices. They recognized the usefulness of Johansens
framework for analyzing stock market and macroeconomic activity it incorporates
dynamic co-movements or simultaneous interactions, allowing the researchers to study
the channels through which macroeconomic variables affected asset pricing, as well as
their relative importance. Their variance decomposition methods, based on a vector
autoregression with orthogonal residuals, showed that macroeconomic factors
explained a substantial part of the variation in stock prices in the medium and short-
runs. Nasseh and Strauss (2000) found that although stock prices were explained by
economic fundamentals in the medium and short-run, the underlying volatility inherent
in stock prices was related to macroeconomic movements in the long-run.
Using US equity market index data, Actert and Racine (1999) examined whether spot
and futures equity markets were cointegrated. Their analytical model was based on the
nonarbitrage, cost-of-carry pricing model of Brenner and Kroner (1995). They assumed
that a stock index and its futures price was cointegrated if the cost of carry was
stationary, otherwise the cointegrating relationship would be trivariate. They found that
the S&P500, associated futures series, and the interest rate were all nonstationary. In
addition, their cointegrating relationships included the index, futures price, and the cost
Mazharul Kazi Ph.D Thesis 93
of carry. Their findings were consistent with the nonarbitrage pricing model and did not
appear to be sensitive to the presence of structural breaks in the series.
Based on practical experience, Duy and Thoma (1998) noticed that although the issue
of identifying cointegrating relationships between time series variables was of
increasing importance, researchers had to reach an agreement on the appropriate
manner of modelling such relationships. They attempted to distinguish between
modelling techniques through a comparison of forecast statistics, at the same time
focusing on the issue of whether or not imposing cointegrating relations via an ECM
(error correction model) improved long-run forecasts. They found that imposing
cointegrating restrictions often improved forecasting power.
Duy and Thoma (1998) constructed a model that took into account zt, an n×1 vector of
nonstationarity, I(1) and time series variables. They assumed that two or more of the
elements of zt form a stationary, I(0), linear combination. Their ECM representing such
a system of variables is:
1 1
1
jE C
t t i t t
i
z w zα γ ε− −=
∆ = + ∆ +∑ (4.01)
where t t
w zβ ′= is the cointegration term;β is an n×1 vector of coefficients in the
cointegrating vector; α is an n×1 vector of coefficients; and EC
tε is a vector of
Gaussian error terms. Duy and Thoma determined the elements of the cointegrating
vector both theoretically and empirically. In doing so they first considered a vector
Mazharul Kazi Ph.D Thesis 94
autoregression (VAR) model in pure differences for examination because if
cointegration was not present in the variables zt then the above error correction models
would be misspecified. This type of model is often recommended when the elements of
zt are difference stationary but not cointegrated. They also considered that a VAR
model in pure levels which estimates a VAR in levels avoids the possibility of
imposing false restrictions on the model. These VARs are:
1
jD
t t t i t
i
z zγ ε−=
∆ = ∆ +∑ (differences) (4.02)
and
1
jL
t t t i t
i
z zγ ε−=
= +∑ (levels) (4.03)
Duy and Thoma (1998) thus tested cointegration within their three systems of equations
by using the Johansen maximum likelihood procedure, which they found satisfactory.
Paul and Mallik (2001) investigated the long-run relationship between three pre-
specified macroeconomic factors and stock prices in the banking and finance sector in
Australia by using cointegration techniques. As a proxy for stock prices, they used time
series data on ASX bank and finance share market indices. All data were quarterly
observations for the period 1/1980 to 1/1999. They measured inflation as the change in
the quarterly consumer price index. They considered three-monthly bank bill rates as
interest rates, and logged seasonally adjusted gross domestic product (GDP) at 1997-98
base prices as a proxy for GDP growth. Upon detection of any cointegrating
Mazharul Kazi Ph.D Thesis 95
relationship, for estimation of their model they constructed an error correction
autoregressive distributed lag (ARDL) model in line with Pesaran and Shin (1995).
Their ARDL error correction model was as follows:
10 1 2 3 1
1 0 0
ln ln lnp r s
t t i t k t
t k l
BP BP INT INFφ φ φ φ− − −− = =
∆ = + ∆ + ∆ + ∆∑ ∑ ∑
4 1 1
0
lnq
t j t t
j
G D P eφ ρ µ− −=
+ ∆ + +∑ (4.04)
where ∆ denotes a first difference operator; µt-1 is an error correction term; p, q, r and s
are the lag lengths; and et is the random disturbance term. Although their results could
not show any significant effect of inflation on stock prices, they found both the interest
rate and GDP growth significant. The interest rate had a negative effect, whereas GDP
growth had a positive effect on the banking and finance sectors stock price.
Earlier, Lim and McNelis (1998) studied the effect of US and Japanese stock market
price movements on that of Australian markets by examining the influence of shocks in
the Japanese Nikkei index and in the US S&P index on the Australian All Ordinaries
index. They considered logged daily data series from 28 September 1990 to 12
September 1996, and used three models including an autoregressive linear model.
More recently, Wongbangpo and Sharma (2002) investigated the role of five
macroeconomic variables, namely, gross national product, consumer price index,
money supply, interest rate, and exchange rate on the stock prices in five ASEAN
countries (Indonesia, Malaysia, the Philippines, Singapore, and Thailand). They
Mazharul Kazi Ph.D Thesis 96
observed both long- and short-term relationships between stock prices and their
macroeconomic variables for the period 1985 to 1996. They used monthly data that
were found to be integrated of order one from the Augmented Dickey-Fuller and
Phillips-Peron tests. They also used maximum likelihood based trace
λ and max
λ statistics
through the cointegration analysis technique of Johansen (1988, 1991) and Johansen
and Juselius (1990) for tracing any causal relationship within the framework of the
VECM. Their system of the short-run dynamics of the stock price series (SPt) was as
under:
1 1 1 1 1 1
1 1 1
k k k
t t i t i i t i i t i
i i i
SP EC SP G N P C PIα β δ θ ξ− − − −= = =
∆ = + + ∆ + ∆ + ∆∑ ∑ ∑
1 1 1 1
1 1 1
k k kS P
i t i i t i i t t
i i i
M S I R E Rρ ω τ ε− − −= = =
+ ∆ + ∆ + ∆ +∑ ∑ ∑ (4.05)
where ECt-1 is the error correction term obtained from the cointegration vector,
, , , , , ,β δ θ ζ ρ ω τ are parameters; k is the lag length; and SP
tε are a stationary random
process with mean zero and constant variance.
Summary
This chapter has reviewed the trend towards analyzing the relationship between the
security market movement and a priori variables, while retaining the basic attributes of
asset pricing theory. As has been seen, many researchers have studied the long-term
relationship between stock indices and macroeconomic variables as well as short-term
dynamics using the cointegration technique. It has become apparent in this chapter that
Mazharul Kazi Ph.D Thesis 97
using the cointegration approach one can efficiently analyse the long-run relationship
between a priori variables that are considered as a proxy for systematic risk factors and
security market prices. We have also seen in this chapter that an additional cross-
country study which traces the interrelationship between various stock markets is useful
for understanding the global impacts on a specific stock markets price movements.
Our contribution here has been to find that an empirical study in asset pricing imposes
no limits within the boundaries of its traditional methods nor models. We deem both
the cointegration method and an autoregressive model based on the market stock prices
is more suitable for us. We will examine the cointegration approach as our preferred
method of analysis in further detail in Chapter 5.
Mazharul Kazi Ph.D Thesis 98
PART III
EMPIRICAL APPROACHES TO SECURITY MARKET
PRICING
The third part of the thesis considers our empirical work. This part consists of two
chapters. Chapter 5 investigates the approaches used to derive our methodological
framework. Chapter 6 systematically derives the methodology suitable for developing
the appropriate model for our research.
Mazharul Kazi Ph.D Thesis 99
CHAPTER 5
METHODOLOGICAL OUTLINE
Many methods are available to analyse empirically the asset pricing mechanism and
determine the relationship, if any, between stock market returns and a priori variables.
As is evident from preceding chapters, the influences of systematic risk factors on the
stock pricing process can be detected by employing various models. A record of the
analytical techniques used in various studies is provided in Table 5.1.
TABLE 5.1 Methods Used for Investigating Asset Pricing
Analytical Methods Recorded Studies
Factor Analysis Ross (1976, 1977); Ross and Roll (1980); Chen, Roll and Ross (1983, 1986); Gehr (1978); Fogler, et al. (1981); Hughes (1985); Beenstock and Chan (1983, 1984); Winkelmann (1984); Dumontier (1986); Kim and Lee (!995); Connor and Korajezyk (1989); McGowan and Francis (1991); Brailsford and Heaney (1998); Connor (1995); Bai and Ng
(2002); and Fama and French (1992, 1993).
Asymptotic Principal Component Chamberlain and Rothschild (1983); Anckonle (1983); Connor and Korajczyk (1986, 1988, 1989); McGowan and Francis (1991); Shukla and Trzcinka (1990); Faff (1988, 1992); and Merville, et al. (2001).
Standard, Cross-Section and Time Series
Regression Analyses
Beggs (1986); Merton (1973); Kritzman (1995); Fam and French
(1992, 1993); Brown and Weinstein (1983); Aitken, et al. (1996); Brailsford and Easton (1991); Faff and Brailsford (1999); Hamao (1988); Faff and Heaney (1999); and Easton and Faff (1994).
Generalised Method of Moments
Zhou (1999); and Velu and Zhou (1999).
Maximum Likelihood Handa and Linn (1993).
Autoregressive Conditional
Heteroskedasticity
Brailsford and Faff (1996)
Cointegration Maysomi and Kho (2000); Johansen (1988, 1991, 1995, 2000); Pesaran and Timmerman (1995); McMillan (2001); Mukharjee and Naka (1995); Cheung and Ng (1998); Ackert and Racine (1999); Nasseh and Stranss (2000); Mukharjee and Hoh (2000); Shamsuddin and Holmes (1997); Cheung and Ng (1998); Paul and Mallik (2001); Roca (1998); and Shamsuddin and Kim (2000, 2003).
Mazharul Kazi Ph.D Thesis 100
Underpinning any methodological approach adopted by empirical studies in finance is
regression analysis. The most common methods are investigated in this chapter to
ascertain the most appropriate for a system which is considered to be a group of
equations containing unknown parameters. In general terms, a system can be estimated
using a number of techniques that take into account the interdependencies among the
equations within the system. This form of system is represented by
( ), ,t t t
F y x β ε= (5.01)
where yt is a vector of endogenous variables; xt is a vector of exogenous variables; and
εt is a vector of possibly serially correlated disturbances. The task of estimation is to
find estimates of the vector of parameters β.
Almost all systems in finance deal with multiple variables so that the methodology falls
within the field of multivariate analysis as well as multiple regression analysis.
Multivariate analysis refers to the analysis of multiple variables singly or in a set of
relationships. Although there are nonlinear relationships, studies for linear relationships
are common. The basis of multivariate analysis is the variate, a linear combination of
variables with empirically determined weights [Hair, et al (1995)]. Usually, a linear
combination of variables with empirically determined weights is the variate value,
whereas multiple regression analysis is a statistical technique that can be used to
analyse a relationship between a single dependent (criterion) variable and several
independent (predictor) variables. The objective of multiple regression analysis is to
use independent variables whose values are known to predict the dependent variable.
Mazharul Kazi Ph.D Thesis 101
The basic motivation for estimating all regression relationships is correlation, which
measures the association between two variables. In regression analysis, the correlation
between the independent variables and the dependent variable provides the basis for
forming the regression variate by estimating regression coefficients (weights) for each
independent variable that maximize the prediction (explained variance) of the
dependent variable.
When more independent variables are added to the variate, the process must also
consider the inter-correlations among independent variables. If the independent
variables are correlated, they share some of their predictive power. Because only the
prediction of the overall variate is used, the shared variance must not be double counted
by using the direct correlations. In these cases, some additional forms of correlation
that represent shared effects are to be calculated for an accurate prediction of the
dependent variable. Partial correlation is the appropriate tool, where correlation of an
independent and dependent variable is obtained by removing the effects of other
independent variables.
To answer any empirical research question and formulate an appropriate model, one
needs to choose an appropriate technique or method for data analysis.
Mazharul Kazi Ph.D Thesis 102
Empirical Methods in Asset Pricing
General Linear Regression
Most of the techniques for asset pricing analyses are special applications of the general
linear regression model (GLM). Within the GLM, a simple form of regression is
A + βX + e = Y (5.02)
where β is the change in Y associated with a one-unit change in X; A is a constant
representing the expected value of Y when X is zero; and e is a random variable
representing the error of prediction.
If X and Y are converted to standard z scores, then zx and zy are measured on the same
scale. The constant A automatically becomes zero, because zy is zero when zx is zero.
Standardising the variance to 1, the closer the β to ± 1 the better the prediction of Y
from X (or vice versa). The equation becomes βZx + e = Zy. When this simple bivariate
form of the GLM adds more independent variables, Xs, to predict Y, it takes the shape
of a multivariate form of the regression:
1
k
i x i y
i
Z e Zβ=
+ =∑ (5.03)
Mazharul Kazi Ph.D Thesis 103
The generalised linear model takes a major leap when the Y side of the equation is
expanded. More than one equation may be required to relate the Xs to the Ys. A full
multivariate form of the regression is
1 1
pk
im x im jm y j m
i j
Z e Zβ γ= =
+ =∑ ∑ (5.04)
where m equals k or p, whichever is smaller, and γ are regression weights for the
standardised Y variate. Generally, there are as many equations as the number of X or Y
variables, whichever is smaller.
Maximum Likelihood
The maximum likelihood (ML) method of estimation is often found useful in stock
pricing analyses. In simple terms, the maximum likelihood estimator (MLE) is that
value of estimated parameter/s that is most likely to have generated the observed sample
values, where most likely is defined as meaning those values of the population
parameters which have the greatest probability of giving rise to the sample values
actually obtained. The advantage of the ML method of estimation is that, under general
conditions, MLEs are consistent and asymptotically efficient. They have desirable large
sample properties. Given the circumstances, MLEs frequently have an important role to
play in multiple regression analysis. MLE is often viewed as an alternative method to
ordinary least squares (OLS), a technique which minimises the sum of squared
residuals for each equation as OLS is MLE when residuals are normally distributed.
Mazharul Kazi Ph.D Thesis 104
Generalised Method of Moments
Many studies analysing asset pricing have used the generalised method of moments
(GMM) in estimating parameters drawn from a random sample, as moments are
essentially the sums of observations. GMM estimation is a direct extension of the
method of moments which is a consistent estimation technique. The method is
applicable both in estimating a single equation and multiple system models. The basis
of this method is that, in random sampling, a sample statistic will converge in
probability to some constant [Greene (2000)]. The GMM estimator belongs to a class
of estimators known as M-estimators that are defined by minimising some criterion
function [EViews (1998, and 2001)]. This is based upon the assumption that the
disturbances in the equations are uncorrelated with a set of instrumental variables.
GMM can be made robust to heteroskedasticity and/or autocorrelation of unknown
form.
Seemingly Unrelated Regression
Sometimes the parameters of a system of equations are estimated by using the
seemingly unrelated regression (SUR) estimation method accounting for
heteroskedasticity and contemporaneous covariance in the errors across equations. SUR
is also known as multivariate regression, or Zellners seemingly unrelated regressions
method. The estimates of cross-equation covariance matrix are based upon parameter
estimates of the unweighted system. The SUR method allows for cross-equation
Mazharul Kazi Ph.D Thesis 105
restrictions on parameters and it is superior to the OLS method in such a situation. In
some special cases for asset pricing analyses this approach is found suitable.
Factor Analysis
The statistical approach called factor analysis has been used to analyse asset pricing
it describes the basis of interrelationships among a large number of financial and
economic a priori variables. The method explains the variables under consideration in
terms of their common underlying dimensions (factors). The objective is to find a way
of condensing the information contained in a number of original variables into a
smaller set of variables (factors), with a minimum loss of information [Hair, et al
(1995)].
Simple factor analysis begins with calculating the equal interval stock returns during
some period and isolating the factors that explain stock market performance. It
computes the correlation between the (daily or specified equal interval) returns of each
stock and the returns on every other stock to isolate the factors. Then it identifies the
groups of stocks that are highly correlated with each other but not with the stocks
outside the group. Similarly, it proceeds to isolate other groups of stocks whose returns
are highly correlated with each other within their own group, until it isolates all the
groups that seem to respond to a common source of risk. Thereafter, factor analysis
attempts to identify the underlying sources of risk for each group. In the first instance,
it relies on intuition to identify the factor that underlies the common variation in returns
among the stocks. Then it tests the intuition. In doing so, the approach defines a
Mazharul Kazi Ph.D Thesis 106
variable that serves as a proxy for the unanticipated change in the factor value. Finally,
it regresses the returns of stocks that appear to depend on the hypothesized factor with
the unanticipated component of the factor value [Kritzman (1995)].
Customarily, factor analysis reveals covariation in returns and attempts to identify the
sources of this covariation, whereas cross-sectional regression analysis requires
specifying the sources of return covariation and attempts to affirm that these sources do
indeed correspond to differences in return. Based upon intuition and prior research,
researchers often infer attributes, rather than factors, that they deem correspond to
differences in stock returns. Once a set of attributes is specified, a measure of
sensitivity to the common sources of risk can be detected. Then we regress the returns
across a large sample of stocks during a given period on the attribute values for each
stock as at the beginning of the period. This regression is repeated over many different
periods. If the coefficients of the attribute values are not zero and are significant in a
sufficiently high number of the regressions, it is concluded that differences in return
across the stocks relate to differences in their attribute values [Kritzman (1995)].
More often than not, security returns used to be characterized by very large cross-
sectional samples (often, more than 10,000 simultaneous return observations) with
strong co-movements. The fundamental sources of these co-movements are not always
obvious and are not easily measurable. Such a statistical system, where a few
unobservable sources of system-wide variation affect many random variables, leads to
factor modelling. Arbitrage pricing theory begins by assuming asset returns follow a
factor model. There exist n assets and k factors; f denotes the k-vector of random
Mazharul Kazi Ph.D Thesis 107
factors; β refers to the n×k matrix of linear coefficients representing assets sensitivities
to movements in the factors (factor loading). One can write an n-vector of returns, r, as
expected returns, E[r], and the sum of two sources of random returns, that is, a factor
return, Bf, and an idiosyncratic return, ε:
[ ]r E r f ε= + Β + (5.05)
where E[f] = 0, E[ε] = 0, and [ ] 0E f ε ′ = . B is defined by the standard linear projection
1[( [ ]) ]( [ ])E r E r f E ff −′ ′Β = − (5.06)
Given a vector of returns, r, and a vector of zero mean-variances, f, the standard linear
projection divides the returns into expected returns, k-linear components correlated
with f, and zero mean idiosyncratic returns uncorrelated with f [Connor and Korajczyk
(1995)].
Additional structural appearances of [ ]r E r f ε= +Β + appear in the literature providing
support for the hypothesis that the idiosyncratic returns are diversifiable risks. In
application, the strict factor models as well as the approximate factor models provide
additional support for this hypothesis. Often, conditional factor models rewrite the
equation with an explicit time subscript to capture the time-varying or dynamic nature
of the return generating process.
Mazharul Kazi Ph.D Thesis 108
Strict Factor Model
Since the factors and idiosyncratic risks are uncorrelated in strict factor models, the
covariance matrix of asset returns denoted by [( [ ])( [ ]) ]E r E r r E r ′∑ = − − is treated as
the sum of two matrices:
[ ]ff V′ ′∑ = ΒΕ Β + (5.07)
where V=E ][ εε ′ and the idiosyncratic returns are assumed uncorrelated with one
another. This means that the covariance matrix of the idiosyncratic risks, V, is a
diagonal matrix. The covariance matrix can be decomposed as the sum of a matrix of
rank k and a diagonal matrix of rank n. This imposes strict restrictions on the
covariance matrix as long as k < n.
This method divides the vector random process into k, common sources of randomness
and, n, asset-specific sources of randomness. Considering n is large and the
idiosyncratic variance of a portfolio with portfolio proportion (weights) as ω,
2 2 2 2
1 1
( m a x )n n
i i i ii
i i
Vω ω ω σ σ ω= =
′ = ≤∑ ∑ (5.08)
Since the portfolio weights sum to one, the average portfolio weight is 1/n. If the
holdings are spread widely over the n assets, then the sum of squared portfolio weights
approaches zero as n goes to infinity. As long as there is an upper bound on the
Mazharul Kazi Ph.D Thesis 109
idiosyncratic variances of individual assets, the idiosyncratic variance of any well-
spread portfolio will be near zero. Thus, given a strict factor model and many assets,
the idiosyncratic returns contain only diversifiable risks [Connor and Korajczyk
(1995)].
Approximate Factor Model
Ross (1976, 1977, 1990) used the factor model assumption to show that idiosyncratic
risks are diversifiable in many asset portfolios. The strict diagonality of V (as it is
assumed that [ ]V E εε ′= is a diagonal matrix) is a sufficient but not a necessary
condition for the diversification argument of Ross. To overcome the condition,
Chamberlain (1983) and Chamberlain and Rothschild (1983) proposed an asymptotic
statistical model for asset returns data called the approximate factor model. This model
retains the diversifiability of idiosyncratic returns but weakens the diagonality
condition on V. It imposes an additional condition that the factor risks are not
diversifiable.
The approximate factor model relies on limiting conditions as the number of assets
grows larger. The model performs a factor decomposition task where the εs are
diversifiable and the fs are pervasive. The diversifiable risk condition is equivalent to a
finite upper bound on the maximum eigenvalue of Vn as n tends to infinity. In addition,
the pervasiveness condition is equivalent to the minimum eigenvalue of nn ΒΒ '
when n
approaches infinity. Here, the covariance matrix of returns is the sum of two
components, 'nnΒΒ and V
n. Thus,
'nnΒΒ has all of its eigenvalues approaching to
Mazharul Kazi Ph.D Thesis 110
infinity, whereas Vn has a bound on all its eigenvalues. Chamberlain and Rothschild
(1983) showed that these bounds carry over to the covariance matrix. The k largest
eigenvalues of the covariance matrix go to infinity with n, and the k+1st largest
eigenvalue is bounded for all n. The model thus proves both the sufficient and
necessary conditions.
Conditional Factor Model
To accommodate time-varying or dynamic conditions in the modelling, the factor
model is mathematically expressed as:
1 1[ ]
t t t t t tr E r f ε− −= + Β + (5.09)
Let 1t−Β be chosen by the conditional projection of rt on ft. That means
1
1 1 1 1[( [ ]) ][( [ ] ]
t t t t t t t t tE r E r f E f f −
− − − −′ ′Β = − so that 1[ ] 0
t t tE fε− ′ = . Now the conditional
covariance matrix is:
1 1 1 1 1t t t t tV− − − − −′Σ = Β Ω Β + (5.10)
where 1 1
[ ]t t t t
E f f− − ′Ω = and 1 1
[ ]t t t t
V E ε ε− − ′=
Even if Vt-1 is diagonal for all t, the system is not statistically identified in this general
form. Connor and Korajczyk (1995) pointed out that to estimate the returns on n
Mazharul Kazi Ph.D Thesis 111
securities for T periods, one has to estimate the n elements of Vt-1, the nk elements of Bt-
1, and the k2 elements of Ωt-1 for each date t. This would provide a total of T(n+nk+ k
2)
parameters to estimate from nT return observations under a structured identifiable
model. To develop a dynamic model from the static model, some but not all of the
rotational indeterminacy of static models must be eliminated. Engle, et al. (1990)
suggested that, conditional on the assumed source of time variation (Bt-1 or Ωt-1) and
the assumed time series properties, it is assumed that the factors follow a GARCH
process. Sentana (1992) indicated that the dynamic structure here eliminates a notable
number of standard rotational indeterminacies of static models. Thus, the conditional
factor models allow for time variation in expected returns so that Et-1[rt] is not constant.
Multi-Factor Models
In their recent study, Merville, Hayes-Yelken and Xu (2001) pointed to a theoretical
perspective of asset pricing approaches. They acknowledged that although the capital
asset pricing model of Sharpe (1964) has limitations in terms of its assumptions and
rational implications in empirical asset pricing literature [Fama and French (1992)], the
multi-factor model that includes the market return should be used in practice to capture
the cross-sectional differences in the average returns of individual stocks. This practice
is generally seen in the arbitrage pricing theory analysis of Ross (1976). As the theory
does not provide any guidelines about factors, researchers often rely on a number of
methods that include statistical factor analysis, principal component analysis, and
attributes of macroeconomic and fundamental variables as asset pricing factors.
Mazharul Kazi Ph.D Thesis 112
With some blurring at the boundaries, the methodology that uses multi-factor models of
asset returns can be divided into three types: (i) macroeconomic factor model; (ii)
statistical factor model; and (iii) fundamental factor model [Connor (1995)]. A
macroeconomic factor model is typically intuitive and the simplest, requiring data on
security returns and macroeconomic variables. The pioneering studies of Chen, Roll
and Ross (1983), Berry, Chan, Chen and Hsieh (1985), and Burmeister and McElroy
(1988) all covered macroeconomic factor modelling which uses time series regression
to estimate security factor betas. The statistical factor model usually follows maximum
likelihood and principal component techniques of data analysis. It requires cross-
sectional or time series samples of security returns to identify the pervasive factors in
the return generating process of securities, and it also uses time series regression to
estimate the factor beta. The fundamental factor model does not requires time series
data or time series regression; rather it relies on attributes relevant to listed companies,
namely, the firm size, the dividend yield, book-to-market ratio, and industry
classification. The fundamental factor model uses observed company attributes as
factor betas, whereas both the macroeconomic and statistical factor models estimate a
firms factor beta by using a time series regression that requires a long and stable
history of returns. Connor (1995) studied the explanatory power of these three types of
multi-factor models and estimated a particular specification of each of the models based
on the work of Connor and Korajczyk (1993). Connor (1995) found that both statistical
and fundamental factor models substantially outperform the macroeconomic model and
the fundamental model slightly outperforms the statistical model. Although Connors
(1995) study was based only on the explanatory power criterion to evaluate the relative
strength of the three approaches to multi-factor modelling, he acknowledged that other
Mazharul Kazi Ph.D Thesis 113
criteria, namely, theoretical consistency and intuitive appeal, are significant in the
literature and deserve attention. In considering other criteria to assess the relative
strength of the models, Connor (1995) observed that the performance of the
macroeconomic factor model became the strongest rather than the weakest.
Time Series Analysis
In general econometrics, the analysis of autocorrelation in disturbances belongs in the
realm of time series analysis. Time series analysis requires some interpretation of both
data generation and sampling. A time series model will typically describe the path of a
variable yt in terms of contemporaneous and perhaps lagged factors xt, disturbances or
innovations εt and its own past yt-1, .
1 2 3 1t t t ty x yβ β β ε−= + + + (5.11)
Essentially, a time series is a single occurrence of a random event. The properties of yt
as a random variable in a cross-section are straightforward. It is common to assume that
disturbances are generated independently from one period to the next.
Cointegration
In the time series area, the most recently developed alternative approach to examine the
relationship between time series variables is cointegration analysis. The cointegration
analysis technique is still evolving and has many forms, as is evident from the literature
Mazharul Kazi Ph.D Thesis 114
which might sometimes seem confusing to the first timer. Often the terms causality and
cointegration are synonymous. The notion of causality has existed as an operational
concept in the empirical macroeconomic and financial literature since Granger (1969)
[Hansen and Mittnik (1998)]. As correlation does not necessarily imply causation in
any meaningful sense of that word, the Granger (1969) approach to the question of
whether the x variable causes y is to see how much of the current y can be explained by
past values of y and then to see whether adding lagged values of x can improve the
explanation. In the view of Granger (1969), y is said to be Granger-caused by x if x
helps in the prediction of y, or equivalently, if the lagged xs are statistically significant.
Usually, two-way causation is checked: x Granger causes y; and y Granger causes x.
The Wald test has been widely used for testing zero restrictions implying Granger non
causality.
It is important to find the causal structure of a vector autoregressive analysis since all
variables are assured to be endogenous in the sense of Sims (1980). During the late
1980s, Engle and Granger (1987) and Granger (1988) introduced a relatively clearer
concept of cointegration and error correction mechanism. Later, Johansen (1988)
proposed a vector autoregressive error correction model based on the concept of
canonical correlation and full information maximum likelihood estimation. Johansen
and Juselius (1990) provided further directions for cointegration analysis and showed
how to handle identification problems and impose necessary restrictions.
A simple exposition of the cointegration approach is that a set of time series variables
are said to be cointegrated if they are integrated of the same order and a linear
Mazharul Kazi Ph.D Thesis 115
combination of them is stationary [Maysami and Koh (2000)]. Such linear
combinations would then point to the existence of a long-term relationship between the
variables [Johansen and Juselius (1990)]. In other words, cointegration means that
despite being individually nonstationary, a linear combination of two or more time
series variables can be stationary in the long-run [Gujarati (1995)]. Although unit roots
or nonstationarity in variables are important for cointegration analysis, Spanos (1990)
argued that the question of whether a priori variables exhibit a unit root is ill-posed, as
it is not generally the case that the unit root found in the autoregressive representation
of a series will model in the context of a vector autoregressive representation. Caporale,
Hassapis and Pittis (1998) showed how the issue of unit roots is an invariant feature of
a time series that relates to long-run causality. They recognized that there are two steps
to model specification, variable selection and the imposition of a structure which draws
upon economic and/or finance theory. Pesaran and Shin (1994) also contributed to the
structural approach to cointegration analysis by developing a method of identification
for the long-run relations.
It may be noted that numerous techniques are available for empirical investigation
using cointegration analysis. We investigate some applications of cointegration analysis
so that we can choose which technique is suitable for our modelling and empirical
analyses.
Mazharul Kazi Ph.D Thesis 116
Applications of Cointegration Analysis
For any cointegration test it is necessary that a unit root in the time series for
nonstationarity is found at the outset. The test statistics can be based on the ordinary
least squares (OLS) results for a suitably specified regression equation. The following
are two forms of Augmented Dickey-Fuller (ADF) [Dickey and Fuller (1979, 1981)]
regression equations for a time series Yt:
0 1 1
1
p
t t j t j t
j
Y Y Yα α γ ε− −=
∆ = + + ∆ +∑ (5.12)
0 1 1 2
1
p
t t j t j t
j
Y Y t Yα α α γ ε− −=
∆ = + + + ∆ +∑ (5.13)
where we assume that t
ε for t = 1,.., N is a white noise process which has no
discernible structure with constant mean or constant variance σ2
and is uncorrelated
with other variables. The first of the above equations has a constant with no trend,
whereas the second equation has a constant with a trend. The number of lagged terms p
is chosen to ensure the errors are uncorrelated. Later, test statistics such as z-test, t-test,
and F-test with or without drift are calculated for both equations against the null
hypothesis to ascertain if there is a unit root. Table 5.2 depicts formats of test statistics
corresponding to each null hypothesis:
Mazharul Kazi Ph.D Thesis 117
TABLE 5.2
ADF Unit Root Testing Formats Null hypothesis | Test Statistic
For equation: 0 1 1
1
p
t t j t j t
j
Y Y Yα α γ ε− −=
∆ = + + ∆ +∑
1
1
0 1
0
0
0
ααα α
=== =
1
1
N
t ratio
F test
α
φ−−
( )
z test
test
Unitroot test zerodrift
τ−−
−
For equation: 0 1 1 2
1
p
t t j t j t
j
Y Y t Yα α α γ ε− −=
∆ = + + + ∆ +∑
1
1
0
0
αα
==
1N
t ratio
α−
z test
testτ−−
Source: Adapted from Dickey and Fuller (1979, 1981), Brooks (2002), EViews, Green (2000), Verbeek (2000) and Dhrymes (1998).
This method is capable of producing test statistics against each null hypothesis
considering both constant with or without trend and even no constant. The constant
with trend and drift are presented as 0 1 2 0α α α= = = and
1 2 0α α= = . Here, 0α is the
drift, 1α represents a unit root, and
2α indicates a trend. The presence of one unit root is
represented by I(1), whereas I(0) stands for no unit root. The presence of a unit root is
derived from test statistics and critical values for the same. When the null hypothesis
cannot be rejected the resulting outcome indicates the presence of a unit root, I(1), and
we say that the series is integrated of order one.
The Engle-Granger (1997) approach to cointegration analysis often uses the
Augmented Dickey-Fuller (ADF) tests on the OLS residuals applying critical values.
The approach requires running a regression of Y1t (being the first element of Yt) against
the other k-1 variables, Y2t, , Ykt, and testing for a unit root in the residuals. If the unit
root hypothesis is rejected, the hypothesis of non cointegration is also rejected.
Mazharul Kazi Ph.D Thesis 118
Phillips and Ouliaris (1990) indicate that the Phillips test is a non parametric approach.
Engle and Granger (1987) considered the same regression equation while undertaking
the Phillips test as in ADF, 0 1
1
k
t t j t j t
j
e e eδ δ η− −=
∆ = + ∆ +∑ , assuming k=0, τ *= t-ratio on
the estimate of δ0, te∆ is the first differenced series, and tη the residuals from the least
squares fit. The Phillips test statistic for a sample of n-observations is then
( )2 2
*
2
2 te
η ω σσττω ω
−= −
∑ (5.14)
where 2 1 2tnσ η−= ∑
and ( )2 1 2
1 1 1
2 ,n k n
t t t j
t j t j
n j kω η ω ηη−−
= = = +
= +
∑ ∑ ∑ (5.15)
employing 1( , )
( 1)
jj k
kω −=
+
where k is chosen sufficiently large to account for all non zero autocorrelations.
Under the null hypothesis of non cointegration, Loughran and Newbold (1997)
observed that both the ADF and Phillips tests provide the same asymptotic (large
sample) distribution. Shamsuddin and Holms (1997) point to a couple of limitations of
ADF, namely, that the arbitrary choice of the dependent variable in the cointegrating
regression often leads to a sequence of residual series based on which variable is
considered as dependent; and the ADF cannot be used to identify the existence of
Mazharul Kazi Ph.D Thesis 119
multiple cointegrating vectors. Their view is that these limitations of ADF can be
overcome by the Johansen procedure.
It is also possible to use the Durbin-Watson (DW) test statistic or the Phillips-Perron
(PP) approach for nonstationarity of residuals. If the DW test is applied to the residuals
of potentially cointegrating regression, it is known as the Cointegrating Regression
Durbin Watson (CRDW). Under the null hypothesis of a unit root in the errors, CRDW
≈ 0 (that is, CRDW is equivalent to zero), so the null of a unit root is rejected if the
CRDW statistic is larger than the relevant critical value which is approximately 0.5
[Brooks (2002)]. Thus, the null hypothesis Ho: ût ~ I(1) can be tested against the
alternative hypothesis H1:ût ~ I(0) using CRDW.
