Systematic risk in the capital asset pricing model for australia a clinical death

ABSTRACT On the ground of a well-known Markowitz 1952’s Modern Portfolio Theory, Sharpe 1964 and Lintner 1965 developed a specific relationship between risk and expected return, which ha

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM

HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES

VIET NAM THE NETHERLANDS

VIETNAM – THE NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

SYSTEMATIC RISK IN THE CAPITAL ASSET PRICING MODEL FOR AUSTRALIA: A CLINICAL DEATH?

BY

NGUYEN CONG THANG

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

Ho Chi Minh City

December 2017

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

SYSTEMATIC RISK IN THE CAPITAL ASSET PRICING MODEL FOR AUSTRALIA: A CLINICAL DEATH?

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

Ho Chi Minh City, December 2017

Nguyen Cong Thang

ACKNOWLEDGEMENTS

I would like to express my special thanks of gratitude to my academic supervisor Dr Duc Vo

He gave me the golden opportunity to do this wonderful project on the topic of capital asset pricing model I know that, for the last 20 years, you has been spending your youth, your effort

to make your life and your future thrive in Australia I appreciate this opportunity

I did not realize that my high school knowledge, my skill I had developed as an Android developer could help me jump over challenges during the process of thesis accomplishment

On that way, I learnt Visual Basic and R and I expect that they are my friends when I struggle with messy data I want to say thanks for my supervisor and for those introducing Beta and R

After all, I leave my last few words to Mom and Dad This thesis is for you This work is my gift to you I have put all great effort to develop and complete this very first academic study From the bottom of my heart, I apologize for your tears I should have focused on getting thing done to have lived happily and planned carefully my future

My dearest loved Mom and Dad! I am still a kid, are not I?

ABBREVIATIONS

C4F: Cahart four-factor model

CAL: Capital allocation line

CAPM: Capital asset pricing model

FF3F: Fama-French three-factor model

GICS: Global Industry Classification Standard

MPT: Modern portfolio theory

ABSTRACT

On the ground of a well-known Markowitz (1952)’s Modern Portfolio Theory, Sharpe (1964) and Lintner (1965) developed a specific relationship between risk and expected return, which has been named as the Sharpe-Lintner Capital Asset Pricing Model (CAPM)

CAPM or the Sharpe-Lintner CAPM is a well-known and most widely used model for estimating a rate of return/cost of capital The CAPM confirms that only systematic risk – denoted by ß (beta), does matter and investors are only compensated for taking systematic risk Since its introduction, many studies have been conducted in an effort to assess the validity of the CAPM in practice Practitioners and regulators around the world including Australia, Germany, New Zealand and United Kingdom employed CAPM as a primary model to estimate asset’s return

However, various studies demonstrated that CAPM appears to underestimate returns for low-beta assets and overestimate returns for high-beta assets The criticism went further as Fama and French (1992) introduced the three-factor model to estimate the asset’s return The Fama-French three-factor model has been proven to work well in the US market and that beta

is alive in the American context However, in contrast to the US market, Vo (2015) argued that the Fama-French three-factor model has been proven to not work well in the Australian context A work by Savor and Wilson (2014) concluded that beta, or systematic risk, is still alive in the US market A similar question is that whether or not beta is still alive in Australia because Vo (2015) has never tested this hypothesis? We are not aware of any study on the issue which has been conducted This study is conducted to fill in the gap

This study examines the validity of the Capital Asset Pricing Model (1965) in the context

of Australia on the ground of the pioneering work by Savor and Wilson (2014) for the US The choice of Australia is important because, among all nations in the Asia-Pacific region, Australia

is one of a few which has required data for the analysis to be conducted

In the heart of the CAPM, beta is considered an important measure of systematic risk which is generally defined as an uncertainty about general economic conditions, such as GNP, interest rates, or inflation From that perspective, a key purpose of this study is to examine and quantify whether or not systematic risk is responsive on the days when macroeconomics news/events are announced or scheduled for announcement

On the ground of Savor and Wilson (2014), four different types of portfolios are considered in this study including: (i) 10 beta-sorted portfolios; (ii) 10 idiosyncratic risk-sorted portfolios (iii) 25 Fama-French size and book-to-market portfolios; and (iv) industry portfolios

In addition, macroeconomic events include announcements in relation to growth, inflation, employment, central bank announcements, bonds, housing, consumer surveys, business surveys and speeches from the Prime Minister or the Governor of the Reserve Bank of Australia Days

with these events are allocated into the group (the so-called a-day) which is separated from the

n-day (non-announcement days) group

In addition, in this study, a sensitivity check, which is beyond Savor and Wilson (2014),

by adopting different definition1 of the a-day group including (i) macroeconomics

announcements which consist of news about growth, inflation, employment, Central Bank, bonds and speeches; (ii) microeconomics announcements which contains news related to housing, consumer survey and business survey; (iii) economics announcements which are basic

news about news about growth, inflation, employment, housing, consumer surveys, business

surveys and speeches; and (iv) financial announcements which are combined by news about Central Bank and bonds

This study is conducted on a sample including more than 2,200 Australian listed firms collected from Bloomberg for the period from 1 January 2007 to 31 December 2016 is employed As such, the total of nearly 2 million observations has been used in this study Using the linear regression with panel-corrected standard errors method and Fama-Macbeth regression across various portfolios, two fundamental findings achieved from this study are as

follows First, there is evidence supporting the presence of systematic risk in the Australian

context Second, the above evidence may disappear when different portfolio formations and

different definitions of macroeconomic events are adopted

In summary, whether or not beta, or systematic risk, is alive in the Australian context depends on how portfolios are formed and macroeconomic events are classified These fundamental issues are generally known as puzzles in asset pricing studies and multi factor model has never been proven to withstand well when different markets/time/techniques are tested

1 An appreciation to an anonymous reviewer who provides critical comments to the previous version of the paper which was presented at the Vietnam’s Business and Economics Research Conference on 16-18 th November 2017

