Multifactor Portfolio Construction:

Transcription

1 Multifactor Portfolio Construction: Does a combined factor approach to portfolio construction offer greater risk-adjusted returns than a single-factor approach? An investigation of equities listed on the Johannesburg Securities Exchange over a 20-year period from 1995 to 2015 Final Research Report Presented to: The Graduate School of Business University of Cape Town In partial fulfilment of the requirements for the degree Master of Business Administration Submitted by: Mohsin Tajbhai Supervisor: David Holland 9 th December 2015

2 I. Abstract This research paper investigates factor investing on the Johannesburg Securities Exchange (JSE) by analysing the effects of eight factors under the broader classification of value, momentum and quality. The investigation is limited to the largest 100 equities by market value listed on the JSE over a 20-year period from The study investigates the relationship of portfolio returns based on exposure to several single factors using a portfolio-based approach with a onemonth rebalancing schedule. In addition, the study investigates benefits of diversification through exposure to several factors using a combined factor approach by ranking portfolios based on several combinations of value, momentum and quality factors. The study employs both information ratios and linear regression against an equally weighted benchmark to measure the risk adjusted returns of combined factor portfolios. The results from the investigation show that over the 20-year period under review all factors tested significantly outperformed an equally weighted benchmark of all of the companies in the sample. The results also indicate a strong presence of a momentum and quality premium on the JSE Copyright over the 20-year period. Lastly the investigation of combined factor UCT strategies identified that a diversified approach to portfolio selection with exposure to a combination of factors produced higher risk adjusted returns. Page-i-

3 II. Acknowledgments I firstly thank God for giving me the ability and the opportunity to complete this MBA. No achievement no matter how great or small can be accomplished without the command of the Lord of the worlds. I thank my supervisor, David Holland, for all his generous assistance, advice and guidance during my work on this thesis. It has been an honour and a privilege to work with a person who has a wealth of knowledge and experience. To my wife Nishaat Limbada, I thank you for your love, support and encouragement during the past two years. Your unconditional love, patience and positive attitude have lightened the load during these last two years. To my parents, I thank you for your support and guidance. I pray that God blesses you with good health and long lives (God willing). Lastly, I thank my spiritual guide and mentor, Sayed Shakeel Ahmed El-Edroos. I have been blessed to have you as a guide and mentor. Your words of encouragement and advice have always motivated me to greater heights. Page-ii-

4 III. Plagiarism Declaration 1. I know that plagiarism is wrong. Plagiarism is to use another s work and pretend that it is one s own. Allowing another to copy my work and use it as their own is also plagiarism. 2. This assignment is my own work. I have not allowed and will not allow anyone to copy my work with the intention of passing it off as his or her own work. 3. I acknowledge that working with someone on my assignment is allowed, but only if a mutual effort is made and, where necessary, different wording and examples are used... 9 th December 2015 Page-iii-

5 Table of Contents I. Abstract... i II. Acknowledgments... ii III. Plagiarism Declaration... iii Table of Contents... iv List of Figures... v List of Tables... vi 1 Introduction Research area and Problem Significance and Purpose Research Questions Research Assumptions Research Scope and limitations Research Copyright Ethics... UCT 5 2 Literature Review Background Single-factor Portfolio Selection Strategies Combined factor strategies Conclusion Methodology Research Approach and Strategy Research Design Sampling Data Analysis Data Reliability Data Validity Research Findings, Analysis and Discussion Size Effect Single-factor Strategies Combined factors Limitations of the research Page-iv-

6 5 Conclusions Research question Research question Research question Recommendations for future Research References Appendix 1. Results of Shapiro-Wilk test for normality Appendix 2. Long Strategy Wilcoxon Rank Sum Test Appendix 3. Linear regression vs equally weighted benchmark List of Figures Figure 1: Research Onion 16 Figure'2':'Diagram'Indicating'Deductive'Approach 17 Figure 3: Graph of portfolio returns ranked according to Market Value 24 Figure Copyright 4: Graph of portfolio returns ranked according to Earnings Yield (EY) UCT 25 Figure 5: Graph of portfolio returns ranked according to Book-to-Market 26 Figure 6: Graph of returns of portfolios ranked according to Dividend Yield (DY) 27 Figure 7: Graph of portfolio returns ranked according to 6-month price momentum 28 Figure 8: Graph of portfolio returns for portfolios ranked according to 12-month price momentum 29 Figure 9: Graph of portfolio returns ranked according to net profit margin 30 Figure 10: Graph of portfolio returns ranked according to Return on Equity (ROE) 31 Figure 11: Graph of Portfolio returns ranked on Return on Invested Capital (ROIC) 32 Figure 12: Returns of portfolios ranked according to 12-month price momentum and earnings yield 38 Figure 13: Returns of portfolios ranked according to ROE and EY 39 Figure 14: Returns of portfolios ranked according to 12-month price momentum and ROE 40 Figure 15: Returns of portfolios ranked according to value momentum and quality 41 Figure 16: Return and Risk for all factors tested using a Long Strategy 45 Figure 17: Long Strategy Excess return and Information ratio (IR) 46 Figure 18: Long/Short Strategy Annual return vs risk 47 Figure 19: Alpha and standard error for factors tested using long strategy 49 Figure 20: Single factors 12-month excess rolling returns vs the equally weighted benchmark 51 Figure 21: Combined factors: 12-month excess rolling retrun vs the equally weighted benchmark 52 Figure 23: Linear regression of Book-to-market vs equally weighted benchmark 65 Figure 24: Linear regression of Dividend Yield vs equally weighted benchmark 65 Page-v-