As most empirical studies in economics and finance involve both nonstationary and
trending variables, differencing and other transformations including seasonal
adjustments for reducing them to stationarity and then analyzing the resulting series are
appropriate. Yet, recent researches show more interesting and efficient analytical ways
for dealing with trending variables that are integrated of order one. In any fully
specified regression model, there is a presumption that the disturbance (tε ) is a
stationary, white noise series. However, this presumption is unlikely to be true if both
endogenous (Yt) and exogenous (Xt) series are integrated. Generally, if two series are
integrated to different orders, then linear combinations of them will be integrated to the
higher of the two orders [Green (2000)]. Thus, if both endogenous (Y) and exogenous
(X) variables at a period (t) are I(1), then we would normally expect Yt β Xt to be of
Mazharul Kazi Ph.D Thesis 120
I(1) regardless of the value of β . On the other hand, if the two series are both I(1),
there may be a β such that:
t t tY Xε β= − is I(0) (5.16)
Intuitively, if the two series are both I(1), then this partial difference between them
might be stable around a fixed mean. The implication of this would be that the series
are drifting together at roughly the same rate. Whenever two series satisfy this
condition, they are said to be cointegrated and the vector [1, β− ] or any multiple of it is
a cointegrating vector. In such a case we can ascertain a long-run relationship between
Yt and Xt and identify the manner in which the two variables drift upward together.
Alternatively, one can observe the short-run dynamics through the relationship between
deviations of Yt from its long-run trend and deviations from Xt from its long-run trend.
In this case, differencing of the data would be counterproductive, since it would
obscure the long-run relationship between Yt and Xt. Studies of cointegration and related
techniques such as vector autoregressive regression (VAR) and vector error correction
model (VECM) are concerned with methods of estimation that preserve information
about both techniques on covariation [Engle and Granger (1987), and Watson (1994)].
VAR is commonly used for forecasting systems of interrelated time series as well as for
analyzing the dynamic impact of random disturbances on the system of variables. The
VAR representation sidesteps the need for structural modelling by treating every
endogenous variable in the system as a function of the lagged variables of all the
endogenous variables. The VAR may be in the form
Mazharul Kazi Ph.D Thesis 121
1 1 ............t t p t p t tY Y Y Xψ ε− −=∏ + +∏ + + (5.17)
where Yt is a k vector of endogenous variables; Xt is a d vector of exogenous variables;
1∏ , , p∏ and ψ are matrices of coefficients to be estimated; and
tε is a vector of
innovations that may be contemporaneously correlated with each other but are
uncorrelated with their own lagged values and uncorrelated with all of the right hand
side variables (EViews).
Since only lagged values of the endogenous variables appear on the right hand side of
each equation, there is no issue of simultaneity and OLS may be the appropriate
technique of estimation. The disturbances are not serially correlated which means it is
unrestrictive since any serial correlation could be absorbed by adding more lagged Ys.
By employing any suitable econometrics software, one can estimate the parameters of
the VAR and then do either cointegration or Granger causality tests to see the long-run
relationship between variables of interests. One can test Granger causality by running a
VAR on the system of equations and testing for zero restrictions on the VAR
coefficients. Often, an alternative approach disregards testing for Granger causality.
The VAR model then generates an error correction mechanism which is accommodated
within the VECM. In this process, it is imperative that lagged length is identified
beforehand. Instead of a trial and error method, the most efficient way of getting the
information about the lagged value could be from either the Akaike Information
Mazharul Kazi Ph.D Thesis 122
Criteria or the Schwarz Bayesian Criteria. In this connection, the residual covariance
Ω is computed as
d e t /t t
t
Tε ε ′Ω = ∑ (5.18)
where tε is the k vector of residuals from k equations or parameters with T
observations . The log-likelihood value, l , is computed assuming a multivariate normal
(Gaussian) distribution as
( ) 1 lo g 2 lo g2 2
T k Tπ= − + − Ωl (5.19)
The two above mentioned information criteria about the length of the lag are as under:
2 / 2 /AIC T n T= − +l (5.20)
2 / log /SBC T n T T= − +l (5.21)
where n = k(d + pk) is the total number of estimated parameters in the VAR. Apart
from these popular criteria, there is another called the Hannan-Quinn Criterion. The
smaller the value of the information criteria, the better is the model.
Further, one might have to compute an impulse response function for each innovation
and endogenous variable pair. Essentially, these trace the effects of a shock to an
endogenous variable on the variables in the VAR. Accordingly, a four-variable VAR
Mazharul Kazi Ph.D Thesis 123
has 16 potential impulse response functions. Variance decomposition provides a
different method of depicting the system dynamics. Variance decomposition gives
information about the relative importance of each random innovation to the variables in
the VAR.
As a VECM is nothing but a restricted VAR that has cointegration restrictions built into
the specification, it is designed for use with nonstationary series that are known to be
cointegrated. The VECM specification restricts the long-term behaviour of the
endogenous variables to converge to their cointegrating relationships while allowing a
wide range of short-term dynamics. The cointegration term is known as the error
correction term since the deviation from long-term equilibrium is corrected gradually
through a series of partial short-term adjustments. We can consider a simple two-
variable system with one cointegrating equation and no lagged difference terms as:
2 , 1 ,t ty yβ= and (5.22)
the VECM as:
1, 1 2, 1 1, 1 1,( )t t t ty y yλ β ε− −∆ = − + and (5.23)
2, 2 2, 1 1, 1 2,( )t t t ty y yλ β ε− −∆ = − + (5.24)
Mazharul Kazi Ph.D Thesis 124
In this model the only right hand side variable 2, 1 1, 1( )t ty yβ− −− is the error correction
term. In long-run equilibrium, this term is zero. However, if y1 and y2 deviated from
long-run equilibrium last period, the error correction term is nonzero and each variable
adjusts to partially restore the equilibrium relation. The coefficients 1λ and
2λ measure
the speed of adjustment. The two endogenous variables y1,t and y2,t will have nonzero
means but the cointegrating equation will have a zero intercept. The VECM will take a
different form by including an intercept term when endogenous variables do not have a
trend. The cointegrating equation has an intercept as
1, 1 2, 1 1, 1 1,
( )t t t t
y y yλ µ β ε− −∆ = − − + and (5.25)
2, 1 2, 1 1, 1 2,( )t t t ty y yλ µ β ε− −∆ = − − + . (5.26)
Both a linear trend and an intercept can be accommodated in the VECM as
1, 1 1 2, 1 1, 1 1,( )t t t ty y yδ λ µ β ε− −∆ = + − − + and (5.27)
2, 2 1 2, 1 1, 1 2,( )t t t ty y yδ λ µ β ε− −∆ = + − − + . (5.28)
It is also possible to include lagged differences instead on the right hand side to meet
the specific research demand.
Mazharul Kazi Ph.D Thesis 125
Given a group of nonstationary series, one may determine whether the series are
cointegrated, and if they are, one can ascertain long-term equilibrium relationships.
Any VAR based cointegration tests can be conducted by using the methodology
developed by Johansen (1988, 1991, 1995a). Johansens method tests the restrictions
imposed by cointegration on the unrestricted VAR involving the series. To explain, we
consider a VAR of order p:1 1 ........t t p t p t ty A y A y Bx ε− −= + + + + , where yt is a k-vector
of nonstationarity; I(1) variables; xt is a d-vector of deterministic variables; and tε is a
vector of innovations. The reduced form of VAR takes the following shape:
1
1
1
p
t t i i t i t t
i
y y y B x ε−
− −=
∆ = + Γ ∆ + +∑∏ (5.29)
where 1
p
i
i
A I=
∏ = −∑ and
1
p
i j
j i
A= +
Γ = − ∑
Grangers representation theorem asserts that if the coefficient matrix Π has reduced
rank r k< , there exist k r× matrices α and β each with rank r such that αβ ′Π = and
tyβ ′ is stationary. Here, r is the number of cointegrating relations and each column of
β is the cointegrating vector. The element of α are known as the adjustment
parameters in the VECM. Johansens method is to estimate the Π matrix in an
unrestricted form, then test whether the restrictions implied by the reduced rank of Π
can be rejected.
Mazharul Kazi Ph.D Thesis 126
Shamsuddin and Holms (1997) applied the Johansen test from their equation
system1 1 1t tx A x ε−= + , where xt was the vector of parameters and εt was the vector of
errors. Subtracting xt-1 from both sides and denoting π = (A1 I), they obtained
expressions such as ∆xt = - T ttx ε+−1 .
As a common practice, the Johansen test involves testing the rank r of Π . The rank of
Π gives the number of independent cointegrating vectors and the following test
statistic is used to determine the rank:
0
0
1
( ) ln (1 )k
t ra ce j
j r
r Tλ λ= +
= − −∑ (5.30)
where ^
jλ =estimated eigenvalues of the ^
Π matrix and T = the effective number of
observations. This is the so-called trace test that checks whether the smaller k-r0
eigenvalues are significantly different from zero. The results on the Johansen trace test
for cointegration are the λ trace statistic. If the null hypothesis 0 0:H r r≤ cannot be
rejected against the alternative hypothesis1 0:H r r k< ≤ , this would mean that there
exists no cointegrating relationship between the variables in the system. Alternatively,
one can test 0 0:H r r≤ versus the more restrictive alternative
0: 1iH r r= + using
0max 0 1
( ) ln(1 )rr Tλ λ += − − (5.31)
Mazharul Kazi Ph.D Thesis 127
This is called the maximum eigenvalue test, as it is based on the estimated
0( 1)r th+ largest eigenvalue. The above two tests are often known as likelihood ratio
tests. They do not have the usual chi-squared distributions; rather, their appropriate
distributions are multivariate extensions of the Dickey-Fuller distributions. As with the
unit root tests, the percentiles of distributions depend on whether a constant and/or a
time trend are included [Verbeek (2000)].
Lutkepohl and Saikkonen (2000) investigated the cointegrating rank of a VAR process
with a time trend. They noticed that standard tests for the cointegrating rank of a VAR
have nonstandard limiting distributions which depend on the characteristics of intercept
terms and time trends in the system. Often these characteristics are unknown. As a
result, they modified the tests that allowed for deterministic linear trends in the data
generation process. Their modified tests were based on the lagrange multiplier principle
and, in contrast to likelihood ratio tests, accounted for the cointegrating rank specified
under the null hypothesis in estimating the trend parameters. Those tests had
nonstandard limiting disturbances which did not depend upon deterministic terms and
had better local power as well as small sample properties in many situations compared
to the competing likelihood ratio tests. Considering a time series of n-dimension,
yt=(y1t,,ynt), t= 1,,T, they generated their initial model:
0 1t ty t xµ µ= + + t = 1, 2, . (5.32)
Mazharul Kazi Ph.D Thesis 128
where 0 ( 1)nµ × and
1( 1)nµ × are unknown parameters which may, of course, be zero,
and xt is an unobservable error process. It is assumed that xt follows a pth-order VAR
process:
1
..........t t t p t p t
x A x A x ε− −= + + + t=1, 2, . (5.33)
where the Aj are (n×n) coefficient matrices. This process was written in error correction
form as:
1
1
1
p
t t j t j t
j
x x x ε−
− −=
∆ = Π + Γ ∆ +∑ t= 1, 2, .. (5.34)
where 1( ....... )n pI A AΠ = − − − − is (n×n), the
1( .... )j j pA A+Γ = − + + , (n×n), are
unknown parameters and ∆ is the usual differencing operator. The error term tε is white
noise, that is, tε ~ (0, )Ω with Ω positive definite.
Andersson and Gredenhoff (1999) examined the usual maximum likelihood procedure
under a fractional equilibrium error situation. Their lagrange multiplier test has the
ability of detecting cointegration relationships when deviations from equilibrium are
persistent.
Otero and Smith (2000) studied the effects of increasing the frequency of observation
and the data span of Johansen cointegration tests. They observed that the ability of the
Johansen tests to detect cointegration is more dependent on the total sample length than
Mazharul Kazi Ph.D Thesis 129
the number of observations. They utilized the procedures of both the Augmented
Dickey-Fuller and the Phillips-Perron tests in their study. They also did simulations by
using both Monte Carlo and relevant empirical applications. The overall results suggest
that when interested in long-run relationships, practitioners ought to rely on data
collected over a long period of time rather than on a large number of observations
collected over a relatively short period.
Pesaran and Shin (1995) acknowledged that economic analysis of long-run
relationships has been the focus of much theoretical and empirical research. They
followed the general practice, instead of cointegration, when the variables in the long-
run relationship were trend stationary. They attempted to de-trend the series and model
the de-trended series as a stationary distributed lag or autoregressive distributed lag
(ARDL) model. Estimation and inference concerning the long-run properties of the
model was then carried out using standard asymptotic normal theory in the context of
the ARDL model [Hendry and Sargan (1984), Wickens and Breusch (1988)]. Typically,
for a large sample, an estimator α is considered a consistent estimator of α when α is
asymptotically unbiased; and var( α ) decreases as the sample size increases [Lardaro
(1993)]. However, the analysis of Pesaran and Shin (1995) became more complicated
when the variables were difference stationary or integrated. They recognized that in the
presence of integrated or I(1) variables the traditional ARDL approach does not work.
Consequently, they choose an alternative estimation and hypothesis testing procedure
for the analysis of those I(1) variables in line with Engle and Granger (1987), Johansen
(1988, 1991), Phillips (1991), Phillips and Hansen (1990), and Phillips and Loretan
(1991).
Mazharul Kazi Ph.D Thesis 130
Hassapis, et al. (1999) showed that sufficient conditions for unit roots found in VAR
systems amount to Granger non causality in any direction among the variables
involved. This implies that a necessary condition for the disappearance of one unit root
in a VAR system implies Granger causality in at least one direction. The researchers
also noticed that causality and cointegration inference is strongly affected by the
omission of an important causal variable. For first order models with non explosive
variables, Granger causality is also sufficient for cointegration.
Lau (1999) attempted to complement empirical studies applying unit root and
cointegration methods and showed that integration and cointegration properties arise
intrinsically in stochastic endogenous growth models under fairly general conditions.
This implies that a unit root has to be present in the autoregressive polynomial of the
variables generated by an endogenous model so as to produce a steady state in the
absence of exogenous elements. The endogenous modelling induces difference
stationarity of the variables even though the external impulses are stationary, and it
leads to the phenomenon of cointegration if the variables satisfy a state-space
representation. The unit root propagation mechanism is the time series analogue of the
constant returns condition. The time series properties of endogenous models combined
with their counterparts for exogenous models lead to testable implications for
distinguishing between these two classes of models.
Johansen (1991) presented new results on maximum likelihood estimators and
likelihood ratio tests for cointegration in Gaussian VAR models with a constant term
Mazharul Kazi Ph.D Thesis 131
and seasonal dummies [Johansen (1991)]. He considered a general VAR model with
Gaussian errors written in error correction form:
1
1
k
t i t i t k t t
i
X X X D µ ε−
− −=
∆ = Γ ∆ + ∏ + Φ + +∑ (t = 1, , T) (5.35)
where Dt are seasonal dummies and µ is a constant term. Further,tε and (t = 1, ..,T)
are independent p-dimensional Gaussian variables with mean zero and variance matrix
Λ . The first k data points X1-k, ., X0 are considered fixed and the likelihood
function is calculated from given values of them. The parameters 1 1,......., , ,k µ−Γ Γ Φ
and Λ are assumed to vary without restrictions, and Johansen formulated the
hypotheses of interest as restrictions on∏ . This model of Johansen (1991) was denoted
by H1 and his hypothesis of at most r cointegration vectors H2 was formulated as:
2 :H α β ′∏ = (5.36)
where β referred to cointegrating vectors, and α, the adjustment coefficients, were p×r
matrices. When comparing models with different numbers of cointegration vectors,
Johansen used the notion H2(r). His objective was to conduct inference on the number
of cointegrating relations as well as their structure without imposing an a priori
structural relationship. This was accomplished by fitting the general VAR model:
1
1
k
t i t i t k t t
i
X X X D µ ε−
− −=
∆ = Γ ∆ + ∏ + Φ + +∑ (5.37)
Mazharul Kazi Ph.D Thesis 132
for describing the variation of data, and formulating questions concerning structural
economic relationship as hypotheses about the parameters of the VAR model. Johansen
(1991) tested these hypotheses using likelihood ratio statistics.
In 2000, Johansen presented an insight into the modelling of cointegration in the VAR
model and examined some results that were obtained from the cointegrating VAR. The
Granger representation theorem was evaluated and the notions of cointegration and
common trends were defined. Johansen defined the related statistical model for
cointegrated I(1) variables, demonstrated how hypotheses on the cointegrating relations
can be estimated under suitable identification conditions in the light of asymptotic
theory, and indicated some applications of the cointegration model in economic
research [Johansen (2000)].
Johansen (2000) also presented some definitions based on the Granger representation
theorem with essential proofs that are suitable for various conditions of modelling
cointegration in a VAR system, and also demonstrated how to express the stochastic
properties of the solution of the autoregressive equations under various assumptions
about the parameters. Johansen derived a reduced error correction model as:
1
1
k
t t i t i t t
i
X X X D ε− −=
∆ = + Γ ∆ + Φ +∑∏ (t = 1, ., T) (5.38)
Mazharul Kazi Ph.D Thesis 133
where Dt are deterministic dummies and the tε are i.i.d. (0, )nN Ω . The equations
determine the process Xt as a function of initial values X0, ., X-k, the ε and
dummies Dt. From the 5.38 equation Johansen then derived a matrix A(z) of n×n order
as:
1
( ) (1 ) (1 )k
i
i
i
A z z I z z z=
= − −∏ − Γ −∑ (5.39)
Allowing ( )A z to denote the determinant, and ( )adjA z the adjoint matrix:
( ( ) )( )
( )
t a d j A zA z
A z
− = (5.40)
The main assumption here is that the polynomial A(z) satisfies the condition ( ) 0A z =
implying either |z| > 1 or z = 1.
Further, based on the Granger representation, Johansen (2000) examined the model
describing an I(1) process having cointegration and αβ ′∏ = . The reduced form of
error correction model (Hr) thus takes the following shape:
t 1
1
Xk
t i t i t t
i
X X Dαβ ε− −=
′∆ = + Γ ∆ + Φ +∑ (t = 1, ..., T) (5.41)
Mazharul Kazi Ph.D Thesis 134
where α and β are n×r, and 1,......, Tε ε are independent Gaussian (0, )nN Ω , and the
variables Dt are deterministic terms. The parameters 1( , , ,......, , , )kα β Γ Γ Φ Ω are freely
varying.
Johansen then formulated a nested sequence of hypotheses 0 .... ....r nH H H⊂ ⊂ ⊂ ⊂ .
The test of Hr in Hn was that there are at most r cointegrating relations. Thus, H0 is just
a VAR model for Xt in differences and Hn the unrestricted autoregressive model for Xt
in levels, and the models in between H1, ,Hn-1 give the possibility to exploit the
information in the reduced rank matrix ∏, and contain information about the long-run
relationships in the economy [Johansen (2000)].
The Johansen approach [Johansen (1988, 1989, 1991, 1995, 2000); Johansen and
Juselius (1988, 1990)] to cointegration analysis seems attractive to many researchers,
although competing approaches like the Engle-Granger two-step method [Engle and
Granger (1987); Granger (1983, 1994)] and the ARDL method [Pesaram and Shin
(1995)] are also popular in practice.
Summary
This chapter investigated various empirical techniques that are suitable for detecting the
relationship between stock market returns or price movements and a priori variables. It
observed that various researchers investigated this relationship by employing different
methods. The analytical periphery depends on the availability of resources including
computer software and data at hand. The recent breakthrough in computing capacity
Mazharul Kazi Ph.D Thesis 135
and speed coupled with the development of statistical and econometric software has
enabled researchers investigating the security pricing process and other relevant issues
to adopt approaches, namely, cointegration analysis, which are more efficient than
traditional approaches such as ordinary regression analysis and/or factor component
analysis. We have observed that, depending on the research objectives and available
resources, different researchers have applied different techniques for cointegration
analysis there is no hard and fast rule as to which technique is to be employed for
investigating both long-run and short-run relationships among pre-identified variables.
In practice, at least three techniques for cointegration analysis are popular we intend
to use Johansen approach.
The empirical methods that are applied and practiced by various researchers suggest
that we can follow any or a combination of techniques for analyzing our research
problems and still achieve results of similar or even better standards. We have
contributed in this chapter by exploring the cointegration technique. In Chapter 6 we
will develop a model suitable to our research purpose.
Mazharul Kazi Ph.D Thesis 136
CHAPTER 6
MODELLING
As stated at the outset, the aim of this thesis is twofold. First, we intend to identify any
long-term relationships between systematic risk factors and Australian stock market
returns based on theories of economics and finance. In line with this research objective,
we have chosen a priori variables to proxy for Australian systematic risk factors. Based
on theoretical and empirical evidence as well as the methodological and analytical
strengths of available econometrics software, we have attempted to develop a model
which makes use of the most recent methodology. We have thus used a modern
empirical approach to ascertain whether a priori variables have influenced Australian
stock market prices movements as well as returns.
Secondly, we aim to study cross-country stock market relationships to determine
whether the Australian stock market is cointegrated with other developed markets
under prevailing global economic and technical settings. Our second analysis is
performed after we have conducted our first analysis the base study using the
method of cointegration. To meet both research objectives, we have developed our
models in terms of cointegration analysis.
Pool of a priori variables
Based on the literature reviews in Chapters 2, 3 and 4, there exists no hard and fast rule
in choosing a priori variables to proxy systematic risk factors for consideration in our
Mazharul Kazi Ph.D Thesis 137
first empirical objective. Often, the decision was made arbitrarily. Yet, it appeared that
the proxy variables considered were based on plausible economic, finance or
accounting theories. Further studies showed that researchers choice of the number of a
priori variables was based on the capabilities of the software that they employed
[Brooks (1997)]. Although rationales for choosing a priori variables may vary from
researcher to researcher, they have to be consistent with prevailing theories, familiar
intuition and research objectives.
Most of the studies in this field, including Roll and Ross (1980), Sinclair (1982), Chen,
et al. (1986), Hamao (1986), Faff (1988), and McGowan and Francis (1991), provided
mixed support for arbitrage pricing theory having three to five proxy variables or
factors. The variables that explained stock returns were changes in industrial
production, inflation, money supply, personal consumption, and interest rates. Fama
(1981) studied the relationship between common stock returns and real economic
variables, namely, capital expenditure, industrial production, gross national product,
money supply, lagged inflation, and the interest rate. Chen, Roll and Ross (1986) used
industrial production, money supply, inflation, the exchange rate, and long- and short-
term interest rates to investigate their effect on stock returns. Chen (1991) found that
the lagged production growth rate, the default premium, the term premium, the short-
run interest rate, and the market dividend-price ratio are the domestic macroeconomic
variables which have a significant direct and indirect impact on changes in stock prices.
Wongbangpo and Sharma (2002) used a model with five macroeconomic a priori
variables gross national product, the consumer price index, money supply, the
Mazharul Kazi Ph.D Thesis 138
nominal interest rate, and the exchange rate to investigate the relationship between
these variables and US stock price indices. Valentine (2000) argued that corporate
profits, interest rates, the exchange rate and overseas influences are relevant to stock
pricing. Maysami and Koh (2000) found six macroeconomic variables to be relevant,
namely, the exchange rate, short- and long-term interest rates, inflation, the money
supply, domestic exports, and industrial production. Paul and Mallik (2001) considered
quarterly observations of inflation, interest rates, and gross domestic product as their
proxy and they found both the interest rate and gross domestic product growth
significant in the Australian banking sector market pricing process.
It is apparent that capital markets around the globe are undergoing tremendous changes
with the impact of the information revolution and changes in household/consumer
tastes. To cope with these changes, market liberalization or financial deregulation often
provides a boost in market activity. Market liberalization also tends to make markets
efficient, where no under- or over-valued securities exist, based on the available
information set [Drew (2000)]. Likewise, available computable general equilibrium
(CGE) models (including ORANI and MONASH) often provide significant inputs to
the development process of an efficient stock market like Australia. Additionally,
global market indices viewed as benchmarks play an important role in the process of
asset pricing.
A list of a priori variables frequently used in various studies is presented in Table 6.1.
Mazharul Kazi Ph.D Thesis 139
TABLE 6.1
Record of A Priori Variables Used in Various Studies Popular a priori variables Referred studies
Industrial Production; Inflation; Interest rates; Treasury
bill rates; Long and short-terms Govt. bond rates;
Corporate bond rates; Equity returns; Per capita
consumption; Risk Premium; Term structure; Foreign
exchange rate; Market indices; Oil price; Overall
production; Gross Domestic Production; Gross national
product; Weekly wages; Money supply, Wholesale sales;
Labour force; Building construction; Federal debt;
Federal budget financing; Housing construction;
Exports and Imports; Population; Inventories; Book-to-
market equity value; Dividend yields; Spread of long-
and short-runs bond yield; High and low-grades bond
yield; Current liabilities; Consumer credit outstanding;
Commercial bank asset value; Money stock; Consumer
price index; Price-earnings ratio; Farm size; Corporate
profit; International or global index
Roll and Ross (1980); Chen, Roll and Ross (1986); Zhou (999); Kim and Lee (1995); Wongbangpo and Sharma (2002); Maysami and Koh (2000); Gehr (1978); Fogler, et al. (1981); Hughes (1985); Beenstock and Chan (1983, 1984); Winkelmann (1984); Dumontier (1986); Kim and Lee (!995); Connor and Korajezyk (1989); McGowan and Francis (1991); Brailsford and Heaney (1998); Connor (1995); Bai and Ng (2002); Fama and French (1992, 1993); Chamberlain and Rothschild
(1983); Anckonle (1983); Connor and Korajczyk (1986, 1988, 1989); McGowan and Francis (1991); Shukla and Trzcinka (1990); Faff (1988, 1992); Merville, et al. (2001); Beggs (1986); Merton (1973); Kritzman (1995); Fam and French (1992, 1993); Brown and Weinstein (1983); Aitken, et al. (1996); Brailsford and Easton (1991); Faff and Brailsford (1999); Hamao (1988); Faff and Heaney (1999); Easton and Faff (1994); Zhou
(1999); Velu and Zhou (1999); Brailsford and Faff (1996); Maysomi and kho (2000); Johansen (1988, 1991, 1995, 2000); Pesaran and Timmerman (1995); McMillan (2001); Mukharjee and Naka (1995); Cheung and Ng (1998); Ackert and Racine (1999); Nasseh and Stranss (2000); Mukharjee and Hoh (2000); Shamsuddin and Holmes (1997); Cheung and Ng (1998); Paul and Mallik (2001); Roca (1998); Shamsuddin and Kim (2000,
2003); Valentine (2000).
We have so far learned that there is a wide range of variables that we can consider as
proxies for systematic risk factors in Australian stock market pricing. To derive the
series of a priori variables, we often need to do some form of transformation or
manipulation of available basic data series. It appears that some types of economic,
financial, and accounting indicators represent the elements of systematic risk that
influence the return generating series of any stock market. Initially, we have decided to
consider a priori variables that have both theoretical and practical implications for our
empirical investigation and analyses. We also understood that transformations might
provide inefficient outcomes when investigating long-run relationships. The
recommendation in the literature is that base level data do not generally provide
information when they are nonstationary. However, a handful of a priori variables were
Mazharul Kazi Ph.D Thesis 140
considered at the beginning based on theoretical consistency that are later reduced to a
workable number based on the data and the analytical capabilities of available
econometric software.
Variables for Modelling
Stock indices represent the returns on the market as a whole. Most stock indices are
price weighted and use market capitalization based weights. These indices act as
market indicators and set up a performance benchmark in the stock market. They thus
provide a sensible and realistic proxy for the security market pricing profile.
Accordingly, we have considered the time series of index values in our analysis
because they represent the Australian stock market performance.
Nearly 100 different Australian Stock Exchange (ASX) share price indices are
currently produced. Some cover shares in specific industries, namely, metal, oil and
gas, banking, insurance, retail, media, and so on. Others cover combinations of these,
namely, all resources, all industrials, all ordinaries, and so on. Further indices cover
combinations of various leaders in the market, namely, S&P/ASX 20, S&P/ASX 50,
S&P/ASX 200, S&P/ASX 300, S&P/ASX MidCap, S&P/ASX SmallCap, and so on. In
each case the value of the stock price index represents the current market capitalization
of all the shares concerned, expressed as a number relative to a base number. Australian
stock indices thus give a quick picture of weighted average price movements in the
market either as a whole or for a sector of the Australian share market [Renton (1998)].
The indices for the Australian stock market are constructed by Standard and Poors,
Mazharul Kazi Ph.D Thesis 141
implying that they are of world standard and accurately represent Australian stock
market performance. Although both the S&P/ASX 200 and the All Ordinaries indices
are comprehensive, we have used the All Ordinaries (ALLORDS). It is a standard
proxy of Australian stock market prices movements/returns over the period lasting from
the 1st quarter of 1983 to the 2
nd quarter of 2002.
As for our a priori variables, we initially selected 15 variables based on both theoretical
and intuitive considerations. For further analysis and modelling we have reduced the
number of a priori variables to six. These ultimate a priori variables for our first
empirical analysis, presented in Chapter 7, are the bank variable interest rate, industrial
production, the exchange rate, the dividend yield, the price-earnings ratio, and the
global stock market index.
For our supplementary objective, that of cross-country stock market cointegration
analysis, we have considered the comprehensive stock market indices of the US, the
UK, Germany, Canada, France, Japan, and Australia. This study concentrated on the
annual data series of all seven stock markets from 1945 to 2002. We have used the
DJIA, FTA, DAX, SBF250, S&P/TSX300 and NIKKEI to represent the stock market
performances of the US, the UK, Germany, France, Canada and Japan respectively for
our second empirical study, presented in Chapter 8.
Mazharul Kazi Ph.D Thesis 142
Modelling
Variables for our first study have drawn upon both prevailing theories and intuition.
Our intention here is to ascertain whether the a priori variables (Xit) have any long-run
effect in explaining the dependent variable (Yt), and if they do, which ones are
influential. To undertake an empirical investigation on the first research question, we
have developed a basic model, which is then extended by necessary transformations to
suit our specific modelling needs.
We start with a simple structural relationship of our variables Yt and Xit may be
presented in either this manner:
1( , .. . ., )t t k tY f X X= (6.01)
or
1
k
t i i t
i
Y X=
= Γ∑ (6.02)
where Yt is the criterion variable representing the security market prices generated from
the quarterly series of the ALLORDS, Xit =X1t, ,Xkt refer to our a priori variables, and
t = study period (that is, 1/1983 to 2/2002 for our first analysis, and 1914 to 2002 for
our second).
Including the error term,tε , in this relationship we have:
Mazharul Kazi Ph.D Thesis 143
( , )t i t t
Y f X ε= (6.03)
The above relationship could also be expressed as:
1
i
k
t it t
i
Y Xδγ ε
=
= ∏ (6.04)
where tε refers to the disturbance or error term; γ is a constant or intercept; and
iδ is
the parameter. Also, here the relationship between predictors or explanatory variables
and between the predictors and the error is assumed to be multiplicative.
Transforming 6.04 into a logarithmic function, the relationship can now be specified in
linear form as:
0
1
k
t i it t
i
y x uµ δ=
= + +∑ (6.05)
where yt = lnYt; µ0 = lnγ; xit = lnXit; δi is the parameter; ut = lnεt; and that ln refers to
natural log.
Further, we can write this equation in another form as:
y X uδ= + (6.06)
Mazharul Kazi Ph.D Thesis 144
where y is of dimension T×1; X is of dimension T×k; δ is of dimension k×1; and u is
of dimension T×1. The represented model is:
t t ty x uδ= + , t = 1, 2, ,T (6.07)
To construct a structural linear regression model, we consider the k regressors or
independent variables, xi, where i=1, ,k and t =1,,T. Through transformation, our
generalized linear multiple regression (GLM) model is:
1 2 2 3 3t t t k kt ty x x x uδ δ δ δ= + + + + +LL (6.08)
The model in matrix form is
11 121 1
2 22 22 2
21 11
1
1
1
k
k
kT TT kTT TkT k
y ux x
xy x u
y ux x
δδ
δ× ×××
= +
L
L
M MM M MM (6.09)
so that the xij element of the matrix X represents the jth
time observation on the ith
variable.
Mazharul Kazi Ph.D Thesis 145
When using a structural model like the GLM, it would be necessary to examine if the
model specifications and variables included in the model are correct. To achieve this
effect a Ramsey Regression Specification Error test is useful. Thus, our generalised
linear model (GLM) or basic structural regression model is
0
1
( | )k
t it i it t
i
E Y X X uµ η=
= + +∑ (6.10)
where ut is assumed to satisfy all the usual assumptions about the regression errors.
In multivariate time series analysis, where the stochastic process is the moving average
process and lags of variables are important to consider in the modelling, the above
basic structural type of model becomes inappropriate. A special type of model is
necessary, one which is valid with autocorrelation and/or other unavoidable
complexities. Although the typical structural approach to simultaneous equation
modelling is often used in describing the basic relationship between endogenous
(dependent) and exogenous (independent) variables of interest in many fields of study,
an autoregressive approach provides outcomes that are more accurate and consistent
with the theory underlying economics and finance. This is because the theory is not
always capable of providing a structured specification to express the dynamic
relationship between variables and/or it becomes complicated and confusing to make
inferences when it is difficult to specify whether endogenous variables appear only on
the left hand side or on both sides of an equation. Additionally, the structural equation
framework is often inadequate in handling problems arising from nonstationarity of the
time series variables, resulting in spurious regressions [Verbeck (2000)].
Mazharul Kazi Ph.D Thesis 146
An autoregressive approach in essence sidesteps the problems associated with structural
equations and endogeneity of variables by modelling all endogenous variables in the
system. By adding additional lagged variables of the dependent variable, this estimation
takes care of serial correlation between the disturbances [Self and Grabowski (2002)].
In order to find long-run relationships between the endogenous and exogenous
variables in our model, we need to accommodate any changes in a variable by reference
to the movements in the current or past values of the variables. A vector autoregressive
(VAR) model incorporates past values of variables into the process. It is a system
which accommodates more than one dependent variable as an alternative to a large-
scale simultaneous equation structural model [Brooks (2002)]. A VAR is a flexible and
generalised model that is extendable to the case where the model needs to include first
difference terms for ascertaining any cointegrating relationships among its variables.