TABLE OF CONTENTS

CHAPTER 1 INTRODUCTION 1

1.1 An overview of asset pricing model 1

1.2 Research questions 3

1.3 Research objectives 3

1.4 A choice of Australia in this study 3

CHAPTER 2 LITERATURE REVIEW 5

2.1 Theoretical literature 5

2.1.1 Modern Portfolio Theory 5

2.1.2 Capital Allocation Line 7

2.1.3 Capital Asset Pricing Model 8

2.1.4 The Downside of the CAPM 11

2.1.5 Fama-French’s Three factor Model 12

2.1.6 Cahart’s Four factor Model 13

2.1.7 Fama-French’s Five factor Model 13

2.2 Empirical literature 14

CHAPTER 3 DATA AND METHODOLOGY 24

3.1 A brief description of the method 24

3.2 Data requirements and data sources 25

3.3 Portfolio constructions 26

3.3.1 Ten beta-sorted portfolios and Ten idiosyncratic risk-sorted portfolios 26

3.3.2 The 25 Fama-French size and book-to-market portfolios 28

3.3.3 Industry portfolios 30

3.4 Calculations of portfolio’s beta and portfolio’s return 30

3.4.1 Pooled regression 30

3.4.2 Fama-MacBeth regression 31

CHAPTER 4 EMPIRICAL RESULTS 32

4.1 Pooled regression’s result 32

4.2 Fama-MacBeth regression’s result 36

4.3 Result’s discussion 39

CHAPTER 5 CONCLUDING REMARKS AND POLICY IMPLICATIONS 40

5.1 Concluding remarks 40

5.2 Policy implications 42

References 44

Appendix 1 48

Appendix 2 52

Appendix 3 53

Appendix 4 54

Appendix 5 58

LIST OF TABLES

Table 2-1 Factor classification 18 Table 2-2 Approaches to Portfolio Formations 22 Table 3-1 Summary of the number of firms in 10 beta-sorted portfolios and in 10

idiosyncratic risk-sorted portfolios 27

Table 3-2 Summary of the number of firms in the 25 Fama-French size and book-to-market

LIST OF FIGURES

Figure 2-1 The attainable E, V combinations 6

Figure 2-2 The Capital Allocation Line 8

Figure 2-3 The strategic investment of investors 9

Figure 2-4 Equilibrium in the capital market 10

1 CHAPTER 1 INTRODUCTION

1.1 An overview of asset pricing model

Since the 1950s, asset pricing has seized great attention from policymakers, academics and practitioners which pushes it to the forefront of finance On the ground of the Modern Portfolio Theory (MPT), Markowitz (1952) presented the efficient frontier to demonstrate the trade-off between return and risk of an investment portfolio Few years later, building on the earlier work of Markowitz (1952), the Capital Asset Pricing Model (CAPM) was developed by Sharpe (1964) and Lintner (1965)

The CAPM gained acceptance for use by academics and practitioners for an extended period of time until the introduction of the three-factor model by Fama and French in 1992 This three-factor model has been widely applied to explain the observed stock returns In addition, various empirical studies provided evidence to argue that the CAPM does underestimate (overestimate) the return for low (high) beta asset However, empirical evidence has generally provided mixed evidence in relation to the validity of CAPM for the purpose of estimating the expected equity return Regardless of the criticism, CAPM still holds its position

of superiority of acceptance and use 74 per cent of 392 United State Chief Financial Officer (CFO) utilized CAPM to evaluate the cost of equity capital (Graham and Harvey, 2001) Similarly, Brounen, Jong and Koedijk (2004) discovered that 43 per cent of 313 European CFO’s decisions used CAPM for the same purpose Mckenzie and Partington (2014) in their report to the Australian Energy Regulator revealed that regulators in Australia, Germany, New Zealand and United Kingdom employed CAPM as a primary model to estimate the cost of equity while regulator in the United State of America utilized Dividend Discount Model (DDM) as the first option and CAPM as the second option Vo (2015), in his recent work, argued that the application of the Fama-French three-factor model into public policy under the context of Australia is not recommended In his study, he gathered weekly data of stock returns

of all listed Australian firms and market return from 1 July 2009 to 31 May 2014 from Bloomberg and utilized Fama-MacBeth (1973)’s two-stage regression technique He suggested three different scenarios to classify raw data as sub samples and five different portfolio formations to put stock in The portfolio formation is initiated from three fundamental ideas: (i) formation based on a number of stocks in each group is equal; (ii) formation based on firm’s market-capitalization; and (iii) formation based on top stock such as top 50 stocks, top 200

stocks Particularly, for each scenario, he adopted five approaches to portfolio formation For each approach, he applied the Fama-MacBeth (1973)’s two-stage regression technique to determine risky factor’s risk premium Finally, his finding showed that the value risk factor is well priced while the size risk factor is not under the context of Australian firms Therefore, the application of Fama-French three-factor model is not appropriate This result is also consistent with Brailsford, Gaunt and O’Brien (2012) and Faff (2004)’s findings

The central piece of the CAPM is its beta A recent work by Savor and Wilson (2014) presented that beta is after all an important measure of systematic risk They found that beta is strongly as well as positively related to average excess return on days when inflation, employment, or Federal Open Market Committee interest rate decisions, which are generally considered sources of systematic risk, are announced Overall, the key contribution of this Savor and Wilson (2014) study is that beta is still alive This simply means that CAPM is still alive at least in the US market From a status quo, our preferred approach is that CAPM and Fama-French three factor model are equally treated Fama-French three-factor model has been proven to work well in the US market However, it has equally been proven to not work well

in the Australian context (Vo, 2015) Equally, beta is still alive in the US market A similar question is that whether or not beta is still alive in Australia? We are not aware of any study on the issue which has been conducted recently This study is conducted to fill in the gap

Due to the foregoing dedicated research, probably, a pattern is observed to have emerged that different asset pricing models are suitable to different countries Therefore, this research raise up a hypothesis that whether the single factor asset pricing model-CAPM is usable or not

in calculation of a return on equity in Asia-Pacific in general or in Australia in particular