7 Figure 25: Linear regression of earnings yield vs equally weighted benchmark 66 Figure 26: Linear regression of 12-month price momentum vs equally weighted benchmark 66 Figure 27: Linear regression of 6-month price momentum vs equally weighted benchmark 67 Figure 28: Linear regression of ROIC vs equally weighted benchmark 67 Figure 29: Linear regression of ROE vs equally weighted benchmark 68 Figure 30: Linear regression of net profit margin vs equally weighted benchmark 68 Figure 31: Linear regression of 12-month price momentum + EY vs equally weighted benchmark 69 Figure 32: Linear regression of 12-month price momentum +ROE vs equally weighted benchmark 69 Figure 33: Linear regression of 12-month price momentum + ROE+ EY vs EW benchmark 70 List of Tables Table 1: Summary of research done on the JSE and emerging markets 15 Table 2: Summary of Single- factor Strategies Long Strategy (Portfolio P1) 33 Table 3: Summary of Single-factor Strategies Long/Short Strategy - Portfolio P1-P5 33 Table 4: Summary of excess returns and Wilcoxon Rank Sum Test 35 Table 5: Wilcoxon Rank Sum-Significance values for all long single-factor strategies 36 Table Copyright 6: Excess return R (P1-EW) Correlation Matrix UCT 36 Table 7: Long/Short Strategies Correlation Matrix 37 Table 8: Long Strategy Summary of Return vs Risk for all factors tested 43 Table 9: Long/Short Strategy Summary of Return vs Risk for all factors tested 44 Table 10: Results of regression against equally weighted benchmark 48 Table 11: 10-Year vs 20-Year comparison 53 Table 12: Shapiro-Wilk test for normality (long strategy) 62 Table 13: Shapiro-Wilk test for normality (long/short strategy) 62 Table 14: Wilcoxon Rank sum test results for all factors tested using long strategy 63 Table 15: Test statistic results Long strategy 64 ' Page-vi-

8 1 Introduction 1.1 Research area and Problem The holy grail for active asset managers is to consistently beat the market and any benchmarks used to measure performance (Muller and Ward, 2013). Despite the idea of efficient market theory proposed by Fama (1970), many active asset managers thrive on identifying areas of inefficiency in markets, which can be capitalised in order to achieve significant excess returns. The key to the success of these active managers is understanding, identifying and capitalising on these inefficiencies before the market does (Ang, 2013). Factor investing is defined as a strategy that aims to harvest inefficiencies or mispricing through exposure to factors (Bender & Briand, 2013). Ang (2013) defines factors by using an analogy of the relationship between nutrients and food. He explains that just as certain foods are rich in different nutrients, securities have varying exposure to different factors. Ang (2013) further explains that factors drive risk premiums and portfolio construction can capitalise on these risk premiums by constructing portfolios of assets with a varying risk profile through exposure to specific factors. This study explores the significance of three groups of factors or styles that have been shown to provide an equity risk premium. These groups of factors include; Value, Momentum and Quality factors. Value investing strategies are defined as strategies that select companies with high intrinsic value relative to their market value (Ang, 2013). Value investors such as Benjamin Graham and Warren Buffet have skilfully beat the market by identifying companies that have high intrinsic value, using factors such as price-to-book, earnings yield and dividend yield as a guide to portfolio selection. Momentum strategies on the other hand capitalise on the over-reaction or under-reaction of the market to prior returns (price momentum) or earnings (earnings momentum). Momentum strategies are defined as strategies that capitalise on stocks that a have a high upward or downward trend by holding winners (stocks with an upward trend) and shorting losers (stocks with a downward trend) (Louis, Jagadeesh & Lakonishok, 1996). In addition, Piotroski, Joos, Monahan, and Lee, (2013) identify that using quality factors as selection criteria provides significantly better performance than that of the market. They define quality factors as factors that highlight a firm s financial performance. In their study on US equities, three financial ratios; Return on Equity (ROE), Return on Assets (ROA) and leverage ratios were used as criteria for portfolio selection. Page-1