Based therefore on the empirical needs of our twofold research objectives, we have
transformed our basic structural model (GLM) into a non structural autoregressive
VAR model as:
0
1
k
t i t i t
i
y y uµ η −=
= + +∑ (6.11)
where yt is 1p× vector of variables that contains all variables (including Xs) as
endogenous and is determined by k lags of all p variables in the system; ut is
1p× vector of error terms; 0µ is 1p× vector of constant term coefficients; and
iη are
Mazharul Kazi Ph.D Thesis 147
p p× matrices of coefficients on the ith
lag of y and t = 1, ., T. A VAR model can
be constructed with first differenced variables if these first differenced variables (∆yt)
are found to be nonstationary (that is, I(1)) and/or the underlying theory permits us to
do so. In such a special case, the VAR would look like
0
1
k
t i t i t
i
y y uµ η −=
∆ = + ∆ +∑ (6.12)
To detect long-run equilibrium relationships as well as short-term dynamic adjustments
among the variables, an error correction term (ECT) is very useful because the
cointegration term (CT) is closely associated with the error correction term or error
correction mechanism (ECT or ECM). Sargan (1964) first applied the concept of ECM
and Engle and Granger (1987) later corrected it for disequilibrium. A vector error
correction model (VECM) is simply a restricted VAR which requires accommodating
the error correction term (ECT/ECM). By identifying the lagged number from either
the Akaike Information Criteria or Schwarz Bayesian Criteria and other suitable
criterion, we have the VECM as
1
0 1
1
p
t t i i t i t
i
y y yµ ε−
− −=
∆ = + + Γ ∆ +∑∏ (6.13)
or, following Johansen, the reduced form of VECM without a constant as:
1
1
1
p
t t i i t i t
i
y y y ε−
− −=
∆ = + Γ ∆ +∑∏ (6.14)
Mazharul Kazi Ph.D Thesis 148
where tε ~ (0, ), 1,...,N t TΩ = ;
1
p
i
i
A I=
∏ = −∑ ; and 1
p
i j
j i
A= +
Γ = − ∑
We note that, at equilibrium, the reduced form of Johansens VECM would be viewed
as 1t ty y −∆ = ∏ . The equation
1
1
p
i i t i
i
y−
−=
Γ ∆∑ would not exist in equilibrium, all t iy −∆
will be zero and setting the error term tε to their expected value of zero will leave
1 0ty − =∏ [Brooks (2002)]. Within the VECM system, П contains both dynamic
adjustment parameters (α ) and cointegrating relationships (β ′ ).
The VECM distinguishes between the short- and long-run forms of causality, which the
standard Granger causality test cannot accomplish. While the joint F-tests on the
coefficients of the differenced terms in the VECM are used to test short-run causality, t-
tests on the coefficients of the lagged ECTs are utilized to determine the long-run
causal relationships since these are derived from the cointegrating relations between the
relevant variables [Self and Grabowski (2002)].
The modified or restricted VAR is the VECM which accommodates the ECM or ECT
often presented in the form that changes in the dependent variable are a function of
both the explanatory variables and the ECT. This captures the level of disequilibrium in
the relationship. This form of VECM (with optional constant µ0) is often written as:
1
0 1
1
p
t i i t i t t
i
y y ECTµ δ ε−
− −=
∆ = + Γ ∆ + +∑ (6.15)
Mazharul Kazi Ph.D Thesis 149
where ECTt-1 contains the error correction terms derived from the cointegrating vectors
and δ records the response of the dependent variable in each period t from equilibrium.
Granger (1988) also recognized this notion by pointing that there are two channels of
causality when using vector error correction (VEC); one is through the lagged values of
the variables and the other is through δ, if δ is significant [Self and Grabowski (2002)].
However, after the cointegrating vector has been normalized, the initial relationship for
yt can be written as
1 1t t i t iy y yη η− −= + +L (6.16)
whilst the equation for ECTt is:
1 1 1t t t i t iECT y y yη η− − −= − − −L (6.17)
Summary
This chapter has developed the models that will be adapted to our specific data series
for empirical analyses in Part IV. As our aim of this research is twofold, we need to be
able to use our models for both purposes. Our models consist of the basic model
(GLM), the VAR and the VECM. The GLM contained both endogenous (yt) and
exogenous (xit) variables, whereas the VAR/VECM model bypassed the structural
GLM model by considering all variables as endogenous yt series. Thus, both yt and xit
series of GLM turned into one endogenous series of yt in VAR/VECM. Our models are:
Mazharul Kazi Ph.D Thesis 150
Structural Model
0
1
( | )k
t i t i i t t
i
E Y X X uµ η=
= + +∑
Non Structural/Autoregressive Models
VAR
0
1
k
t i t i t
i
y y uµ β −=
= + +∑
or
0
1
k
t i t i t
i
y y uµ β −=
∆ = + ∆ +∑
VEC
1
0 1
1
p
t i i t i t t
i
y y E C Tµ δ ε−
− −=
∆ = + Γ ∆ + +∑ ;
or
1
1
1
p
t t i i t i t
i
y y y ε−
− −=
∆ = + Γ ∆ +∑∏
Depending on the nonstationarity condition of our variables, model estimation
criterion, econometrics software packages and research objectives, we will use one or a
combination of suitable model/s from the above pool of models for our empirical
analyses in Chapters 7 and 8.
Mazharul Kazi Ph.D Thesis 151
PART IV
METHODOLOGY
As the objective of this thesis is twofold, it is necessary to perform two separate
analyses. The first analysis examines which a priori variables are cointegrated with
Australian stock market returns, while the second analysis tests both long-run cross-
country relationships and short-term dynamics. Part IV thus contains two chapters.
Chapter 7, to address the first objective, carries out the necessary empirical analysis and
reports the outcomes. Chapter 8, to fulfill the second objective, analyses relevant data
and details the results.
Mazharul Kazi Ph.D Thesis 152
CHAPTER 7
(EMPIRICAL ANALYSIS 1)
ANALYSIS OF SYSTEMATIC RISKS FOR THE
AUSTRALIAN STOCK MARKET
In the literature on investment risk, the total risk is divided into two components,
namely, systematic risk and unsystematic risk. This classification is conceptualised by
the standard deviation of an investment return from the point of view of diversification.
The diversifiable risk is the unsystematic risk, while non diversifiable risk is systematic
risk. The systematic risk principal states that the reward for bearing risk depends only
on the systematic risk of an investment and thus the expected return on a risky asset
depends only on that assets systematic risk. Accordingly, systematic risk is also
market risk.
From the literature review we have seen that asset pricing theories do not specify the
underlying economic forces or systematic risk factors that drive securities prices. The
empirical analysis uses particular software and depends on both the availability of data
and established statistical criteria that are frequently used in the selection of variables.
Accordingly, the rationale for the selection of variables is essentially based on intuitive
financial theory [Chen et. al. (1986), Mukharjee and Naka (1995), McMillan (2001)].
Although the preliminary a priori variables for this study were pre-identified based on
theory as well as intuition and, accordingly, necessary data had been collected, some
prior tests needed to be carried out to select the appropriate variables for cointegration
analysis.
Mazharul Kazi Ph.D Thesis 153
Data and Variables
The preliminary variables identified for the basic model are a priori variables that have
been derived from both theory and practice. Theories in economics and finance suggest
that macroeconomic variables such as growth in gross domestic product, changes in
industrial production, the interest rate, the unemployment rate, the exchange rate, and
net exports are likely to be relevant in predicting stock market performance. Likewise,
principles of accounting as well as financial theories point to the importance of
corporate profits, the price-earnings (P/E) ratio, the dividend yield, and global equity
indices for the local stock market as significant determinants of stock market
performance. To address the first objective of this research, 15 initial a priori variables
were considered. The data were assembled from various sources. The data on gross
domestic product (GDP), per-capita GDP (GDPPC), the industrial production index
(IPI), the manufacturing commodity price index (MPI), the unemployment rate (UR),
imports and exports to derive net exports (NX), and the consumer price index (CPI)
were collected from the Australian Bureau of Statistics (ABS). The data on the M3
money supply (MS), the standard variable bank interest rate (BVIR), the 11AM cash
rate (IR11AM) and net exports (NX) were collected from both the ABS and the
Reserve Bank of Australia. The net export (NX) figures were derived from the imports
and exports of all goods and services of Australia during the period under study
as ( )NX EXPORTS IMPORTS= − . Company or corporate profits (CP), the price
earnings or P/E ratio (PER), dividend yields (DY), and the Australian to US dollar
exchange rate (ER) were obtained from the Reserve Bank of Australia. Data on the
Morgan Stanley capital international world index (MSCI) was used as a proxy for
Mazharul Kazi Ph.D Thesis 154
global equity market influences and was obtained from the Morgan Stanley worldwide
website.
The All Ordinaries index (ALLORDS) was used to obtain a measure of the market
price movements of Australian securities since this index is comprehensive and
available for our both analyses. Share prices were collected from the Securities Industry
Research Centre of the Asia-Pacific and ABS sources. Depending on the availability of
data, we constructed quarterly time series for the period from the first quarter of 1983
to the second quarter of 2002 (1/1983to 2/2002). Most of the data series were balanced
except for corporate or company profits where there were a few missing observations.
Given the unavailability of a longer period and given a lower frequency of data, the
quarterly data seemed satisfactory for the intended empirical analyses.
Because the selection of variables for inclusion in a model is an important step in the
analysis, to ensure that the model neither includes any irrelevant variable nor omits any
important variable and that the parameters are stable, several statistical tests were
performed. The first was to use the correlation values of a priori variables and stock
market returns from Annexure 1. Other diagnostic tests such as multi-collinearity were
also taken into account (Annexure 1). However, some of the variables were eliminated
as they were thought to be irrelevant. Accordingly, our initial 15 a priori variables were
reduced to six for the final analysis, based on its statistical correlations with the
Australian security market price movement, possible multi-collinearity among the
variables, and both theoretical and intuitive weights. These six final a priori variables
are industrial production, the bank variable interest rate, corporate profits, the dividend
Mazharul Kazi Ph.D Thesis 155
yield, the price earnings ratio, and MSCI. We used logged values of all series of
variables.
To validate the above variable selection decision, a data reduction procedure was
followed using the SPSS 11 statistical software to develop an idea about the number of
components or factors based on eigenvalue. This procedure extracted one component or
factor based on eigenvalues of more than 0.1 from the initial 16 variables, including the
ALLORDS. Graph 7.1 plots the extracted components/factors, while Table 7.1 reports
the component matrix.
GRAPH 7.1
Component/Factor Plot
Scree Plot
Component Number
16151413121110987654321
Eig
en
va
lue
500000
400000
300000
200000
100000
0
Mazharul Kazi Ph.D Thesis 156
TABLE 7.1
Component Matrix Raw Rescaled
Component Component
1 1
LNALLORDS 622.675 1.000 LNMSCI .282 .952
LNGDPPC .084 .919 LNGDP .126 .893
LNIPI .085 .839 LNER -.106 -.793 LNCP .302 .769
LNMS .290 .757 LNMPI .080 .681 LNCPI .091 .677
LNBVIR -.254 -.609 LNIR11AM -.244 -.592
LNDY -.135 -.545 LNUR -.083 -.467
LNPER .079 .199
LNNX .253 .158 Extraction Method: Principal Component Analysis.
a 1 components extracted.
The descriptive statistics of the seven variables (which include the six a priori and the
ALLORDS return) are reported in Table 7.2. Table 7.2 shows no major discrepancies,
meaning that the selected seven variables are consistent with conventional research norms.
TABLE 7.2
Descriptive Statistics LNALLORDS LNBVIR LNCP LNDY LNIPI LNMSCI LNPER
Mean 7.570596 2.097070 8.472301 1.373518 4.570115 5.974402 0.012789 Median 7.539122 1.991859 8.504892 1.340665 4.551235 5.997631 0.011196 Maximum 8.101006 2.904633 9.234838 1.903102 4.774069 6.548168 0.337367
Minimum 6.894670 1.448980 7.473069 0.946497 4.326778 5.174042 -0.521375 Std. Dev. 0.315845 0.487376 0.430099 0.199241 0.123883 0.344317 0.130132 Skewness 0.063872 0.394773 -0.284654 0.758343 -0.171726 -0.234526 -0.772756 Kurtosis 2.080722 1.586991 2.358626 3.123477 2.202477 2.329086 6.608913
Jarque-Bera 2.368826 7.204940 2.022549 6.367846 2.073507 1.842875 42.38536 Probability 0.305926 0.027256 0.363755 0.041423 0.354604 0.397947 0.000000
Sum 499.6593 138.4066 559.1719 90.65220 301.6276 394.3106 0.844104 Sum Sq. Dev. 6.484260 15.43977 12.02406 2.580292 0.997554 7.706014 1.100737
Observations 66 66 66 66 66 66 66
Mazharul Kazi Ph.D Thesis 157
Hypothesized Relationships of Variables
We aim to investigate the interrelationship between Australian security market returns,
the set of a priori variables that represent the money market, the goods market, the
foreign exchange market and the global stock market performance. Here we have
considered industrial production (IPI) to represent the goods market. The money market
is represented by the bank variable interest rate (BVIR) which is also linked to the
exchange rate (ER) representing the foreign exchange market. The security market is
represented by the stock price index (ALLORDS), which is also linked to the dividend
yield (DY) and the price earnings ratio (PER). The global stock market is represented
by the performance of the global index MSCI which might impact under globalization
on individual markets. In line with previous studies (referred to in earlier chapters), we
expect that our a priori variables will show a linkage with stock market performance.
This study will thus investigate both the long-run and short-run relationship between
the Australian stock market price movements or returns ( ALLORDS∆ ) and the changes
in selected six a priori variables (BVIR, IPI, CP, DY, MSCI and PER). Usually, the
simple model is specified in the following form:
( , , , , , )t t t t t t t
ALLORDS f BVIR CP DY IPI MSCI PER∆ = ∆ ∆ ∆ ∆ ∆ ∆ (7.01)
The relationship between interest rates and stock prices from the perspective of asset
portfolio allocation is commonly negative. An increase in interest rates raises the
required rate of return, which in turn inversely affects the value of the asset. Measured
as opportunity cost, the nominal interest rate will affect investors decision on stock
Mazharul Kazi Ph.D Thesis 158
holdings. A rise in the opportunity cost will, however, motivate investors to substitute
shares for other assets in their portfolios. Also, an increase in interest rates may cause a
recession and thus cause a decline in future corporate profitability, as corporate
financial costs may rise due to increased interest payments. Further, higher interest
rates have a discouraging effect on mergers, acquisitions and buyouts. Interest rates
might have a positive relationship with stock returns, as an increase in the rate of
interest raises the opportunity cost of holding cash and is likely to lead to a substitution
effect between stocks and other interest bearing securities/accounts. Changes in interest
rates are expected to affect the discount rate in the same direction through their effect
on the nominal risk-free rate [Mukharjee and Naka (1995)]. However, nominal interest
rates are assumed to contain information about future economic conditions and to
capture the state of investment opportunities in stocks; in addition, studies have found
that short-term interest rates have a significant negative influence on the stock market.
Thus, we hypothesize a negative relationship between interest rates (BVIR) and stock
prices (ALLORDS).
We hypothesize a positive relationship between ALLORDS and industrial production
(IPI), as IPI typically has a positive relationship with the stock market. An increase in
production is likely to influence stock prices through its positive impact on gross
domestic product and corporate profitability. An increase in output may increase
expected future cash flows and thereby raise stock prices, while the opposite effect
would occur in a recession.
Mazharul Kazi Ph.D Thesis 159
We hypothesize a negative relationship between the exchange rate and the stock price,
as changes in the exchange rate affect the performance of the stock market through its
impact on net exports and thus the current account position of an export focused
country like Australia. A number of commentators have suggested that currency
depreciation will have a favourable impact on a domestic stock market [Ma and Kao
(1990), Mukharjee and Naka (1995)]. Thus, an Australian dollar depreciation against
the US dollar implies that products exported from Australia become cheaper in the
world market. If the demand for exportable items is elastic, the volume of exports from
Australia would increase, which in turn causes higher cash flows, corporate profits and
stock prices of Australian companies. However, the contrary relationship may be found
under different circumstances.
Movements in the dividend yield are thought to be related to long-run business
conditions and they represent a predictable component of stock market returns.
We hypothesize that the dividend yield has a positive relationship with stock prices.
However, in the short-run, the price would drop immediately after the dividend payout
for a specific stock due to speculation about the lack of an immediate profit-taking
opportunity and/or a longer holding period to receive another dividend payout.
We hypothesize a positive relationship between corporate profits and market stock
price, because it captures predictable elements in future return series. This often relates
to the price-earnings ratio which boosts the confidence of investors by encouraging
them to invest in the stock market.
Mazharul Kazi Ph.D Thesis 160
Finally, we consider that the global stock market price index (MSCI) has some impact
on a developed stock market like Australia. Changes in the MSCI could pose either a
direct or indirect impact on the local stock market depending on the trading relationship
with other markets. We hypothesize a direct and positive relationship between
ALLORDS and MSCI.
The hypothesized relationship may hold or be rejected depending on the methodology
used in the analysis. In this connection, we note that, amongst others, Verbeck (2000)
pointed to the fact that to apply standard estimation or testing procedures in a dynamic
model, it is typically required that the variables are stationary since regressing a
nonstationary I(1) endogenous variable like LNALLORDS on nonstationary I(1)
exogenous variables such as LNIPI, LNBVIR, LNCP, LNDY, LNPER, and LNMSCI
may lead to so called spurious regression when we use the standard regression model,
where estimators and test statistics are misleading. An exception to this rule occurs
when two or more I(1) variables are cointegrated, meaning that there is a linear
combination of these nonstationary I(1) variables which is stationary I(0). In such cases
a long-run relationship between these variables exists and this also has implications for
the short-run behaviour of the I(1) variables. To capture this relationship for both the
long- and short-runs, there has to be some mechanism that drives the variables and their
short-run unstable behaviors to their long-run equilibrium relationship. This mechanism
is modelled by an error correction mechanism, in which the equilibrium error also
drives the short-run dynamics of the series. Accordingly, we intend to employ the error
correction model discussed in Chapter 6 for the empirical examination.
Mazharul Kazi Ph.D Thesis 161
Unit Root and Breakpoint Tests
For cointegration analysis, it is important to check whether all the time series variables
are nonstationary with unit root I(1) and stationary with unit root I(0) at difference
before using them in cointegration analyses. This is because the standard inference
procedures that are relevant to the standard regression model do not apply in
cointegration analysis it bypasses the standard regression modelling structure.
Cointegration analysis requires us to use only those variables that are nonstationary
with unit root I(1). We have tested the stationarity of all these series using EViews
econometric software. We tested for unit roots both in levels and first differences for all
three possible states of the model in relation to intercept and trend. The tested models
were with intercept but no trend, intercept with trend, and no intercept or trend.
We used both the Augmented Dickey-Fuller and the Phillips-Perron test procedures in
EViews3.1. Results of the unit root tests are presented in Table 7.3.
TABLE 7.3
Unit Root Test Results
Augmented Dickey-Fuller (ADF) Phillips-Perron (PP) Variables Intercept,
No trendA Intercept
with trendB No intercept,
No trendC Intercept, No trendA
Intercept with trendB
No intercept, No trendC
At level LNALLORDS -0.7693 -2.5204 1.6447 -0.9158 -3.1529 1.8879 LNBVIR -1.0780 -2.8031 -0.7916 -1.1332 -2.2060 -1.1586 LNMSCI -2.0051 -2.5200 2.4765 -2.7021 -3.4094 2.6647 LNIPI -1.5737 -3.0819 3.2505 -1.9306 -3.1799 4.5530 LNER -1.5525 -2.1974 0.3993 -1.6695 -2.2856 0.4539 LNDY -2.0014 -2.6032 -0.5466 -2.4353 -2.7896 0.8648
LNPER -1.8346 -2.9635 0.2156 -1.8475 -2.5987 0.4182 At 1st diff. ∆LNALLORDS -6.7377 -6.6898 -5.9812 -11.9180 -11.8279 -11.0715 ∆LNBVIR -4.5835 -4.6497 -4.6254 -6.0566 -6.0173 -6.0533 ∆LNMSCI -6.0321 -6.1883 -5.2225 -9.1863 -9.3631 -8.4619 ∆LNIPI -4.6453 -4.7518 -3.3183 -9.1809 -9.2496 -7.5641 ∆LNER -3.9624 -3.9067 -3.8699 -7.8623 -7.8127 -7.8161 ∆LNDY -4.9831 -4.9657 -5.0108 -8.3097 -8.2527 -8.3542 ∆LNPER -4.4205 -4.3887 -4.4167 -6.5535 -6.5088 -6.5723
Note: MacKinnon Critical values at level: for model A. -2.9851; model B. -3.469; model C. -1.9439, and at 1st difference: for model A. -2.8955; model B. -3.4626; model C. -1.9445
Mazharul Kazi Ph.D Thesis 162
The test results are compared against the MacKinnon (1991) critical values for the
rejection of the null hypothesis of no unit root. Table 7.3 shows that all seven a priori
variables are integrated of order one I(1) in levels and of order zero I(0) in first
differences, meaning that they are nonstationary in levels and stationary in first
differences.
We plotted the residuals of the Australian stock market data in Graph 7.2 to identify
whether there was any break in the series owing to the stock market crash of October
1987. We also graphed the residuals of the a priori variables series in Graph 7.3. Both
graphs indicate that all series contain an outlier in the fourth quarter of 1987 which
suggests a break point dummy variable for the analysis to capture the impact of the 12
October 1987 stock market crash.
GRAPH 7.2
Plot of Residuals and Two StandardError Bands
Quarters
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
0.0
0.1
0.2
0.3
1985Q4 1988Q2 1990Q4 1993Q2 1995Q4 1998Q2 2000Q4 2001Q4
Mazharul Kazi Ph.D Thesis 163
GRAPH 7.3
We also conducted a statistical test called the Chow Breakpoint test using EViews4 to
ascertain if the null hypothesis of no significant break in 1987.4 data series can be
rejected. The Chow Breakpoint test produced an F-statistic of 6.547176 (probability
0.000015) and a log likelihood ratio statistic of 41.71134 (probability 0.000001), as
reported in Table 7.4.
TABLE 7.4
Chow Breakpoint Test Chow Breakpoint: 1987:4
F-statistic 6.547176 Probability 0.000015
Log likelihood ratio 41.71134 Probability 0.000001
Note: Critical values of F-statistic (1df) are 2.71, 3.84 and 6.63 at 10%, 5% and 1% significance levels respectively.
The Chow Breakpoint test rejected the null hypothesis of no-effect of the October 1987
(1987.4) stock market crash in the movement of Australian stock prices against the
alternative hypothesis. Thus the breakpoint test implies that the October 1997 stock
market crash is significant for consideration in our analysis. We have included a break-
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
84 86 88 90 92 94 96 98 00 02
RESIDLNIPIRESIDLNMSCIRESIDLNBVIR
RESIDLNERRESIDLNDYRESIDLNPER
RESIDUL PLOT OF A PRIORI VARIABLES
Mazharul Kazi Ph.D Thesis 164
point dummy (which takes the value one for the 4th quarter of 1987 and zero elsewhere)
as an exogenous variable in the model for the investigation of the cointegrating
relationship between the Australian stock market returns and our a priori variables.
Lag Length Selection
As the autoregressive model is sensitive to the lag operator chosen, we have to
ascertain the appropriate lag length before we conduct the cointegration analysis in line
with Johansen. We have used both Microfit 4.1 and EViews4 to determine the optimal
lag length based on the Akaike Information Criteria (AIC) or Schwarz Bayesian
Criteria (SBC) criteria.
Both the likelihood ratio (LR) and the adjusted LR statistics for selecting the optimal
VAR order by using Microfit 4.1 suggested that optimal lag length should be one.
The same conclusion was suggested by the SBC tests through employing EViews 4.
The results are provided in Tables 7.5A and 7.5B.
TABLE 7.5A Test Statistics and Choice Criteria for Selecting the Order of the VAR Model --------------------------------------------------------------------------------------------------------------------------------------------
List of variables included in the unrestricted VAR: LNALLORDS LNBVIR LNCP LNDY LNIPI LNMSCI LNPER =================================================================================== Order LL AIC SBC LR test Adjusted LR test 5 922.6844 677.6844 419.1023 ------ ------
4 853.5152 657.5152 450.6496 2χ ( 49)= 138.3383[.000] 58.9639[.156]
3 766.6831 619.6831 464.5339 2χ ( 98)= 312.0025[.000] 132.9847[.011]
2 711.1570 613.1570 509.7242 2χ (147)= 423.0547[.000] 180.3184[.032]
1* 632.4021 583.4021 531.6857 2χ (196)= 580.5645[.000] 247.4537[.007]
0 -22.6793 -22.6793 -22.6793 2χ (245)= 1890.7[.000] 805.8838[.000]
Test Statistics and Choice Criteria for Selecting the Order of the VAR Model using Microfit4. Probability in [ ].
Mazharul Kazi Ph.D Thesis 165
TABLE 7.5B VAR Lag Order Selection Criteria Endogenous variables: LNALLORDS LNBVIR LNCP LNDY LNIPI LNMSCI LNPER Exogenous variables: C Sample: 1983:1 2002:2 Included observations: 61
Lag LogL LR FPE AIC SC HQ
0 301.7397 NA 1.50E-13 -9.663595 -9.421364 -9.568662 1 646.5622 599.1998 9.29E-18 -19.36269 -17.42484* -18.60323 2 719.4107 109.8700 4.52E-18 -20.14461 -16.51114 -18.72062 3 779.4621 76.78697 3.71E-18 -20.50695 -15.17786 -18.41843 4 866.4853 91.30303* 1.51E-18* -21.75362 -14.72890 -19.00057*
5 934.6668 55.88645 1.53E-18 -22.38252* -13.66219 -18.96494
* indicates lag order selected by the criterion LR: sequential modified LR test statistic (each test at 5% level)
FPE: Final prediction error AIC: Akaike information criterion SC: Schwarz information criterion HQ: Hannan-Quinn information criterion
Test Statistics and Choice Criteria for Selecting the Order of the VAR Model using EViews4
Even though the SBC and Hannan-Quinn Criterion (HQ) criteria suggested a higher lag
length, we could not take the risk of over-parameterization by considering higher lags
because our data series are quarterly and we have a short time series with only 78
observations, out of which 61 are balanced. We therefore chose one lag for the model
which also seemed in conformity with finance theory and investors intuition that the
stock market is very volatile.
Cointegration Analysis
For the cointegration analysis, we considered a vector autoregressive (VAR) model
which has a constant (but no trend) and included the breakpoint dummy as exogenous,
as in the following:
0
1
k
t i t i t t
i
y y D uµ β ϕ−=
= + + +∑ (7.02)
Mazharul Kazi Ph.D Thesis 166
where ( , , , , , , )t
y LNALLORDS LNBVIR LNCP LNDY LNIPI LNMSCI LNPER ′= a 7×1
vector of I(1) variables considered as endogenous in the model; Dt is a vector of
breakpoint dummy exogenous variable; µ0 is a constant and ut is white noise.
In order to perform Johansens cointegration analysis we have converted the VAR into
a vector error correction model (VECM) by incorporating error correction term or error
correction mechanism (ECM-1) into the system. Thus, our transformed VECM is
1
0 1
1
p
t i i t i t t t
i
y y E C M Dµ δ ϕ ε−
− −=
∆ = + Γ ∆ + + +∑ (7.03)
or 1
0 1
1
p
t i i t i t t t
i
y y y Dµ α β ϕ ε−
− −=
′∆ = + Γ ∆ + + +∑ (7.04)
where t
ε ~ (0, )iidN Ω .
Considering the previously identified lag length one as the order of the VAR, we have
employed EViews4 to perform our analysis following Johansen (1991). Accordingly,
we have performed a likelihood ratio (LR) test, the maximum eigenvalue (max
λ ) test
and the trace (trace
λ ) test for the cointegration analysis. The cointegration results along
with test statistics are presented in Table 7.6.
Mazharul Kazi Ph.D Thesis 167
TABLE 7.6
Cointegration Results Hypothesis Critical Value
Null Alternative
Test Statistic 5% 1%
Eigenvalue
(A) Cointegration testa
Test Statistic: Maximal Eigenvalue (max
λ )
r = 0 r = 1 82.50867 45.28 51.57 0.724508 r ≤ 1 r = 2 56.39844 39.37 45.10 0.585725 r ≤ 2 r = 3 37.22766 33.46 38.77 0.441043 r ≤ 3 r = 4 30.79637 27.07 32.24 0.381955
r ≤ 4 r = 5 14.77187 20.97 25.52 0.206110 r ≤ 5 r = 6 6.831499 14.07 18.63 0.101243 r ≤ 6 r = 7 0.393115 3.76 6.65 0.006124
Test Statistic: Trace (trace
λ )
r = 0 r ≥ 1* 228.9276 124.24 133.57 0.724508 r ≤ 1 r ≥ 2 146.4189 94.15 103.18 0.585725 r ≤ 2 r ≥ 3 90.02050 68.52 76.07 0.441043 r ≤ 3 r ≥ 4 52.79285 47.21 54.46 0.381955
r ≤ 4 r ≥ 5 21.99648 29.68 35.65 0.206110 r ≤ 5 r ≥ 6 7.224615 15.41 20.04 0.101243 r ≤ 6 r = 7 0.393115 3.76 6.65 0.006124
(B) The long-run equationb
LNALLORDS(3.5776) = 0.3557LNBVIR(5.4923) + 1.2869LNCP(24.4554) +
0.8600LNDY(4.7723) 4.24174LNIPI(8.1860) 0.9201LNMSCI(2.3277)
0.0047LNPER(0.0010) or
LNALLORDS(3.5776) + 0.3557LNBVIR(5.4923) 1.2869LNCP(24.4554)
0.8600LNDY(4.7723) + 4.24174LNIPI(8.1860) + 0.9201LNMSCI(2.3277) +
0.0047LNPER(0.0010) = 0
r = the number of cointegrating vectors. a. Optimal lag structure is 1 and the VAR contains a constant without trend and breakpoint dummy as exogenous to the model. b. The cointegrating vector is normalised on the Australian stock price index (LNALLORDS). The LR test statistics, given in parentheses, are used to test the null hypothesis that each coefficient is statistically zero. The test statistic is
asymptotically distributed as a chi-square distribution with 1 degree of freedom. The critical values of chi-square distribution at 5% and 10% significance levels are 3.841 and 2.706 respectively.
It is evident from the results in Table 7.6 that we can reject the null hypothesis of
0r = against the alternative 1r = from the max
λ test. The same outcome is achieved
from the trace
λ test which has rejected 0r = against 1r ≥ . We have ascertained that in
our model only one stationary linear combination of variables is cointegrated in the
Mazharul Kazi Ph.D Thesis 168
long-run. Coefficients of the cointegrating equation (B) in Table 7.6 are normalized
by11
S Iβ β = as an identification process of the Johansen (1995) procedure, since the
long-run multiplier matrix Пy does not generally lead to a unique choice for the
cointegrating relations. The identification of β in y y
α β ′Π = requires at least r
restrictions per cointegrating relation (r). As we have found that r =1, one restriction
should be enough to identify the cointegrating relationship which is the normalizing
restriction applied to the LNALLORDS variable. LNALLORDS is considered as the
cointegrating equation, because it is the vector that contains the maximum eigenvalue.
Although the normalization is convenient from the mathematical point of view, it may
not always be meaningful otherwise. It has an advantage because such a normalization
is made without assuming anything about which variables are cointegrated, that is, it
serves the purpose without normalizing β.
It appears from the likelihood ratio (LR) test results of restrictions concerning each
variable in equation (B) of Table 7.6 that we can reject the null hypothesis of no
significance in relation to four a priori variables, including interest rate (LNBVIR),
corporate profit (LNCP), dividend yield (LNDY) and industrial production (LNIPI) at
the 5% level. Although in terms of LR test results both LNMSCI and LNPER do not
seem significant even at the 10% level, the global stock market index is significant
when we look at the t-statistic (2.7196) for LNMSCI. Respective t-statistics for
LNBVIR, LNCP, LNDY, LNIPI, LNMSCI, and LNPER are 3.5762, 11.3262, 4.4858,
4.2826, 2.7196 and 0.0484. From these results we can see that four to five of our a
priori variables have significant long-run influence on Australian stock price
movements or returns.
Mazharul Kazi Ph.D Thesis 169
Our results therefore suggest that although the linear combination of all our modelled
variables are found cointegrated, all variables are not equally influential. The
significantly influential a priori variables in the long-run cointegrating relationship for
the Australian stock market are the bank variable interest rate, corporate profitability,
dividend yield, and industrial production. The global stock market index also has some
influence. The price-earnings ratio seems to have insignificant influence from the
results of both the likelihood ratio and t tests statistics of EViews4.
Results of dynamic time series models and their corresponding error correction
mechanisms (ECM -1) are presented in Table 7.7.
Mazharul Kazi Ph.D Thesis 170
TABLE 7.7
Multivariate Dynamic Time Series (Short-Run) Models Entry Coefficient Standard Error t statistic[probability]
LHS variable: ∆LNALLORDS
∆LNALLORDS(-1)
∆LNBVIR(-1)
-0.4259
-0.2139
0.0391
0.1091
-4.8797
-1.9595
∆LNCP(-1) 0.0239 0.0362 0.6593
∆LNDY(-1) 0.3247 0.2233 1.4540
∆LNIPI(-1) 0.7837 0.1938 1.2332
∆LNMSCI(-1) 1.0964 0.6355 4.8393
∆LNPER(-1) -0.1174 0.0727 -1.6148
ECM(-1) -0.0401 0.0391 -1.0270
CHSQ(1) 0.6852 [0.4078]
LHS variable: ∆LNBVIR
∆LNALLORDS(-1) ∆LNBVIR(-1)
-0.1143 0.3978
0.1018 0.1279
-1.1231 3.1116
∆LNCP(-1) -0.0378 0.0424 -0.8914
∆LNDY(-1) -0.0087 0.2616 -0.0334
∆LNIPI(-1) 0.1963 0.7446 0.2636
∆LNMSCI(-1) 1.0964 0.2654 0.7091
∆LNPER(-1) -0.1174 0.0852 -0.4702
ECM(-1) 0.1048 0.0458 2.2884
CHSQ(1) 4.9172 [0.0266]
LHS variable: ∆LNCP
∆LNALLORDS(-1)
∆LNBVIR(-1)
0.7380
0.5079
0.2997
0.3764
1.3494
3.5824
∆LNCP(-1) 0.4471 0.1248 2.8126
∆LNDY(-1) 2.1662 0.7701 -0.6995
∆LNIPI(-1) -1.5333 2.1920 2.0367
∆LNMSCI(-1) 1.5916 0.7815 -0.6752
∆LNPER(-1) -0.1694 0.2508 -0.3242 ECM(-1) -0.0104 0.1348 -6.9414
CHSQ(1) 20.4846 [0.0000]
LHS variable: ∆LNDY
∆LNALLORDS(-1)
∆LNBVIR(-1)
-0.0674
0.15245
0.1061
0.1332
-0.6360
1.1443
∆LNCP(-1) 0.0288 0.0442 0.6529
∆LNDY(-1) 0.1948 0.2726 0.7147
∆LNIPI(-1) 0.1240 0.7758 0.1599
∆LNMSCI(-1) 0.1138 0.2766 0.4115
∆LNPER(-1) -0.0047 0.0888 -0.0529
ECM(-1) -0.0102 0.0477 -0.2135
CHSQ(1) 0.0470 [0.08284]
LHS variable: ∆LNIPI
∆LNALLORDS(-1)
∆LNBVIR(-1)
-0.0081
-0.0331
0.0183
0.0230
-0.4431
-1.4404 ∆LNCP(-1) -0.0005 0.0076 -0.0640
∆LNDY(-1) 0.0354 0.0471 0.7519
∆LNIPI(-1) 0.0454 0.1339 0.3389
∆LNMSCI(-1) 0.0673 0.0478 1.4104
∆LNPER(-1) 0.0075 0.0153 0.4876
ECM(-1) 0.0010 0.0083 1.2111
CHSQ(1) 1.3005 [0.2541]
LHS variable: ∆LNMSCI
∆LNALLORDS(-1)
∆LNBVIR(-1)
0.0786
-0.1339
0.1000
0.1254
0.7871
-1.0672
∆LNCP(-1) -0.0384 0.0416 -0.9227
∆LNDY(-1) 0.1190 0.2566 0.4637
∆LNIPI(-1) 0.2943 0.7303 0.4030
∆LNMSCI(-1) 0.2641 0.2604 1.0144
∆LNPER(-1) -0.0592 0.0836 -0.7081 ECM(-1) 0.0414 0.0450 0.9220
CHSQ(1) 0.8991 [0.3568]
LHS variable: ∆LNPER
∆LNALLORDS(-1)
∆LNBVIR(-1)
0.2271
-0.3612
0.1416
0.1779
1.6040
-2.0305
∆LNCP(-1) 0.0087 0.0590 0.1483
∆LNDY(-1) 0.9716 0.3639 2.6697
∆LNIPI(-1) -0.7002 1.0358 -0.6760
∆LNMSCI(-1) 0.0060 0.3693 0.5433
∆LNPER(-1) 0.6022 0.1185 5.0807
ECM(-1) -0.0715 0.0637 -1.1223
CHSQ(1) 1.2293 [0.2675]
Note: Critical values for t-statistics (2-sided test) are 1.64, 1.96 and 1.58 at10%, 5% and 1% significance levels respectively.
