Asset Growth and Cross-Sectional Stock Returns on the Johannesburg Stock Exchange

Transcription

1 Asset Growth and Cross-Sectional Stock Returns on the Johannesburg Stock Exchange A research report Presented to The Graduate School of Business University of Cape Town In partial fulfilment of the requirements for the Masters of Business Administration Degree By Samukelo Sifiso Zwane MBA full-time 2016 Supervised by: Sean Gossel

2 ABSTRACT The study investigates the relationship between asset growth and the cross-section of stock returns on the Johannesburg Stock Exchange (JSE). Asset growth is compared with book-to-price ratio and cash-flow-to-price ratio, which previous researchers have found to be significant at explaining the cross-section of stock returns on the JSE. The analysis is conducted on annual stock returns over the period 2000 to 2015, which has been broken down into two sub-periods: 2000 to 2007 and 2008 to The overarching aim of the study is to determine whether asset growth can be considered as one of the significant factors when explaining the cross-section of stock returns on the JSE. Additionally, the study aims to identify whether the level of significance and predictive power of style attributes changes over different business cycle phases. The study follows a similar method to Fama and Macbeth (1973). A univariate analysis is performed through cross-sectional forward regression on asset growth, book-to-price ratio and cash-flow-to-price ratio. Results are analysed over the three periods, with and without systematic risk (beta). The results of the study show that asset growth is significant at explaining cross-section of stock returns. This is in line with findings by Hoffman (2012), which found that there is a weak relationship between asset growth and cross-sectional stock returns. Even though, asset growth is significant when explaining cross-section of stock returns, the predictive power of asset growth is weak. Taking into Copyright account systematic risk does not change the significance of asset UCT growth when explaining the cross-section of stock returns. The paper contributes to literature by showing that the level of significance of asset growth in explaining cross-section of stock returns varies over different business cycle phases in the JSE. In the periods 2000 to 2007 and 2000 to 2015, asset growth was not significant at explaining the crosssection of stock returns. However, asset growth was significant at explaining the cross-section of stock returns over the period 2008 to Book-to-price ratio and cash-flow-to-price ratio are found to be significant when explaining the cross-section of stock returns, confirming results by Hodnett (2010). Book-to-price ratio and cashflow-to-price are found to be significant when explaining the cross-section of stock returns over the periods 2000 to 2015 and 2000 to What s more, the predictive power of the variables is high over both periods, even though the accuracy of the prediction is low. The variables are not significant at explaining the cross-section of stock returns over the period 2008 to 2015 and their predictive power is low. Introducing beta in the analysis does not change the variables level of significance. Asset growth, book-to-price and cash-flow-to-price are all significant when explaining cross-section of stock returns over different business cycle phases. Introducing systematic risk does not change the level of significance of the variables. Even though, the variables can have predictive power in some periods, none of the variables is significantly accurate at predicting forward returns in a univariate framework.

3 PLAGIARISM DECLARATION I know that plagiarism is wrong. Plagiarism is to use another s work and pretend that it is one s own. I have used a recognised convention for citation and referencing. Each significant contribution and quotation from the works of other people has been attributed, cited and referenced. I certify that this submission is my own work. I have not allowed and will not allow anyone to copy this essay with the intention of passing it off as his or her own work. Signature: Name: Samukelo Sifiso Zwane Exam Number: ZWNSAM002 Date: 7 th December 2016

4 ACKNOWLEDGEMENTS A special thanks to my wife and daughter for supporting me throughout this challenging MBA year. I love you guys. I would like to acknowledge my supervisor, Dr Sean Gossel, for guidance and encouragement throughout the write up of this paper. The speed at which you responded to my queries and questions made the write up process efficient. Your knowledge and insights have been invaluable. I would also like to thank Mary Lister for her patience in checking my references and formatting the paper. A special acknowledgement to the Graduate School of Business Library for the use of Datastream terminals. To my MBA classmates, it was the most challenging and insightful year. Thank you for sharing this experience with me. Finally, I would like to acknowledge my entire family (My Father and Mother, Brothers, Sisters and In-laws) for their constant support throughout this challenging MBA year. You all played a special role in making this year a success. 4 P a g e

5 TABLE OF CONTENTS ABSTRACT... 2 PLAGIARISM DECLARATION... 3 ACKNOWLEDGEMENTS... 4 TABLE OF CONTENTS... 5 LIST OF TABLES... 8 LIST OF FIGURES... 9 CHAPTER 1: INTRODUCTION RESEARCH AREA PROBLEM STATEMENT PURPOSE, SIGNIFICANCE OF STUDY AND GAP IN LITERATURE RESEARCH QUESTIONS RESEARCH ETHICS CHAPTER 2: LITERATURE REVIEW MODERN PORTFOLIO THEORY AND THE CAPITAL ASSET PRICING MODEL Copyright INTRODUCTION... UCT EMPIRICAL EVIDENCE IN DEVELOPED MARKETS EMPIRICAL EVIDENCE IN DEVELOPING MARKETS EFFICIENT MARKET HYPOTHESIS BEHAVIOURAL FINANCE INTRODUCTION EMPIRICAL EVIDENCE IN DEVELOPED MARKETS EMPIRICAL EVIDENCE IN DEVELOPING MARKETS SUMMARY OF LITERATURE REVIEW CHAPTER 3: METHODOLOGY AND DATA METHODOGY DATA HARVESTING EXPLORATORY VARIABLES DATA ADJUSTMENTS COMPLETENESS LIQUIDITY OUTLIERS STANDARDISATION P a g e