To shed light on the controversy about Sharpe-Lintner version of CAPM in the context

of Australia, this study bases on the pioneering work by Savor and Wilson (2014) for the US This research utilizes daily data for more than 2,200 Australian listed firms are collected from Bloomberg for the period from 1 January 2007 to 31 December 2016 Days with

announcements (the a-day) in relation to growth, inflation, employment, central bank announcements, bonds, housing, consumer surveys, business surveys and speeches from the

Prime Minister or the Governor of the Reserve Bank of Australia scheduled to be announced

are allocated into the group which is separated from the n-day (non-announcement days) group

Moreover, various portfolios are considered in this study including: (i) 10 beta-sorted portfolios; (ii) 10 idiosyncratic risk-sorted portfolios (iii) 25 Fama-French size and book-to-market portfolios; and (iv) industry portfolios Portfolio’s return is considered in two

dimensions: value-weight and equal-weight based direction In relation to methodology, the linear regression with panel-corrected standard errors method and Fama-MacBeth regression are both employed The structure of thesis is represented as follow: Chapter 2 is about Literature Review Data and methodology are discussed in the Chapter 3 Chapter 4 considers emperical results Concluding remarks and policy implications are put in the Chapter 5

1.3 Research objectives

This study is conducted to achieve the following research objectives:

 A confirmation of the validity/non-validity of employing the Capital Asset Pricing

Model (CAPM) in Australia on the ground of its Beta following Savor and Wilson (2014) approach

 The robustness of empirical findings in relation to the validity of the CAPM using Savor and Wilson (2014) approach when various portfolio formations and detailed classification of macroeconomic events are considered

1.4 A choice of Australia in this study

It is optimal if this study is conducted using data from Vietnam However, a preliminary analysis indicates that a substantially large volume of data is required for this type of study In addition, one of the key cornerstones of this empirical study is the availability of various announcements in relation to macroeconomic issues such as economic growth, money supply, unemployment and the others Unfortunately, this type of data is not publicly and substantially available in Vietnam

From 30 countries including in the Asia Pacific region, Australia is the best candidate at least on the following aspects: (i) a substantially large volume of data for listed firms are

available (more than 2,200 listed firms for more than 20 years of data); (ii) announcements of macroeconomic issues are publicly available and they are transparently recorded; (iii) Australia

is by all means a small, open, and advanced economy in the region; and (iv) support from the access of data is available and confirmed As such, Australia is selected for the purpose of this study

2 CHAPTER 2 LITERATURE REVIEW

2.1 Theoretical literature

2.1.1 Modern Portfolio Theory

Markowitz (1952) suggested Modern Portfolio Theory (MPT) which is one of the two standard asset pricing theories as an explanation of investment behavior The MPT is constructed from the rule (so-called expected returns-variance of returns rule or E-V rule) that the investor probably does consider expected returns a desirable thing and variance of returns

an undesirable thing and focuses on the stage of originating the relevant beliefs and ending with the choice of portfolio.2 In his interesting note, he stated that this rule has some advantages

to shed light on risk-averse investor’s behavior (i.e minimizing variance of returns for given expected returns and maximizing expected returns for given variance of returns) and to imply diversification

Followed by E-V rule, a rational investor, instead of allocating all his funds in a security with the greatest discounted value, he would diversify his fund among all those securities with give maximum expected returns However, the return, by itself, is not a constant number overtime and always influenced by specific-firm characteristics and nonspecific-firm characteristics Thus, as a matter of fact, investors have to bear risk Moreover, the portfolio with maximum expected return is not necessarily the one with minimum risk That is, there is

a rate at which investors would make a trade-off between expected return and risk To a averse investor, he would minimize risk for given expected return and maximize expected return for given risk Graphically, his best selection of risk-expected return combination is demonstrated by the curve which begins from A and ends at B in the following figure The risk factor is plotted on vertical line while the expected return plotted on the horizontal line

2 A belief includes a set of expected returns of securities and covariance between two any returns’ securities

Figure 2-1 The attainable E, V combinations

 µi: Expected return of security i

 Xi: The fraction of the investor's funds invested in security i

 σij: Covariance between security i and j

In terms of diversification, the E-V rule also suggests a guide to the right kind of diversification The diversification process is not as simple as increasing the number of securities in the portfolio One direction of diversification of a set of sixty different railway securities is different from the same size one with railroad, public utility, mining, manufacturing, construction and real estate,…One plausible reason is that securities in the same industry probably tends to move together greater than those in different industry Another explanation is that from equation (2), the bigger covariance is, the larger the variance of portfolio is That is, the latter is better than the former Finally, Markowitz concluded that a risk-averse investor probably follows the strategy of minimizing risk for given expected return

and maximizing expected return for given risk and, after all, right kind of diversification is better

2.1.2 Capital Allocation Line

Tobin (1958), in his attractive paper, developed his Separation Theorem to investigate the operation of the capital market While Markowitz focused on the risky assets and diversification, he took one step back for a broader view His idea, through Separation Theorem, stated that an investor allocates his wealth not only on the risky assets but also the riskless one It is said that the risky asset features for the equity market and riskless assets does for the bond market so his finding is probably one of the connections between the stock market and bond market For the graphical relationship, the Capital Allocation Line (CAL) was introduced as an appropriate nominator

Given the expected return of the risky assets and riskless assets are E(Rp) and Rf while their risks are measured by standard deviation denoted by σp and σf, respectively It is could be inferred that the σf is zero because by definition, the riskless assets produces a certain future return and their covariance – cov(p, f) - is zero, too.3 Suppose that an investor places a proportion of his wealth (α) in the risky assets the remainder (1 – α) in the riskless assets According to Markowitz, his expected return and risk are yielded:

𝐸(𝑅𝑐) = 𝛼𝐸(𝑅𝑝) + (1 − 𝛼)𝑅𝑓 (3)

𝜎𝑐 = √𝛼2σ𝑝2 + (1 − α)2σ𝑓2+ 2α(1 − α)cov(p, f) = 𝛼𝜎𝑝 (4)