Mazharul Kazi Ph.D Thesis 171
The identified long-run cointegrating relation amongst seven variables in this study
from the perspective of the Australian stock market is plotted in Graph 7.4:
GRAPH 7.4
State of Equilibrium Pricing in the Australian Stock Market
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
-1.0
-0.8
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
84 86 88 90 92 94 96 98 00
Cointegrating relation of Australian Stock Market and apriori variables
Note: The cointegration plot shows the pattern of integration in the long-run.
Taking ∆LNALLORDS as the left hand side variable in the short-run model in Table
7.7 (which may be thought as the dependent variable in structural time series), it may
be suggested that the Australian stock market is dynamic and has been continually
corrected from its own disequilibrium of the previous quarter at a speed of 4%, while
all individual variables are contributing to the process of adjustment for equilibrium.
The bank interest rate (∆LNBVIR), global influence (∆LNMSCI) and the previous
performance of Australian market itself (∆LNALLORDS) are found significant in the
dynamic adjustment process, although the error correction mechanism (ECM1) is not
very efficient. The interest rate (∆LNBVIR) and company profits (∆LNCP) are found
to significantly contribute towards long-run equilibrium as their related error correction
mechanisms are significant.
Mazharul Kazi Ph.D Thesis 172
From another angle, coefficients of our long-run parameter β ′ upon normalization for
LNALLORDS are 0.3557, 1.2869, 0.8600, 4.2418, 0.9201 and 0.0047 for
LNBVIR, LNCP, LNDY, LNIPI, LNMSCI and LNPER respectively. Resulted
corresponding t-statistics are 3.5762, 11.3262, 4.4858, 4.2826, 2.7196 and 0.0484.
Thus, our estimated normalized ALLORDS
β with corresponding t-values in the
parentheses is:
11
( 3.5762)21
(11.3262)31
41 (4.4858)
51 ( 4.2826)
61 ( 2.7196)
71( 0.0484)
1.000000
0.3557
1.2869
0.8600
4.2418
0.9201
0.0047
ALLORDS
βββ
β ββββ
−
−
−
−
− = = − − −
(7.05)
This suggests that the LNBVIR, LNCP, LNDY, LNIPI and LNMSCI variables are
significant in the long-run cointegrating relationship for Australia as they are
significant when compared with the critical value for the t-statistic (1.96) at the 5%
significance level.
The short-run dynamic system provided us with coefficients of α corresponding to
∆LNALLORDS, ∆LNBVIR, ∆LNCP, ∆LNDY, ∆LNIPI, ∆LNMSCI and ∆LNPER.
The coefficients of α in respective order are 0.0401, 0.1048, 0.0104, 0.0102,
0.0010, 0.0414, and 0.0715. Corresponding t-values for α are 1.0270, 2.2884,
Mazharul Kazi Ph.D Thesis 173
6.9414, 0.2135, 1.2111, 0.9220 and 1.1223 respectively. This information can be
presented as:
( 1.0270)11
(2.2884)21
( 6.9414)31
41 ( 0.2135)
51 (1.2111)
61 (0.9220)
71( 1.1223)
0.0401
0.1048
0.0104
0.0102
0.0010
0.0414
0.0715
ααα
α αααα
−
−
−
−
− − = = − −
(7.06)
wherein the ECM-1 for the LNALLORDS is ALLORDSα =
11α = 0.0401, which is the
adjustment parameter in the cointegrating equation for Australia. The t-statistic in
parentheses corresponding to11α indicates that ECM-1 for LNALLORDS is not very
significant although the linear combination of all variables are found cointegrated. This
might mean that the Australian stock market is yet to be efficient in terms of its auto
correction.
We have obtained the estimates of the short-run parameters for the Australian market
ALLORDSΓ as 0.4259, 0.2139, 0.0239, 0.3247, 0.7837, 1.0964, and 0.1174 for
∆LNALLORDS -1, ∆LNBVIR -1, ∆LNCP -1, ∆LNDY -1, ∆LNIPI -1, ∆LNMSCI -1 and
∆LNPER -1 respectively. The corresponding t-statistics for ALLORDSΓ are 4.8797,
1.9595, 0.6593, 1.4540, 1.2332, 4.8393, and 1.6148. This suggests that in the process
of the short-run adjustment for the Australian stock market, ∆LNALLORDSt-1,
∆LNBVIRt-1 and ∆LNMSCIt-1 are significant at the 5% level. This means that
Mazharul Kazi Ph.D Thesis 174
Australian stock market prices are being adjusted each quarter dominantly by the
influences of the markets own performance as well as interest rate and global stock
market movements of previous quarter.
We thus present the short-run estimated parameter ALLORDSΓ as:
( 4.8797)
( 1.9595)
(0.6593)
(1.4540)
(1.2332)
(4.8393)
( 1.6148)
0.4259
0.2139
0.0239
0.3247
0.7837
1.0964
0.1174
ALLORDS
−
−
−
− − Γ = −
(7.07)
Based on the above results, we present the estimated model (VECM) for Australia in
equations 7.08 and 7.09. Our estimated model takes the following shape:
∆LNALLORDSt = −0.0401*[ 1*LNALLORDS -1 −0.3557*LNBVIR -1 +
1.2869*LNCP -1 + 0.8600*LNDY -1 − 4.2418*LNIPI-1 −
0.9201*LNMSCI-1 − 0.0047*LNPER-1]
[−0.4259*∆LNALLORDS-1 0.2139* ∆LNBVIR-1 +
0.0239* ∆LNICP-1 +0.3247* ∆LNDY-1+ 0.7837* ∆LNIPI-1 +
1.0964* ∆LNMSCI-1 − 0.1174*∆LNPER-1 ]. (7.08)
or
Mazharul Kazi Ph.D Thesis 175
∆LNALLORDSt = − 0.0401*LNALLORDS -1 + 0.0143*LNBVIR -1 −0.0516*LNCP-1
− 0.0345*LNDY -1 + 0.1701*LNIPI-1 + 0.0369*LNMSCI-1 +
0.0002*LNPER-1 + 0.4259* ∆LNALLORDS-1 + 0.2139*∆LNBVIR-1 +
0.0239*∆LNCP-1 0.3247* ∆LNDY-1 − 0.7837*∆LNIPI -1
1.0964*∆LNMSCI -1 + 0.1174*∆LNPER -1 . (7.09)
In long-run equilibrium, the second part of the above model would not exist; therefore,
our solved model in reduced form is as follows:
∆LNALLORDSt = −0.0401*LNALLORDS -1 + 0.0143*LNBVIR -1 − 0.0516*LNCP-1
− 0.0345*LNDY -1 + 0.1701*LNIPI -1 + 0.0369*LNMSCI -1 +
0.0002*LNPER -1 (7.10)
These results are interesting and useful in understanding the Australian stock market
pricing mechanism as well as its return generating process. The cointegrating analysis
presented in Table 7.6 shows that all our variables are cointegrated in the long-run,
although only four to five variables are influential in the pricing process. These
variables are interest rate, corporate profit, dividend yield, industrial production, and
global stock market movements.
Although the presence of cointegration among variables does not necessarily imply that
all variables will contribute the process of integration equally, some might like to see
what happens if the insignificant variable/s is/are dropped from the model and rerun the
analysis. We have done so by dropping our insignificant variable the price earnings
ratio (LNPER). This additional cointegration analysis shows that there exists a
Mazharul Kazi Ph.D Thesis 176
cointegrating relationship between the variables considered and that all variables are
significantly influential. This additional analysis thus corroborates the earlier claim that
our a priori variables as proxy to systematic risk factors for Australian stock market are
being priced in the long-run.
Caution
The results of this analysis should not be treated as conclusive for an investment. Apart
from understanding Australian stock market pricing based on the contributions of our
significant a priori variables, there remains other important issues that affect the return
generating process. These issues are the cost of equity capital, asset valuation, industry
analysis, a firms management and operational efficiency analysis, and so on.
An Australian investor should consider all relevant sources of information when
making an investment decision. Even within the arena of systematic risk analysis for
assets, one should identify both business and financial risks and analyse them while
selecting a stock in their portfolio for investment. The business risk of a firm depends
on the systematic risk of a firms assets. The greater a firms business risk, the greater
the firms cost of equity. The other component in the cost of equity is determined by a
firms financial structure which also needs assessment while selecting stocks for an
investment portfolio. This extra risk that arises from the use of debt financing is called
the financial risk of a firms equity. The well known propositions of Franco Modiglini
and Merton Mills are useful in analysing the value of a firm as well as its cost of equity.
Their first proposition is that the value of a firm is independent of a firms capital
structure, while their second proposition states that the cost of equity depends on the
Mazharul Kazi Ph.D Thesis 177
required rate of return on a firms assets, its firms cost of debt, and its firms debt-
equity ratio. When investigating portfolio investment, an investor should analyse two
types of risks, the total systematic risk of the firms equity consisting of business
risk, and financial risk [Ross, et. al. (1998)].
It is more useful for our purposes to know that the linear combination of modelled
variables contains a relationship the variables are cointegrated and our a priori
variables explain the price movements as well as the return generating process of the
Australian stock market in both the long- and short-runs.
Summary
In this chapter, we performed necessary analyses to answer the research question of
whether some of our identified a priori variables as proxy to systematic risk factors for
Australian stock market explain return generating as well as pricing process. We
followed the model that was presented in Chapter 6 and used the most current
econometrics software procedures for the empirical analysis. The outcome is in line
with current theory, yet interestingly different on some points. We found that four to
five of our identified a priori variables explain the Australian stock market pricing
mechanism these variables have a long-term relationship with Australian stock
returns. Further interpretations of the results will be discussed in the research
conclusion Chapter 9. We now progress to our second analysis, in Chapter 8.
Mazharul Kazi Ph.D Thesis 178
CHAPTER 8
(EMPIRICAL ANALYSIS 2)
INTERDEPENDENCE OF STOCK MARKETS UNDER
GLOBALIZATION
The common intuition is that under the prevailing globalized world economic order, the
stock markets of most developed nations are becoming increasingly interrelated.
Capital and investment funds now flow freely from one country to another. Due to
internet technology and the speedy transfer of information the world has become one
interdependent market place. As a result, it is likely that the stock market performance
of one country will be influenced by that of others. This is especially true for efficient,
developed markets.
We aim to examine the common intuition from an Australian perspective. Our
investigation was carried out, in the first instance, in a straightforward way the aim
was to ascertain whether the Australian stock market is cointegrated with the US, UK,
Canadian, German, French and Japanese stock markets. Based on annual data series
from 1945 to 2002 we have studied both the long-run relationships and the short-run
dynamics operating between the seven markets under the global economic and financial
system.
In addition, we have investigated the issue of global integration and considered the
impact of foreign exchange on the model, in order to understand the exchange rate risk
Mazharul Kazi Ph.D Thesis 179
element. We have analyzed two models: the structural model, and the TWI weighted
autoregressive model.
Background
Although we aim to examine empirically whether the common intuition about the
integrated global financial market stands, the prevailing literature needs to be reviewed.
The theoretical basis of the common intuition is traced in early mean-variance analyses
of international share price integration or international asset pricing analyses, namely,
the international asset pricing model (IAPM) and international arbitrage pricing theory
(IAPT) [Solnik (1974a), Sutlz (1981)]. IAPM is an extended version of the capital asset
pricing theory (CAPM) of Sharpe (1963, 1964) and Lintner (1965) while IAPT is
influenced by the arbitrage pricing theory (APT) of Ross (1976), Roll (1977), and Roll
and Ross (1980). The portfolio theory of Markowitz (1952) has been the foundation of
all asset pricing theories. Other studies aimed to assess integration patterns of global
stock markets through correlation and other methods. These include Grubel (1968),
Levy and Sarnat (1970), Solnik (1974b, 1983), Sutlz (1981), Lessard (1973), Ripley
(1973), Panton, et al (1976), Eun and Resnick (1984), Errunza (1983), Chan, et. al
(1992), Maldanado and Saunders (1981), Phillippatos, et al (1983), Arshanapalli and
Doukas (1993), Aggarwal and Rivoli (1980), Bekart and Harkey (1995), Eun and
Beswick (1984), Janakiramannan and Lamba (1998), Joen and Chiang (1991), and Cha
and Oh (2000).
Mazharul Kazi Ph.D Thesis 180
Mills and Mills (1991), Kasa (1992), Cheung and Mak (1992), Arshanapalli and
Doukes (1993), Nasseh and Strauss (2000), Lim and McNelis (1998), Maysami and
Kho (2000), Leong and Felmingham (2001), and Shamsuddin and Kim (2003),
amongst others, have studied the interrelationship between various markets by applying
cointegration analysis to international share prices. Further support for studying the
interdependence of share markets is found in studies on market efficiency or efficient
market hypothesis, including those of Roca (1999), Roca, et al (1998) and Chan and Oh
(2000).
Generally, any potential gain from the international diversification of a portfolio is
inversely related to the extent of stock market integration. A low correlation between
returns of national and overseas indices allows investors to minimize portfolio risk
through international diversification. Thus, an analysis of the long-run co-movement of
national stock prices with that of overseas stock prices and the short-run temporal
relationship between the two is important for managing an international portfolio.
Additionally, interdependent international stock prices reflect economic integration in
the form of trade linkages and foreign direct investment. Co-movement of underlying
macroeconomic variables across nations may lead to co-movement of stock prices
[Bracker, et al. (1999)]. Still, foreign exchange risk is thought to be an important
consideration by some investors buying foreign stocks.
Studies on stock market integration may be classified into two categories, both based
on the foreign exchange risk component: (i) studies that considered the exchange rate
factor in the analysis [Taylor and Tonks (1989), Bekaert and Harvey (1997)]; and (ii)
Mazharul Kazi Ph.D Thesis 181
studies that ignored the exchange rate risk component. The second category may be
further divided into two sub-groups, based on their assumptions: (i) ignore the
exchange rate and use local currency in the analysis because the foreign exchange risk
can be hedged while investing in overseas markets [Bracker, et al (1999), Ragunathan,
et al (1999)]; and (ii) ignore the exchange rate risks because they are priced in
international asset markets [Dumas and Solnik (1995), Iorio and Faff (2000), Khoo
(1994), and Choi, et al (1998)].
In this chapter, we investigate the global stock market integration issue considering
both ideas on the exchange rate risk element. We therefore have two analyses. One
considers the idea that the foreign exchange component is important for consideration
in assessing the stock market integration under globalization. The second follows the
idea that the foreign exchange component is irrelevant. In our first analysis, we have
used structural and autoregressive approaches to empirical analysis. In our second
analysis we did not go as far as exploring other effects of risks including exchange rate
risk, currency depreciation and impact of inflation. Instead, we have investigated the
cross-country interrelationship in a straightforward way, that is, by using the
cointegration technique of Johansen, based on an autoregressive error correction model.
For our first analysis, we have deliberately ignored the exchange rate component
because of the belief that an astute investor would hedge the exchange risk this is a
diversifiable risk factor is most cases. We acknowledged Solnik (1983) who showed
that if a factor model is believed to hold when asset returns are expressed in some
arbitrary chosen currency, the factor structure as well as its major conclusions is
Mazharul Kazi Ph.D Thesis 182
invariant to the currency chosen. For analogous reasons, we have also ignored the
possible influence of the share return parity. Our first analysis is straightforward in that
it aims to ascertain whether the Australian stock market is integrated with selected six
overseas markets and it aims to identify the short-run dynamics behind it.
For our second analysis, we have used the trade weighted index (TWI) of Australias
major trading partners to represent the foreign exchange risk element. We undertook
investigations considering structural and autoregressive models. The structural model
incorporated exchange risk and share-return parity issues. In analyzing our TWI
adjusted structural model, the ordinary least squares (OLS) estimation technique is
followed.
Data and Variables
We have considered the national stock index of seven developed countries Australia,
the US, the UK, Canada, Germany, France, and Japan. As the S&P/ASX 200 index or
accumulated index series was unavailable for the entire period under investigation, we
used the All Ordinaries index (ALLORDS) to represent the performance of Australian
stock market. Accordingly, our data series consist of the yearly index values of the
ALLORDS, DJIA, FTA, SBF250, DAX, TSX300, and NIKKEI for Australia, the US,
UK, Canada, Germany, France, and Japan respectively. The data have been gathered
from sources that include the Australian Bureau of Statistics, the Reserve Bank of
Australia, and the worldwide website of Harcourt College Publishers. The period of
study is based on the collected data series 1945 to 2002. Annual data series are
Mazharul Kazi Ph.D Thesis 183
considered for the purposes of studying both the long-run relationships as well as the
short-run dynamics operating between the selected seven stock markets. In all of our
analyses we have used logged values of selected variables.
Preliminary Tests
At the outset, the data for this study were checked from descriptive statistical
viewpoints. These are presented in Table 8.1.
TABLE 8.1
Descriptive Statistics AUSTRALIA USA UK CANADA GERMANY FRANCE JAPAN
Mean 5.245621 6.221484 4.771890 4.228871 3.296500 6.437833 6.334756 Median 5.023436 6.369301 4.203049 4.893704 3.285783 6.305764 6.502100 Maximum 8.119666 9.349852 8.083964 8.245610 6.100207 9.097584 10.56916 Minimum 2.860394 3.999668 3.083139 0.548757 0.415283 4.594600 2.789827
Std. Dev. 1.522504 1.415010 1.524519 2.228121 1.359712 1.328497 2.619283 Skewness 0.345815 0.467197 0.809055 -0.053329 0.188878 0.393691 0.073577 Kurtosis 2.016226 2.338045 2.311779 1.821935 2.065949 1.912129 1.403392
Jarque-Bera 5.362856 4.862655 11.46589 5.188750 3.764519 6.687736 9.533426 Probability 0.068465 0.087920 0.003238 0.074693 0.152246 0.035300 0.008508
Sum 466.8603 553.7121 424.6982 376.3695 293.3885 572.9671 563.7933
Sum Sq. Dev. 203.9858 176.1982 204.5260 436.8778 162.6959 155.3116 603.7366
Observations 89 89 89 89 89 89 89
It appears from Table 8.1 that our data series consisting of 89 observations have
probabilities of being significant at around the 10% level in terms of various
preliminary statistical measures, including the mean, the standard deviation, skewness,
kurtosis, and the Jaque-Bera statistic.
Mazharul Kazi Ph.D Thesis 184
Analysis 1
In selecting the optimal lag length for our first analysis, we have conducted the
necessary tests using EViews4, as presented in Tables 8.2A and 8.2B. The results of the
correlogram in Table 8.2A indicates that the appropriate lag length is two, while the lag
length selection test statistics of LR, FPE, SBC and HQ in Table 8.2B clearly suggest
that a lag of one period is the optimal lag length. Although the AIC and correlogram
criteria suggested differently, the other tests suggested that a lag of one year is
appropriate. Accordingly, we have taken the lag length as one in the model for analysis.
TABLE 8.2A
Correlogram of Residuals Sample: 1945 2002
Included observations: 58
Autocorrelation Partial Correlation Lag AIC PAC Q-Stat Prob
. |*******| . |*******| 1 0.943 0.943 54.324 0.000 . |*******| . | . | 2* 0.887 -0.029 103.17 0.000 . |****** | . | . | 3 0.836 0.021 147.34 0.000
. |****** | .*| . | 4 0.779 -0.075 186.47 0.000 . |****** | . | . | 5 0.724 -0.020 220.88 0.000 . |***** | . | . | 6 0.674 0.009 251.24 0.000
. |***** | . | . | 7 0.624 -0.019 277.82 0.000 . |**** | . | . | 8 0.575 -0.025 300.83 0.000 . |**** | .*| . | 9 0.518 -0.101 319.92 0.000
. |**** | . | . | 10 0.461 -0.047 335.31 0.000 . |*** | . | . | 11 0.410 0.021 347.77 0.000 . |*** | .*| . | 12 0.356 -0.061 357.37 0.000
. |** | . | . | 13 0.310 0.034 364.77 0.000 . |** | .*| . | 14 0.256 -0.118 369.94 0.000 . |** | . | . | 15 0.209 0.036 373.48 0.000
. |*. | . | . | 16 0.168 0.003 375.83 0.000 . |*. | .*| . | 17 0.121 -0.076 377.08 0.000 . |*. | . | . | 18 0.081 0.024 377.65 0.000
. | . | . | . | 19 0.048 0.008 377.86 0.000 . | . | . | . | 20 0.014 -0.029 377.87 0.000 . | . | . | . | 21 -0.014 0.026 377.89 0.000
. | . | .*| . | 22 -0.051 -0.124 378.14 0.000 .*| . | . | . | 23 -0.089 -0.036 378.92 0.000 .*| . | . |*. | 24 -0.108 0.109 380.11 0.000
Mazharul Kazi Ph.D Thesis 185
TABLE 8.2B
Lag Length Selection VAR Lag Order Selection Criteria Endogenous variables: AUSTRALIA USA UK CANADA GERMANY FRANCE JAPAN
Exogenous variables: C Date: 11/22/03 Time: 17:51 Sample: 1945 2002 Included observations: 58
Lag Log likelihood LR FPE AIC SBC HQ
0 -124.1068 NA 2.17E-07 4.520924 4.769598 4.617787 1 199.0015 557.0832* 1.72E-11* -4.931085 -2.941691* -4.156176* 2 237.1254 56.52862 2.69E-11 -4.556048 -0.825936 -3.103094 3 283.5171 57.58967 3.58E-11 -4.466106 1.004725 -2.335107 4 337.6588 54.14168 4.53E-11 -4.643406 2.568145 -1.834361 5 414.9343 58.62285 3.75E-11 -5.618425* 3.333844 -2.131336
* indicates lag order selected by the criterion LR: sequential modified likelihood ratio (LR) test statistic (each test at 5% level) FPE: Final prediction error AIC: Akaike information criterion SBC: Schwarz information criterion
HQ: Hannan-Quinn information criterion
We could not find any significant effect of the stock market crash of October 1987 in
our series from the Chow Breakpoint test [Chow (1960)]. This is because we failed to
reject the hypothesis of no breakpoint effect in the 1987 data even at the 10% level. We
observed the F-statistic 0.814644 with p-value 0.580112, and the log likelihood ratio
statistic 7.068216 with corresponding p-value of 0.421812. These are not significant at
acceptable levels. As a result, we did not consider any exogenous dummy variable in
the model.
Furthermore, to undertake cointegration analysis in the spirit of Johansen, we have
conducted the necessary unit root tests using both the ADF [Dickey and Fuller (1979,
1981)] and PP [Phillips and Perron (1988)] test procedures. These are presented in
Table 8.3.
Mazharul Kazi Ph.D Thesis 186
TABLE 8.3
Unit Root Test Results
Augmented Dickey-Fuller (ADF) Phillips-Perron (PP) Variables Intercept,
No trendA Intercept
with trendB No intercept,
No trendC Intercept, No trendA
Intercept with trendB
No intercept, No trendC
At level
AUSTRALIA -0.0861 -2.2462 3.2128 -0.3195 -2.9038 3.0873
USA -0.0151 -1.1831 3.4482 -0.2368 -1.5074 3.7666
UK 0.0253 -2.2697 2.6029 -0.2431 -2.6681 2.4349
CANADA -0.9264 -2.0602 2.1724 -1.0656 -2.2391 2.3153
GERMANY -0.9647 -2.8672 0.9940 -1.1078 -3.0808 0.8983
FRANCE -0.7003 -2.7838 3.8898 -0.9017 -3.6897 3.6777
JAPAN -2.6085 -0.0516 1.6254 -2.6789 -0.1857 1.7593
At 1st diff.
∆AUSTRALIA -4.9478 -4.4088 -3.7166 -8.8982 -8.8172 -7.7549
∆USA -4.1458 -4.0870 -2.9673 -7.6233 -7.5492 -6.3772
∆UK -5.7611 -5.7935 -4.4700 -8.3472 -8.2862 -7.6105
∆CANADA -3.9954 -3.8950 -3.2572 -7.8111 -7.7395 -7.0719
∆GERMANY -5.1658 -5.0782 -4.8364 -8.2049 -8.1195 -8.0446
∆FRANCE -5.4494 -5.3924 -3.5407 -8.5461 -8.4742 -7.0986
∆JAPAN -3.1398 -3.9737 -2.6973 -6.0568 -6.7411 -5.5385
Note: MacKinnon Critical values at level: for model A. -2.9851; model B. -3.462; model C. -1.9439, and at 1st difference: for model A. -2.8955; model B. -3.4626; model C. -1.9440
The test results reported in Table 8.3 are compared against the MacKinnon (1990)
critical values for rejection of the null hypothesis of no unit root. It clearly suggests that
all of the seven series are integrated to order one I(1) in levels and are of order zero I(0)
in first differences, meaning that they are nonstationary in levels and stationary in first
differences.
Cointegration Analysis
To determine both the long- and short-run relationships operating between the
Australian stock market and those of the six other major markets, we have treated all of
the seven series or variables as an endogenous series:
Mazharul Kazi Ph.D Thesis 187
( , , , , , , )ty AUSTRALIA USA UK CANADA GERMANY FRANCE JAPAN ′=
where the logged stock market indices of seven countries are assumed to be integrated
of order one, I(1), and also they are thought to be cointegrated. We have adapted the
generalised vector autoregressive (VAR) model with k lags containing all variables (g
=7), as under:
1 1 2 2t t t k t k ty y y y u− − −=Φ +Φ + +Φ +L (8.01)
where yt is a vector of order 7 1× , Φ is a 7 7× matrix, and ut is a 7 1× vector. Johansen
cointegration requires us to convert the above VAR into an error correction form. The
vector error correction model (VECM) takes the following form:
1 1 1 ( 1)t t k t k t k ty y y y ε− − − − −∆ =∏ +Γ∆ + +Γ ∆ +L (8.02)
where 1
( )k
i gi
Iβ=
∏ = ∑ − is the long-run coefficient matrix of the lagged yt, 1
( )i
i j gj
Iβ=
Γ = ∑ −
is a coefficient matrix of k-1 lagged difference variables, and ∆yt, on the right hand side
of the equation refers to short-run dynamics. Accordingly, the reduced form VECM for
the seven cross-country cointegration analysis is:
1
1
k
t t k i t i ti
y y y ε−
− −=
∆ = ∏ + ∑ Γ ∆ + (8.03)
Mazharul Kazi Ph.D Thesis 188
where αβ ′∏ = as per Johansen (1988, 1996, 2000).
By considering 1st order VAR and employing EViews4 to perform the analysis
following Johansen (1988, 1991, 1995a, 1995b, 1996, 1999, 2000), we have obtained
test results. It is evident from the results that although we cannot reject the null
hypothesis of no cointegration from the test result ofmaxλ , we can reject the null
hypothesis of 0r = from the traceλ test against the alternative 1r = . As per the trace
tests results there are at most two cointegrating equations. Since the first equation
provides the relationship from the perspective of the Australian stock market, we can
say that the Australian stock market is cointegrated with the other six markets. In other
words, movements in Australian stock prices are influenced by major developed
markets. Graph 8.1 plots the cointegrating relationship between Australia and the other
six markets under study.
GRAPH 8.1
Pattern of Australian Stock Market Integration under Globalization
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
1950 1960 1970 1980 1990 2000
Cross-Country Cointegrating relation for AUSTRALIA
Mazharul Kazi Ph.D Thesis 189
For convenience, the coefficients of the first cointegrating equation have been
normalized. To ascertain the strength of the relationship and the extent of influences,
the likelihood ratio (LR) test results are produced by imposing appropriate restrictions
in the model solutions. We have used EViews4 to perform the necessary tests in this
regard. The cointegration results along with test statistics are presented in Table 8.4.
TABLE 8.4
Cointegration Results Hypothesis Critical Value
Null Alternative
Test Statistic At 5% level At 1%
Eigenvalue
(A) Cointegration testa
Test Statistic: Maximal Eigenvalue (maxλ )
r = 0 r = 1 41.7013 45.28 51.57 0.5128 r ≤ 1 r = 2 33.4656 39.37 45.10 0.4384 r ≤ 2 r = 3 25.7108 33.46 38.77 0.3581 r ≤ 3 r = 4 17.8657 27.07 32.24 0.2651 r ≤ 4 r = 5 8.9440 20.97 25.52 0.1429 r ≤ 5 r = 6 5.5932 14.07 18.63 0.0919 r ≤ 6 r = 7 3.4890 3.76 6.65 0.0584
Test Statistic: Trace (traceλ )
r = 0 r ≥ 1* 136.7697 124.24 133.57 0.5128
r ≤ 1 r ≥ 2* 95.0683 94.15 103.18 0.4384 r ≤ 2 r ≥ 3 61.6028 68.52 76.07 0.3581 r ≤ 3 r ≥ 4 35.8919 47.21 54.46 0.2651 r ≤ 4 r ≥ 5 18.0262 29.68 35.65 0.1429 r ≤ 5 r ≥ 6 9.0822 15.41 20.04 0.0919 r ≤ 6 r = 7 3.4890 3.76 6.65 0.0584
(B) The long-run equationb
AUSTRALIA(8.1862) = 0.2088USA(0.4818) 0.8222UK(8.1032) +
0.3154CANADA(1.6272) 0.2155GERMANY(2.3006) + 0.0572FRANCE(0.0354)
0.0461JAPAN(0.2461)
or
AUSTRALIA(8.1862) 0.2088USA(0.4818) 0.8222UK(8.1032) +
0.3154CANADA(1.6272) 0.2155GERMANY(2.3006) + 0.0572FRANCE(0.0354)
0.0461 JAPAN(0.2461) = 0
r *= the number of cointegrating vectors.
a. Optimal lag structure is one and the VAR contains constant without trend and breakpoint dummy as exogenous to the model.
b. The long-run cointegrating vector is normalised on the logged Australian All Ordinaries index (AUSTRALIA). The Likelihood
Ratio (LR) test statistics, given in parentheses, are used to test the null hypothesis that each coefficient is statistically zero. The test
statistic is asymptotically distributed as a chi-square distribution with one degree of freedom. The critical values of chi-square
distribution at 5% and 10% significance levels are 3.841 and 2.706 respectively.
Mazharul Kazi Ph.D Thesis 190
TABLE 8.5
Multivariate Dynamic Time Series (Short-Run) Models
Item Coefficient Standard Error t-ratio LHS variable: ∆AUSTRALIA
∆AUSTRALIA(-1)
∆USA(-1)
0.1096
-0.0352
0.2322
0.2760
0.4722
-0.1276
∆UK(-1) -0.0399 0.1822 -0.2188
∆CANADA(-1) -0.1340 0.1320 -1.0153
∆GERMANY(-1) 0.0155 0.0922 0.1676 ∆FRANCE(-1) -0.2354 0.2804 -0.8396
∆JAPAN(-1) -0.0237 0.1082 -0.2191
ECM(-1) -0.3286 0.1659 -1.9812
F-Statistic 1.0998
LHS variable: ∆USA
∆AUSTRALIA(-1)
∆USA(-1)
-0.1242
0.2359
0.1861
0.2213
-0.6672
1.0660
∆UK(-1) 0.1104 0.1461 0.7560
∆CANADA(-1) 0.0198 0.1058 0.1875
∆GERMANY(-1) 0.0812 0.0739 1.0985
∆FRANCE(-1) -0.4521 0.2248 -2.0112
∆JAPAN(-1) -0.0669 0.0868 -0.7714
ECM(-1) 0.1171 0.1330 0.8807
F-Statistic 1.2186
LHS variable: ∆UK ∆AUSTRALIA(-1)
∆USA(-1)
0.1790
0.5074
0.2779
0.3304
0.0644
1.5355
∆UK(-1) -0.1539 0.2181 -0.7054
∆CANADA(-1) -0.1397 0.1580 -0.8842
∆GERMANY(-1) 0.1141 0.1104 1.0342
∆FRANCE(-1) -0.7435 0.3357 -2.2151
∆JAPAN(-1) -0.0421 0.1295 -0.3253
ECM(-1) 0.0483 0.1986 0.2433
F-Statistic 1.1926
LHS variable: ∆CANADA
∆AUSTRALIA(-1)
∆USA(-1)
-0.1502
-0.0834
0.2886
0.3432
-0.5205
-0.2430
∆UK(-1) 0.0618 0.2266 0.2730
∆CANADA(-1) -0.1844 0.1641 -1.1233
∆GERMANY(-1) 0.1610 0.1146 1.4051
∆FRANCE(-1) 0.1979 0.3486 0.5675 ∆JAPAN(-1) 0.0818 0.1345 0.6082
ECM(-1) -0.3848 0.2062 -1.8689
F-Statistic 1.3739
LHS variable: ∆GERMANY
∆AUSTRALIA(-1)
∆USA(-1)
-0.7132
0.0783
0.4086
0.4858
-1.7457
0.1613
∆UK(-1) 0.6448 0.3207 2.0105
∆CANADA(-1) 0.0470 0.2324 0.2023
∆GERMANY(-1) -0.0960 0.1623 -0.5918
∆FRANCE(-1) 0.0270 0.4935 0.0547
∆JAPAN(-1) -0.1870 0.1905 -0.9818
ECM(-1) 0.4064 0.2919 1.3921
F-Statistic 0.8088
LHS variable: ∆FRANCE
∆AUSTRALIA(-1) ∆USA(-1)
0.0796 -0.0601
0.1893 0.2251
0.4203 -0.2673
∆UK(-1) 0.1244 0.1486 0.8371
∆CANADA(-1) -0.0891 0.1076 -0.8278
∆GERMANY(-1) 0.0709 0.0752 0.9428
∆FRANCE(-1) -0.2445 0.2287 -1.0695
∆JAPAN(-1) 0.0037 0.0882 0.0415
ECM(-1) -0.1372 0.1352 -1.0147
F-Statistic 0.8210
LHS variable: ∆JAPAN
∆AUSTRALIA(-1)
∆USA(-1)
-0.4194
-0.1847
0.3136
1.3729
-1.3374
-0.4953
∆UK(-1) 0.5669 0.2462 2.3028
∆CANADA(-1) 0.0772 0.1783 0.4330
∆GERMANY(-1) 0.0734 0.1245 0.5895
∆FRANCE(-1) -0.0198 0.3788 -0.0523 ∆JAPAN(-1) 0.0807 0.1462 0.5523
ECM(-1) 0.5157 0.2241 2.3014
F-Statistic 1.3077
Mazharul Kazi Ph.D Thesis 191
It appears from the LR test statistics and the estimated coefficients corresponding to
each market in (B) of Table 8.4 that the movements of UK stock prices significantly
influence the movement in Australian stock prices at the 5% level. The Australian
market is corrected at the rate of 33% from its previous years disequilibrium as shown
in the dynamic solution presented in Table 8.5.