6 3.3 DATA BIAS DATA SNOOPING LOOK AHEAD BIAS SURVIVORSHIP BIAS CHAPTER 4: DESCRIPTIVE STATISTICS AND UNIVARIATE ANALYSIS DESCRIPTIVE STATISTICS UNIVARIATE ANALYSIS UNIVARIATE ANALYSIS: FAMA-MACBETH METHOD UNADJUSTED RETURNS UNIVARIATE ANALYSIS: FAMA-MACBETH METHOD RISK-ADJUSTED RETURNS STRENGTH OF FORECASTING ABILITY ASSUMPTIONS AND BIASES OF ORDINARY LEAST SQUARES REGRESSION RESEARCH FINDINGS, ANALYSIS AND DISCUSSIONS INTRODUCTION FAMA-MACBETH UNADJUSTED RETURNS FULL PERIOD: 2000 TO SUB-PERIOD 1: 2000 TO SUB-PERIOD 2: 2008 TO COMPARISON OF FAMA-MACBETH UNADJUSTED AND BETA-ADJUSTED RESULTS FULL PERIOD: 2000 TO SUB-PERIOD 1: 2000 TO SUB-PERIOD 2: 2008 TO STRENGTH OF FORECASTING ABILITY FULL PERIOD: 2000 TO SUB-PERIOD 1: 2000 TO SUB-PERIOD 2: 2008 TO SUMMARY AND CONCLUSION FUTURE RESEARCH BIBLIOGRAPHY APPENDIX 1-STYLE VARIABLES AND COMPUTATION APPENDIX 2-DATA DISTRIBUTION BOOK TO SHARE PRICE VALUE CASH TO SHARE PRICE VALUE GROWTH(COOPER) P a g e

7 GROWTH (HOFFMAN) P a g e

8 LIST OF TABLES Table 1: Data variables Table 2: Descriptive statistics Table 3: Fama-MacBeth univariate results Full Period Table 4: Fama-MacBeth univariate results sub-period Table 5: Fama-MacBeth analysis univariate results sub-period Table 6: Fama-MacBeth Unadjusted versus Beta Adjusted results-full Period Table 7: Fama-MacBeth Unadjusted versus Beta Adjusted results-sub period Table 8: Fama-MacBeth Unadjusted versus Beta Adjusted results-sub period Table 9: Strength and accuracy of forecasting ability- Full Period Table 10: Strength and accuracy of forecasting ability- sub period Table 11: Strength and accuracy of forecasting ability- sub period P a g e

9 LIST OF FIGURES Figure 1: Histogram: Book to share price-full Period Figure 2: Histogram: Book to share price-sub period Figure 3: Histogram: Book to share price-sub period Figure 4: Histogram: Cash flow to share price-full Period Figure 5: Histogram: Cash flow to share price-sub period Figure 6: Histogram: Cash flow to share price-sub period Figure 7: Histogram: Growth (Cooper)-Full Period Figure 8: Histogram: Growth (Cooper)-sub period Figure 9: Histogram: Growth (Cooper)-sub period Figure 10: Histogram: Growth (Hoffman) Full Period Figure 11: Histogram: Growth (Hoffman)-sub period Figure 12: Histogram: Growth (Hoffman) - sub period P a g e

10 CHAPTER 1: INTRODUCTION 1.1 RESEARCH AREA Markowitz (1952) laid down the foundation of modern portfolio theory (MPT) through his seminal paper Portfolio Selection. The paper asserts that rational investors will invest in an efficient portfolio; this is a portfolio which maximises return for a given level of risk or minimizes risk for a given level of return. Building on Markowitz seminal paper, the Capital Asset Pricing Model (CAPM) is used to determine the prices of securities and the inputs in the mean-variance optimising algorithm under the Markowitz formulation (Sharpe, 1964). The main assumption underlying MPT and the CAPM is that financial markets are efficient (Fama, 1965). Therefore, investors cannot use historical prices to predict future prices of securities. However, over the years, a number of scholars have proved that the CAPM does not hold (Blume & Friend, 1973; Douglas, 1968; Miller & Scholes, 1972). The main reasons for this is that the efficient market hypothesis as defined by Fama (1965) does not hold in most markets and investors do not behave rationally. The breakdown of CAPM means the inputs into the mean-variance optimising algorithm of Markowitz s formulation break down. Additionally, it means that investors are able to predict Copyright expected stock returns using historic stock prices or other UCT factors. This has led to the popularity of active asset management, where the aim of the investor is to outperform the market or a stated benchmark. Behavioural finance theory aims to provide explanations for why investors behave irrationally (Ricciardi & Simon, 2000). It posits that investors are susceptible to a number of biases which result in irrational behaviour. These result in a breakdown of the Markowitz mean variance optimisation theorem, CAPM and efficient market hypothesis (EMH) which are based on the assumption that investors behave rationally. Some of the biases which have been uncovered by researchers include overconfidence, framing and question wording, loss aversion and disposition effect (Kahneman & Tversky, 1979; Shiller, 1987; Tversky & Kahneman, 1986). Due to the breakdown of the CAPM, analysts and researchers have been continuously researching for variables which can explain the expected returns of stocks better than beta. Investors have identified a number of variables which have significant explanatory power on the expected returns on stocks compared to beta (Basu, 1977; Haugen & Baker, 1996; Litzenberger, 1979). Some of these variables include price-to-earnings ratio, book-to-market value, dividend yield and the size of the business. The ultimate aim of uncovering these variables is to enable investors to design portfolios which will outperform their respective benchmarks. 10 P a g e