Where:

 E(Rc): Expected return of the combination

 E(Rp): Expected return of risky assets

 Rf: Expected return of riskless assets

 σc: Standard deviation of the combination

Extracting α from equation (4) and substituting for it in equation (3) That yields:

From the equation (5), it is inferred that there is a linear line is drawn in the σc, E(Rc) space

Figure 2-2 The Capital Allocation Line

Source: Tobin (1958)

2.1.3 Capital Asset Pricing Model

Sharpe (1964) re-employed the hypothesis of E-V rule of risk-averse investor and portfolio’s expected return and its risk manipulation (Markowitz, 1952) and Tobin (1958)’s finding of wealth allocation of an investor into risky asset and riskless one in order to examine further the operation of capital market as investors physical interact His interesting note

probably could be divided into three sub sections: (i) the optimal investment policy; (ii) the

equilibrium of the capital market and (iii) the capital assets’ price

In relation to the optimal investment policy, basing on E-V rule, he pointed out that a rational investor is likely to pick up efficient portfolios which are X, B, A, θ and Y in the following figure Remarkably, all those points lie on the same line – the investment opportunity curve.4

4 Markowitz (1952) named it efficient E, V combinations

Figure 2-3 The strategic investment of investors

PA and Pθ at C, D and F, respectively The C, D and F portfolio all offer the same risk but their expected returns are not equal That of F portfolio is higher than that of D and that of D is higher than that of C Thus, F is chosen or investor’s portfolio is reflected by Pθ

Next, in terms of the equilibrium of the capital market, Sharpe (1964) stated that it is a consequence of investors’ optimal investment policy and their physical interaction Particularly, to an investor, a combination of stocks in the portfolio F is likely to bring attractive expected return as compared to portfolio C, D Therefore, each investor wants to own that such portfolio F and reject portfolio C, D As a matter of fact, to stocks in the portfolio F, higher

demand at any given current price leads to less stocks’ expected return Due to lower stocks’ expected returns, expected return of the F portfolio is lower too This leads to portfolio F become inefficient or its position shifts to the left while its risk does not vary Similarly, by the same arguments, portfolio C, D become efficient or its position shifts the right while its risk does not vary Consequently, the movement of those portfolios makes the investment opportunity curve to be flatter This result is represented in the following figure

Figure 2-4 Equilibrium in the capital market

Source: Sharpe (1964)

Finally, regarding to the capital assets’ price, the researcher proposed the terminology of systematic risk denoted by β which is defined as the response of stock’s return with respect to return of efficient portfolio in order to explain expected stock return in equilibrium It is also inferred that β, by itself, is a component of the asset’s total risk On the ground of the idea of slope of the tangent line of investment opportunity curve at θ equal to the slope of Pθ yields the following equation:

𝐸(𝑅𝑖) = 𝑃 + (𝐸(𝑅𝑔) − 𝑃)𝛽𝑖𝑔 (6)

Where:

 E(Ri): Expected return of stock i

 P: The pure interest rate

 E(Rg): Expected return of efficient portfolio g

 βig: The response of stock’s return with respect to return of efficient portfolio According to the above equation, given the pure interest rate and expected return of efficient portfolio g, low β asset is probably to associate with low expected return and vice versa This finding is also consistent with risk-return trade off regime reflected through Markowitz (1952)’s efficient E-V combination curve

2.1.4 The Downside of the CAPM

In 2002, Estrada, following Bawa and Lindenberg (1977) and Hogan and Warren (1974)’s finding, criticized that probably CAPM is not an appropriate model to manipulate assets return due to the fact that variance of return employed as a measure of risk He pointed out that, in practice, it is probably not all assets returns follow the property of symmetry and normality which CAPM relies on Therefore, he suggested downside beta as an alternative outlet for risk In his research, the downside beta is defined as the ratio of co-semi-variance between assets return and market return to the market’s semi-variance of return The key point differentiates his approach in calculation of beta from that mostly mentioned in textbook is that the former one only takes care of returns below sample mean in contrast to the latter one does both The reason behind originated from the fact that investors do not worry about returns jumps above the sample mean In order to demonstrate convincible justification, he utilized Morgan Stanley Capital Indices database of emerging markets including monthly return of 27 countries One of the regression analysis is made in which mean return is the regressand while four risk variables beta, downside beta, semi-deviation and standard deviation are regressors

As a matter of fact, the result showed that only downside beta is the only significant one Put

it differently, downside beta is better than beta in terms of assets return explanation To deepen this finding, the researcher also divided 27 countries into three equal-member groups ranked from highest beta to the lowest beta The result also reaffirms the foregoing regression’s result that downside beta dominates beta in terms of assets return explanation

2.1.5 Fama-French’s Three factor Model

Fama and French (1992), in their remarkable paper, showed a voice that the market ß as employed alone probably has no ability to describe stock return of nonfinancial firms listed on the NYSE, AMEX and NASDAQ in the period of 1963-1990 This study attracts academics’ and practitioners’ attention at that time because its finding contradicts traditional wisdom about the role of market ß Indeed, following Bhandari (1988) and Banz (1981)’s study, they stated that the existing negative relationship between size (ME-equals to stock prices times number

of share) and average stock return Moreover, they also found that a robust positive relationship between book-to-market equity (BE/ME) and average stock return (Stattman 1980; Rosenberg, Reid and Lanstein 1985; Chan, Hamao and Lakonishok 1991) On that basic, in 1993, Fama

and French argued that there are three variables make stock return deviate around its mean: (i)

Rm-Rf: the excess return between market portfolio and risk-free rate, called market risk

premium; (ii) HML: the return of high book-to-market ratio portfolio less low book-to-market

ratio portfolio, called value premium and (iii) SMB: the difference in return between small capitalization portfolio and big capitalization portfolio, called size premium All of them play

as a risk factor in the sense that they capture variation in stock return Put it differently, the expected stock return is explained by market risk premium, high minus low and small minus big Based on the following argument, Fama and French suggested another asset pricing model called Fama-French three-factor model or FF3F, for short The model is expressed as follow:

E(Ri) = Rf + (E(Rm) – Rf)βmkt + E(SMB)βsmb + E(HML)βhml

Where:

 E(Ri): Expected return of stock i

 Rf: The risk-free rate

 βmkt: The response of stock’s return with respect to return of market portfolio

 βsmb: The factor loading of stock on SMB factor

 βhml: The factor loading of stock on HML factor

 E(Rm): Expected return of market portfolio

 E(SMB): The difference in expected return between small capitalization portfolio

and big capitalization portfolio

 E(HML): The expected return of high book-to-market ratio portfolio less low

book-to-market ratio portfolio

2.1.6 Cahart’s Four factor Model

Jegadeesh and Titman (1993) investigated firms with significant profits listed on NYSE and AMEX from 1965 to 1989 for the explanation of stock return Based on the examination, they suggested a strategy for holding stock is that purchase stocks performed well and selling those did poorly in the last 6-month will generate a significant return in the next 6-month As

a matter of fact, this strategy originated from delayed price reactions to firm-specific information instead of being implied by stocks’ systematic risk or their delayed reaction to common risk factor Four year later, their work in combination with FF3F was inherited by Carhart (1997) He constructed a model in order to describe the return of mutual fund equity named after him called Cahart four-factor model or C4F, for short The model is expressed as follows:

E(Ri) = Rf + (E(Rm) – Rf)βmkt + E(SMB)βsmb + E(HML)βhml + E(WML)βwml

Where:

 E(WML): The difference in expected return between diversified winner portfolio

and looser portfolio Other factors are defined similarly in the FF3F model

2.1.7 Fama-French’s Five factor Model

Fama and French (2015) based on the dividend-discount model (DDM) to argue that some other factors are able to explain the share price With a bit of manipulation, they express the relation between expected return and expected investment, book to market ratio and expected investment as follows:

 Mt: The total market value of the firm’s stock

 Yt + τ: The total equity earnings

 dBt + τ: Change in the total book equity

 r: The long-term average expected stock return

 t, τ: Indicator of time period

Dividing both side of the foregoing equation by book equity That yields:

𝑀𝑡

𝐵𝑡 = ∑

𝐸(𝑌𝑡+𝜏−𝑑𝐵𝑡+𝜏) (1+𝑟)𝜏

∞ 𝜏=1

Fama and French stated that the above equation produces three implication about

expected returns: (i) keep other things constant, a higher book-to-market equity ratio pertains

to higher expected return; (ii) keep other things constant, a higher expected earnings yields a higher expected return and (iii) keep other things constant, a higher expected growth in book equity related to lower expected return Due to those implications, the authors concluded that investment and profitability are likely able to describe expected stock return Their finding is also consistent with the previous findings such as Novy-Marx (2013); Haugen and Baker (1996); Fairfield, Whisenant and Yohn (2003); Titman, Wei and Xie (2004) The work of Fama and French (2015) probably could be expressed by the following equations:

Rit– Rft= αi + b i(Rmt– Rft) + s iSMBt+ hiHMLt+ r iRMWt+ c iCMAt

Where:

 Rit: The time-series return of stock or portfolio i

 Rft: The risk-free rate return

 Rmt: The return on the value-weighted market portfolio

 SMBt: The difference in return between small capitalization portfolio and big

capitalization portfolio

 HMLt: The return of high market ratio portfolio less low

book-to-market ratio portfolio

 RMWt: The difference between return on the diversified portfolios of stocks

with robust and weak profitability

 CMAt: The difference between the return on diversified portfolios of stocks of

low and high investment firms

2.2 Empirical literature

In the empirical studies focusing on capital asset pricing model, undoubtedly, some of the most well-known are the work of Fama and MacBeth (1973) and Jensen, Black, & Scholes (1972) In relation to Fama and MacBeth’s study, in order to examine the validity of CAPM

on practice, they verified the risk-return tradeoff on the New York Stock Exchange They observes that, on average, a linear positive relation between the two Moreover, they also confirmed CAPM’s hypothesis that except for beta, there is no measure of risk which possibly could systematically influence the return In addition, Shapiro and Lakonishok (1984), in an effort to revive CAPM, spilt their data on market excess return That is, they classified the

observations into up market group (when market return is greater than the risk-free rate) and into down market group (when market return is less than the risk-free rate) They discovered that positive significant estimated beta’s coefficient is associated with the former group and negative significant estimated beta’s coefficient for the latter one In a different language, as CAPM’s prediction, high beta stocks do better in up market and worse in down market than low beta Few years later, Pettengill, Sundaram and Mathur (1995), in the same data classification approach, continued to reaffirm the findings for the US market As a result, they are considered as key contributors to the variant of CAPM, the so-called conditional capital asset pricing model.

Since the work by Shapiro and Lakonishok (1984) got published, research area on asset pricing has witnessed a significant number of studies to have been conducted Among the studies, Lam (2001) for Hong Kong Stock Exchange perhaps is probably a prominent study in Asia Pacific region This study demonstrated that beta and return are positive correlated in the up-market period while this relationship turns out to be negative in the down-market period In addition, Tanga and Shumb (2003) investigated the conditional capital asset pricing model by employing data at global scale which includes many countries in the Pacific Ocean region.7Similar to Shapiro and Lakonishok (1984), the results confirmed that beta and return moved in the same direction as market return excess the risk-free rate In short, findings from those studies suggested that beta was a useful risk measure Not only advocate of the CAPM was discovered in the Stock Exchange but also a majority of regulators in Australia, Germany, New Zealand, USA, Canada and UK still employ it to measure the cost of equity (Sudarsanam, Kaltenbronn, & Park, 2011).8

In contrast, various studies have demonstrated their concerns of the CAPM which

originated from its simplicity and implication AL-Qudah and Laham (2013), in their study on

the determinants of stock return of 48 industrial companies listed in the Amman Stock Exchange from January 2000 to December 2009, argued that the statistically significant impact

of systematic risk on stock return could not be established in the study Moreover, employing data gathered from FAME database over the period of April 2000 to June 2007, Ramlogan and Bhatnagar considered that the Fama-French three-factor model (FF3F) was superior to explain the stock returns than the CAPM in United Kingdom stock market