The first cointegration equation contains both a long-run relationship and the short-run
dynamics of the VECM, 1 1t i iy y yα β − −′∆ = + Γ ∆ , where ∆yt =∆AUSTRALIA,
normalized β ′ = 1, 0.21, 0.82, 0.32, 0.22, 0.06, 0.05 and corresponding t-statistics
for the US, the UK, CANADA, GERMANY, FRANCE, and JAPAN are 1.51551,
8.89530, 2.62820, 2.66750, 0.31799, and 1.02482 respective. The estimated β for
Australia and the estimated α with its corresponding t-statistics in parentheses are
provided in the following:
11 11
( 1.516)21 21
( 8.895)31 31
41 (2.628) 41
51 51( 2.668)
61 61(0.318)
71 71( 1.025)
1.00
0.21
0.82
0.32 ;
0.22
0.06
0.05
AUSTRALIA
β αβ αβ α
β β α αβ αβ αβ α
−
−
−
−
− − = = = − −
( 1.981)
(0.881)
(0.243)
( 1.866)
(1.392)
( 1.015)
(2.301)
0.33
0.12
0.05
0.38
0.41
0.14
0.52
−
−
−
− = − −
(8.04)
The resultant t-statistics for AUSTRALIAβ in the above equation suggests that three markets
are influential for Australia in the long-run as they are significant at 5% levels. These
three markets are UK, Canada, and Germany. In the same cointegrating vector for
Australia, α = (0.33, 0.12, 0.05, 0.38, 0.41, 0.14, 0.52) where ECM-1 = 11α = 0.33
Mazharul Kazi Ph.D Thesis 192
is relevant to the long-run equation for Australia. Corresponding t-values of α are
1.98124, 0.88070, 0.24332, 1.86585, 1.39216, 1.01473 and 2.30141. From t-statistics
for α in the above equation, it is evident that the error correction mechanism ECM-1
for Australia is significant at the 5% level.
The coefficients of the short-run adjustment parameters 1i y−Γ ∆ are 0.11, 0.04, 0.04,
0.13, 0.02, 0.24, and 0.02, with t-values as 0.47218, 0.12758, 0.21883, 1.01529,
0.16760, 0.83956, and 0.21907 respectively. Based on these results, the solved
model is presented in equations 8.05 and 8.06:
∆AUSTRALIA t = 0.33*[AUSTRALIA -1 + 0.21*USA -1 0.82*UK -1
+ 0.32*CANADA -1 0.22*GERMANY-1 + 0.06*FRANCE -1
0.05*JAPAN -1] + [0.11*∆AUSTRALIA -1 0.04*∆USA -1
0.04*∆UK -1 0.13*∆CANADA -1 + 0.02*∆GERMANY -1
0.24*∆FRANCE -1 0.02*∆ JAPAN -1] (8.05)
∆AUSTRALIA t = 0.33*AUSTRALIA -1 0.07*USA -1 + 0.27*UK -1
0.10*CANADA -1 + 0.07*GERMANY-1 0.02*FRANCE -1
+ 0.02*JAPAN -1 + 0.11*∆AUSTRALIA -1 0.04*∆USA -1
0.04*∆UK -1 0.13*∆CANADA -1 + 0.02*∆GERMANY -1
0.24*∆FRANCE -1 0.02*∆ JAPAN -1. (8.06)
The estimated long-run model for Australia in equilibrium, where the short-run
equation does not exists, is presented by equation 8.07:
∆AUSTRALIA t = 0.33*AUSTRALIA -1 0.07*USA -1 + 0.27*UK -1
0.10*CANADA -1 + 0.07*GERMANY-1 0.02*FRANCE -1
+ 0.02*JAPAN -1. (8.07)
Mazharul Kazi Ph.D Thesis 193
The above solved model (8.07) indicates that Australian stock market returns are
influenced by the previous years stock market performance of the US at 7%, the UK at
27%, Canada at 10%, Germany at 7%, France at 2%, Japan at 2%, and Australian
market performance at 33%. It is evident from the t-tests for AUSTRALIAβ presented in the
parentheses of equation 8.04 that in addition to the UK, the Canadian and German
markets also have significant influence on the Australian stock market in the long-run.
Both the LR test statistics in (B) of Table 8.4 and the t-value corresponding to the UK
in equation 8.04 suggest that the influence of the UK market alone is highly significant
for Australia. During this process, the Australian market is corrected from its own
preceding disequilibrium at a speed of 33%.
We have ascertained that the movement of Australian stock market prices is linked with
developed overseas stock markets in the long-run as the linear combination of all
markets under investigation is found cointegrated. Yet, only one to three overseas
markets are found significantly influencing the Australian stock returns based on LR
and t-tests statistics. As the t-test is also relevant, we find the UK, Canada and
Germany are significant at 5% levels, while insignificant markets in the cointegrated
equation are the US, France and Japan.
Supplementary Tests
To corroborate the outcomes we undertook additional analysis by considering only the
significant variables. We have re-estimated the equation dropping insignificant
markets. The additional cointegration analysis was set on two lags because this
Mazharul Kazi Ph.D Thesis 194
appeared to be reasonable. The trace test indicates the presence of one cointegrating
vector at both 5% and 1% significance levels, meaning that all three markets are
cointegrated. Estimation of the long-run equation shows that all cointegrated markets
are significant at 5% levels from the resulted t-statistics. From the LR test it appears
that all variables are significant at 10% levels. The t-test result corroborates the earlier
findings that the UK market is the most dominating market for Australia, although the
LR test indicated otherwise. As per the LR test results, both the German and Canadian
markets are significant at 5% levels, while the UK market is significant at 10% levels.
The German market is found dominating within the cointegration relationship. It is
worth noting that the additional test result corroborated our findings that all three
markets, namely, the UK, Canadian, and German markets, are linked with the
Australian market in the long-run and they appeared to be significant at acceptable
levels.
Analysis 2
Here we have followed two test procedures using TWI adjusted data. The first followed
a structural regression method on the adjusted data, while the second followed the same
cointegration approach used in our first analysis. As we aim to verify if the selected
overseas stock markets are linked to the Australian market even when the foreign
exchange risk component is taken into consideration, we used the trade weighted index
(TWI) as proxy for our adjusted data series. We have considered the relative weight of
each country in the TWI composed by the Reserve Bank of Australia [Becker and
Davies (2002)]. The percentage weights in TWI for the US dollar, UK pound, Canadian
Mazharul Kazi Ph.D Thesis 195
dollar, and Japanese yen in 2002-2003 were 15.07, 5.14, 1.57, and 17.21 respectively.
Since both France and Germany now use common currency, the TWI weight for the
European euro was considered for these countries. The percentage weight of euro for
all common currency affiliated European countries (except the UK) was 12.40 in terms
of the Australian dollar. Since we are interested in the TWI for euro relevant to our
study, we have readjusted the weight based on the % trading relationships. The %
positions of the average international trades (exports and imports) between Australia
and both France and Germany are 13.8% and 23.4% respectively. Accordingly, our
adjusted TWI values for France and Germany become 1.7 (=12.4 × 13.8%) and 3.0
(=12.4 × 23.4%) respectively. For both analyses, the natural logarithms of data are
used.
Structural Approach
We considered a structural regression model in the following manner:
1
p
A j j
j
R T R T W I=
= + ∆∑ (8.08)
where AR is the return on the Australian market, jR refer to returns in selected overseas
stock markets, 1,( , )j pT t t= L is the weight of corresponding currency of selected
overseas markets in the TWI, and TWI∆ is expected depreciation of the TWI or error
term. Again, 1 1 2 2 n nt e t e t e TWI+ + + =∆L and
1 2
1
1n
i n
i
t t t t=
= + + + =∑ L for all currency
specific countries in TWI for Australia where ( , , )i i nT t t= L . The above model is
Mazharul Kazi Ph.D Thesis 196
derived from the notion of share-return parity reflected in the generalised nature
equation 1 2 1 1 2 2 1 1 2 2( ) ( ) ( )n A n n n nt t t R t R t R t R t e t e t e+ + + = + + + + + + +L L L .
To estimate equation 8.08 we have applied both the ordinary least squares (OLS) and
generalized method of moments (GMM) approaches. Both the OLS and GMM
estimations results in relation to equation 8.08 are presented in Tables 8.6 and 8.7:
TABLE 8.6
Dependent Variable: RAUSTRALIA
Method: Least Squares (OLS)
Sample(adjusted): 1946 2002
Included observations: 57 after adjusting endpoints
Variable Coefficient Std. Error t-Statistic Prob.
RUSA 0.028059 0.152705 0.183749 0.8550
RUK 0.104235 0.024912 4.184088 0.0001
RCanada 0.018028 0.005080 3.548912 0.0009
RFrance 0.224051 0.077894 2.876367 0.0059
RGermany 0.001662 0.001833 0.906372 0.3691
RJapan -0.042002 0.041872 -1.003104 0.3206
C 0.002888 0.003341 0.864475 0.3915
R-squared 0.568987 Mean dependent var 0.011297
Adjusted R-squared 0.517265 S.D. dependent var 0.031709
S.E. of regression 0.022031 Akaike info criterion -4.678109
Sum squared residual 0.024269 Schwarz criterion -4.427208
Log likelihood 140.3261 F-statistic 11.00096
Durbin-Watson stat 2.398591 Probability of F-statistic 0.000000
Mazharul Kazi Ph.D Thesis 197
TABLE 8.7 Dependent Variable: RAUSTRALIA Method: Generalized Method of Moments (GMM) Sample(adjusted): 1946 2002 Included observations: 57 after adjusting endpoints Kernel: Bartlett, Bandwidth: Fixed (3), No prewhitening Simultaneous weighting matrix and coefficient iteration
Convergence achieved after: 1 weight matrix, 2 total coefficient iterations Instrument list: RUSA RUK RCANADA RFRANCE RGERMANY RJAPAN
Variable Coefficient Std. Error t-Statistic Prob.
RUSA 0.028059 0.143685 0.195284 0.8460
RUK 0.104235 0.017845 5.841229 0.0000
RCanada 0.018028 0.005290 3.407572 0.0013
RFrance 0.224051 0.090421 2.477875 0.0166
RGermany 0.001662 0.000563 2.951147 0.0048
RJapan -0.042002 0.035899 -1.170002 0.2475 C 0.002888 0.002527 1.142799 0.2586
R-squared 0.568987 Mean dependent var 0.011297 Adjusted R-squared 0.517265 S.D. dependent var 0.031709 S.E. of regression 0.022031 Sum squared resid 0.024269 Durbin-Watson stat 2.398591 J-statistic 1.56E-32
The results of the OLS estimation provided in Table 8.6 indicate that coefficients of
three variables are significant at 5% levels based on t-statistics. These variables are the
stock market returns of the UK, Canada, and France. From the results of the GMM
estimation in Table 8.7 it appears that coefficients of four variables are significant at
5% levels. These markets are the UK, Canada, France and Germany. From the t-
statistics we can see that out of four markets, the UK market is dominating in the
process maintaining linear relationship. Although the R2 and adjusted R
2 for both tests
do not show very strong fits, other parameters, including the log likelihood, Durbin-
Watson, F-tests statistics and probability in both tests, are within satisfactory levels.
The above structural analysis gives us information about the linkages between the
Australian stock market and other developed overseas markets under globalization with
due consideration given to the exchange risk component. We understand that three to
Mazharul Kazi Ph.D Thesis 198
four overseas markets are linked with the Australian stock market. These are the UK,
Canada, France and Germany, with the UK market the most influential.
Autoregressive Approach
Using both the Augmented Dickey-Fuller and Phillips-Peron procedures, we have
utilised our adjusted series in natural logarithms to identify if they are integrated to
order one in levels and to order zero in first differences. We found all variables are I(1)
in levels and I(0) in first differences. We also tested for appropriate lag length using
standard techniques of likelihood ratio, Akaike Information Criterion, Schwartz
Bayesian Information Criterion, Hannan-Quinn Criterion and Final Prediction Error.
We observed that appropriate lag length is one.
We progress with our analysis using the cointegration approach of Johansen, same as
for our first analysis. Both trace and max-eigenvalue tests show the existence of
cointegrating relationship at the 5% levels. The tests also indicate that the linear
combination of all variables is cointegrated in the long-run while there are short-term
dynamics within the markets for error correction. The cointegration analysis in our
second analysis provides results that are different from our first analysis. From the t-
statistics it appears that within the cointegrating equation five markets are significant
for Australia at the 5% level. These are the US, the UK, Canada, Germany and Japan.
From the likelihood ratio (LR) tests it is revealed that only three markets are significant
between the 5% and 10% levels. These markets are the UK, Germany and Japan, with
the UK the most significant at the 10% level and remainder markets significant at the
Mazharul Kazi Ph.D Thesis 199
5% level. With respect to short-run dynamic adjustment, the Australian market is being
auto corrected at the speed of 9.4%. Still, the error correction mechanism ECM-1 for
Australia is insignificant.
Supplementary Tests
As the error correction mechanism ECM-1 for Australia is insignificant and not all
markets are sufficiently significant, we have further tested our model by dropping the
insignificant market/s. To find the error correction mechanism ECM-1 for Australia as
well as considered markets significant enough, we required three extra tests. The first
extra test indicated that although five overseas markets are cointegrated with the
Australian market and all variables are significant at 5% levels from their
corresponding t-statistics, the ECM-1 for Australia was insignificant. From LR test
results we found that the US market is insignificant as well. We therefore ran our model
again dropping the US from the equation. The second extra test result indicated that we
should run our model one more time dropping the Japanese market because it seemed
insignificant in addition to the insignificant error correction mechanism ECM-1 for
Australia. We therefore conducted our final test by considering only three overseas
markets, namely, the UK, Canada and Germany. We obtained the ECM-1 for Australia
as significant at 5% levels, while all variables are also significant at 5% levels from
both t-test and LR test statistics. The linear combination of the markets of Australia, the
UK, Canada and Germany are also found cointegrated. The results of all undertaken
cointegration tests in this part of the analysis are provided in Table 8.8.
Mazharul Kazi Ph.D Thesis 200
TABLE 8.8
Items
Test on Model 1 for Australia
Test on Model 2 for Australia
Test on Model 3 for Australia
Test on Model 4 for Australia
Variables in model
AUSTRALIA, USA, UK, CANADA, FRANCE, GERMANY, JAPAN
AUSTRALIA, USA, UK, CANADA, GERMANY, JAPAN
AUSTRALIA, UK, CANADA, GERMANY, JAPAN
AUSTRALIA, UK, CANADA, GERMANY
Test relationship
Cointegrated Cointegrated Cointegrated Cointegrated
Significance of ECM-1 for Australia
Not significant Not significant Not significant Significant
Overseas markets Significance from t- test at 5% level
USA, UK, CANADA, GERMANY, JAPAN
UK, CANADA, GERMANY, JAPAN
UK, CANADA, GERMANY, JAPAN
UK, CANADA, GERMANY
Overseas markets
Significance from LR test at 5% level
GERMANY,
JAPAN
UK, CANADA,
GERMANY, JAPAN
UK,
CANADA, GERMANY
UK,
CANADA, GERMANY
Overseas markets Significance from LR test at 10% level
UK, GERMANY, JAPAN
UK, CANADA, GERMANY, JAPAN
UK, CANADA, GERMANY
UK, CANADA, GERMANY
Australian markets significance from
LR test at 5% level
Not significant
Not significant
Significant
Significant
It appears from Table 8.8 that the result of the cointegration tests on Model 4 is very
satisfactory from every aspect. All modelled variables are cointegrated as well as
significant. The error correction mechanism ECM-1 for Australia is significant at 5%
levels from both LR and t-tests. The complete test results are provided in Table 8.9:
Mazharul Kazi Ph.D Thesis 201
TABLE 8.9
Vector Error Correction Estimates
Sample(adjusted): 1947 2002 Included observations: 56 after adjusting endpoints
Standard errors in ( ) and t-statistics in [ ]
Cointegrating Equation: Equation for Australia
AUSTRALIA (-1) 1.000000
UK(-1) -0.937789
(0.06728) [-13.9396]
CANADA(-1) 0.393356 (0.10603) [ 3.70977]
GERMANY(-1) -0.320276
(0.08100)
[-3.95390]
C -4.134737
Error Correction: D(AUSTRALIA) D(UK) D(CANADA) D(GERMANY)
Cointegrating Equation 1 -0.428136 0.083469 -0.466556 0.541700 (0.16804) (0.21344) (0.19543) (0.29481) [-2.54782] [ 0.39105] [-2.38730] [ 1.83746]
D(AUSTRALIA(-1)) 0.087336 -0.059234 -0.075148 -0.790582 (0.20737) (0.26340) (0.24117) (0.36381) [0.42116] [-0.22488] [-0.31159] [-2.17308]
D(UK(-1)) -0.127143 -0.050664 0.045764 0.735829
(0.17052) (0.21660) (0.19832) (0.29917)
[-0.74560] [-0.23390] [ 0.23076] [ 2.45958]
D(CANADA(-1)) -0.195267 -0.180547 -0.182048 0.061660
(0.13014) (0.16530) (0.15135) (0.22832) [-1.50045] [-1.09222] [-1.20280] [ 0.27006]
D(GERMANY(-1)) -0.005672 0.152919 0.144629 -0.064803 (0.08775) (0.11146) (0.10205) (0.15394) [-0.06464] [ 1.37200] [ 1.41721] [-0.42095]
C 0.085081 0.080176 0.080938 0.047939 (0.02677) (0.03400) (0.03113) (0.04696)
[ 3.17871] [ 2.35825] [ 2.60008] [ 1.02089]
Mazharul Kazi Ph.D Thesis 202
The final cointegration relationship can be seen in Graph 8.2.
GRAPH 8.2
Final Relationship for Australia under Prevailing Influences of Globalization
-.5
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
-.5
-.4
-.3
-.2
-.1
.0
.1
.2
.3
.4
1950 1960 1970 1980 1990 2000
Cointegrating relation between three significant overseas markets and the Australian market
The above results are in line with the common intuition about the present global
system. The significant stock markets in our cointegration analysis infer that the UK,
Canadian, and German stock markets are moving in the same way as Australias.
Australian investors have thus little scope to diversify their portfolios by including UK,
Canadian and German stocks when Australian stocks are bearish. In contrast, there will
be opportunities for effective diversification vis-a-vis the US, French and Japanese
markets. Investors should consider additional issues such as the cost of overseas
capital, stock values, and so on. Attention should therefore be focused on analyses of
fundamentals and related risk factors including foreign exchange risk, overseas
inflation, and specific industry risk factors.
Mazharul Kazi Ph.D Thesis 203
Summary
This chapter has examined the cointegrating relationship between Australian stock
price movements and that of six overseas markets using the Johansen cointegration
approach. The analysis confirmed that the Australian stock market has a long-run
relationship with overseas equity markets. The significant overseas markets for
Australia are the UK, Canada and Germany. The Australian stock markets error
correction mechanism is found to be significant at the 5% level.
Although developed stock markets are cointegrated with each other, Australian
investors can still diversify their portfolio risks by considering US, French and
Japanese stocks because these markets are insignificant for Australia in the
cointegrating equation. In addition to knowing about the nature of global stock market
co-movements, Australian investors diversifying their portfolios should consider issues
such as the cost of foreign capital, fundamental values of international stocks, foreign
exchange, and other relevant risks. Upon appropriate assessment of all risk factors,
including both systematic and unsystematic risks, prudent investors should be able to
form effective portfolios that include both local and international stocks and thus for
better returns.
Mazharul Kazi Ph.D Thesis 204
PART V
CONCLUSION
This is the last part of the thesis and consists of one chapter only. Chapter 9 discusses
our outcomes, provides some interpretations and makes recommendations for future
research.
Mazharul Kazi Ph.D Thesis 205
CHAPTER 9
CONCLUSION AND RECOMMENDATION
Outcomes
The aim of this thesis was twofold. First, we aimed to find out which systematic risk
factors are priced in the long-run return generating process in the Australian stock
market. Secondly, we attempted to ascertain if the Australian stock market is integrated
under the influence of globalization. To answer these two questions we focused on
finding reasonable empirical answers. The literature review of asset pricing led us to
ask the two questions. It is clear from the finance literature that the stock market and its
pricing behaviour is highly relevant to both investors and researchers. The performance
of any stock market will always be volatile. And yet, investors are faced with much
risks exposure. To survive the process and gain from their investment, investors need to
be efficient in managing their investment risk.
There have been many studies done on stock market volatility and associated topics.
The various theories provide clear information about the various types of risks
associated with investing in stocks or equity. Investment risk is of two generic types,
based on efficacy of diversification. The type of risk that can be diversified away is
called unsystematic risk, while the other type that cannot be diversified away is called
systematic risk. Unsystematic risk is also asset-specific or unique risk. The
unsystematic risk is negligible for a well diversified portfolio. Essentially all of the risk
Mazharul Kazi Ph.D Thesis 206
in such a portfolio is systematic. In this context, the systematic risk principle states that
the expected return on a risky asset depends only on the assets systematic risk as the
unsystematic risk is supposed to be eliminated by diversification. Although the
systematic risk is thought to be the market risk, the beta coefficient does not refer to the
entirety of systematic risk. The beta coefficient only refers to the amount of systematic
risk that is present in a particular risky asset relative to that of an average risky asset.
Diversifiable risk ought to be managed for efficiency and improved performance under
the competitive market environment. Non diversifiable risk, on the other hand, is
beyond the scope of a firms normal operations. Thus, studying the systematic risk of
an investment is an important issue to investors, one which requires special attention
their investment decision making.
The movement of stock prices has been the focus of many finance studies. Current
theories have already answered vital questions associated with asset pricing and
investment decision making. Still, answers to certain questions are inadequate and often
they require thorough research. We therefore aimed to extend current knowledge by
identifying systematic risk factors significant for the Australian stock market and
examining the nature of stock market integration under the influence of globalization.
We used the most up-to-date methods of empirical data analysis. We also used up-to-
date econometrics knowledge and software. An intensive literature review was
conducted to help identify the most common systematic risk factors. The empirical
investigation conducted to ascertain the causal relationship between Australian stock
Mazharul Kazi Ph.D Thesis 207
market returns and the common systematic risk factors and also the movements of
overseas markets under globalization was carried out through testing our generated
models. We acknowledged that the idea of causality of any nature must come from
reality, as empirical or statistical relationships per se cannot logically explain causation
[Gujarati (1995)]. Accordingly, to ascribe causality we opted to appeal to theory. In
assessing statistical relationships among variables, we essentially dealt with randomness
that is nonstationary by its nature and developed autoregressive models to that effect.
This research questions considered here evolved from both theory and real life
situations relevant to many investors and fund managers in the Australian marketplace.
Theoretical and practical aspects have been intensively explored to help answer the
questions of which systematic risk factors are priced in the Australian stock market and
whether there is any cross-country relationship between Australian and other developed
markets in the both long- and short-terms.
We saw from our reviews in Chapters 2, 3 and 4 that many researchers have dealt with
the problems of identifying systematic risk factors to explain stock returns. Most have
followed the Roll and Ross method, in line with arbitrage pricing theory (APT) rather
than the capital asset pricing model of Sharpe and Lintner. Researchers who conducted
studies in line with APT followed two different approaches in dealing with factors.
Some researchers considered the nature of common factors as latent, while others
preferred to pre-specify their variables by considering these factors as a priori. In this
research we treated our factors as pre-specified or a priori. We saw in the literature
review chapters that modern researchers have been attempting to find out if a priori
Mazharul Kazi Ph.D Thesis 208
variables can explain the returns generating process of a stock market considering the
comprehensive market index to proxy for their market returns.
It was also seen from the literature review in Chapters 3 and 4 that different researchers
used different methodologies in their analytical process, including factor analysis,
principal component analysis, general regression analysis, generalized method of
moments, time series regression analysis, causality analysis, and residual based as well
as autoregressive cointegration approaches. From available methods of cointegration
analysis, the lately developed approach of Johansen is considered more suitable over
earlier ones. This is because the cointegration approach of Johansen has the built-in
capacity to show both the long-run relationship between variables and the short-run
dynamics of the model. Thus, we used the Johansens cointegration method for the
purpose of both our analyses in Chapters 7 and 8, generating reliable results by
identifying which of our a priori variables are influencing the stock market returns in
both the long- and short-runs, and providing clarity about the dynamics of our models. The
outcomes are in line with both theory and general intuition in finance and investment.
In Chapter 7 we ascertained that at least four a priori variables are significant at the 5%
level, including the interest rate, corporate profit, dividend yield and industrial
production. Although the likelihood ratio test indicated that both the global stock
market index and price-earnings ratio are insignificant even at the 10% level, yet the
global stock market index was found significant at the 5% level, along with the interest
rate, corporate profit, dividend yield, industrial production, and price-earnings ratio
from the t-statistics. We observed that the linear combination of all our modelled
Mazharul Kazi Ph.D Thesis 209
variables is cointegrated even though not all variables are significantly influential.
We concluded that four to five of our a priori variables have significant long-run
influences on Australian stock price movements or the return generating process and
they are priced in the market.
The findings of our second analysis in Chapter 8 demonstrated that in longer run, the
Australian stock market is linked to these major developed markets: the US, the UK,
Canada, Germany, France and Japan. All were found cointegrated with the Australian
stock market. The likelihood ratio test result indicated that within the global stock
market set-up, the significant overseas markets for Australia are the UK, Canada and
Germany, with the UK being heavily influential.
Through the literature reviews of Chapters 2, 3, and 4 we established both theoretical
and practical knowledge about variables to proxy systematic risk factors. Initially we
identified a larger number of proxy variables, which we then reduced for efficient
modelling. We specified the a priori variables to consider in our appropriate model. In
Chapter 6, we developed suitable models in both structural and autoregressive fashions.
We tested our models empirically in Chapters 7 and 8. The outcomes of Chapter 7
showed us that Australian stock market price movements respond to changes in our a
priori variables and some of these systematic risk factors are influential in the long-run
while short-run dynamics are also present. We learnt from Chapter 8 about the nature
of the global stock market integration mechanism. We saw that the Australian stock
market is integrated with major overseas markets under the influence of globalization,
with some overseas markets more influential than others.
Mazharul Kazi Ph.D Thesis 210
Recommendations
We recommend that Australian investors and fund managers closely watch any sharp
movement in the a priori variables which we found influential for Australia. As the
scope of this thesis was necessarily limited, our outcomes do not suggest that
Australian investors invest exclusively on the basis of this study there are other
types of risks that also warrant proper investigation before making any investment
decision. Some of these risks are company specific, while others are industry specific.
Investors own studies on the fundamental valuation of assets and the costs of capital
are important for superior investment returns. The lessons of this thesis will only be
effective if investors treat the outcomes here as complementary to other types of
knowledge, knowledge which an Australian investor should pursue as part of their
ongoing investment decision making processes.
Future Research
Many other questions remain unresolved; these can be carried over for further study.
An interesting area of future study relates to the long-run relationship between
Australian and emerging stock markets. Additional research projects may be developed
on matters such as asset pricing bubbles and asset valuation problems, testing the
efficient market hypothesis for emerging markets, and how to set a universal best
practice norm for global stock trading.
Mazharul Kazi Ph.D Thesis 211
Bibliography
Abadir, K. M.; Hadri, K. and Tzavalis, E., 1999. The influence of VAR dimensions on estimator bias, Econometrica, vol. 67, pp.
163-181.
Abdullah, D. A. and Hayworth, S. C., 1993. Macroeconomics of stock price fluctuations, Quarterly Journal of Business and
Economics, vol. 32(1), pp. 49-63.
Abel, A. B., 1999. Risk premia and term premia in general equilibrium, Journal of Monetary Economics, vol. 43(1), pp. 3-33.
Abelson, P. and Joyeux, R., (Ed.), 2000. Economic Forecasting, Allen and Unwin, Australia.
Ackert, L. F. and Racine, M. D., 1999. Stochastic Trends and Cointegration in the Market for Equities, Journal of Economics and
Business, vol. 51, pp. 133-143.
Adams, P. and Parmenter, B., 2000. Forecasting the Australian economy: the role of the MONASH model, in: Abelson, P. and
Joyeux, R., Ed., Allen & Unwin, Australia.
Adams, P. D.; Dixon, P. B.; McDonald, D.; Meagher, G. A. and Parmenter, B. R., 1994. Forecasts for the Australian economy
using the MONASH model, International Journal of Forecasting, vol. 10, pp. 181-189.
Adler, M. and Dumas, B., 1983. International portfolio selection and corporation finance: A synthesis, Journal of Finance, vol. 46,
pp. 925-984.
Agarwal, R. and Rivoli, P., 1989. The relationship between the US and four Asian stock markets, ASEAN Economic Bulletin, vol.
6(1), pp. 110-117.
Ahn, S. K. and Reinsel, G. C., 1990. Estimation for partially non-stationary multivariate autoregressive models, Journal of
American Statistical Association, vol. 85, pp. 813-823.
Aiken, L. S. and West, S. G., 1991. Multivariate Regression: Testing and Interpreting Interactions, Saga Publication, CA, USA.
Aitken, M., et. al., 1996. Price clustering on the Australian stock exchange, Pacific-Basin Finance Journal, vol. 4 (2-3), pp. 297-
314.
Akaike, H., 1973. Information Theory and Extension of the Maximum Likelihood Principle, In: Petrov, B. N. and Csaki, F., Edt.
Proceeding of the Second International Symposium on Information theory, Budapest, pp. 267-281.
Akaike, H., 1974. A new Look at the Statistical Model Identification, IEEE Transactions on Automatic Control Annual
Conference, vol.19 (6), pp. 716-723.
Akaike, H., 1987. Factor analysis and AIC, Psychometrika, vol. 52, pp. 317-332.
Alexander, G. J. and Francis, J. C., 1986. Portfolio Analysis, Prentice Hall, USA.
Allen, D. E. and Prince, R., 1995. The Winner-Loser Hypothesis: some preliminary Australian evidence on the impact of changing
risk, Applied Economics Letters, vol. 2, pp. 280-283.
Anckonie, A. III., 1983. A contribution to the factor identification problem associated with the application of the arbitrage pricing
theory, A paper presented at the Eastern Finance Association Annual Meeting (21 April 1983), in Lee (1991), Advances in
Quantitative Analysis of Finance and Accounting, vol. 1 (A), pp. 25-43.
Anderson, D.; Lunch, A. and Mathiou, N., 1990. Behaviour of CAPM anomalies in small firms: Australian evidence, Australian
Journal of Management, vol. 15, pp. 1-38.
Anderson, T. W., 1951. Estimating linear restrictions on regression coefficients for multivariate normal distributions, Annals of
Mathematics and Statistics, vol. 22, pp. 327-351.
Anderson, T. W., 1984. An introduction to Multivariate Statistical Analysis, Wiley, New York.
Andersson, M. K. and Gredenhoff, M. P., 1999. On the maximum likelihood cointegration procedure under a fractional equilibrium
error, Economics Letter, No. 65, pp. 143-147.
Andrews, D. W. K., 1991. Heteroskedasticity and Autocorrelation Consistent Covariance Matrix Estimation, Econometrica, vol.
59(3), pp. 817-858.
Ankie, H.., 1997. Globalisation and the Postcolonial World The New Political Economy of Development, Macmillan, London,
pp. 114-131.
Apreda, R.., 1999. Dynamic arbitrage gaps for financial assets: in a nonlinear and chaotic price adjustment, Journal of
Multinational Financial Management, vol. 9, pp. 441-457.
Ardalan, K., 1999. The no-arbitrage condition and financial markets with transaction costs and heterogeneous information: The bid-
ask spread, Global Finance Journal, vol. 10 (1), pp. 83-91.
Arshanapalli, B., and Doukas, J., 1993. International stock market linkages: evidence from the pre- and post-October 1997 period,
Journal of Banking and Finance, 17, pp. 197-208.
Australian Stock Exchange (ASX), website: www.asx.com.au
Bac, K. andPlisa, S. R., 1991. On the fundamental theorem of asset pricing with a finite state space, Journal of Mathematical
Economics, vol. 20, pp. 1-18.
Bai, J. and Ng, S., 2002. Determining the number of factors in approximate factor models, Econometrica, vol. 70(1), pp. 191-221.
Bakshi, G. S. and Chen, Z., 1996. Asset pricing without consumption or market portfolio data, Working Paper, Ohio State
University.
Balcer, Y. and Loppman, S.A., 1984. Technological expectations and adoption of improved technology, Journal of Economic
Theory, vol. 34, pp. 292-318.
Balke, N. S., and Wynne, M. A., 2000. An equilibrium analysis of relative price changes and aggregate inflation, Journal of
Monetary Economics, vol. 45(2), pp. 269-292.
Ball, R. and Finn, F. J., 1989. The effect of block transactions on share prices: Australian evidence, Journal of Banking and
Finance, vol. 13, pp. 397-419.
Mazharul Kazi Ph.D Thesis 212
Ball, R.; Brown, P. and Finn, F. J., 1977. Share capitalisation changes, information and the Australian equity market, Australian
Journal of Management, vol. 2, pp. 105-125.
Baltagi, B. H., 1995. Econometric Analysis of Panel Data, John Wiley, UK.
Balvers, R. J.; Cosimano, T. F. and McDonald B., 1990. Predicting Stock Returns in an Efficient Market, Journal of Finance, vol.
45, pp. 1109-1128.
Banerjee, A.; Dolado, J. J.; Galbraith, J. W. and Hendry, D. F., 1993. Cointregration, Error Correction, and the Econometric
Analysis of Non-Stationary Data, Oxford University Press, Oxford.
Banz, R., 1981. The relationship between return and market value of common stock, Journal of Financial Economics, pp. 3-18.