11 1.2 PROBLEM STATEMENT Studies by Page and Palmer (1991), Harvey (1995) and Van Rensburg and Robertson (2004), among others, have proved that the CAPM does not hold on the Johannesburg Stock Exchange (JSE). Therefore, beta is not significant in explaining the cross-section of stock returns. Instead, a number of variables have been determined as significant when explaining the cross-section of stock returns on the JSE. Investors use these variables when designing models that can better project expected stock returns (Auret & Sinclaire, 2006; Hodnett, 2010). Additionally, understanding influences on stock returns helps investors to appraise historic portfolio performances much better. Without a clear understanding of factors that influence the returns of stocks, investors will not be able to confidently design portfolios that can outperform their stated benchmarks. Therefore, this paper aims to add to a continuing body of research information about factors that have significant influence on the cross-section of stock returns on the JSE. More specifically, it will determine whether asset growth is significant in influencing these returns. 1.3 PURPOSE, SIGNIFICANCE OF STUDY AND GAP IN LITERATURE Daniel and Titman (1997) argue that attributes themselves, rather than the covariance structure of returns, explains the cross-sectional variation in stock returns. Characteristics such as such as size, leverage, Copyright past returns, dividend-yield, earning-to-price-ratios and book-to-market-ratios UCT have been found to be significant in explaining cross section of stock returns. Van Rensburg and Robertson (2004) show that price-to-net asset value, dividend yield, price to earnings, cash-to-price, price-toprofit and size better explain the cross-section of JSE stock returns. Hodnett (2010) analysed 38 firm-specific variables to determine their significance in explaining the cross-section of stock returns. Book-value to price and cash flow to price were found to the most significant firm-specific attributes in explaining the cross-section of returns on the JSE. This paper recognises the results by Hodnett (2010). Therefore, the paper will use these two significant factors from this study. In addition, the paper includes asset growth (Asset (t)/asset (t-1))-1) as determined by Cooper, Gulen and Schill (2008), and asset growth [(Asset (t)/asset (t-1))] as determined by Hoffman (2012) as significant when explaining the cross-section of stock returns. Cooper et al. (2008) carried out a study on the New York Stock Exchange (NYSE) and found that asset growth was more significant in explaining cross-sectional stock returns compared to any other variables which had been determined before. Hoffman (2012) carried out a study on the JSE and found that asset growth was not significant in determining cross-sectional stock returns. Therefore, the ultimate aim of the study is to confirm the JSE study by Hoffman (2012) and to extend literature by applying the new metric as used by Cooper et al. (2008) in the NYSE on the JSE. Active asset managers can use the paper when deciding whether asset growth should be considered one of the significant factors that can help predict expected stock returns on the JSE. 11 P a g e

12 In academia, the paper will help update literature on style attributes which can explain the crosssection of returns on the JSE. Furthermore, future research will be able to take into account asset growth as one of the attributes in explaining the cross-section of stock returns. 1.4 RESEARCH QUESTIONS The primary question for this research is: Is asset growth one of the significant factors in explaining the cross-section of stock returns on the JSE? The resulting sub questions based on the primary question are: 1. Is asset growth (Asset (t)/asset (t-1))-1) as determined by Cooper et al. (2008) significant in explaining stock returns on the JSE? 2. Is asset growth (Asset (t)/asset (t-1)) as determined by Hoffman (2012) significant in explaining cross-sectional returns on the JSE? 3. Are book to price and cash flow to price as determined by Hodnett (2010) significant in explain the cross section of stock returns on the JSE 4. Do the explanatory powers of asset growth, book to price and cash flow to price change between the and 2008 to 2015 business cycle phases? 1.5 RESEARCH ETHICS This paper uses secondary data from DataStream in the UCT Graduate School of Business library. Therefore, ethical clearance is not required. However, the data will be used in a very ethical manner. Any manipulations which happen to the data will be stated. The paper is structured as follows: Chapter 2 provides the theoretical and empirical literature relating to MPT and the CAPM, EMH and behavioural finance. Chapter 3 details the research methodology used in the research, adjustments to data, possible biases and how they have been addressed. Chapter 4 provides descriptive statistics of the data. Further literature on the univariate model, adjusting the univariate model for systematic risk and testing the predictive power of univariate models. Chapter 5 analyses the findings, draws conclusions and provides suggested areas for future research. 12 P a g e