7 This list contains Japan, Canada, US, Hong Kong, Singapore and Taiwan

8 See Appendix 3 for more detail

Daniel, Titman and Wei (2001) employed monthly data of listed firm on Tokyo Stock Exchange which accounts for nearly 85 per cent of the total market size of Japan in the period

of 1971-1997 to emphasize the role of characteristic model over the factor model in the context

of Japan Due to the size and book to market ratio observations, they classified firms into 25 portfolios to test the validity of CAPM and FF3F alike One of their findings demonstrated that CAPM was likely not appropriate while FF3F probably is the case in Japan In terms of CAPM, basing on the fact that the 25 intercepts implied by OLS estimation method on those 25 portfolios were jointly statistically significant different from zero, therefore, they argued that some firms located in the small portfolios and high book to market ratio portfolios earned very high CAPM risk-adjusted abnormal return Put it differently, the stock excess return could not totally embraced by systematic risk

In addition, Choudhary and Choudhary (2010), in their empirical paper, also concluded the same result in relation to the validity of CAPM from 1996 to 2009 in Bombay Stock Exchange They stated that the stock excess return was not certainly explained by systematic risk Particularly, higher ß was not associated with higher expected return and the intercept was statistically significant different from zero Those findings contradict the implication of traditional CAPM that the market risk premium should be positive and the intercept should be zero In conclusion, this research is consistent with Daniel, Titman and Wei (2001), Bajpai and Sharma (2015) Similarly, the risk-return tradeoff suggested by traditional CAPM was also not observed on the Malaysian Stock Exchange from 2001 to 2013 (Mollik, 2014) or Nigerian stock exchange (Oke, 2013; Olakojo and Ajide, 2010)

Visually, the traditional CAPM is likely to be an elegant effective model to express return

in terms of risk Howerer, the anti-CAPM movement was initiated since it has given birth by Sharpe (1964), Lintner (1965) One of the earliest effort in discovering stock return’s explanation was conducted by Basu (1977) He observed firm’s performance traded on NYSE from 1956 to 1971 and went to a conclusion that low P/E probably associated with high return and vice versa In short, a propotion of stock return is reflected by P/E or this price-earning ratio should be a determinant of stock return At the same time, Roll (1977) stated that although market portfolio is an efficient portfolio which implied by Markowitz’s viewpoint, the formation of market portfolio is a challenge because every asset must be included Moreover,

he also concluded that an examination of linear relation between expected return and beta is only verified if and only if the true market porfolio is employed in the examination Therefore, the validity of CAPM equation probably is still questionable Being inspired by Roll, some

evidence against CAPM has bloomed (Shanken, 1985; Banz, 1981; Reinganum, 1981; Amihud

& Mendelson, 1986; Fama and French, 1992).9

In short, for a general view of the invalidity of the CAPM in the real world, Fernandez (2015) criticized that is an absurb model due to the fact that its assumptions are unrealistic For example, investors are homogenous Investors are guided by E-V maxim Moreover, Yang and Chae (2008) were also in the same side They argued that transaction cost, investor irrationality and missing risk factors which were not mentioned in the literature implythe failure of CAPM

As such, CAPM has been mirroring the complexity of real world by senseless simplification

In the asset pricing research field, the solution to the asset return is a thorny and attractive subject In response to the appeal, a majority of researches have been conducted to figure out the cross-sectional predictors of stock return Subrahmanyam (2010) contributed 50 variables, McLean & Pontiff (2015) discovered 82 factors, Green, Hand and Zhang (2013) offered 330 firm specific characteristics Remarkably, Harvey, Liu and Zhu (2015) utilized more strict statistical criterion than normal to filter truly significant factors.10 Generally, they segmented factors as follows:

Table 2-1 Factor classification

Source: Harvey, Liu and Zhu (2015)

In the development progress of asset pricing, probably, it is a deficiency without mention the seminal work of Fama and French because it covered all studies had been done during two decades pertain to asset pricing models One common feature is that these researches attempted

to relate size, book-to-market ratio, earnings-price ratio and leverage to stock return For a comprehensive view of stock return determinants for US market, in 1992, Fama and French demonstrated that cross-sectional variation in stock return could be captured by size and book-to-market ratio Their results, in short, narrowed down from four variables to two variables due

to size and book-to-market ratio reflect the effect of earnings-price ratio while leverage is contained by book-to-market ratio One year later, they pushed this finding further through the construction of small minus big (SMB) portfolio and high minus low (HML) portfolio whose fundamental initiated from allocation firm’s stock to appropriate sub portfolios with respect to theirs book-to-market ratio and size independently Moreover, they also argued that apart from beta, SMB and HML portfolios should be a proxy for risk According to them, the single factor asset pricing model needed to be upgraded to a new one As such, FF3F was proposed However, the FF3F is criticized for data mining-based formation instead of being employed theoretical foundation (MacKinlay, 1995; Kogan & Tian, 2015; Wang & Wu, 2011) Similarly, the C4F is also not recommended in practise because it was given birth by bottom up approach Turning to the researches which dedicated to the validity of FF3F in the practise, it could

be seen that although FF3F is likely to outperform CAPM in the sense of stock return’s explanation, the former model is still far from level of large scale deployment

O'Brien, Brailsford and Gaunt (2008) stated that their findings advocated the superiority

of FF3F The research utilized data in the period of 1982-2006 from Australian Stock Exchange which contains approximately 98 per cent of all listed Australian firms The dataset played as

an advantage over the perious papers which limited by time spanning and market coverage alike in the sense of the authors The 25 Fama and French size and book to market portfolio formation was employed Particularly, on the one hand, in terms of book to market, stocks were divided into five segments with each segment compressed the same numbers of observations