Barone-Adesi, G., 1985. Atbitrage equilibrium with skewed asset returns, Journal of Financial and Quantitative Analysis, vol. 20,
pp. 299-313, in Faff (1993).
Been, W., Glosten, L. R. and Jagannathan, R., 1990. Predictable Variations in Stock Index returns, Journal of Finance, vol. 44, pp.
1177-1189.
Beenstock, M. and Chan, K. F., 1983. The factor structure of security returns in the UK, Working Paper, City University Business
School, London.
Beenstock, M. and Chan, K.F., 1984. Testing asset pricing theories in the UK securities market 1962-1981, Working Paper, City
University Business School, London.
Beggs, J. J., 1986. A simple exposition of the arbitrage pricing theory approximation, Australian Journal of Management, vol. 11,
pp. 13-22.
Bekaert, G. and Harvey, C.R., 1995. Time varying world market integration, Journal of Finance, vol. 1(2), pp. 403-444.
Bekaert, G. and Harvey, C.R., 1997. Emerging equity market volatility, Journal of Financial Economics, vol. 43, pp. 29-77.
Benjamin, G., 1973. The intelligent investor, 4th Ed., Harper and Row Publishers, New York.
Berkowitz, J. and Kilian, L., 2000. Recent Developments in Bootstrapping Time Series, Econometric Review, vol. 19, pp. 1-48.
Bernoulli, D., 1738. A selection of seminal articles in asset pricing and derivative valuation, in volume II of Dimdon and
Mussavian (Eds.), Foundation of Finance, Darmouth Publishing (1999), UK.
Berry, M. A.; Burmeister, E. and McElroy, M. B., 1988. Sorting out risks using known APT factors, Financial Analysts Journal,
vol. 44(2), pp. 29-42.
Black, F., 1972. Capital market equilibrium with restricted borrowing, Journal of Business, vol. 45 (3), pp. 444-455.
Black, F., 1974. International capital market equilibrium with investment barriers, Journal of Financial Economics, vo. 1 (4), pp.
337-352.
Black, F., 1986. Noise, Journal of Finance, vol. 41, pp. 529-543.
Black, F., 1993. Beta and Return, Journal of Portfolio Management, vol. 20(1), pp. 8-18.
Black, F.; Jensen, M. C. and Scholes, M. S., 1972. The capital asset pricing model: Some empirical tests, in Jensen, M.C. (Ed.),
Studies in the Theory of Capital Markets, Praeger, New York.
Black, F.; Jensen, M. C. and Scholes, M., 1972. The capital asset pricing model: Some empirical tests, in Jensen, Ed., Studies in the
theory of capital markets, Praeger, New York; Also in Fama and French (1993).
Blackman, S. C.; Holden, K. and Thomas, W., 1994. Long term relationship between international share prices, Applied Financial
Economics, vol. 4(4), pp. 297-304.
Bodie, Z., et. al., 1998. Essentials of Investments, 3rd Ed., McGraw-Hill, USA
Bossaert, P., 1988. Common Nonstationarity Components of Asset Prices, Journal of Economic Dynamics and Control, vol. 12, pp.
347-364.
Boswijk, H. P., 1996. Testing identifiability of cointegrating vectors, Journal of Business and Economic Statistics, vol. 14, pp. 153-
160.
Bower, D. H.; Bower, R. S. and Logue, D. E., 1992. A primer on Arbitrage Pricing Theory, In: Stern, J. M. and Chew, D. H., (Ed.),
The Revolution in Corporate Finance, Basil Blackwell, Cambridge, pp. 98-106.
Box, G. E. P. and Jenkins, G. M., 1970. Time Series Analysis, Forecasting and Control, San Francisco: Holden Day, in Engle and
Granger (1987).
Box, G. E. P. and Tiao, G. C., 1981. A canonical Analysis of Multiple Time Series with Applications, Biometrika, vol. 64, pp. 355-
365.
Brailford, T. and Easton, S., 1991. Seasonality in Australian share price indices between 1936 and 1957, Accounting and Finance,
vol. 31, pp. 69-85.
Brailford, T. J. and Faff, R. W., 1996. An evaluation of volatility forecasting techniques, Journal of Banking and Finance, vol. 20
(3), pp. 419-438.
Brailford, T. J. and Josev, T., 1997. The impact of the return interval on the estimation of systematic risk, Pacific-basin Financial
Journal, vol. 5(3), pp. 357-376.
Brailsford, T. and Heaney, R., 1998. Investments, Harcourt Brace, Sydney
Brealy, R. A., 1983. An Introduction to Risk and Return from Common Stocks, 2nd Ed., The MIT Press, Cambridge.
Breeden, D. T.; Gibbons, M. R. and Litzenberger, R. H., 1989. Empitical tests of the consumption based CAPM, Journal of
Finance, vol. 44, pp. 231-262.
Breen, W.; Glosten, L. R. and Jagannathan, R., 1990. Predictable Variations in Stock Index returns, Journal of Finance, 44,
pp.1177-1189.
Brennan, M. J., 1970. Taxes, market valuation and corporate financial policy, National Tax Journal, vol. 23, pp. 417-427.
Brennan, M. J.; Chordia, T. and Subrahmanyam, A., 1998. Alternative factor specifications, security characteristics, and the cross-
section of expected returns, Journal of Financial Economics, vol. 49, pp. 345-373.
Brenner, R. J. and Kroner, K. F., 1995. Arbitrage, cointegration, and testing the unbiasedness hypothesis in financial markets,
Journal of Financial and Quantitative Analysis, vol. 30 (1), pp. 23-42.
Brigham, E. F. and Gapenski, L. C., 1997. Financial Management: Theory and Practice, 8th Ed., Harcourt Publishers, USA.
Mazharul Kazi Ph.D Thesis 213
Brocato, J., 1994. Evidence on adjustments in major national stock markets linkages over the 1980s, Journal of Business Finance
and Accounting, vol. 21, pp. 643-647.
Brooks, C., 2002. Introductory Econometrics for Finance, Cambridge University Press, UK.
Brower, D.; Brower, R. and Logue, D., 1992. A primer on arbitrage pricing theory, In: Stern, J. M. and Chew, D. H., Ed., The
Revolution in Corporate Finance, Basil Blackwell, Oxford, pp. 98-106.
Brown, C. and Chan, R., 1999. Value-at-Risk Models and options Portfolios, Economic Papers, vol. 18(4), pp. 107-121.
Brown, D. H.; Bower, R. S. and Logue, D. E., 1984. A primer on arbitrage pricing theory, Midland Corporate Finance Journal,
vol. 2(3), pp. 31-40.
Brown, D. H.; Bower, R. S. and Logue, D. E., 1992. A primer on arbitrage pricing theory, In: Stern, J. M. and Chew, D. H. (Ed.),
The Revolution in Corporate Finance, 2nd (Ed.), Basil Blackwell, Cambridge, pp. 98-106.
Brown, S. J. and Otsuki, T., 1990. Macroeconomic factors and the Japanese equity markets: the CAPMD project, In: Elton and
Gruber (Eds.), Japanese capital markets, Harper and Row, New York.
Brown, S. J. and Otsuki, T., 1993. Risk premia in Pacific-Basin capital markets, Pacific_Basin Finance Journal, vol. 1 (3), pp. 235-
261.
Brown, S. J. and Weinstein, M. I., 1983. A New Approach to Testing Asset Pricing Models: The Bilinear Paradigm, Journal of
Finance, vol. 38, pp. 711-744.
Brown, S. J. and Weinstein, M. I., 1985. Derived Factors in Event Studies, Journal of Financial Economics, vol. 14, pp. 491-495.
Brown, S. J., 1989. The number of factors in security returns, Journal of Finance, vol. 44(5), pp. 1247-1252.
Brown, S. J.; Goetzmann, W. N. and Grinblatt, M., 1997. Positive Portfolio Factors, Preliminary Article, pp.1-21.
Bulkley, G. and Taylor, N., 1996. A cross-section test of the present value model, Journal of Empirical Finance, vol. 2 (4), pp.
295-306.
Callahan, C. M. and Mohr, R. M., 1989. The Determinants of Systematic Risk: A Synthesis, The Financial Review, vol. May, pp.
157-181.
Campbell, J. Y. and Hamao, Y., 1992. Predictable Stock Returns in the United States and Japan: A study of the long-term capital
integration, Journal of Finance, vol. 47, pp. 43-67.
Campbell, J. Y. and Shiller, R. J., 1987. Cointegration and tests of present value model, Journal of Political Economics, vol. 95, pp.
1062-1088.
Campbell, J. Y. and Shiller, R. J., 1988. Interpreting Cointegrated Models, Journal of Economic Dynamics and Control, vol. 12,
pp. 505-522.
Campbell, J. Y., 1987. Stock Returns and the Term Structure, Journal of Financial Economics, vol. 18, pp. 373-400.
Campbell, J. Y., 2003. A multivariate model of strategic asset allocation, Journal of Financial Economics, vol. 67 (1).
Campbell, J.A.; Lo, A. W. and Mackinlay, A.C., 1997. The Econometrics of Financial Markets, Princeton, Princeton University
Press, New Jersey.
Caporale, G. M.; Hassapis, C. and Pittis, N., 1998. Unit roots and long-run causality; investigating the relationship between output,
money and interest rates, Economic Modelling, vol. 15, pp. 91-112.
Cattell, R. B., 1971. Abilities Their Structure Growth and Action, Houghton Miffin, New York.
Cattell, R. B., 1978. The Scientific Use of Factor Analysis, Plenum, New York.
Cha, B. and Oh, S., 2000. The relationship between developed equity markets and the Pacific Basin emerging equity markets,
International Review of Economics and Finance, vol. 9(4), pp. 299-322.
Chamberlain, G., 1983. Funds, factors, and diversification in arbitrage pricing models, Econometrica, vol. 51(5), pp. 1305-1323.
Chamberline, G. and Rothschild, 1983. Arbitrage, factor structure, and mean variance analysis on large asset market,
Econometrica, vol. 51, pp. 1281-1304.
Chan, C. K. and Chen, N. F., 1991. Structural and return characteristics of small and large firms, Journal of Finance, vol. 46, pp.
1467-1484.
Chan, C. K.; Gup, B. E. and Pan, M. S., 1992. An empirical analysis of stock prices in major Asian markets and US, Financial
Review, vol. 27(2), pp. 289-307.
Chan, K.; Chen, N. and Hsieh, D., 1983. An exploratory investigation of the firm size effect, Working Paper, CRSP, University of
Chicago.
Chan, L. K. C. and Lakonishok, J., 1993. Are the Reports of Betas Death Premium? Journal of Portfolio Management, vol. 19(4),
pp. 51-62.
Chang, A. C. S., 1995. The UK stock market and economic factors: a new approach, Journal of Business Finance and Accounting,
vol. 22(1), pp. 129-142.
Chang, Y. and Park, J. Y., 2003. Index models with integrated time series, Journal of Econometrics, vol. 114, pp. 73-106.
Chen, B., 1995. Long-run purchasing power parity: Evidence from some European Monetary System Countries, Applied
Econometrics, vol. 27, pp. 377-383.
Chen, F. N., 1991. Financial investment opportunity and the macroeconomy, Journal of Business, vol.46, pp. 529-554.
Chen, N. F. and Ingersoll, J. E., 1983. Extra pricing in linear factor models with finitely many assets: A note, Journal of Finance,
vol. 38(3), pp. 985-988.
Chen, N. F.; Roll, R. and Ross, S. A., 1983. Economic forces and the stock market: testing the APT and alternative asset pricing
theories, Working Paper 119, CRSP, in Lee (1991), Advances in Quantitative Analysis of Finance and Accounting, vol. 1 (A),
pp. 25-43.
Chen, N. F.; Roll, R. and Ross, S. A., 1986. Economic forces and the stock market, Journal of Business, vol. 59(3), pp. 383-403.
Cheung, Y. L. and Mak, S. C., 1992. The international transmission of stock market fluctuation between the developed markets and
the Asian pacific markets, Applied Financial Economics, vol. 2(1), pp. 43-47.
Cheung, Y. W. and Lai, K. S., 1993. Finite sample sizes for Johansens Likelihood Ratio tests for Cointegration, Oxford Bulletin of
Economics and Statistics, vol. 55, pp. 313-328.
Mazharul Kazi Ph.D Thesis 214
Cheung, Y. W., and Ng, L. K., 1998. International evidence on the stock market and aggregate economic activity, Journal of
Empirical Finance, vol. 5, issue 3, pp. 281-296.
Chin, D. M., and Miller, P. J., 1998. Fixed vs. floating exchange rates: A dynamic general equilibrium analysis, European
Economic Review, vol. 42(7), pp. 1221-1249.
Cho, D.; Elton, E. and Gruber, M., 1984. On the robustness of the Roll and Ross arbitrage pricing theory, Journal of Finance and
Quantitative Analysis, vol. 19, pp. 1-10.
Chong, Y. T., and Hendry, D. F., 1986. Econometric evaluation of linear macroeconomic models, Review of Economic Studies, vol.
53(4), pp. 671-690.
Chow, G. C., 1960. Tests of equality between sets of coefficients in two linear regressions, Econometrica, vol. 28(3), pp. 591-605.
Christou, C. and Pittis, N., 1999. Forward versus reverse regression and cointegration, Economics Letters, vol. 65, pp. 157-163.
Chung, P. J. and Liu, D. J., 1994. Common stochastic trends in Pacific Rim stock markets, Quarterly Review of Economics and
Finance, vol. 34(3), pp. 241-259.
Clare, A. D., Mars, M. and Thomas, S. H., 1995. The integration and efficiency of international bond markets, Journal of Business
Finance and Accounting, vol. 22(2), pp. 313-322.
Clark, S. A., 2000. Arbitrage approximation theory, Journal of Mathematical Economics, vol. 33, pp. 167-181.
Cochrane, D. and Orcutt, G., 1949. Application of Least Squares Regression to Relationships Containing Autocorrelated Error
Terms, Journal of the American Statistical Association, vol. 44, pp. 32-61.
Cochrane, J. H., 1991. Production-Based Asset Pricing and the Link between Stock Return and Economic Fluctuations, Journal of
Finance, vol. 46, pp.209-238.
Cohen, D., 1998. The Wealth of the World and the Poverty of Nations, translated by Jacqueline Lindenfeld, The MIT Press,
Cambridge, pp. 53-54.
Connor, G. and Korajczyk, R. A., 1986. Performance measurement with the arbitrage pricing theory: A new framework for
analysis, Journal of Financial Economics, vol. 15, pp. 373-394.
Connor, G. and Korajczyk, R. A., 1988. Risk and return in an equilibrium APT: Application of new test methodology, Journal of
Financial Economics, vol. 21(2), pp. 255-289.
Connor, G. and Korajczyk, R. A., 1989. An intertemporal equilibrium beta pricing model, Review of Financial Studies, vol. 2, pp.
373-392.
Connor, G. and Korajczyk, R. A., 1993. A Test For the Number of Factors in Approximate Factor Model, Journal of Finance, vol.
48(4), pp. 1263-1292.
Connor, G. and Korajczyk, R. A., 1995. The arbitrage pricing theory and multifactor models of asset returns, In: Operations
Research and Management Science: Finance, North-Holland, Amsterdam, vol. 9, pp. 87-144.
Connor, G., 1984. A unified beta pricing theory, Journal of Economic Theory, vol. 34, pp. 13-31.
Connor, G., 1985. Arbitrage Pricing Theory: The Way Forward, Australian Journal of Management, vol. 10(1), pp. 109-130.
Connor, G., 1995a. The three types of factor models: A comparison of their explanatory power, Financial Analysts Journal, vol.
51, pp. 42-46.
Conway, G. and Reinganum, M. R., 1988. Stable factors in security returns: Identification using cross validation, Journal of
Business and Economic Statistics, vol. 6(1), pp. 1-15.
Copland, T. E. and Weston, J. F., 1988. Financial Theory and Corporate Policy, Reading, Wesley, USA
Corhay, A.; Tournai, R. and Urbain, J. P., 1993. Common stochastic trends in European stock markets, Economic Letters, vol. 42,
pp. 385-390.
Cox, J. C. and Ross, S. A., 1976. The valuation of options for alternative stochastic process, Journal of Financial Economics, vol.
7, pp. 229-263.
Cox, J. C.; Ingersoll, J. E. Jr. and Ross, S., 1985. An intertemporal general equilibrium model of asset prices, Econometrica, vol.
53(2), pp. 363-384.
Crowder, W. J., 1994. Foreign Exchange Market Efficiency and Common Stochastic Trends, Journal of International Money and
Finance, vol. 13, pp. 551-564.
Cuthbertson, K.; Hall, S. G. and Taylor, M. P., 1992. Applied Econometric Techniques, Harvester Wheatsheaf, New York.
Daniel, C., 1998. The Wealth of the World and the Poverty of Nations, translated by Jacqueline Lindenfeld, The MIT Press,
Cambridge.
Das, D. K., 1993. International Finance: Contemporary issues, Routledge, UK.
Daugaard, D., and Valentine, T., 1995. Financial Risk Management: A practical approach to derivatives, Harper Educational
Publishers, pp. 31-41.
Davidson, L. S. and Froyen, R.T., 1982. Monetary policy and stock returns: Is the stock market efficient? Federal Reserve Bank of
St Luis Review, No. 64, pp. 3-12.
Davis, K. and Lim, G. C., 1999. Managing Risks: What have we learnt? An Overview, Economic Papers, ESA, vol. 18(4), pp. 2-
6.
Debrue, G., 1959. Theory of Value, John Wiley and Sons, New York.
Delbaen, F. and Schachermayer, W., 1994. A general version of the fundamental theorem of asset pricing, Mathematical Annals,
vol. 300, pp. 463-520.
DeLima, P. J. F., 1997. On the robustness of nonlinearity tests to moment condition failure, Journal of Econometrics, vol. 76 (1-2),
pp. 251-280.
Dhakal, D., Kandil, M. and Sharma, S. C., 1993. Causality between the money supply and share prices: a VAR investigation,
Quarterly Journal of Business Economics, vol. 32(3), pp. 52-74.
Dhrymes, P. J., 1970. Econometrics: Statistical Foundations and Applications, Harper and Row, N.Y.
Dhrymes, P. J., 1998. Time Series, Unit Roots, and Cointegration, Academic Press, USA.
Mazharul Kazi Ph.D Thesis 215
Dhrymes, P. J.; Friend, I. and Gultekin, N. B., 1984. A critical reexamination of the empirical evidence in the Arbitrage Pricing
Theory, The Journal of Finance, vol. 39 (2), pp. 323-350.
Dhrymes, P. J.; Friend, I.; Gultekin, M. N. and Gultekin, N. B., 1985. New tests of APT and their implication, The Journal of
Finance, vol. 40 (3), pp. 659-675.
Diba, B. T. and Grossman, H. I., 1988. The theory of rational bubbles in stock prices, Economic Journal, vol. 98, pp. 746-754.
Dickey, D. A. and Fuller, W. A., 1979. Distribution of Estimators for Autoregressive Time Series Regression with a Unit Root,
Journal of the American Statistical Association, vol. 74, pp. 423-431.
Dickey, D. A. and Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time-series with a unit-root, Econometrica, vol.
49, pp. 1057-1072.
Dickey, D. A. and Pantula, S., 1987. Determining the order of differencing in autoregressive process, Journal of Business and
Economic Statistics, vol. 5, pp. 455-461.
Dickey, D. A.; Jensen, D. W. and Thornton, D. L., 1991. A primer on cointegration with an application to money and income,
Federal Reserve Bank of St. Louis Review, pp. 58-78.
Dimson, E. and Mussavian, M., 1999. Three centuries of asset pricing, Journal of Banking and Finance, vol. 23, pp. 1745-1769.
Dixon, P. B. and Permenter, B. R., 1990. Dynamics in ORANI: A large scale general equilibrium model, Working Paper, No. 1,
Institute of Applied Economic and Social Research, University of Melbourne, Australia.
Drew, M. E., 2000. Market efficiency, superannuation and funds management: an economic analysis of the performance and
persistence of Australian superannuation fund managers, PhD Thesis, Department of Economics, University of Queensland,
pp. 38-39.
Duffie, D., 1992. Dynamic Asset Pricing Theory, Princeton University Press, Princeton, N.Y.
Duffie, D., 1996. Dynamic Asset Pricing Theory, Princeton University Press, Princeton, N.Y.
Dumontier, M., 1986. Arbitrage pricing models: An empirical study on the Paris market, Finance, vol. 7, pp. 7-21.
Dunbar, C. G., 2000. Factors affecting investment bank initial public offering market share, Journal of Financial Economics, vol.
55, pp. 3-41.
Duy, T. A., and Thoma, M. A., 1998. Modeling and Forecasting Cointegrated Variables: some Practical Experience, Journal of
Economics and Business, vol. 50, pp. 291-307.
Dybvig, P. H. and Ross, S. A., 1985. Yes, the APT is Testable, Journal of Finance, vol. 40, pp. 1173-1188.
Dybvig, P. H. and Ross, S. A., 1987. Arbitrage, In: Eatwell, J., et al., (Ed.), The New Palgrave: Finance, MacMillan, New York.
Dybvig, P. H., 1983. An explicit bound on individual assets deviations from APT pricing in a finite economy, Journal of Financial
Economics, vol.12, pp. 483-496.
Easton, S. A. and Faff, R. W., 1994. An investigation of the robustness of the day-of-the-week effect in Australia, Applied
Financial Economics, vol. 4 (2), pp. 99-110.
Efron, B. and Tibshirani, R. J., 1993. An introduction to Bootstrap, Chapman & Hall, N.Y.
Ehrhardt, M. C., 1987. A Mean-Variance Derivation of Multi-Factor Equilibrium Model, Journal of Financial and Quantitative
Analysis, vol. 22(2), pp. 227-236.
Elder, J., 1997. Estimating the arbitrage pricing theory with observed macro factors, Economic Letters, vol. 55 (2), pp. 241-246.
Elliott, G.; Rothenberg, T. J. and Stock, J. H., 1996. Efficient Tests for an Autoregressive Unit Root, Econometrica, vol. 64(4), pp.
813-836.
Ellis, C. D., et al., (Ed.), 1989. Classics: Another Investors Anthology, Dow Jones-Irwin, Illinois, USA.
Ellis, C. D., et al., (Ed.), 1991. Classics II: Another Investors Anthology, Irwin, Illinois, USA.
Elton, E. J. and Gruber, M. J., (ed.), 1972. Security Evaluation and Portfolio Analysis, Prentice-Hall, Englewood Cliffs, New
Jersey, USA.
Elton, E. J., Gruber, M. J. and Blake, C., 1995. Fundamental economic variables, expected returns, and bond fund performance,
Journal of Finance, vol. 50, issue 4, pp. 1229- 1256.
Elton, E. J.; Gruber, M. J. and Blake, C., 1991. Modern Portfolio Theory and Investment Analysis, John Wiley, NY.
Enders, W., 1995. Applied Econometric Time Series, John Wiley & Sons, New York.
Engel, C., 1992. On the foreign exchange risk premium in a general equilibrium model, Journal of International Economics, vol.
32(3-4), pp. 305-319.
Engle, R. F. and Granger, C. W. J., 1987. Co-integration and Error Correction: Representation, Estimation and Testing,
Econometrica, vol. 55(2), pp. 251-276.
Engle, R. F. and Granger, C. W. J., 1991. Long-run Econometric Relations, Reading in Cointegration, Oxford University Press,
Oxford.
Engle, R. F. and Yoo, B. S., 1987. Forecasting and testing in cointegrated systems, Journal of Econometrics, vol. 35, pp. 143-159.
Engle, R. F. and Yoo, B. S., 1991. Cointegrated economic time series: an overview with new results, In: Engle, R. F. and Granger,
C. W. J. (Eds.), Long Run Economic Relationships, Oxford University Press, Oxford, pp. 237-266.
Engle, R. F.; Granger, C. W. J. and Hallman, J. J., 1989. Merging short- and long-run forecasts: An application of seasonal
cointegrating to monthly electricity sales forecasting, Journal of Econometrics, vol. 40(1), pp. 45-62. Also in Engle, R. F. and
Granger, C. W. J., (Ed.), Long-Run Economic Relationships: Readings in Cointegration, Oxford university Press, NY.
Epps, T. W. and Kramer, C. F., 1995. A variance test of the linear factor representation of returns and equilibrium pricing
restrictions, Working Paper, University of Virginia, USA.
Errunza, V., 1983. Emerging markets: A new opportunity for improving global portfolio performance, Financial Analysts Journal,
vol. 39, pp. 1311-1324.
Estep, T.; Hansen, N. and Johnson, C., 1983. Sources of value and risk in common stocks, Journal of Portfolio Management, vol.
9(4), pp. 5-13.
Eun, C. S. and Resnick, B. G., 1984. Estimating the correlation structure of international share prices, Journal of Finance, vol. 39,
pp. 1311-1333.
Mazharul Kazi Ph.D Thesis 216
EViews3, 1998. Users Guide, Quantitative Micro Software, USA.
Eviews4, 2001. Users Guide, Quantitative Micro Software, LLC, website: www.eviews.com
Faff, R. W. and Brailsford, T. J., 1999. Oil price risk and the Australian stock market, Journal of Energy Finance and
Development, vol. 4, pp. 69-87.
Faff, R. W. and Heaney, R., 1999. An examination of the relationship between Australian industry equity returns and expected
inflation, Applied Economics, vol. 31 (8), pp. 915-933.
Faff, R. W., 1988. An empirical test of the arbitrage pricing theory on Australian stock returns 1974-1985, Accounting and
Finance, November, pp. 23-43.
Faff, R. W., 1990. An empirical test of arbitrage equilibrium with skewed asset returns: Australian evidence, Working Paper No.
32, Department of Accounting and Finance, Monash University, in Faff (1993).
Faff, R. W., 1991. A likelihood ratio test of the zero-beta CAPM in Australian equity returns, Accounting and Finance, vol. 31, pp.
88-95.
Faff, R. W., 1992. A multivariate test of an equilibrium APT with time varying risk premia in Australian equity market, Australian
Journal of Management, December, vol. 17 (2), pp. 233-258.
Faff, R. W., 1993. An empirical test of arbitrage equilibrium with skewed asset returns: Australian evidence, Asia Pacific Journal
of Management, vol. 10 (2), pp. 195-211.
Fama, E. F. and French, K. R., 1988. Dividend yields and expected stock returns, Journal of Financial Economics, vol. 22, pp. 3-
25.
Fama, E. F. and French, K. R., 1989. Business conditions and expected returns on stocks and bonds, Journal of Financial
Economics, vol. 25, pp. 23-49.
Fama, E. F. and French, K. R., 1992. The economic fundamentals of size and book-to-market equity, Working Paper, Graduate
School of Business, University of Chicago, Chicago, IL, USA.
Fama, E. F. and French, K. R., 1992a. The cross-section of expected stock returns, Journal of Finance, vol. 47(2), pp. 427-465.
Fama, E. F. and French, K. R., 1992b. The economic fundamentals of size and book-to-market equity, Working Paper, Graduate
School of Business, University of Chicago, Chicago, Also in: Fama, E. F. and French, K. R., 1993.
Fama, E. F. and French, K. R., 1993. Common risk factors in the returns on stocks and bonds, Journal of Financial Economics, vol.
33(1), pp. 3-56.
Fama, E. F. and French, K. R., 1995. Size and book-to-market factors in earnings and returns, Journal of Finance, vol. 50, pp. 131-
155.
Fama, E. F. and French, K. R., 1996. Multifactor explanations of asset pricing anomalies, The Journal of Finance, March, vol. LI
(1), pp. 55-84.
Fama, E. F. and Gibbons, M., 1984. A comparison of inflation forecasts, Journal of monetary Economics, vol. 13, pp. 327-348.
Fama, E. F. and MacBeth, J. D., 1973. Risk, Return and Equilibrium: Empirical Tests, Journal of Political Economy, vol., 81 (3),
pp. 607-636.
Fama, E. F. and Schwert, W., 1977. Asset Returns and Inflation, Journal of Financial Economics, vol. 5, pp. 115-146.
Fama, E. F., 1965. The behaviour of stock market prices, Journal of Business, pp. 34-105.
Fama, E. F., 1968. Risk, return, and equilibrium: some clarifying comments, Journal of Finance, vol. 23(1), pp. 29-40. Also in:
Elton, E. J. and Grubar, M. J., 1972, (Ed.), Security Evaluation and Portfolio Analysis, Pentice Hall, USA, pp. 559-571.
Fama, E. F., 1981. Stock returns, real activity, inflation, and money, The American Economic Review, vol. 71, pp. 545-565.
Fama, E. F., 1990. Stock returns, expected returns and real activity, Journal of Finance, vol. 45, pp. 1089-1108.
Ferreira, E. and Regulez, M. 1996. A note on cointegration and control, Journal of Economic Dynamics & Control, vol. 20, pp.
963-966.
Ferson, W. E. and Harvey, C. R., 1991. The variation of economic risk premiums, Journal of Political Economy, vol. 99, pp. 385-
415.
Ferson, W. E. and Harvey, C. R., 1993. The risk and predictability of international equity returns, Review of Financial Studies, vol.
6, pp. 527-566.
Ferson, W. E. and Harvey, C. R., 1994. Sources of risk and expected returns in global equity markets, Journal of Banking and
Finance, vol. 18(4), pp. 775-803.
Ferson, W. E., 1990. Are the latent variables in time-varying expected returns compensation for consumption risk? Journal of
Finance, vol. 45, pp. 397-430.
Ferson, W. E.; Sarkissian, S. and Simin, T., 1999. The alpha factor asset pricing model: A parable, Journal of Financial Markets,
vol. 2, pp. 49-68.
Finn, M. G., 1999. An equilibrium theory of nominal and real exchange rate comovement, Journal of Monetary Economics, vol.
44(3), pp. 453-475.
Fisher, L. and Lorie, J. H., 1977. A Half Century of Returns on Stocks and Bonds, Chicago University Press, Chicago, USA.
Flannery, M, J. and James, C. M., 1984. The Effects of Interest Rate Changes on the Common Stock Returns of Financial
Institutions, The Journal of Finance, vol. 39, pp. 1141-1153.
Fogler, H. R.; John, K. and Tipton, J., 1981. Three factors, interest rate differentials and stock groups, Journal of Finance, May, pp.
323-335.
Frankfurter, G. M., and Phillips, H. E., 1996. Normative implications of equilibrium models: Homogenous expectations and other
artificialities, Journal of Economic Behaviour and Organization, vol. 31(1), pp. 67-83.
French, K. R.; Schwert, G. S. and Stambaugh, R. F., 1987. Expected stock returns and volatility, Journal of Financial Economics,
vol. 19, pp. 3-30.
Fry, M. J., 1997. In favour of financial liberalization, The Economic Journal, vol. 107, pp. 754-770.
Fuller, W. A., 1976. Introduction to Statistical Time Series, John Wiley, New York.
Mazharul Kazi Ph.D Thesis 217
Fung, H. G. and Lie, C. J., 1990. Stock market and economic activities: a causal analysis, In Khee and Chang (Eds.), Pacific-basin
capital market research, Amsterdam, North Holland.
Gehr, A. Jr., 1978. Some tests of the arbitrage pricing theory, Journal of the Midwest Finance Association, pp. 91-105.
George, B. and Nick, T., 1996. A cross-section test of the present value model, vol.2, issue 4, Journal of Empirical Finance, pp.
295-306.
George, T. J. and Hwang, C-Y., 1998. Endogenous market statistics and security pricing: An empirical investigation, Journal of
Financial Markets, vol. 1, pp. 285-319.
Geske, R. and Roll, R., 1983. The fiscal and monetary linkage between stock returns and inflation, Journal of Finance, vol. 38, pp.
1-33.
Geweke, J. and Zhou, G., 1996. Measuring the pricing error of the arbitrage pricing theory, Review of Financial Studies, vol. 9, pp.
553-583.
Gibbons, M. R. and Ferson, W., 1985. Testing asset pricing models with changing expectations and an unobservable market
portfolio, Journal of Financial Economics, vol. 14, pp. 217-236.
Gibbons, M. R., 1981. Empirical examination of the return generating process of arbitrage pricing theory, Working Paper, Stanford
Business School.
Gibbons, M. R., 1982. Multivariate tests of financial models: A new approach, Journal of Financial Economics, vol. 10, pp. 3-27.
Gibbons, M. R., Ross, S. A. and Shanken, J., 1989. A test of efficiency of a given portfolio, Econometrica, vol. 57 (5), pp. 1121-
1152.
Gilles, C. and LeRoy, S. F., 1991. On the Arbitrage Pricing Theory, Economic Theory, vol. 1(3), pp. 213-229.
Gjerde, O. and Saettem, F., 1999. Causal Relations Among Stock Returns and Macroeconomic Variables in a Small Open
Economy, Journal of International Financial Markets, Institutions and Money, vol. 9, pp. 61-74.
Goldsworthy, B.; Lewis, D., and Shuetrim, G., 2000. APRA and the Financial System Inquiry, Working Paper, No. 3, Australian
Prudential Regulation Authority, Australia.
Golsten, L. R.; Jagannathan, R. and Runkel, D. E., 1993. On the relation between the expected value and the volatility of the
nominal excess return on stocks, Journal of Finance, vol. 48(5), pp. 1779-1801.
Gonzalez-Rivera, G., 1998. Dynamic asset pricing and statistical properties of risk, Journal of Economics and Business, vol. 50 (5),
pp. 461-470.
Granger, C. W. J. and Konishi, T., 1992. Separation in cointegrated systems, Mimeo, University of California, San Diego.
Granger, C. W. J. and Lin, J., 1995. Causality in long run, Economic Theory, vol. 11, pp. 530-536.
Granger, C. W. J. and Newbold, P., 1974. Spurious Regressions in Econometrics, Journal of Econometrics, vol 2, pp. 111-120.
Granger, C. W. J. and Newbold, P., 1977. Forecasting Economic Time Series, New York Academic Press., USA.
Granger, C. W. J. and Weiss, A. A., 1983. Time Series Analysis of Error-Correcting Models, In: Karlin, S.; Amemiya, T. and
Goodman, L. A., (Ed.), Studies in Econometrics, Time Series and Multivariate Statistics, pp. 255-278, Academic Press, New
York.
Granger, C. W. J., 1969. Investigating Causal Relations by Econometric Models and Cross-Spectra. Methods, Econometrica, vol.
37(3), pp. 424-438.
Granger, C. W. J., 1981. Some properties of Time Series Data and Their Use in Econometric Specification, Journal of
Econometrics, vol. 16, pp. 121-130.
Granger, C. W. J., 1983. Cointegrated Variables and Error-Correcting Models, Discussion Paper, No. 83(13), University of
California, San Diego, USA.
Granger, C. W. J., 1986. The developments in the study of cointegrated economic variables, Oxford Bulletin of Economics and
Statistics, vol. 48, pp. 213-228.
Granger, C. W. J., 1988. Causality, cointegration and control, Journal of Economic Dynamics and Control, vol. 12, pp. 551-559.