13 CHAPTER 2: LITERATURE REVIEW 2.1 MODERN PORTFOLIO THEORY AND THE CAPITAL ASSET PRICING MODEL INTRODUCTION The foundation of MPT was laid down by Markowitz (1952), who defines an efficient portfolio as one that maximises return for a given level of risk or minimizes risk for a given level of return. The inputs into Markowitz s formulation are expected returns of stocks and the covariance structure of the stock returns. Through the use of quadratic programming, an efficient frontier is computed, and all efficient portfolios are then found to lie on the efficient frontier. Markowitz (1952) asserts that a rational investor will want to invest in a portfolio that lies on the efficient frontier since it a welldiversified portfolio which minimises risk for a given level of return. However, Tobin (1958) argues that investors may not only want to invest in a risk-efficient portfolio but may also want to hold a risk-free asset for liquidity purposes. Tobin posits that risk-averse investors tend to hold cash while more risk-seeking investors will hold more risky securities. Accordingly, Tobin argues that investors can reduce the risk of inefficient portfolios by including riskfree securities. Treynor (1961), Sharpe (1964), Lintner (1965a), Lintner (1965b) and Mossin (1966) subsequently developed the Capital Asset Pricing Model (CAPM) for pricing risky securities. According to CAPM, investors are exposed to two types of risks: systematic (undiversifiable) risk and unsystematic risk (diversifiable risk). Since unsystematic risk can be diversified away, investors cannot be compensated for taking on unsystematic risk in their portfolios. On the other hand, systematic risk is the only risk which is relevant to a portfolio and for which investors should receive compensation through higher returns. Whereas Markowitz (1952) presents the relationship between the expected return and total risk (variance) of a portfolio, CAPM presents a relationship between the expected return and the beta (systematic risk) of a portfolio EMPIRICAL EVIDENCE IN DEVELOPED MARKETS The CAPM of Sharpe (1964) and Lintner (1965) and the EMH of Fama (1970, 1991) argue that the expected return on stocks depends on the stock s beta (non-diversifiable risk). The hypothesis is supported by earlier research by Black, Jensen and Scholes (1972), who show that before 1963, a simple positive relationship between beta and stock returns existed. Subsequent studies have identified a number of anomalies in the CAPM-EMH. Initially, studies by Lintner (1965a), Sharpe (1966) and Jensen (1968) provided evidence that the CAPM model holds for individual stocks. By regressing individual stock returns on stock betas, they were able to prove that the intercept is equivalent to the risk-free rate of return and the slope of the line was equal to the excess return. 13 P a g e

14 Since stock betas can be extended to portfolio betas, Friend and Blume (1970), Blume (1970) and Black et al. (1972) extended the evidence of the CAPM to portfolios. This provided more stable results since estimates of beta for diversified portfolios are more stable than estimates of beta for individual stocks. Fama and Macbeth (1973) estimate month by month cross-section regressions of monthly returns on beta to address the problems caused by correlation of residuals when determining whether the CAPM holds. Fama and Macbeth (1973) conclude that the CAPM and EMH hold for the NYSE. Early tests that refuted the CAPM can be traced back to Douglas (1968).Using monthly and quarterly data, Douglas (1968), shows that beta is not the only factor that is responsible for systematically contributing to average stock returns as stated by the two-factor CAPM. Miller and Scholes (1972) use a different technique from Douglas (1968) to reinforce the fact that there seem to be other factors in addition to beta that may be responsible for the average stock returns, therefore disputing the CAPM. Regressions of cross-sectional stock returns on beta consistently found that the intercept is slightly lower than the average risk-free rate and the slope of the security market line is less than the average excess market return, therefore disputing the CAPM. These results are true for Douglas (1968), Black, Jensen and Scholes (1972), Miller and Scholes (1972) and Blume and Friend (1973). Studies on cross-sectional returns on stocks have uncovered a number of variables that explain expected return on stocks better than the stock s beta. Basu (1977), for example, shows that (P/E) ratio provides a better measure of expected stock returns compared to the stock s beta. The study provides further evidence that the CAPM does not hold in developed markets. Additional studies by Litzenberger (1979) and Bhandari (1988) also provide evidence that the CAPM does not hold in developed markets. Litzenberger (1979) uncovers a strong positive relationship between dividend yield and expected return on the NYSE during the period between 1936 and 1977, while Bhandari (1988) reports that there is a significant relationship between debt to equity ratio and stock returns. Banz (1981) shows that there is a size effect when determining expected returns on stocks, providing further evidence that the CAPM does not hold. Banz (1981) paper reveals that, on average, smaller firms have higher risk-adjusted returns than larger firms. Fama and French (1993) reveal that size and book to market value of stocks captured the cross-sectional returns more effectively than beta. However, Kent and Titman (1997) argue that after controlling for size and book to market value, stock returns are more correlated to betas based on the Fama and French (1993) factors. Haugen and Baker (1996) found that past returns, trading volume, return on equity and price/earnings are the strongest predictors of stock returns and they are stable from one country to the next. Hong, Lim and Stein (2000) show that expected stock returns associated with momentum strategies decrease with size and analyst coverage, instead of the beta of the stock. Doukas and McKnight (2005) provide further evidence of the momentum effect by showing that it also applies in Europe. 14 P a g e