On the other hand, in relation to size aspect, stocks were allocated into five portfolios in which the first portfolio embraced stocks which account for the largest 75 per cent of market capitalisation, the second portfolio made up the remaining largest 15 per cent of market capitalisation, the third portfolio contained the remaining largest 5 per cent of market capitalisation and the fourth and the fifth portfolio emcompassed the remaining largest 3 per cent and 2 per cent of market capitalisation, respectively As a result, there were 25 smaller portfolios originated from the intersection of 5 portfolios on the book to market axis and 5 portfolios on the size axis For those 25 portfolios, the CAPM regression was only able to explain return in 6 ones due to the fact that the intercepts in the remains were statistically

significant different from zero Put it differently, the pricing error signal exists in the CAPM

or beta coefficient, by itself, cannot explain stock return In contrast, nearly all factor loadings

of market factor, size factor and value-growth factor were statistically different from zero Moreover, the statistics of intercepts exhibit opposite resuts which implied that those three foregoing factors capture stock return In short, FF3F outperformed CAPM

Similarly, the research of Phong and Hoang (2012) for listed firms on Vietnam’s stock market from Jan 2007 to Dec 2011 also produced the same conclusion Using different portfolio formation as compared to those of O'Brien, Brailsford and Gaunt (2008), the authors divided dataset into six portfolios independently sorted on size and book-to-market ratio including: B/H, B/M, B/L; S/H, S/M and S/L and embeded FF3F and CAPM in them Following the magnitude of adjusted-R2 and level of the number of significant estimates, the researchers posited that CAPM was likely to be crowded out by FF3F At the same time, the research of Wasiuzzaman and Monfared (2012) in Malaysia which employed similar data classification and comparable benchmarks also generated a consistent result

In contrast, Gharghori, Lee and Veeraraghavan (2009) concluded the same result through they demonstrated concerns in relation to the FF3F’s application In their study, for each variable of interest (e.g market equity, book to market, earings to price, cashflow to price, leverage and share turnover), stocks were allocated into sextiles After that, the CAPM regression and FF3F regression with GMM were made on six portfolios, respectively The results which compressed by estimates, their t-statistic value, adjusted-R2 and Wald test posited that the FF3F played a better role in stock return’s explanation as compared to CAPM at least

in the size effect, book to market effect, earnings to price effect and cashflow to price effect Nevertheless, the FF3F cannot fit returns in some cashflow to price effect-differentiated portfolios As such, it is far from conclusion that FF3F could be deployed in Australia as it used to be in the US

Focusing on the invalidity of FF3F on practice, Vo (2015) posited that FF3F probably was an seminal effort for academic purposes but the adoptation of the model into practice was problematic; and as such, the model is recommeneded in the context of Australia The conclusion extracted from the numerical results of factor loadings and intercepts in the FF3F model were not in line with the expectation Particularly, from the dataset covered stock return

of Australian listed firms in the five-year period 2009-2014, the author employed five different data classification approaches adopted in the previous to build up tested portfolios for the verification of FF3F More importantly, prior to the portfolio formation process, observations

were filtered by three distinguish scenarios That is, for each scenario, five different data classifcation approaches were deployed later Under each test portfolio, he utilized Fama-MacBeth regression to figure out the estimated coefficients of market factor, size factor and value factor In addition, the researcher also argued that those estimated coefficients must be positive number and the intercepts were not statistically significant different from zero in all portfolios to reflect the nature tradeoff between return on equity and factor However, this prerequisites were not satisfied because of the HML’s factor loading exhibited inverse relation

in nearly all tested portfolios In relation to two other factors, market’s factor loading demonstrated the same phenomenon in the two first scenarios, whereas negative SMB’s factor loading was observed in third scenario Hence, it is likely reasonable to argue that the adoption

of FF3F may lead to serious causes The following table contains various portfolio formation approaches summaried by Vo (2015)

Table 2-2 Approaches to Portfolio Formations

1

All stocks are ranked by size and sorted in five portfolios with each portfolio containing the same number of stocks Stocks by book-to-market ratios (lowest to highest) are ranked and quintile portfolios of equal numbers of stocks are formed

2

Each firm (largest to smallest) by market capitalisation is ranked and then assigned to one of five size portfolios The largest size portfolio contains the first n number of stocks that make up 75% of total market capitalisation The second portfolio contains the next

n number of stocks that make up the next 15% of total market capitalisation The next

3 portfolios contain the next 5%; 3%; and 2% of total market capitalisation These market capitalisation breakpoints are argued to parallel the findings of Fama and French (2006) For value factor, portfolios are constructed using book-tomarket breakpoints determined on the basis of sorts on the top 200 stocks and subsequently applied to the full sample of stocks

3

For a size factor, each stock is first ranked by market capitalisation (largest to smallest) The largest size portfolio contains the largest 50 stocks The second size portfolio contains the next 150 stocks (i.e stocks 51–200) The third and fourth size portfolios contains the next 100 and 200 stocks The fifth size portfolio contains all other listed stocks For a value factor, breakpoints for book-to-market value are determined on the basis of the top 200 stocks and then applied to the full sample of stocks

4

For a size factor, the approach is similar to Approach 3 However, for a value factor, this approach adopts an allocation in which stocks are allocated into quintile portfolios where each portfolio contains the same number of stocks

5

Method 5 is the same portfolio construction approach as Method 4 but on a reduced sample of stocks Specifically, stocks with a price of less than $0.20 are excluded from the sample

Source: Vo (2015)

Following those dedicated researches, from the set of surveyed countries, one pattern emerges that the traditional CAPM is applicable to a number of nations while it is not recommended on the remains Even in the context of US and Japan – two leader nations in the Asia Pacific region, the validity of CAPM on practice is still a controversial topic In addition, the study of FF3F’s adoption left a confusing picture at least in the context of Australia However, CAPM has equally been proven to not work well in the Australian context Therefore, this research rises up a hypothesis that whether the single factor asset pricing model-

CAPM is usable or not in calculation of a return on equity in Asia-Pacific in general or in Australia in particular