Granger, C. W. J., 1991. Developments in the study of cointegrated economic variables, In: Engle, R. F. and Granger, C. W. J.,
(Ed.), Long-Run Economic Relationships: Readings in Cointegration, Oxford University Press, pp. 65-80.
Green, W. H., 2000. Econometric Analysis, 4th (Ed.), Prentice Hall International, USA.
Grinbatt, M. and Titman, S., 1983. Factor pricing in a finite economy, Journal of Financial Economics, vol.12(2), pp. 497-507.
Grubel, H. G., 1968. Internationally diversified portfolios: welfare gains and capital flows, American Economic Review, vol. 5, pp.
1299-1314.
Gruber, M., 1996. Another puzzle: The growth in activity managed mutual funds, Journal of Finance, vol. 51(3), pp. 783-810.
Gujarati, D., 1992. Essentials of Econometrics, McGraw-Hill Inc., USA, pp.379-398, 404-430.
Gujarati, D., 1995. Basic Econometrics, McGraw-Hill inc., USA, pp. 497-530, 540-576.
Gultekin, M. N. and Gultekin, N. B., 1983. Stock market seasonality: International evidence, Journal of Financial Economics, vol.
12, pp. 469-481.
Gultekin, N. B., 1983. Stock market returns and inflation: Evidence from other countries, Journal of Finance, vol. 38, pp. 49-65.
Hagastom, R. G. Jr., 1995. The Warren Buffett Way - Investment Strategies of the Worlds Greatest Investor, John Wiley, NY.
Hair, J.F.; Anderson, R. E. and Tatham, R. L., 1995. Multivariate Data Analysis with readings, Prentice Hall, USA.
Hall, S. G. and Wickens, M. R., 1993. Causality in integrated systems, Discussion Paper, No. 27-93, Centre for Economic
Forecasting, London Business School, UK.
Hamao, Y., 1986. An empirical examination of the arbitrage pricing theory using Japanese data, Working Paper, Yale School of
Management, Yale University, USA, in Das, Dilip K.,Ed., 1993.
Hamao, Y., 1988. An empirical examination of Arbitrage Pricing Theory: Using Japanese data, Japan and the World Economy, vol.
1, pp. 45-61.
Hamao, Y., 1988a. An empirical examination of Arbitrage Pricing Theory: Using Japanese data, Working Paper, University of
California, USA.
Mazharul Kazi Ph.D Thesis 218
Hamao, Y.; Masulis, R. and Ng, V., 1990. Correlation in price changes and volatility across international stock markets, Review of
Financial Studies, vol. 3, pp. 281-307.
Hamilton, J. D., 1989. A new approach to the economic analysis of nonstationary time series and the business cycle, Econometrica,
vol. 57(2). Pp. 357-387.
Hamilton, J. D., 1994. Time Series Analysis, Princeton University Press, New Jersey, USA.
Handa, P. and Linn, S. C., 1993. Arbitrage pricing with estimation risk, Journal of Finance and Quantitative Analysis, vol. 28 (1),
pp. 81-100.
Hansen, G. and Kim, J. R., 1996. The reliability of the Johansen procedure, Discussion Paper, No. 91/1996, Institute of Statistics
and Econometrics, Christian Albrechts University at Kiel, Germany.
Hansen, G.; Kim, J. R. and Mittnik, S., 1998. Testing cointegrating coefficients in vector autoregressive error correction models,
Economics Letters, No. 58, pp. 1-5.
Hansen, L. P. and Jagannathan, R., 1991. Implications of security market data for models of dynamic economies, Journal of
Political Economy, vol. 99, pp. 225-262.
Hansen, L. P. and Jagannathan, R., 1997. Assessing specification errors in stochastic discount factor model, Journal of Finance,
vol. 52, pp. 557-590.
Hansen, L. P., 1982. Large sample properties of the generalized method of moments estimators, Econometrica, vol. 50, pp. 1029-
1054.
Hansen, P. R. and Johansen, S., 1998. Workbook on cointegration, Oxford University Press, Oxford.
Harbo, I.; Johansen, S.; Nielsen, B. and Rahbek, A., 1998. asymptotic inference on cointegrating rank in partial systems, Journal of
Business Economics and Statistics, vol. 16, pp. 388-399.
Harcourt College Publishers, Online data service, web: www.economagic.com
Harris, R. J., 1985. A Primer of Multivariate Statistics, Academic Press, New york.
Harrison, J. M. and Kreps, D. M., 1979. Martingales and arbitrage in multi-period securities markets, Journal of Economic Theory,
vol. 20, pp. 381-408.
Harvey, C. R. and Kirby, C. M., 1995. Analytical tests of factor pricing models, Working Paper, Duke University, Durham, NC.
Harvey, C. R. and Kirby, C. M., 1995a. Analytical Tests of Factor Pricing Models, Preliminary Article, pp.1-39.
Harvey, C. R., 1989. Time-varying conditional covariances in tests of asset pricing models, Journal of Financial Economics, vol.
24, pp. 289-317.
Harvey, C. R., 1990. On the linearity of conditionally expected market returns, Working Paper, Duke University.
Harvey, C. R., 1991. The world price of covariance risk, Journal of Finance, vol. 46, pp. 111-157.
Hassapis, C.; Pittis, N. and Prodromidis, H., 1999. Unit root and Granger causality in the EMS interest rates: the German
Dominance Hypothesis revisited, Journal of International Money and Finance, vol. 18, pp. 47-73.
Hawawini, G., 1984. European equity markets: Price behaviour and efficiency, Monograph Series in Finance and Economics,
Salmon Brothers Center for the Study of Financial Institutions, New York University, USA.
Hawawini, G., 1993. Market efficiency and equity pricing: international evidence and implications for global investing, in Das,
D.K., 1993.
He, J. and Ng, L. K., 1994. Economic factors, fundamental variables and equity returns, Journal of Business, vol. 67, pp. 599-609.
He, J.; Kan, R.; Ng, L. and Zahang, C., 1996. Tests of the relationships among economic factors, fundamental variables and equity
returns using multi-factor models, Journal of Finance, vol. 51, pp. 1891-1908.
Heaney, R. and Hooper, V., 1998. An examination of the variation in equity market returns and volatility in Asia Pacific region, A
Seminar Paper, presented at the First Semester Seminar Series 1998 of Department of Accounting and Finance, Macquarie
University, Sydney.
Hendry, D. A. P. and Sargan, P., 1984. Dynamic Specifications, Chapter 18, in Handbook of Economics, vol. II, ed. Griliches, Z.
and Intriligator, M., North Holland.
Hendry, D. F. and Juscilius, K., 2000. Explaining Cointegration Analysis, Part I, Energy Journal, vol. 21, pp. 1-41.
Hermes, N. and Lensink, R., 2000. Financial system development in transition economies, Journal of Banking and Finance, vol.
24, pp. 507-524.
Hewarathna, R. and Silvapulle, P., 1996. An empirical investigation of the relationship among real, monetary and financial
variables: Australian evidence, Discussion Papers, Series A, No. 96.04, School of Economics and Commerce, Latrobe
University, Australia.
Hicks, J. R., 1946. Value and Capital, Oxford University Press, Oxford.
Hiemstra, C. and Jones, J. D., 1997. Another look at long memory in common stock returns, Journal of Empirical Finance, vol. 4
(4), pp. 373-401.
Hirt, G. A. and Block, S. B., 1996. Fundamentals of Investment Management, 5th Ed., McGraw-Hill, USA.
Hoffmann, M., 2001. Long run recursive VAR modeling and QR decompositions, Economics Letters, vol. 73, pp.15-20.
Hoogvelt, A., 1997. Globalisation and the postcolonial World The new political economy of development, Macmillan, London,
pp. 114-131.
Horn, R. V., 1993. Statistical Indicators for Economic & Social Sciences, Cambridge University Press, UK.
Huang, R. D. and Jo, H., 1995. Data frequency and the number of factors in stock returns, Journal of Banking and Finance, vol. 19
(6), pp. 987-1003.
Huang, R. D. and Kracaw, W. A., 1984. Stock market returns and real activity: a note, Journal of Finance, vol. 39, pp. 267-273.
Hubarman, G., 1982. A simple approach to arbitrage pricing, Journal of Economic Theory, vol. 28, pp. 183-191.
Huffman, G. W., 1985. Adjustment costs and capital asset pricing, The Journal of Finance, July, vol. XL (3), pp. 691-709.
Hughes, P. J., 1982. A test of the arbitrage pricing theory, Working Paper, University of British Colombia.
Hughes, P. J., 1985. A Test of Arbitrage Pricing Theory Using Canadian Security Returns, Canadian Journal of Admin. Science,
vol. 1, No. 2, pp.195-214.
Mazharul Kazi Ph.D Thesis 219
Hylleberg, S.; Engle, R. F.; Granger, C. W. J. and Yoo, B. S., 1990. Seasonal integration and cointegration. Journal of Economics,
vol. 44, pp. 215-238.
Ingersoll, J. E., 1987. Theory of Financial Decision Making, In: Rowman and Littlefield, New York.
Jagannathan, R. and Viswanathan, S., 1988. Linear factor pricing, term structure of interest rates and the small firm anomaly,
Working Paper, No. 57, Department of Finance, Northwestern University, Evanstone, Il, USA.
Jagannathan, R. and Wang, Z., 1993. The CAPM is Alive and Well, unpublished paper, Carlton School of Management, University
of Minisota, Mineapolis, MN.
Jagannathan, R. and Wang, Z., 1996. The conditional CAPM and the cross-section of expected returns, Journal of Finance, vol. 51,
pp. 3-53.
Janakiramannan, S. and Lamba, A. S., 1998. An empirical examination of linkages between Pacific-Basin stock markets, Journal of
International Financial Markets, vol. 8, pp. 155-173.
Janet, L., 1994. Benjamin Graham on value investing: Lesson from the Dean of Wall Street, Dearborn Financial Publishing, USA.
Jobson, J. D., 1982. A Multivariate Linear Regression Test for the Arbitrage Pricing Theory, The Journal of Finance, vol. 37 (4),
pp. 1037-1042.
Joen, B. N. and Chiang, T. C., 1991. A system of stock prices in world stock exchanges: common stochastic trends for 1975-1990?
, Journal of Economics and Business, vol. 43, pp. 329-338.
Johansen, S. and Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration with application to the demand
for money, Oxford Bulletin of Economics and Statistics, vol. 52, pp. 169-210.
Johansen, S. and Juselius, K., 1991. Some structural hypotheses in a multivariate cointegration analysis of the purchasing power
parity and uncovered interest parity for UK, Manuscript for the Journal of Econometrics.
Johansen, S. and Juselius, K., 1992. Testing structural hypotheses in a multivariate cointegration analysis of the PPP and UIP for
UK, Journal of Econometrics, vol. 53, pp. 211-244.
Johansen, S. and Juselius, K., 1994. Identification of the long-run and short-run structure: An application to the IS/LM model,
Journal of Econometrics, vol. 63, pp. 7-36.
Johansen, S. and Lando, D., 1996. Multi-period models as cointegration models, University of Copenhagen.
Johansen, S. and Schaumberg, E., 1999. Likelihood analysis of seasonal cointegration, Journal of Econometrics, vol. 88, pp. 301-
339.
Johansen, S. and Swensen, A. R., 1999. Testing some exact rational expectations in vector autoregressive models, Journal of
Econometrics, vol. 93, pp. 73-91.
Johansen, S., 1988. Statistical analysis of cointegration vectors, Journal of Economic Dynamics and Control, vol. 12, pp. 231-254.
Johansen, S., 1988a. The mathematical structure of error correction models, Contemporary Mathematics, vol. 80, pp. 259-386.
Johansen, S., 1990. A representation of vector autoregressive processes integrated of order 2, Manuscript for the Journal of
Econometric Theory.
Johansen, S., 1991. Estimation and Hypothesis Testing of Cointegration Vectors in Gaussian Vector Autoregressive Models,
Econometrica, vol. 59(6), pp. 1551-1580.
Johansen, S., 1991a. The Power Function for the Likelihood Ratio Test for Cointegration, In: Gruber, J., Ed., Economic Decision
Models: New Methods of Modeling and Application.
Johansen, S., 1991b. The statistical analysis of I(2) variables, Discussion Paper, University of Copenhagen.
Johansen, S., 1992. Determination of cointegration rank in the presence of a linear trend, Oxford Bulletin of Economics and
Statistics, vol. 54(3), pp. 383-397.
Johansen, S., 1992a. Cointegration in partial system and the efficiency of single equation analysis, Journal of Econometrics, vol.
52, pp. 389-402.
Johansen, S., 1994. The role of the constant and linear terms in cointegration analysis of nonstationary time series, Econometric
Reviews, vol. 13, pp. 205-231.
Johansen, S., 1995. Likelihood-based Inference in Cointegrated Vector Autoregressive Models, Oxford University Press, Oxford.
Johansen, S., 1995a. Identifying Restrictions of Linear Equations with Applications to Simultaneous Equations and Cointegration,
Journal of Econometrics, vol. 69, pp. 111-132.
Johansen, S., 1996. Likelihood based inference in Cointegrated Vector Autoregressive Models, Oxford University Press, Oxford.
Johansen, S., 2000. Modelling of cointegration in the vector autoregressive model, Economic Modelling, vol. 17, pp. 359-373.
Johnston, J., 1984. Econometric Methods, McGraw-Hill, New York.
Jones, D., 2000. Emerging problems with the Basel Capital Accord: Regulatory capital arbitrage and related issues, Journal of
Banking and Finance, vol. 24, pp. 35-58.
Joreskog, K. G. and Goldberger, A. S., 1975. Estimation of a model with multiple indicators and multiple causes of a single latent
variable, Journal of American Statistical Association, vol. 70, pp. 631-639.
Joreskog, K. G., 1977. Factor analysis by least squares and maximum likelihood methods, In: Enslein, K.; Ralslon, A. and Wilf, H.
S. (Edt.), Statistical Methods for Digital Computers, Wiley, New York.
Juttner, D. J., 1995. International Finance and Global Investments, 3rd Ed., Longman.
Kaiser, H. F., 1958. The varimax criterion for analytic rotation in factor analysis, Psychometrika, vol. 23, pp. 187-200.
Kasa, K., 1992. Common stochastic trends in international stock markets, Journal of Monetary Economics, vol. 29, pp. 95-124.
Kazi, M. H., 1997. Economic development in regional perspective: policy implication for Australia, M.Com (Hons.) Thesis, School
of Economics & Finance, UWS, Australia.
Kim, D. and Lee, C., 1995. On the number of factors and the role of market returns in linear factor models, in Lee, Cheng-few, Ed.,
Advances in Quantitive Analysis of Finance and Accounting, vol. 3(A), JAI press Inc., USA, pp. 15-47.
King, B. F., 1966. Market and industry factors in stock price behaviour, Journal of Business, vol. 39, pp. 139-190.
Kleibergen, F. and VanDick, H., 1994. Direct cointegration testing in error correction model, Journal of Economics, vol. 63(1), pp.
61-103.
Mazharul Kazi Ph.D Thesis 220
Kline, P., 1997. An easy guide to factor analysis, Routledge, London.
Koch, R., 1994. The investors guide to selecting shares that perform: ten ways that work, Financial Times, Pitman Publishing,
UK.
Koedijk, K. G. and Wolft, C. C. P., 1996. Exchange rate returns, news, and risk premia, Economic Letters, vol. 50(1), pp. 127-134.
Kolb, R. W. and Rodriguez, R. J., 1996. Financial Management, 2nd Ed., Blackwell, USA.
Konishi, T.; Ramey, V. A. and Granger, C. W. J., 1993. Stochastic trends and short-run relationships between financial variables
and real activity, Working Paper, No. 4275, NBER.
Kornish, Laura J., 1999. On optimal replacement thresholds with technological expectations, Journal of Economic Theory, vol. 89,
pp. 261-266.
Kreps, D. M., 1981. Arbitrage and equilibrium in economics with infinitely many commodities, Journal of Mathematical
Economics, vol. 8, pp. 15-35.
Kritzman, M., 1993. What practitioners need to know about factor models, Financial Analysts Journal, vol. May-June, pp. 14-17.
Kritzman, M., 1995. The Portable Financial Analyst- what practitioners need to know, In: The Financial Analysts Journal, The
Association of Investment Management and Research, Probus Publishing, USA, pp. 103-111.
Kritzman, M., 1995a. The Problem Financial Analyst: What Practitioners Need to Know, Probus Publishing, USA.
Kruskal, J. B., 1978. Factor analysis and principal components, In: Kruskal, W.H. and Tanur, J.M., International Encyclopedia of
Statistics, The Free Press, pp. 307-328.
Kupper, E. F., 1999. Risk Management in Banking, APRA, Australia.
Kwaitkowski, D.; Phillips, P. C. B.; Schmidt, P. and Shin, Y., 1992. Testing the Null Hypothesis of Stationarity Against the
Alternative of a Unit Root, Journal of Econometrics, vol. 54, pp.169-178.
Kwan, C. C.Y., 1999. A note on market-neutral portfolio selection, Journal of Banking and Finance, vol. 23 (5), pp. 773-800.
Kwon, C. S. and Shin, T. S., 1999. Cointegration and Causality between Macroeconomic Variables and Stock Market Returns,
Global Finance Journal, vol. 10, pp.71-81.
Kwon, C. S.; Shin, T. S. and Bacon, F. W., 1997. The effect of macroeconomic variables on stock market returns in developing
markets, Multinational Business Review, vol. 5(2), pp. 63-70.
Lajeri, F. and Dermine, J., 1999. Unexpected inflation and Bank Stock Returns: The Case of France 1977-1991, Journal of Banking
and Finance, vol. 23, pp. 939-953.
Lardaro, L., 1993. Applied Econometrics, Harper Collins College Publishers, USA.
Lau, S-H. P., 1999, I(0) ln: integration and cointegration out: Time series properties of endogenous growth models, Journal of
Econometrics, vol. 93, pp. 1-24.
Lawley, D. N. and Maxwell, A. E., 1971. Factor Analysis as a Statistical Method, 2nd Ed., Butterworth, UK.
Leamer, E. E., 1983. Lets Take the Con out of Econometrics, American Economic Review, vol. 23, No. 1, pp. 31-43.
Lehmann, B. N. and Modest, D., 1988. The Empirical Foundation of the Arbitrage Pricing Theory, Journal of Financial
Economics, vol. 21(2), pp. 213-254.
Leong and Felmingham, 2001. The interdependence of stock markets in the developed economies of East Asia, Draft Paper,
School of Economics, University of Tasmania, Australia.
Leudecke, B. P., 1984. Arbitrage pricing theory: The way forward? Working Paper, Australian Graduate School of Management,
In: Faff (1988).
Levy, H. and Sarnat, M., 1970. International diversification of investment portfolios, American Economic Review, vol. 60, pp. 668-
675.
Lewellen, J., 1999. The time-series relations among expected return, risk and book-to-market, Journal of Financial Economics, vol.
54(1), pp. 5-43.
Li, H. and Maddala, G. S., 1996. Bootstraping Time Series Models, Economic Reviews, vol. 15, pp. 115-158.
Li, Y., 1998. Expected stock returns, risk premiums and volatilities of economic factors, Journal of Empirical Finance, vol. 5(2),
pp. 69-97.
Lim, G. C. and Kells, S. S., 1995. Portfolio implications of an equity rain in Australia, The Economic Record, vol. 71, pp. 367-378.
Lim, G. C. and McNelis, P. D., 1998. The Effect of the Nikkei and the S&P on the All-Ordinaries: A Comparision of Three
Models, International Journal of Finance and Economics, vol.3, pp. 217-228.
Linn, S. and Chong, E., 1985. The empirical implications of approximate factor structure in large asset market, Working Paper,
University of Iowa.
Lintner, J., 1965. The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets, Review
of Economics and Statistics, vol. 47(1), pp. 13-37.
Lintner, J., 1965a. Security prices, risk, and maximal gain from diversification, Journal of Finance, vol. 20(4), pp. 387-615.
Litterman, R. B. and Weis, L. M., 1985. Money, Real Interest Rates, and Outputs: A Reinterpretation of Postwar US Data,
Ecinometrica, vol. 53, pp. 129-156.
Loewenstein, M. and Willard, G. A., 2000. Rational equilibrium asset-pricing bubbles in continuous trading models, Journal of
Economic Theory, vol. 91, pp. 17-58.
Lutkepohl, H. and Saikkonen, P., 1997. Testing for the cointegrating rank of a VAR process with a time trend, Discussion Paper,
No. 79, Humboldt-University, Berlin, Germany
Lutkepohl, H. and Saikkonen, P., 1999. A lag augmentation test for the cointegrating rank of a VAR process, Economics Letters,
vol. 63, pp. 23-27.
Lutkepohl, H. and Saikkonen, P., 2000. Testing for the cointegrating rank of a VAR process with a time trend, Journal of
Econometrics, vol. 94, pp. 177-198.
Lutkepohl, H., 1993. Introduction to multiple time series analysis, Springer Verlag, New York.
Ma, C. K. and Kao, G. W., 1990. On exchange rate changes and stock price reactions, Journal of Business Finance and
Accounting, vol. 17(3), pp. 441-449.
Mazharul Kazi Ph.D Thesis 221
MacBeth, J. D., 1975. Tests of the two parameter model of capital market equilibrium, PhD Dissertation, University of Chicago,
IL.
MacKinlay, A. C. and Richardson, M. P., 1991. Using generalized method of moments to test mean-variance efficiency, Journal of
Finance, vol. 46, pp. 511-527.
MacKinlay, A. C., 1995. Multifactor models do not explain deviations from the CAPM, Journal of Financial Economics, vol. 38,
pp. 3-28.
MacKinnon, J. G., 1990. Critical values for Cointegration Tests, In Modelling Long-Run Economic Relationships, Engle and
Granger (ed.), Oxford University Press.
MacKinnon, J. G., 1991. Critical values for cointegration tests, In: Engle, R. F. and Granger, C. W. F. (Eds.), Long-Run Economic
Relationships, Oxford University Press, Oxford.
Maldanado, R. and Saunders, A., 1981. International portfolio diversification and the inter-temporal stability of international stock
market relationships, Financial Management, vol. 10, pp. 54-63.
Mallik, G., Walker, G. and Sim, S., 2001. Inflation, interest rates and bank stock prices: a test for four Singapore banks,
unpublished paper, corresponding author: Mallik, G., [email protected].
Maloney, G., 1999. Risk-Lessons from the stock market, Economic Papers, ESA, vol. 18(4), pp. 134-146.
Mankiw, N. G.; Romer, D. and Shapiro, M. D., 1985. An unbiased relaxation of stock market volatility, Journal of Finance, vol.
XL (3), pp. 677-689.
Markowitz, H. M., 1952. Portfolio Selection, Journal of Finance, vol. 7, pp. 77-91.
Markowitz, H. M., 1959. Portfolio Selection: Efficient Diversification of Investments, Wiley, New York.
Markowitz, H. M., 1983. Nonnegative or Not Negative: A question about CAPM, Journal of Finance, vol. 38(2), pp. 283-295.
Markowitz, H. M., 1990. Foundation of portfolio theory, in Maler, K-G., 1992, Nobel Lectures in Economic Sciences 1981-1990,
World Scientific, USA, pp. 279-287.
Marshall, D. A., 1992. Inflation and Asset Returns in Monetary Economy, Journal of Finance, vol. 47, pp. 1315-1342.
Masih, A. M. M. and Masih, R., 1997. A comparative analysis of the propagation of stock market fluctuations in alternative models
of financial links, Applied Financial Economics, vol. 7, pp. 59-74.
Masih, A. M. M. and Masih, R., 1997a. Dynamic linkages and the propagation mechanism driving major international stock
market: An analysis of the pre- and post- crash eras, Quarterly Review of Economics and Finance, vol. 37(4), pp. 859-885.
Mayadunne, G., et al., 1995. An empirical investigation of stock persistence in economic time series, The Economic Record, vol.
17, pp. 145-156.
Mayers, D., 1972. Non-marketable assets and capital market equilibrium under uncertainty, in Jensen, M.C. (Ed.), Studies in the
Theory of Capital Markets, Praeger, New York.
Mayo, H. B., 1997. Investments: An Introduction, 5th Ed., The Dryden Press, USA.
Maysami, R. C. and Kho, T. S., 2000. A vector error correction model of the Singapore stock market, International Review of
Economics and Finance, vol. 9, pp. 79-96.
McAleer, M.; Pagan, A. R., and Volker, P. A., 1985. What Will Take the Con Out of Econometrics? American Economic Review,
vol. 75, No. 3, pp. 293-307.
McCulloch, R. and Rossi, P. E., 1991. A bayesian approach to testing the arbitrage pricing theory, Journal of Econometrics, vol. 49
(1-2), pp. 141-168.
McElroy, M. B. and Burmeister, E., 1988. Arbitrage Pricing Theory as a restricted nonlinear multivariate regression model, Journal
of Business and Economic Statistics, vol. 6(1), pp. 29-42.
McGowan, C. B. and Francis, J. C., 1991. Arbitrage pricing theory factors and their relationship to macroeconomic variables, in
Lee, Cheng F., Ed., Advances in Quantitative Analysis of Finance and Accounting, vol. 1(A), JAI Press Inc., USA, pp. 25-43.
McKenzie, J., et. al., 1995. MINITAB for Windows- Statistical Software (Student Edn.), USA.
McMillan, D. G., 2001. Cointegration Relationships between Stock Market Indices and Economic Activity: Evidence from US
Data, Discussion Paper, Issue No. 0104, Centre for Research into Industry, Enterprise, Finance and the Firm (CRIEFF),
University of St. Andrews, Scotland.
Merrill Lynch HSBC Australia Pty Ltd, 2001. Local Market (ASX) Indices, website: www.mlhsbc.com.au
Merton, R. C., 1973. An intertemporal capital asset pricing model, Econometrica, vol. 41, pp. 867-887.
Merton, R. C., 1977. A reexamination of capital asset pricing model, In: Friend, I. and Miller, J. (Eds.), Risk and Returns in
Finance, Ballinger, Cambridge.
Merton, R. C., 1980. On estimating the expected return on the market: an explanatory investigation, Journal of Financial
Economics, vol. 8, pp. 323-362.
Merville, L. J.; Hayes-Yelken, S., and Xu, Y., 2001. Identifying the Factor Structure of Equity Returns, The journal of Portfolio
Management, pp. 51-61.
Michiel, J. and Smith, J. G., (Ed.), 1998. Globalization, Growth and Governance: Creating an Innovative Economy, Oxford
University Press.
Microfit4.0, 1997. Working with Microfit 4.0 Interactive Econometric Analysis, Pesaran, M. H. and Pearan, B., (Ed.), Oxford
University Press, UK.
Mills, T. C. and Mills, A. G., 1991. The international transmission of bond market movements, Bulletin of Economic Research, vol.
43, pp. 273-283.
Mills, T. C., 1999. The Econometric Modelling of Financial Time Series, 2nd Ed., Cambridge University Press, Cambridge.
Mittnik, S.; Kim, J. R. and Rachev, S. T., 1996. Chi-square type distributions for heavy-tailed variables, Discussion Paper, No.
94/1996, Institute of Statistics and Econometrics, Christian Albrechts University at Kiel, Germany.
Mjelde, J. W., et. al., 2002. Understanding Cointegration: An Application to the Western United States Electricity Market, The
Electricity Journal, Elsevier Science, pp. 81-90.
Mazharul Kazi Ph.D Thesis 222
Mjelde, J. W.; Bessler, D. A. and Jerko, C. A., 2002. Understanding Cointegration: An Application to the Western United States
Electricity Market, The Electricity Journal, pp. 81-90.
Modiglini, F. and Miller, M., 1958. The cost of capital, corporation finance, and the theory of investment, American Economic
Review, vol. 48, pp. 261-297.
Mookerjee, R. and Yu, Q., 1997. Macroeconomic variables and stock prices in a small open economy: The case of Singapore,
Pacific Basin Finance Journal, vol. 5, pp. 377-388.
Morgan Stanley Capital International Inc. (MSCI) website: www.msci.com/mxci/index.htm
Mossin, J., 1966. Equilibrium in a capital asset market, Econometrica, vol. 34 (4), pp. 768-783.
Mukharjee, T. K. and Naka, A., 1995. Dynamic relations between macroeconomic variables and the Japanese stock market: an
application of a vector error-correction model, The Journal of Financial Research, vol. 18 (2), pp. 223-237.
Murphy, C., 2000. Forecasting the Australian economy using MM2, in: Abelson, P. and Joyeux, R.,2000, Ed., Allen & Unwin,
Australia, pp. 75-115.
Nabeya, S. and Tanaka, K., 1990. Limiting powers of unit root tests in time series regression, Journal of Econometrics, vol. 46, pp.
247-271.
Nardari, F. and Zhou, G., 1999. What is the decomposition of the market wealth portfolio in the CAPM?, Working Paper,
Washington University.
Nasseh, A. and Strauss, J., 2000. Stock prices and domestic and international macroeconomic activity: a cointegration approach,
The Quarterly Review of Economics and Finance, vol. 40(2), pp. 229-245.
Nawalkha, S. K., 1997. A multibeta representation theorem for linear asset pricing theories, Journal of Financial Economics, vol.
46, pp.357-381.
Neusser, K., 2000. An algebraic interpretation of cointegration, Economics Letters, vol. 67, pp. 273-281.
Newey, W. K. and West, K. D., 1987. A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance
matrix, Econometrica, vol. 55(3), pp. 703-708.
Newey, W. K., 1985. Generalized method of moments specification testing, Journal of Econometrics, vol. 29, pp. 229-256.
Ng, V.; Engle, R. F. and Rothschild, M., 1992. A multi-dynamic-factor model for stock returns, Journal of Econometrics, vol. 52
(1-2), pp. 245-266.
Norusis, M. J., 1998. SPSS 8.0: Guide to Data Analysis, Prentice-Hill Inc., New Jersey, USA.
Ohlson, J. A. and Garman, M.B., 1980. A dynamic equilibrium for Ross arbitrage model, Journal of Finance, vol. 35, pp. 675-684.
Oldfield, Jr. G. S. and Rogalski, R. J., 1981. Treasury Bill Factors and Common Stock Returns, Journal of Finance, vol. 36, pp.
337-350.
Osterwald-Lenum, M., 1992. A note with quintiles of the asymptotic distribution of the maximum likelihood cointegration rank test
statistics, Oxford Bulletin of Economics and Statistics, vol. 54(3), pp. 461-471.
Otero, J. and Smith, J., 2000. Testing for cointegration: power versus frequency of observation-further Monte Carlo results,
Economics Letters, No. 67, pp. 5-9.
Pagan, A. R., 1987. Three Econometric Methodologies: A Critical Appraisal, Journal of Economic Surveys, vol. 1(1), pp. 3-24.
Panton, D. B., Lessig, V. P. and Joy, O. M., 1976. Co-movements of international equity markets: a taxonomic approach, Journal
of Finance and Quantitative Analysis, vol. 11, pp. 415-432.
Patterson, K., 2000. An Introduction to Applied Econometrics: A Time Series Approach, Macmillan, UK.
Paul, S. and Mallik, G., 2001. Macroeconomic Factors and the Bank and Finance Stock Prices: The Australian Experience and
Lessons for GCC Countries, Conference Paper, International Conference on Structure, Performance and Future of Financial
Institutions in Member States of GCC, Quarter, 7-9 April 2001.
Pearce, D. K. and Roley, V., 1983. Stock prices and economic news, Journal of Business, vol. 58, pp. 49-67.
Peiro, A., 1999. Skewness in financial returns, Journal of Banking and Finance, vol. 23 (6), pp. 847-862.
Peirson, G.; Brown, R.; Easton, S. and Howard, P., 1998. Business Finance, 7th Ed., McGraw-Hill , Sydney.
Perron, P. and Campbell, J. Y., 1993. A note on Johansens cointegration procedure when trends are present, Empirical Economics,
vol. 18, pp. 777-789.
Perron, P., 1988. Trends and random walks in macroeconomic time series: further evidence from a new approach, Journal of
Economic Dynamics, vol. 12, pp. 297-332.
Pesaran, M. H. and Shin, Y., 1994. Long-Run Structural Modelling, Discussion Paper, No. 94-29, Department of Applied
Economics, University of Cambridge.
Pesaran, M. H. and Shin, Y., 1995. An Autoregressive Distributed Lag Modelling Approach to Cointegtation Analysis, Working
Paper, Amalgamated Series No. 9514, Department of Applied Economics, University of Cambridge, UK.
Pesaran, M. H. and Timmerman, A., 1992. A simple non-parametric test of predictive performance, Journal of Business and
Economic Statistics, vol. 10(4), pp. 461-465.
Pesaran, M. H. and Timmerman, A., 1995. Predictability of Stock Returns: Robustness and Economic Significance, Journal of
Finance, vol. 50, pp. 1201-1228.
Pesaran, M. H. and Timmerman, A., 2000. A Recursive Modelling Approach to Predicting UK Stock Returns, Economic Journal,
vol. 110, pp. 159-191.
Pesaran, M. H.; Shin, Y. and Smith, R., 2000. Structural analysis of vector error correction models with exogenous I(1) variables,
Journal of Econometrics, vol. 97, pp. 293-343.
Peters, E. E., 1991. Chaos and order in the capital markets: A new view of cycles, prices and market volatility, John Wiley & Sons,
New York.
Phillip, P. C. B. and Oliaris, S., 1990. Asymptotic Properties of Residual Based Tests for Cointegration, Econometrica, vol. 58(1),
pp. 165-193.
Phillip, P. C. B. and Perron, P., 1988. Testing for the Unit Root in Time Series Regression, Biometrika, vol. 75, pp. 335-346.
Mazharul Kazi Ph.D Thesis 223
Phillipatos, D. B.; Christofi, A. and Christofi, P., 1983. The inter-temporal stability of international stock market relationships:
Another view, Financial Management, vol. 12, pp. 63-69.
Phillips, P. C. B. and Darlauf, S. N., 1986. Multiple Time Series Regression with Integrated Processes, Review of Economic
Studies, vol. 53, pp. 473-496.
Phillips, P. C. B. and Hansen, B., 1990. Statistical Inference in Instrumental Variables Regression with I(1) Processes, Review of
Economic Studies, vol. 57, pp. 99-125.
Phillips, P. C. B. and Loretan, M., 1991. Estimating Long Run Economic Equilibria, Review of Economic Studies, vol. 58, pp. 405-
436.
Phillips, P. C. B. and Perron, P., 1988. Testing for Unit Root in time series regression, Biometrika, vol. 75, pp. 335-346.
Phillips, P. C. B. and Solo, V., 1992. Asymptotic for Linear Process, Annals of Statistics, pp. 971-1001.
Phillips, P. C. B., 1987. Time Series Regression with a Unit Root, Econometrica, vol. 55, pp. 277-301.
Phillips, P. C. B., 1991. Optimal Inference in Cointegrated Systems, Econometrica, vol. 59, pp. 283-306.