15 More recent evidence on the inability of beta to predict cross-sectional stock returns compared to other variables comes from Pontiff and Woodgate (2008) and, Cooper et al. (2008). Pontiff and Woodgate (2008) show that share issuance has a high predictive power on cross-sectional stock returns compared to size and book to market ratios on the US stock market. A study by Cooper et al (2008) found that between 1968 and 2002 on the NYSE, firms with lower asset growth earned an annualised risk-adjusted return of 9.1%. On the other hand, over the same period, firms with higher asset growth earned a -10.4% return EMPIRICAL EVIDENCE IN DEVELOPING MARKETS De Villiers, Pettit and Affleck-Graves (1986) investigated whether the small firm effect exists on the JSE as Banz (1981) reported it did on the NYSE. Their study revealed that it did not. Instead, it showed that large firms tend to perform better than small firms on the JSE. It did not link expected returns with beta of stocks. Therefore, the study did not provide evidence that the CAPM existed. Affleck-Graves, Bradfield and Barr (1988) found that the CAPM provided a reasonable explanation for the cross-section of returns on the JSE, but did not provide a reasonable explanation for the cross-section of returns on gold stocks. Therefore, the CAPM could partially explain the crosssection of returns on the JSE. After numerous studies on the price/earnings and size effect in developed markets, Page and Palmer (1991) studied the JSE to determine if these effects existed. The paper found that the price/earnings effect existed while the size effect did not exist. This provided further evidence that the CAPM did not provide an explanation for the cross-sectional returns on the JSE. Harvey (1995) studies 18 emerging market countries to test whether stock market expected returns can be explained by the capital asset pricing model. The results show that beta is unable to explain any cross-sectional variations in expected returns. Harvey (1995) asserts that emerging markets are inefficient, which results in a breakdown of the CAPM. Van Rensburg and Robertson (2004) examine whether firm-specific attributes can explain the crosssection of stock returns outside the CAPM framework. The results show that price-to-net asset value, dividend yield, price-to-earnings, cash flow-to-price, price-to-profit and size were all significant in explaining cross-section of stock returns. However, using multifactor testing, only size and price-to-earnings ratio were found to be significant in explaining cross-sectional returns on stocks, therefore confirming that the CAPM does not provide an explanation for the cross-section of portfolio returns. Auret and Sinclair (2006) use the same data as Van Rensburg and Robertson (2004) to determine if book to market value is significant in explaining cross-section of stock returns on the JSE. They found that even though book-to-market value has significant explanatory power compared to size and price-to-earnings, including book-to-market value in the multivariate model does not improve the explanatory power of the combined variables. This reinforces previous findings that the CAPM model does not provide a good explanation for the cross-section of stock returns. 15 P a g e

16 Hodnett (2010) uses linear and non-linear techniques to determine the predictors of cross-sectional returns on the JSE. By analysing 38 firm-specific attributes that could explain cross-sectional stock returns, the results show that there are various factors that could significantly explain crosssectional stock returns under the univariate framework. However, in a multivariate framework, book-value to price and cash flow- to price are found to be the most significant factors in predicting cross-sectional stock returns. This shows that the CAPM model does not provide a significant explanation for cross-section of stock returns on the JSE compared to other firm-specific variables such as book-value to price and cash flow to price value. The findings above pave the way for this thesis s main question: Can asset growth be considered as one of the firm-specific attributes that we can use to explain the cross-section of stock returns on the JSE just as it is on the NYSE, as determined by Cooper et al. (2008)? 16 P a g e

17 2.2 EFFICIENT MARKET HYPOTHESIS MPT and the CAPM are built on the assumption that markets are efficient. Fama (1965) shows that markets follow a random walk and for the first time defines the efficient market concept. Samuelson (1965) provides further evidence that stock price movements follow a random walk and, therefore, provides evidence that markets are efficient. According to Fama, a market is efficient if it fully reflects all available information. However, this is an extreme case and it is highly unlikely that any market will be able to satisfy the criteria. Therefore, Fama (1969) splits the definition of market efficiency into three forms: weak, semi-strong and strong. Furthermore, Fama specifies tests that can be used to assess whether a market is efficient or not. Fama (1969) asserts that weak form market efficiency implies that current stock prices fully reflect all information contained in the past history of prices. This implies that investors cannot use historic information about share prices to make investments that will perform better than the market. To test weak form efficiency, you have to prove that the market follows a random walk process. The semi-strong form asserts that current stock prices fully reflect all historic information on stock prices and all publicly available information. The implication is that investors cannot use publicly available information to make investments that will perform better than the market. Therefore, to test the semi-strong hypothesis, you have to prove that asset pricing models fully reflect all publicly available information. Finally, strong form market efficiency states that stock prices reflect all information: historic information on share prices, publicly available information and privately available information. Therefore, to test the strong form market hypothesis, you have to prove that investors who are privy to private information have higher yielding portfolios than the market. According to Fama (1969), tests for market efficiency are cumulative. For a market to be strong form efficient, it has to pass tests of weak form efficiency and semi-strong form efficiency. Therefore, if a market fails weak form efficiency then it cannot be either semi-strong-form efficient or weak-form efficient. The efficient market hypothesis has been criticised by a number of researchers. Fama (1969) asserts that certain conditions have to exist for the efficient market hypothesis to hold. Two of the conditions are: there are no transaction costs; and all market information is available to all market participants free of any charges. Grossman and Stiglitz (1980) argue that these conditions are unrealistic and proposes an asset pricing model that reflects an equal degree of disequilibrium. The model reflects only partially prices of informed individuals (arbitrageurs), so that those who incur costs to receive information are rewarded through higher returns. Shiller (2003) argues that the efficient market hypothesis does not provide an explanation for investors bidding up stock prices until there is a market bubble and the subsequent high volatility experienced when markets crash. Debondt, Forbes, Hamalaimen and Muradoglu (2010) posit that research into finance should focus on understanding the behavioural biases of investors instead of trying to understand the behaviour of investments. Debondt et al. (2010) argue that this will provide an understanding of the irrational behaviour by investors and investments that has resulted in the breakdown of theories such as the efficient market hypothesis model and the CAPM. 17 P a g e