3 CHAPTER 3 DATA AND METHODOLOGY

3.1 A brief description of the method

The main purpose of this research is to consider the validity of CAPM under the context

of Australia This study adopts the approach from Savor and Wilson (2014) is that all

observations are separated into a-day (announcement days) group as it is related to days on which growth, inflation, employment, central bank announcements, bonds, housing, consumer surveys, business surveys and speeches from the Prime Minister or the Governor of the Reserve Bank of Australia scheduled to be announced and n-day (non-announcement days) group as it

pertains to the other days In short, the CAPM equation under nested data separation is represented as follows:

 D Dummy variable D = 1 if day is a-day and vice versa

The daily return is measured by the difference in natural logarithmic of two continuous stock close prices Mathematically, the daily return is expressed as follows:

Ri,t = ln(Close Pricei,t) - ln(Close Pricei,t-1) Concerning regression technique, this study employs linear regression with panel-corrected standard errors method The rationale behind is that it fixes heteroscedasticity and contemporaneously correlated across panels of disturbances

Moreover, to verify the validity of CAPM in Australia, this research also employs the Fama and MacBeth (1973) regression for three reasons: (i) it is a practical way to figure out how the risk factors describe the asset return; (ii) it is developed by Professor Eugene F Fama

is known as “The Father of Finance” and inventor of Fama-French three-factor model and (iii) this regression technique is also proposed by Savor and Wilson (2014), Vo (2015) and Brailsford, Gaunt and O'Brien (2012) Technically, the Fama and MacBeth regression

comprises two stages In the first-stage, to each single time period, a cross-sectional regression

is performed Next, in the second-stage, those first stage-estimated coefficients are averaged across time to obtain the final estimates In the scope of this research, the Fama and MacBeth regression is utilized in the two following models:

RA

i, t +1 – RA

f, t + 1 = α0 + γ0βA

i, t + µi, t + 1 and

RN

i, t +1 – RN

f, t + 1 = α1 + γ1βN

i, t + µi, t + 1 Where RA

i, t +1 – RA

f, t + 1 is the daily expected excess return in a-day, RN

i, t +1 – RN

f, t + 1 is the daily expected excess return in n-day βAi, t is the asset’s market beta in a-day and βNi, t is the asset’s market beta in n-day (both estimated over the previous year using daily returns)

3.2 Data requirements and data sources

Data covers 2,200 Australian listed firms on Australian Securities Exchange from Bloomberg from 1 January 2007 to 31 December 2016.11 Two additional required inputs for CAPM, being the risk-free rate and the market returns, are also collected from the same source and the same period The risk-free rate’s proxy-Commonwealth Government bonds from Reserve Bank of Australia with the maturity of 10 years-is adopted in this study The market return is the return summation of all listed Australian firms collected by this research.12 Moreover, the dataset also records news (day of issue, impact) in relation to event types

such as growth, inflation, employment, central bank announcements, bonds, housing, consumer surveys, business surveys and speeches from the Prime Minister or the Governor of the Reserve Bank of Australia Days contain at least one of those event types are called announcement days

or a-days and vice versa Moreover, their impacts on market-moving potential are collected

and classified into three levels: (i) high expected impact; (ii) medium expected impact and (iii) low expected impact In the case of more than one impact levels are presented, the highest one would be chosen Those data are available and judged by Forex Factory website (http://www.forexfactory.com/) In addition, for consistency, this segment of dataset spans from1 January 2007 to 31 December 2016

11 From more than 2,200 Australian listed firms, this research removed firms whose industrial sectors are not recorded Moreover, research also did the same to firms which had less than 25% to total observations Although this criterion is likely arbitrary, an unbiased view is the top priority However, as market return is calculated, all of those rejected firms are accounted to keep it as closed as possible to its theoretical version

12 The Australian Securities Exchange’s index could be referred to the S&P/ASX 20, S&P/ASX 50, S&P/ASX

200, S&P/ASX 300 and All Ordinaries

3.3 Portfolio constructions

Beta is after all a crucial measure of systematic risk and it is at the heart of the CAPM which can be used to estimate asset return In this paper, the beta estimates are used as an important determinant of asset returns in announcement days as compared to other days In the context of the United State of America, Savor and Wilson argued that stock market beta was strongly related to average returns in ten-beta sorted portfolio, the 25 Fama and French size and book-to-market portfolio, ten-idiosyncratic risk sorted portfolio, industry portfolio and even for non-equity asset such as government bonds and currency carry-trade portfolio Therefore, it is essentially to demonstrate CAPM’s performance on those portfolios if the targeted market is not US However, due to data is not available for government bonds and currency carry-trade portfolio, this study adopts the first fourth portfolios to examine the validity of CAPM in the context of Australia

3.3.1 Ten beta-sorted portfolios and Ten idiosyncratic risk-sorted portfolios

In relation to 10 beta-sorted portfolios, each firm in the dataset is allocated to the

corresponding portfolio by the following procedure Initially, for each year, separately for

announcement day and non-announcement day, individual stock market beta was estimated using one year of daily returns, then firm was sorted into deciles due to the estimated beta In

a simple language, all firms in the dataset were allocated into one of the ten portfolios For each year, the first portfolio was featured by the 10 per cent lowest-beta while the tenth portfolio comprises the top 10 per cent highest-beta This kind of construction procedure is repeated every year Technically, the one year rolling beta was produced using the ordinary least square for the following equation:13

Rit – Rft= αit + b i(Rmt– Rft) + µitSimilarly, the idiosyncratic risk-sorted portfolio was constructed in the same manner However, it is noted that for this portfolio formation, every time the individual stock market beta was generated, the standard deviation of return residual of the foregoing equation was record, too Then, firm was also sorted into deciles due to that statistical number Particularly, for a year, the first portfolio contains top 10 per cent lowest-standard deviation of return residual Meanwhile, the materials of tenth portfolio are top 10 per cent highest-standard deviation of return residual The idiosyncratic risk-sorted portfolios were rebalanced annually

13 According to the central limitation theorem, the one year rolling beta is estimated only if the minimal observation-30-is satisfied