Phoebus, D., 1998. Time Series, Unit Roots, and Cointegration, Academic Press, USA.
Phoebus, D., 2002. Time Series, Unit Roots and Cointegration, Academic Press, USA.
Pinches, G. E., 1994. Financial Management, Harper Collins College Publishers, NY.
Pindyck, R. S. and Rubinfeld, D. L., 1991. Econometric Models and Economic Forecasts, McGraw-Hill, NY.
Poon, S. and Taylor, S. J., 1991. Macroeconomic factors and the UK stock market, Journal of Business Finance and Accounting,
vol. 18(5), pp. 619-636.
Popper, K., 1934 (Trans., 1959). The Logic of Scientific Discovery, Harper and Row, New York, In: Dimson in Mussavian (1999).
Poterba, J. M. and Summers, L. H., 1988. Mean Reversion in Stock Prices, Journal of Financial Economics, vol. 22, pp. 27-59.
Powers, M. R., 1995. A theory of risk, return and solvency, Insurance: Mathematics and Economics, vol. 17 (2), pp. 101-118.
Priestley, R., 1996. The arbitrage pricing theory, macroeconomic and financial factors, and expectations generating processes,
Journal of Banking and Finance, vol. 20 (5), pp. 869-890.
Ramsey, J. B., 1969. Tests for specification errors in classical linear least-squares regression analysis, Journal of the Royal
Statistical Society, vol. 31(2), pp. 350-371.
Rao, C., 1973. Linear Statistical Inference and its Applications, 2nd Ed., Wiley, New York.
Rault, C., 2000. Non-causality in VAR-ECM models with purely exogeneous long-run paths, Economics Letters, vol. 67, pp. 121-
129.
Reilly, F. K. and Norton, E. A., 1999. Investments, 5th Ed., Dryden-Harcourt, USA.
Reimers, H. E., 1991. Comparisons of tests for multivariate cointegration, Working Paper, No. 58, Institute of Statistics and
Econometrics, University of Kiel.
Reinganum, M., 1981. The Arbitrage Pricing Theory: Some Empirical Results, Journal of Finance, vol. 36, pp. 313-322.
Reinsel, G. C. and Ahn, S. K., 1988. Asymptotic properties of likelihood ratio test for cointegration in nonstationary vector AR
model, Technical Report, Department of Statistics, University of Wisconsin, Madison.
Reinsel, G. C. and Ahn, S. K., 1992. Vector AR models with unit roots and reduced rank structure: estimation, likelihood ratio test,
and forecasting, Journal of Time Series Analysis, vol. 13, pp. 353-375.
Reinsel, G. C., 1991. Elements of multivariate time series analysis, Springer Verlag, New York.
Reisman, H., 1988. A general approach to the arbitrage pricing theory (APT), Econometrica, vol. 56, pp. 473-476.
Reisman, H., 1992. Intertemporal arbitrage pricing theory, Review of Financial Studies, vol. 5, pp. 105-122.
Reisman, H., 1992a. Reference variables, factor structure, and the approximate multibeta representation, Journal of Finance, vol.
47, pp. 1303-1314.
Reisman, H., 1992b. The APT with proxies, Working Paper, Technion, Haifa.
Renton, N. E., 1998. Understanding the stock exchange, 3rd Ed., Griffin Press, Australia.
Reserve Bank of Australia, website: www.rba.gov.au
Reyes, M. G., 1999. Size, time-varying beta, and conditional heteroscedasticity in UK stock returns, Review of Financial
Economics, vol. 8, pp. 1-10.
Ripley, D., 1973. Systematic elements in the linkage of national stock market indices, The Review of Economics and Statistics, 15,
pp. 356-361.
Robinson, J., 2000. In CNNFN Market Call Aug. 22. website: www.cnnfn.cnn./2000/08/22/cashflow/robinson_ceo/index.htm
Roca, E. D., 1998. Are equity markets interdependent? Evidence from three groups of countries, Ph.D Thesis, School of
International Business, Griffith University, Australia.
Roll, R. W. and Ross, S. A., 1980. An empirical investigation of the arbitrage pricing theory, Journal of Finance, vol. 35, pp. 1073-
1103.
Roll, R. W. and Ross, S. A., 1983. Regulation, the Capital Asset Pricing Model, and the Arbitrage Pricing Theory, Public United
Fortnightly, vol. 111(11), pp. 22-28.
Roll, R. W. and Ross, S. A., 1984. The Arbitrage Pricing Theory Approach to Strategic Portfolio Planning, Financial Analysts
Journal, vol. 40(3), pp. 14-26.
Roll, R. W. and Ross, S. A., 1984a. A critical reexamination of the empirical evidence on arbitrage pricing theory: A reply, Journal
of Finance, vol. 39(2), pp. 347-350.
Roll, R. W., 1977. A critique of the asset pricing theorys test: Part I, Journal of Financial Economics, vol. 4, pp. 129-176.
Roll, R. W., 1988. The international crash of October 1987, Financial Analysts Journal, vol. 44, pp. 19-35.
Rosenberg, B. and Marathe, V., 1975. Common factors in security returns: microeconomic determinants and macroeconomic
correlations, proceedings of the Seminar on the Analysis of Security Prices, pp. 61-115.
Ross, S. A. and Zisler, R., 1991. Risk and Return in Real Estate, Journal of Real Estate Financial Economics, vol. 4(2), pp. 175-
190.
Ross, S. A., 1976. The arbitrage theory of capital asset pricing, Journal of Economic Theory, vol. 13, pp. 341-360.
Mazharul Kazi Ph.D Thesis 224
Ross, S. A., 1977. Return, risk, and arbitrage, In: Risk and Return in Finance, I, ed. by Friend, I. and Bicsler, J.I., Cambridge,
Massachusetts, USA.
Ross, S. A., 1978. A simple approach to the valuation of risky streams, Journal of Business, vol. 51, pp. 453-475.
Ross, S. A., 1990. Arbitrage and the APT: some new results, Working Paper, School of Management, Yale University, New
Haven, CT.
Ross, S. A.; Westerfield, R. W. and Jordan, B. D., 1998. Fundamentals of Corporate Finance, 4th (Ed.), Irwin McGraw-Hill, USA.
Roy, A. D., 1952. Safety first and the holding of assets, Econometrica, vol. 20, pp. 431-439.
Rubinstein, M., 1974. An aggregation theorem for security markets, Journal of Financial Economics, vol. 1, pp. 225-244.
Saikkonen, P. and Lutkepohl, H., 1999. Local power of likelihood ratio tests for the cointegrating rank of a VAR process,
Econometric Theory, vol. 15, pp. 50-78.
Saikkonen, P. and Luukkonen, R., 1997. Testing cointegration in infinite order vector autoregressive process, Journal of
Econometrics, vol. 81, pp. 93-126.
Saikkonen, P., 1992. Estimation and testing of cointegrated systems by an autoregressive approximation, Econometric Theory, vol.
8, pp. 1-27.
Samuelson, P. A., 1965. Proof That Properly Anticipated Prices Fluctuate Randomly, International Management Review, vol. 6,
pp. 41-49.
Samuelson, P. A., 1969. Lifetime portfolio selection by dynamic programming, Review of Economics and Statistics, vol. 51, pp.
239-246.
Sargan, J. D., 1964. Wages and Prices in the United Kingdom: A study in Econometric Methodology, originally published and
reproduced in Wallis, K. F. and Hendry, D. F., eds., Quantitative Economics and Econometric Analysis, Basil Blackwell,
Oxford, 1984.
Schmidt, P. and Phillips, P. C. B., 1992. LM tests for a unit root in the presence of deterministic trends, Oxford Bulletin of
Economics and Statistics, vol. 54, pp. 257-287.
Schotman, P. C., 2001. When unit roots matter: excess volatility and excess smoothness of long-term interest rates, Journal of
Empirical Finance, vol. 8, pp. 669-694.
Schwarz, Z. G., 1978. Estimating the dimension of a model, Annals of Statistics, vol. 6, pp. 461-464.
Seber, G. A. F., 1984. Multivariate Observations, Wiley, New York.
Self, S. and Grabowski, R., 2002. The economic impact of gender participation by level of education: A time series analysis, The
Journal of Developing Areas, vol. 36, pp. 57-79.
Sentana, E., 1992. Identification of multivariate conditionally heteroskedastic factor models, Discussion Paper, No. 139, London
School of Economics, UK.
Shakla, R. and Trzcinka, C., 1990. Sequential tests of arbitrage pricing theory: A comparisons and maximum likelihood factors,
Journal of Finance, vol. 45, pp. 1541-1564.
Shamsuddin, A. F. and Holmes, R., 1997. Cointegration test of the monetary theory of inflation and forecasting accuracy of the
univariate and vector ARMA models of inflation, Journal of Economic Studies, vol.24, No. 5, pp. 294-306.
Shamsuddin, A. F. and Kim, J. H., 2000. Integration and Interdependence of Stock and Foreign Exchange Markets: An Australian
Perspective, Conference Paper, Australian Finance and Banking Conference, Sydney.
Shamsuddin, A. F. and Kim, J. H., 2003. Integration and Interdependence of Stock and Foreign Exchange Markets: An Australian
Perspective, International Financial Markets, Institutions and Money, vol. 13, pp. 237-254.
Shanken, J. and Weinstein, M. I., 1990. Macroeconomic variables and asset pricing: Estimations and tests, Working Paper,
University of Rochester, NY.
Shanken, J., 1982. The arbitrage pricing theory: Is it testable? Journal of Finance, vol. 37(5), pp. 1129-1140.
Shanken, J., 1985. Multi-beta CAPM or equilibrium APT: A reply, journal of Finance, vol. 40(4), pp. 1189-1196.
Shanken, J., 1985a. Multivariate tests of the zero-beta CAPM, Journal of Finance and Economics, vol. 14, pp. 327-348.
Shanken, J., 1987. Multivariate proxies and asset pricing relations: living with the roll critique, Journal of Financial Economics,
vol. 18 (1), pp. 91-110.
Shanken, J., 1987a. Non-synchronous data and the covariance factor structure of returns, Journal of Finance, vol. 42, pp. 221-231.
Shanken, J., 1990. Intertemporal Asset Pricing: An Empirical Investigation, Journal of Econometrics, vol. 45, pp. 99-120.
Shanken, J., 1992. The current state of the arbitrage pricing theory, Journal of Finance, vol. 47(4), pp. 1569-1574.
Shanken, J., 1992a. On the estimation of beta pricing models, Review of Financial Studies, vol. 5, pp. 1-34.
Shanken, J., 1996. Statistical methods in tests of portfolio efficiency: a synthesis, In: Maddala, G. S. and Rao, C. S. (Ed.),
Handbooks of Statistics: Statistical Methods in Finance, vol. 14, North Holland, Amsterdam, pp. 693-711.
Sharpe, W. F., 1963. A simplified model of portfolio analysis, Management Science, vol. 9, pp. 277-293.
Sharpe, W. F., 1964. Capital asset prices: A theory of market equilibrium under conditions of risk, Journal of Finance, vol. 19(3),
pp. 425-442.
Sharpe, W. F., 1966. Security prices, risk and maximal gains from diversification: Reply, Journal of Finance, vol. Dec., pp. 743-
744.
Sharpe, W. F., 1978. Investments, Prentice Hall, Englewoog Cliffs, New Jersey, USA.
Sharpe, W. F., 1984. Factor Models, CAPMs and the APT, Journal of Portfolio Management, vol. 11(1), pp. 21-25.
Sharpe, W. F., 1990. Capital asset prices with and without negative holdings, in Maler, K-G.,1992, Nobel Lectures in Economic
Sciences 1981-1990, World Scientific, USA, pp. 303-331.
Sharpe, W. F.; Gordon, A. J. and Jeffery, B. V., 1995. Investment, 5th Ed., Prentice Hall, USA.
SHAZAM Econometrics Software, 2001. SHAZAM Users Reference Manual, Version 9, Norwest Econometrics Ltd., Canada.
Shukla, R. and Trzcinka, C., 1990. Sequential tests of the arbitrage pricing theory: A comparison of principal components and
maximum likelihood factors, The Journal of Finance, vol. XLV (5), pp. 1541-1564.
Siebert, H., (Ed.), 1991. Capital Flows in the World Economy Symposium 1990, JCB MOHR (PAUL SIEBECK), Germany.
Mazharul Kazi Ph.D Thesis 225
Silvapulle, P., 1996. A score test for seasonal fractional integration and cointegration, Discussion Papers, Series A, No. 96.01,
School of Economics and Commerce, Latrobe University.
Sims, C. A., 1980. Macroeconomics and Reality, Econometrica, 58, pp. 1-48.
Sims, C. A.; Stock, J. H. and Watson, M. W., 1990. Inference in linear time series models with some unit roots, Econometrica, vol.
58, pp. 113-144.
Sinclair, N. A., 1982. An empirical test of Arbitrage Pricing Theory, Ph.D Thesis, Australian Graduate School of Management,
University of New South Wales.
Sinclair, N. A., 1984. Aspects of the factor structure implicit in the Australian industrial equity market: february 1958 to August
1977, Australian Journal of Management, vol. 9, June.
Sinclair, N. A., 1987. Multifactor asset pricing models, Accounting and Finance, vol. 27, pp. 17-37.
Sivalingam, G., 1990. Modern Portfolio Management, Longman, New York.
Smithson, C. W.; Schraudenbach, L. S., and Smith, C. W., 1992. Financial Price Risk Evidenced in Share Price Behavior, Working
Paper, Risk Management Research, Chase Manhattan Bank, N.A, New Jersey 08540.
Snell, A., 1999. Testing for r versus r-1 cointegrating vectors, Journal of Econometrics, vol. 88, pp. 151-191.
Snow, K. N., 1991. Diagnosing asset pricing models using the distribution of asset returns, Journal of Finance, vol. 46, pp. 955-
983.
Solnik, B. H., 1974. An equilibrium model of the international capital market, Journal of Economic Theory, vol. 8 (4), pp. 500-524.
Solnik, B. H., 1974a. The international pricing of risk: An empirical investigation of world capital market structure, Journal of
Finance, vol. 29, pp. 48-54.
Solnik, B. H., 1974b. Why not diversify internationally rather than domestically? Financial Analysts Journal, vol. 30, pp. 48-54.
Solnik, B. H., 1977. Testing international asset pricing: Some pessimistic views, Journal of Finance, vol. 32, pp. 503-511.
Solnik, B. H., 1983. International Arbitrage Theory, Journal of Finance, vol. 38, pp. 449-459.
Stambaurgh, R. F., 1982. On the exclusion of assets from tests of the two-parameter model: A sensitivity analysis, Journal of
Financial Economics, vol. 10, pp. 237-268.
Standard and Poor, 2001. Setting the Standard: Understanding Indices, website: www.standardandpoor.com.au
StatSoft Inc., 2002. Structural Equation Modeling, website: www.statsoftinc.com
Steven, J., 2002. Applied multivariate statistics for social sciences, Lawrence Erlbaum Associations Publishers, New jersey.
Stock, J. H. and Watson, M. W., 1988. Variable Trends in Economic Time Series, Journal of Economic Perspectives, vol. 2(3),
pp.147-174.
Stock, J. H. and Watson, M. W., 1988a. Testing for common trends, Journal of American Statistical Association, vol. 83, pp. 1097-
1107.
Stock, J. H. and Watson, M. W., 1989. New indexes of Coincedent and Leading Economic Indicators, in NMER Macroeconomics
Annual 1989, ed. by Blanchard and Fisher, MIT Press, Cambridge.
Stock, J. H. and Watson, M. W., 1993. A simple Estimator of Cointegrating Vectors in Higher Order Integrated Systems,
Econometrica, vol. 61(4), pp. 783-820.
Stock, J. H. and Watson, M. W., 2001. Vector Autoregressions, Journal of Economic Perspectives, vol. 15(4), pp. 101-115.
Stock, J. H. and Watson, M. W., 2003. Introduction to Econometrics, Addison Wesley, USA.
Stock, J. H., 1987. Asymptotic Properties of Least Squares Estimates of Cointegrating Vectors, Econometrica, vol. 55, pp. 1035-
1056.
Stock, J. H., 1990. A class of Tests for Integration and Cointegration, Harvard University Mimeo, USA.
Stock, J. H., 1994. Unit Roots, Structural Breaks, and Trends, In: Engle, R. and McFadden, D., (Ed.), Handbook of Econometrics,
vol. 4, Elsevier, Amsterdam.
Studenmund, A. H., and Cassidy, H. J., 1992. Using Econometrics: A practical Guide, Harper Collins Publishers, USA.
Szpiro, G. G., 1991. Noise in autoregressive time-series, Economic Letters, vol. 37(1), pp. 45-50.
Tabachmick, B. G. and Fidell, L. S., 1996. Using Multivariate Statistics, 3rd Ed., Harper Collins College Publishers, pp. 815-818.
Tan, K. J., 1991. Risk return and the three-moment capital asset pricing model: another look, Journal of Banking and Finance, vol.
15(2), pp. 449-460.
Tanchen, G., 1986. Statistical properties of generalized method of moment estimators of structural parameters obtained from
financial market data, Journal of Business and Economic Statistics, vol. 4, pp. 397-425.
Taylor, M. P. and Peel, D. A., 1998. Periodically collapsing stock price bubbles: a robust test, Economics Letters, vol. 61, pp. 221-
228.
Taylor, M. P., 1987. Risk Premia and Foreign Exchange A multiple time series approach to testing uncovered interest parity,
Weltwirtschaftliches Archive, vol. 123(4), pp. 579-591.
Theil, H., 1971. Principles of Econometrics, John Wiley, New York.
Thornton, D. I. and Batten, D. S., 1985. Lag length selection and granger causality between money and income, Journal of Money,
Credit and Banking, vol. 17, pp.164-178.
Tobin, J., 1958. Liquidity preference as a behaviour toward risk, Review of Economic Studies, vol. 25, pp. 65-86.
Toda, H. Y., 1994. Finite sample properties of likelihood ratio tests for cointegrating ranks when linear trends are present, Review
of Economics and Statistics, vol. 76, pp. 66-79.
Toda, H. Y., 1995. Finite sample performance of likelihood ratio tests for cointegrating ranks in vector autoregressions,
Econometric Theory, vol. 11, pp. 1015-1032.
Trainor, J. L. and Black, F., 1973. How to use security analysis to improve portfolio selection, Journal of Business, vol. 46, pp. 66-
86.
Treynor, J. L., 1961. Toward a theory of market value of risky assets, In: Dimson, E. and Mussavian, M., (1999).
Truett, L. J. and Truett, D. B., 1980. Managerial Economics, South-Western Publishing Co., Ohio, USA.
Trzcinka, C., 1986. On the number of factors in the arbitrage pricing model, The Journal of Finance, vol. 41(2), pp. 347-368.
Mazharul Kazi Ph.D Thesis 226
United Nations, 1992. International Monetary and Financial Issues for the 1990s, vol. 1, pp. 56-57.
Valelentine, T. J., 2000. Share prices and fundamentals, In: Albert, P. and Joyeux, R., Ed., Economic Forecasting, Allen & Unwin,
pp. 175-193.
Velu, R. and Zhou, G., 1999. Testing multi-beta asset pricing models, Journal of Empirical Finance, vol. 6, pp. 219-241.
Verbeck, M., 2000. A Guide to Modern Econometrics, John Wiley, Chichester, UK.
VonNewmann, J. and Morgenstern, O., 1944. Theory of Games and Economic Behavior, princeton University Press (1947), USA.
Wald, A., 1950. A note on the identification of economic relations, In: Koopmans, T. C. (Ed.), 1950. Statistical Inference in
Dynamic Economic Models, pp. 238-234, Wiley, NY; Also in: Anderson, T. W. (Ed.), 1955. Selected papers in Statistics and
probability by Abraham Wald, McGraw-Hill, NY.
Warga, 1989. Experimental design in tests of linear factor models, Journal of Business and Economic statistics, vol. 7, pp. 91-198.
Watson, D. S., 1963. Price Theory and Its Uses, Houghton Mifflin Company, Boston, USA.
Watson, M. W., 1994. Vector autoregression and cointegration, In: Engel, R. F. and McFadden, D. L. (Ed.), Handbook of
Econometrics, vol. 4, Elsevier Science, North-Holland, Amsterdam, pp. 2843-2915.
Wegge, L. L., 1996. Local identifiability of the factor analysis and measurement error model parameter, Journal of Econometrics,
vol. 70(2), pp. 351-382.
Wei, K. C. J., 1988. An asset pricing theory unifying the CAPM and APT, Journal of Finance, vol. 43(4), pp. 881-892.
Wei, K. C. J.; Lee, C. G. and Chen, A. H., 1991. Multivariate Regression Tests of the Arbitrage Pricing Theory: The Instrumental
Variables Approach, Review of Quantitative Finance and Accounting, vol. 1, pp. 191-208.
Wheatley, S. M., 1989. A critique of latent variable tests of asset pricing models, Journal of Financial Economics, vol. 23, pp. 325-
338.
White, H., 1980. A heteroskedasticity-consistent covariance matrix estimator and direct test for heteroskedasticity, Econometrica,
vol. 48, pp. 817-838.
Wickens, M. R. and Breusch, T. S., 1988. Dynamic Specification, theLong Run Estimation of the Transformed Regression models,
The Economic Journal, vol. 98, pp. 189-205.
Wickens, M. R., 1993. Interpreting cointegrating vectors and common stochastic trends, Discussion Paper, No. 14-93, Centre for
Economic Forecasting, London Business School, UK.
Williams, J. T., 1977. Capital asset prices with heterogeneous beliefs, Journal of Financial Economics, vol. 5, pp. 219-239.
Wilson, J. W., 1999. Turmoil in Financial Market: Should we be Surprised? Economic Papers, EGA, vol. 18(4), pp. 7-19.
Winkelmann, M., 1984. Testing APT for German stock market, Paper presented at the 11th Annual Meeting of the European
Finance Association, Manchester.
Wongbangpo, P. and Sharma, S. C., 2002. Stock market and macroeconomic fundamental dynamic interactions: ASEAN5
countries, Journal of Asian Economics, vol. 13, pp. 27-51.
Wood, J., 1991. A cross-sectional regression test of the mean-variance efficiency of an Australian value weighted market portfolio,
Accounting and Finance, vol. 31, pp. 96-105.
Yoo, B. S., 1985. Multi-cointegrated Time Series and Generalized Error Correction Model, Discussion Paper, University of
California, San Diego, USA.
Yoo, B. S., 1985a. Multi-cointegrated Time Series and Generalized Error Correction Models, Manuscript, University of California,
San Diego, Also in: Engle, R. and Granger, C. W. J., 1997.
Young, S. D.; Berry, M. A.; Harvey, D. W. and Page, J. R., 1987. Systematic risk and accounting information under the arbitrage
pricing theory, Financial Analysts Journal, vol. 43, pp. 73-76.
Zechner, J., 1996. Financial market- product market interactions in industry equilibrium: Implications for information acquisition
decision, European Economic Review, vol. 40(3-5), pp. 883-896.
Zhou, G. and Velu, R., 1999. Testing multi-beta asset pricing models, Journal of Empirical Finance, vol. 6, pp. 219-241.
Zhou, G., 1994. Analytical GMM tests: Asset pricing with time-varying risk premium, Review of Financial Studies, vol. 7, pp. 687-
709.
Zhou, G., 1999. Security factors as linear combinations of economic variables, Journal of Financial Markets, vol. 2, pp. 403-432.
Mazharul Kazi Ph.D Thesis 227
Annexure 1
Pair-wise Correlation Matrix on Base Data
ALLORDS BVIR CP CPI DY ER GDP GDPPC IPI IR11AM M3 MPI MSCI NX PER UR
ALLORDS 1.000
BVIR -0.685 1.000
CP 0.838 -0.654 1.000
CPI 0.908 -0.749 0.775 1.000
DY -0.547 0.626 -0.415 -0.377 1.000
ER -0.767 0.436 -0.562 -0.640 0.563 1.000
GDP 0.968 -0.744 0.863 0.945 -0.478 -0.738 1.000
GDPPC 0.971 -0.711 0.869 0.929 -0.468 -0.748 0.998 1.000
IPI 0.953 -0.731 0.856 0.974 -0.408 -0.697 0.987 0.982 1.000
IR11AM -0.674 1.000 -0.652 -0.742 0.615 0.410 -0.738 -0.704 -0.724 1.000
M3 0.951 -0.772 0.843 0.954 -0.473 -0.721 0.994 0.985 0.982 -0.770 1.000
MPI 0.905 -0.733 0.781 0.998 -0.358 -0.653 0.939 0.924 0.973 -0.726 0.948 1.000
MSCI 0.985 -0.737 0.843 0.944 -0.555 -0.749 0.981 0.978 0.974 -0.730 0.974 0.939 1.000
NX -0.301 0.143 -0.425 -0.182 0.206 0.181 -0.310 -0.335 -0.276 0.164 -0.258 -0.170 -0.276 1.000
PER 0.535 -0.729 0.365 0.619 -0.573 -0.512 0.571 0.542 0.568 -0.725 0.597 0.619 0.585 -0.135 1.000
UR -0.482 -0.205 -0.317 -0.298 -0.162 0.385 -0.434 -0.483 -0.423 -0.245 -0.369 -0.311 -0.414 0.318 0.245 1.000
Mazharul Kazi Ph.D Thesis 228
Common Sample Correlation Matrix on Base Data
Annexure 1 (Contd.)
ALLORDS BVIR CP CPI DY ER GDP GDPPC IPI IR11AM M3 MPI MSCI NX PER UR
ALLORDS 1.000
BVIR -0.703 1.000
CP 0.838 -0.654 1.000
CPI 0.842 -0.838 0.775 1.000
DY -0.534 0.599 -0.415 -0.315 1.000
ER -0.676 0.427 -0.562 -0.429 0.588 1.000
GDP 0.956 -0.770 0.863 0.929 -0.434 -0.625 1.000
GDPPC 0.961 -0.731 0.869 0.902 -0.421 -0.636 0.997 1.000
IPI 0.924 -0.783 0.856 0.960 -0.351 -0.529 0.985 0.977 1.000
IR11AM -0.700 1.000 -0.652 -0.836 0.599 0.423 -0.768 -0.728 -0.781 1.000
M3 0.936 -0.799 0.843 0.957 -0.424 -0.611 0.993 0.981 0.985 -0.797 1.000
MPI 0.839 -0.824 0.781 0.996 -0.291 -0.447 0.924 0.897 0.959 -0.822 0.953 1.000
MSCI 0.977 -0.777 0.843 0.906 -0.545 -0.638 0.975 0.970 0.957 -0.775 0.966 0.901 1.000
NX -0.317 0.138 -0.425 -0.161 0.211 0.189 -0.336 -0.369 -0.293 0.142 -0.273 -0.141 -0.293 1.000
PER 0.396 -0.714 0.365 0.516 -0.542 -0.414 0.455 0.417 0.446 -0.715 0.494 0.519 0.465 -0.108 1.000
UR -0.350 -0.340 -0.317 -0.062 -0.276 0.204 -0.281 -0.345 -0.252 -0.343 -0.204 -0.078 -0.251 0.320 0.460 1.000
Mazharul Kazi Ph.D Thesis 229
Pair-wise Correlation Matrix on LNDATA
Annexure 1 (Contd.)
LNALLORDS LNBVIR LNCP LNCPI LNDY LNGDP LNER LNGDPPC LNIPI LNIR11AM LNMPI LNMS LNMSCI LNNX LNPER LNUR
LNALLORDS 1.000
LNBVIR -0.690 1.000
LNCP 0.804 -0.699 1.000
LNCPI 0.929 -0.750 0.755 1.000
LNDY -0.534 0.653 -0.383 -0.372 1.000
LNGDP 0.943 -0.792 0.839 0.945 -0.482 1.000
LNER -0.720 0.529 -0.496 -0.604 0.591 -0.723 1.000
LNGDPPC 0.940 -0.764 0.844 0.920 -0.481 0.996 -0.742 1.000
LNIPI 0.946 -0.762 0.840 0.972 -0.408 0.988 -0.675 0.979 1.000
LNIR11AM -0.678 1.000 -0.697 -0.741 0.638 -0.782 0.495 -0.752 -0.752 1.000
LNMPI 0.927 -0.740 0.763 0.998 -0.358 0.943 -0.618 0.919 0.973 -0.731 1.000
LNMS 0.935 -0.796 0.796 0.987 -0.425 0.981 -0.656 0.963 0.989 -0.788 0.985 1.000
LNMSCI 0.987 -0.729 0.808 0.963 -0.527 0.953 -0.690 0.942 0.963 -0.719 0.960 0.962 1.000
LNNX 0.174 0.045 -0.122 0.255 0.235 0.229 -0.058 0.223 0.224 0.036 0.248 0.234 0.158 1.000
LNPER -0.006 0.079 -0.222 -0.002 -0.052 -0.036 0.040 -0.048 -0.055 0.082 -0.012 -0.007 0.009 0.194 1.000
LNUR -0.456 -0.135 -0.275 -0.303 -0.128 -0.422 0.404 -0.479 -0.418 -0.185 -0.318 -0.329 -0.390 -0.277 0.085 1.000
Mazharul Kazi Ph.D Thesis 230
Common Sample Correlarion Matrix on LNDATA
Annexure 1 (Contd.)
LNALLORDS LNBVIR LNCP LNCPI LNDY LNGDP LNER LNGDPPC LNIPI LNIR11AM LNMPI LNMS LNMSCI LNNX LNPER LNUR
LNALLORDS 1.000
LNBVIR -0.622 1.000
LNCP 0.766 -0.690 1.000
LNCPI 0.656 -0.725 0.612 1.000
LNDY -0.592 0.589 -0.392 -0.119 1.000
LNGDP 0.867 -0.730 0.775 0.910 -0.296 1.000
LNER -0.739 0.562 -0.709 -0.506 0.474 -0.722 1.000
LNGDPPC 0.890 -0.679 0.789 0.857 -0.297 0.992 -0.755 1.000
LNIPI 0.811 -0.713 0.753 0.940 -0.203 0.991 -0.660 0.976 1.000
LNIR11AM -0.607 0.999 -0.681 -0.716 0.589 -0.715 0.543 -0.664 -0.701 1.000
LNMPI 0.655 -0.721 0.633 0.997 -0.102 0.913 -0.544 0.863 0.943 -0.711 1.000
LNMS 0.738 -0.753 0.680 0.991 -0.191 0.955 -0.572 0.913 0.972 -0.743 0.988 1.000
LNMSCI 0.956 -0.737 0.726 0.773 -0.619 0.908 -0.713 0.905 0.868 -0.726 0.765 0.834 1.000
LNNX 0.092 0.158 -0.122 0.216 0.271 0.221 -0.137 0.228 0.217 0.169 0.211 0.201 0.070 1.000
LNPER -0.043 -0.062 -0.345 0.083 -0.124 -0.087 0.246 -0.148 -0.098 -0.069 0.067 0.046 0.018 0.207 1.000
LNUR -0.420 -0.308 -0.275 -0.076 -0.229 -0.338 0.382 -0.434 -0.319 -0.328 -0.101 -0.142 -0.241 -0.325 0.408 1.000
Mazharul Kazi Ph.D Thesis 231
Pair-wise Correlation Matrix on 1st Differenced Data
Annexure 1 (Contd.)
DALLORDS DBVIR DCP DCPI DDY DER DGDP DGDPPC DIPI DIR11AM DMPI DMSCI DNX DUR DPER DMS
DALLORDS 1.000
DBVIR -0.046 1.000
DCP -0.056 0.005 1.000
DCPI 0.021 -0.115 -0.113 1.000
DDY -0.426 0.224 0.139 -0.012 1.000
DER -0.037 -0.166 -0.043 -0.047 -0.103 1.000
DGDP -0.057 0.117 0.235 0.004 0.191 0.034 1.000
DGDPPC -0.047 0.117 0.233 0.014 0.186 0.018 0.989 1.000
DIPI -0.119 0.094 0.128 -0.054 0.211 -0.156 0.401 0.407 1.000
DIR11AM -0.040 0.996 -0.009 -0.115 0.231 -0.192 0.105 0.107 0.084 1.000
DMPI 0.071 0.204 0.233 -0.085 0.286 -0.242 -0.118 -0.093 0.226 0.194 1.000
DMSCI 0.488 -0.051 -0.152 -0.064 -0.803 0.137 -0.045 -0.047 -0.110 -0.059 -0.080 1.000
DNX 0.056 -0.113 -0.192 -0.002 0.067 -0.135 -0.121 -0.130 -0.055 -0.097 0.171 -0.024 1.000
DUR -0.154 -0.270 -0.229 0.076 -0.279 -0.131 -0.375 -0.380 -0.028 -0.271 -0.124 0.075 0.050 1.000
DPER 0.092 -0.106 0.001 0.015 -0.377 -0.001 0.054 0.044 -0.090 -0.113 -0.101 0.321 -0.067 0.194 1.000
DMS -0.098 0.037 0.095 -0.092 0.099 -0.027 0.228 0.173 0.136 0.074 -0.097 0.065 -0.065 -0.070 0.085 1.000
Mazharul Kazi Ph.D Thesis 232
Common Sample Correlation Matrix on 1st Differenced Data
Annexure 1 (Contd.)
DALLORDS DBVIR DCP DCPI DDY DER DGDP DGDPPC DIPI DIR11AM DMPI DMSCI DNX DUR DPER DMS
DALLORDS 1.000
DBVIR 0.020 1.000
DCP -0.056 0.005 1.000
DCPI 0.021 -0.158 -0.113 1.000
DDY -0.440 0.175 0.139 -0.015 1.000
DER -0.046 -0.049 -0.043 -0.067 -0.164 1.000
DGDP -0.077 0.136 0.235 0.010 0.252 -0.021 1.000
DGDPPC -0.076 0.129 0.233 0.024 0.263 -0.028 0.995 1.000
DIPI -0.144 0.071 0.128 -0.053 0.305 0.012 0.475 0.484 1.000
DIR11AM 0.029 0.996 -0.009 -0.157 0.181 -0.054 0.127 0.120 0.048 1.000
DMPI 0.044 0.225 0.233 -0.083 0.331 -0.222 -0.123 -0.109 0.195 0.200 1.000
DMSCI 0.527 -0.022 -0.152 -0.066 -0.826 0.199 -0.057 -0.065 -0.152 -0.043 -0.084 1.000
DNX 0.097 -0.138 -0.192 -0.002 0.081 -0.115 -0.123 -0.128 -0.070 -0.118 0.206 -0.037 1.000
DUR -0.160 -0.363 -0.229 0.075 -0.248 -0.092 -0.371 -0.372 -0.045 -0.366 -0.141 0.050 0.027 1.000
DPER 0.107 -0.108 0.001 0.014 -0.367 0.008 0.054 0.048 -0.113 -0.119 -0.099 0.307 -0.085 0.175 1.000
DMS 0.027 0.180 0.095 -0.147 0.188 -0.001 0.389 0.345 0.173 0.161 -0.052 -0.049 -0.184 -0.237 0.028 1.000