18 For the purposes of this paper, it is sufficient to show that the semi-strong form market hypothesis does not hold. Therefore, having shown that the CAPM model does not hold in developing and developed markets, this means that beta cannot be used to determine expected returns on stocks. Furthermore, it means other variables are more significant when explaining cross-sectional returns on stock compared to beta. Therefore, given that there a number of studies which have found that the JSE is not efficient, the main question this paper asks that is relevant to the JSE is as follows: Is asset growth a significant factor is estimating expected stock returns? 18 P a g e

19 2.3 BEHAVIOURAL FINANCE INTRODUCTION Unlike conventional finance theories that embrace modern portfolio theory and efficient market hypothesis, behavioural finance explores psychological and sociological influences on financial decision making by individuals, groups and organisations (Ricciardi & Simon, 2000). Ultimately, behavioural finance aims to provide an explanation for market anomalies, speculative market bubbles and market crashes. Behavioural finance does not assume that investors act rationally. Instead, it posits that investors decisions may be influenced by emotions and psychological biases (Ricciardi & Simon, 2000). Furthermore, the researchers state that finance is still a centrepiece of behavioural finance. However, the behavioural aspects of psychology and sociology are brought in to provide further insight EMPIRICAL EVIDENCE IN DEVELOPED MARKETS Behavioural finance emerged with the introduction of prospect theory by Kahneman and Tversky (1979). Prospect theory was presented as an alternative to expected utility theory, which was a basis for modern portfolio theory, particularly the design of the efficient frontier by Markowitz (1952). Expected utility theory uses the law of diminishing marginal utility to explain risk aversion among investors. On the other hand, prospect theory states that investors are risk-averse when faced with decisions Copyright regarding gains but use the law of diminishing marginal disutility UCT when faced with losses. The loss aversion function is non-linear under prospect theory. It asserts that the disutility derived from losses is larger than the utility derived from gains for the same size of losses or gains. This implies that investors give more weight to losses than gains. Therefore, when faced with an opportunity that may result in an equal likelihood of gaining or losing their wealth, investors will seek to avoid the loss. Regret theory was introduced by Loomes and Sguden (1982). It states that when investors make a decision, they take into account the feelings of regret in case they make the wrong choice. This results in investors holding losers longer and selling winners much earlier. Framing decisions or questions was introduced by Tversky and Kahneman (1986). It states that investor preferences will be influenced by whether a decision is framed as a loss or a gain. This is contrary to the principle of invariance, which underlies the rational theory of choice. Non-linear preferences were uncovered through the use of Allais s (1953) paradox. Allais s paradox uses an example to show that the incremental difference between probabilities of 0.99 and 1.00 has more impact on preferences than the difference between 0.10 and Source of dependence was uncovered by Ellesberg (1961). It states that people s willingness to bet not only depends on the magnitude of the uncertainty involved, but it also depends on the source of uncertainty. 19 P a g e

20 Risk-seeking theory, introduced by Kahneman and Tversky (1979), asserts that people prefer a small probability of winning a large price prize than the expected value which may involve large probability of losing. Additionally, risk-seeking is more pronounced when people have to choose between a certain loss and a high probability of a bigger loss. Shiller (1987) shows that investors were over-confident in their ability to determine the direction of market movement after the 1987 market crash. Most investors cited that their gut feeling told them markets would rebound to their original levels after the crash. There was no supporting empirical evidence. Gender over-confidence was confirmed by Barber and Odean (2001) after investigating investment decisions by men versus women. They find that men trade more than women, therefore exhibiting over-confidence in their decision making. Loss-aversion theory by Tversky and Kahneman (1991) states that losses and disadvantages have a larger impact on preferences compared to gains and advantages. Representativeness is proven by Shefrin and Statman (1995), who show that investors make decisions based on a company s good reputation as opposed to companies that offer guarantees of higher future earnings. It turned out that the companies which had high book-to-market values tended to produce low returns in the short term. Nofsinger Copyright and Rui (2007) analyse the trading decisions of Chinese UCT investors by studying their brokerage accounts. The studies show that Chinese investors display three biases: (1) the disposition effect, which means they tend to sell stocks that have appreciated in price but not stocks that have depreciated in price; (2) over-confidence, which means they make frequent trades; (3) representativeness bias, which means they appear to believe that past returns are representative of future returns. Economou, Kostakis and Philippas (2011) conducted studies on the herding behaviour among investors in the Portuguese, Italian, Spanish and Greek markets. The studies confirm that investors exhibit herding behaviour during periods of extreme uncertainty and volatility. 20 P a g e

21 2.3.3 EMPIRICAL EVIDENCE IN DEVELOPING MARKETS Representativeness is shown to be prevalent on the JSE when Page and Way (1993) studied the trading pattern of 204 securities between July 1974 and June The study shows that investors on the JSE tend to overact to the most recent information. M kombe and Ward (2002) studied the performance of stocks after the Initial Public Offering (IPO) and discovered that over-confidence is one of the factors that affects investors returns. Between 1995 and 2006, the three-year performance of IPOs underperformed the market by 50% on an abnormal return basis and 47% on a cumulative abnormal return basis. Gilmour and Smith (2002) found herding among institutional investor on the JSE. Furthermore the degree of herding was found to increase with increases in the uncertainty around future prices (risk). However, at extremely high volatility levels, the degree of herding reduced. Prayag and Van Rensburg (2004) show that analysts tend to be over-optimistic when forecasting earnings estimates of companies listing on the JSE between 1990 and However, the accuracy of these earnings estimates increases as the announcement date approaches. Disposition effect and a strong desire to cut losses was determined to be the key motivation for investment managers of listed property funds to sell their holdings (Lowies, Hall, & Cloete, 2013). Furthermore, due to the challenges and time spent purchasing a property, the investment managers were Copyright shown to be loss-averse. Decisions on whether to purchase a property UCT investment were found to be influenced by framing. The existence of behavioural biases among JSE investors shows that investors are irrational. Therefore, the efficient market hypothesis and the CAPM, which assume that investors are rational, do not apply on the JSE. This paves the way for this thesis s main question. Essentially, this question is as follows: Is asset growth a significant factor in explaining the cross-return of stocks on the JSE? 21 P a g e

22 2.4 SUMMARY OF LITERATURE REVIEW The literature starts by providing a comprehensive review of modern portfolio theory and explores the mean-variance optimisation algorithm determined by Markowitz. It explores the implications of an efficient portfolio to the investor. The literature then links Tobin s separation theorem and the efficient risky portfolio. It uses literature to explore reasons for investors to invest in risk-free assets and risky efficient portfolio. The CAPM is introduced as a tool that is used to price securities and provide inputs into the Markowitz mean-variance optimisation problem. Empirical evidence that supports the use of the CAPM in developing and developed countries is explored. It provides reasons for the breakdown of the CAPM in developing and developed countries. The breakdown of the CAPM and the use of beta to explain cross-sectional returns is linked with the emergence of studies that found other variables to explain cross-sectional returns more accurately than beta. Literature on the factors which explain the cross-section of stock returns better than beta is also explored. The efficient market hypothesis is introduced as a foundation for the CAPM and Markowitz mean variance formulation. Early literature which supported the fact that the market is efficient is explored. Additionally, more recent literature which does not support the existence of an efficient market is also explored. Behavioural finance theories that aim to explain why irrational behaviour among investors results in the breakdown of the CAPM and EMH are explored for both developed markets and emerging markets. Throughout, links are made between the main question and the literature review. 22 P a g e

23 CHAPTER 3: METHODOLOGY AND DATA 3.1 METHODOGY The methodology followed in this paper is similar to Fama and French (2008). It uses cross-sectional regression analysis; calculation of sorted returns ranked by explanatory variables; and time series and cross-sectional correlation analysis. Time series and cross-sectional correlation analysis is carried out to determine if there is consistency in the relationship between cross-sectional stock returns and the chosen explanatory variables. A number of risk measures are applied to the sorted returns to determine if the risk profiles of the different stocks can provide explanations for the differences in cross-sectional returns. The sorted returns analysis is carried out for different subperiods. This is to determine if the explanatory variables maintain their explanatory power despite the time period involved. The null hypothesis under each regression analysis is based on the EMH. As a result, the null hypothesis is that there is no relationship between the expected stock returns and explanatory variable. The market is deemed efficient if future returns cannot be determined from historic values of any of the explanatory variables since the information used is publicly available. Rejection of the null hypothesis implies that there is a relationship between stock returns and the explanatory variable. Furthermore, by rejecting the null hypothesis, the research will confirm that the EMH does not hold for the JSE. Most of the relationships between the stock returns and explanatory variables have been documented before. Therefore, the paper not only tests the null hypothesis, but it also determines the alignment of the relationships with previous studies. 3.2 DATA HARVESTING The data used to conduct the analysis was obtained from Datastream. Annual data of closing share prices, share-price-to-book ratios, share-price-to-cash-flow ratios, book-asset-values and betas were collected from Datastream. There were 162 stocks listed on the JSE All Share Index as at December Microsoft Excel was used to calculate the desired attributes to be modelled. Table 1 details the codes which were used to extract the data and the formulae used to calculate the desired variables in Microsoft Excel. 23 P a g e

24 Table 1: Data variables Description Datastream code Formulae Stock return PI Book-to-share-price PTBV 1 PTBV Cash-to-share-price PC 1 PC Asset growth (Cooper) Asset (Hoffman) growth DWTA DWTA Beta BETA BETA ( PI(T) PI(T 1) ) 1 (1) (2) (3) ( DWTA(T) DWTA(T 1) ) 1 (4) ( DWTA(T) DWTA(T 1) ) (5) The full data period runs from 1 January 2000 to 31 December The period was chosen to cover the recent domestic and international financial crises and two business cycle phases. Additionally, the period was chosen so that is was not too long, making the study irrelevant, or too short, compromising Copyright the credibility of the results. To determine whether UCT the explanatory power of the independent variables changes over different business cycle phases, the full data period was divided into two sub-periods. Sub-period 1 is from 1 January 2000 to 31 December 2007; sub-period 2 is from 1 January 2008 to 31 December 2015; and the full data period is from 1 January 2000 to 31 December Sub-periods 1 and 2 were chosen to take account of the impact of the financial crises on the relationship between stock returns and the explanatory variables EXPLORATORY VARIABLES Hodnett (2010) tested 38 attributes to determine if they were significant when explaining crosssectional stock returns (listed in Appendix 1). Book-value-to-price and cash-flow-to-price were found to be the most significant style attributes when explaining cross-sectional stock returns. This study recognises the findings by Hodnett (2010). Therefore, the study will not re-test the variables which were found to be insignificant when explaining cross-sectional returns by Hodnett(2010). Furthermore, this study recognises the fact that Hodnett (2010) did not test asset growth as one of the significant style attributes when explaining cross-sectional returns. Therefore, asset growth, book-value-to-price and cash-flow-to-price are the explanatory variables that will be tested for significance when explaining the cross-section of stock returns. The study by Hodnett (2010) divided the style attributes into six categories, as follows: fundamental values relative to share; solvency and liquidity ratios; fundamental growth; operating performance margins; size and return momentum; and consensus analyst forecast estimate. 24 P a g e

25 According to Hodnett (2010), fundamental value relative to share price separates growth stocks from value stocks. Growth stocks have higher fundamental values per share compared to value stocks. Solvency and liquidity ratios provide information about the financial state of the company at a point in time. Fundamental growth style variables include growth in cash, earnings, profit margins, dividends and sales. Operating profit margins include ratios that provide information on the return generated by the company from its assets. Size and return momentum style attributes include variables that are price-sensitive. Consensus analyst forecasts estimates include analyst forecast of earnings and dividends. This study tests three style attributes to determine if they are significant when explaining the crosssectional stock returns. These variables are: book-value-to-price, cash-flow-to-price, asset growth as determined by Hoffman (2012) and Cooper et al. (2008). The papers by Hoffman (2012) and Cooper et al. (2008) came to different conclusions. Cooper et al. (2008) studied the NYSE and found that asset growth as determined by equation 4 was the most significant variable when explaining the cross-section of stock returns on the NYSE. Hoffman (2012), on the other hand, found that asset growth as determined by equation 5 was not significant in determining the cross-section of stock returns on the JSE. Both asset growth variables as defined by Cooper et al. (2008) and Hoffman (2012) fall under the operating profit margin category as defined by Hodnett (2010), and provide an indication of how well the company generates returns from its assets. Therefore, the aim of this study is to confirm the results from Hoffman (2012) and to apply the new metric as used by Cooper et al. (2008) to the JSE. 3.3 DATA ADJUSTMENTS The initial sample of 162 stock returns was reduced to 132 by removing stocks with insufficient return observations. To address possible biases on the data set, the following adjustments were made: COMPLETENESS The data used to conduct the analysis is subject to the level of completeness of Datastream as at December Some stocks, which were listed as at December 2015, were not listed for the entire the observation period. Therefore, in line with Haugen and Baker(1996), incomplete data was addressed by assigning the average population mean over the exposure. This method introduces bias to the data as information which is incomplete may have been available during the period of testing. However, it would have been replaced with a population mean. Nevertheless, Haugen and Baker (1996) argue that this is the most appropriate way of addressing missing observations since it does not remove complete data records because of a few missing observations. 25 P a g e

26 3.3.2 LIQUIDITY Highly liquid and tradable stocks result in robust results. In contrast, illiquid stocks that are traded thinly may result in skewed data. All stocks listed on the JSE have to meet minimum liquidity criteria, which are reviewed every six months. According to the Ground Rules for Managing the FTSE/JSE Africa Index Series 2016: Securities which do not turnover at least 0.5% of their shares in issue, after the application of any free float restrictions, per month in at least ten of the 12 months prior to a semi-annual review in March and September will not be eligible for inclusion in the indexes (except the FTSE/JSE Fledgling Index and selected specialist indexes) for the next six months (FTSE Russell, 2016, p. 14). However, some of the stocks that fall under the FTSE/JSE Fledgling Index, the FTSE/JSE Secondary Market Indexes, the FTSE/JSE Property Loan Stock Index and the FTSE/JSE Property Unit Trust Index may not be liquid since they are exempted from meeting minimum liquidity requirements. However, these stocks make up a small proportion of the JSE All Share Index and have an even smaller market capitalisation compared to the rest of the stocks on the index. As a result, they are not expected to bias the results significantly OUTLIERS Outliers are usually due to extreme events that may not be repeated in the future, and may result in biased results. To improve the reliability of the results, it is necessary to address outliers. Therefore, in accordance with Baars (2014), outliers are addressed through a Winsorisation procedure where a mean and standard deviation of the variables for all stocks for each year is calculated, and a limit of three standard deviations from the mean is set. All stocks with a mean in excess of three standard deviations from the mean are then temporarily removed from the sample. For the remaining stocks, a new mean and standard deviation is calculated, and the stocks which were temporarily removed from the sample are then reintroduced into the sample with their values pegged at three times standard deviations of the new sample STANDARDISATION For the firm-specific variables, a standardisation procedure was performed to facilitate comparisons to be made between the different variables. The variables were standardised to a zero mean and standard deviation equal to one, paving the way for a comparison between the magnitude of the slopes estimated in the cross-sectional regression analysis. 26 P a g e