9 In other studies Van der Hart, Slagter, and Van Dijk (2003) and identify that the single-factor strategies discussed above can be enhanced by combining factors. They use strategies that combine value, momentum and quality factors, creating new factors with diversified exposure. Van der Hart et al. (2003) show that a diversified strategy that uses momentum and value significantly outperforms both an equally weighted benchmark as well as value and momentum strategies used independently. 1.2 Significance and Purpose Despite the abundance of papers on combined factor strategies in developed markets and emerging markets, only a few papers have statistically justified the use of a combined factor strategy in the Johannesburg Security Exchange (JSE). This study aims to fill this gap by investigating both single factor portfolio selection strategies based on value, momentum and quality as well as combined factor strategies that implement combinations of value, momentum and quality. An explanatory study is defined by Cooper & Schindler (2003 p. 97) as a study that attempts to identify key variables affecting data values. This study takes a typical explanatory route to investigate the relationship between value, momentum and quality factors and the returns of equities listed on the JSE. In addition, the study statistically verifies whether adopting a diversified approach to portfolio selection through the use of a combined factor strategy can improve a portfolio s risk-adjusted returns. 1.3 Research Questions The aim of the research paper is to add to the wealth of knowledge on factor investing and combined factor investing by examining equities listed on the JSE. The scope of the research is to investigate the largest 100 stocks by market value listed on the JSE over the last 20 years, with the goal of understanding the following research questions: Research Question 1 Is there evidence that supports the view that adopting a single-factor portfolio selection strategy based on value, momentum or quality will outperform the market, represented by an equally weighted portfolio of South African stocks? Page-2

10 This question aims at investigating whether factors that represent value, momentum and quality can be used successfully as single-factor screens to select portfolios that outperform the market, represented by an equally weighted portfolio of stocks from the sample. An equally weighted portfolio was selected as the benchmark instead of the J203. The J203 is a market capitalisation weighted index of the largest 160 companies on the JSE, which is biased towards the higher capitalisation equities such as Naspers, MTN, Sasol etc., (54% of the J203 is dominated by the top 10 equities by market cap). Because of the market capitalisation bias we have not used the J203 in any comparative studies in this research paper. We believe that in using an equally weighted portfolio as a benchmark for comparison, a better and unbiased view of a factor s stock-picking prowess is achieved. In addition, Van der Hart et al. (2003) cite short selling constraints in emerging markets as a reason for the use of an equally weighted benchmark rather than a zero-investment a strategy Research Question 2 Is there evidence that any one single-factor strategy outperforms every other single-factor strategy? This purpose of this question is to establish if one single-factor strategy exists that outperforms every other single-factor strategy, i.e., is there statistical evidence that momentum strategies significantly outperform either value or quality single-factor strategies or vice versa. The key is to establish if a dominant strategy exists in the South African Market Research Question 3 Is there evidence that suggests using a multifactor approach to portfolio construction offers higher risk-adjusted returns? This research question covers the primary goal of this study, which is to investigate whether adopting a multifactor strategy that uses combinations of value, momentum and quality can enhance a portfolio s performance. The ultimate goal of this research question is to build a case either for or against diversifying by exposure to several factor premiums rather than just a singlefactor. a A zero-investment strategy is a long/short strategy that takes a long position on winners and a short position on losers. b Return on Invested Capital = NOPAT/invested capital. Since financials have extremely high-invested capital they tend to have very low ROIC which would skew results. Hence for this study, all financials from factor models that use ROIC have been omitted. ROE is a better measure of Page-3

11 1.4 Research Assumptions Several assumptions were made in order to simplify or ensure credibility of the results. These assumptions are discussed below: No account of transactional fees has been allowed for this is an unrealistic assumption as in reality transactional fees can significantly reduce overall returns; however, because this study is a comparative study of different selection strategies, it was assumed that all the strategies tested incur zero or similar transactional costs. The dataset was assumed to be free from survivorship bias. Survivorship bias is the tendency to select only companies that have been listed for a prescribed period, or that have existed for a certain period. In reality, active managers cannot know which companies will be delisted in the future, and hence survivorship bias in the data could lead to errors. In this study, the elimination of survivorship bias was ensured by using databases such as INET BFA, Bloomberg and Datastream that capture all companies regardless of when they were listed or delisted, or for how long they have existed. All companies listed during the period under review have been considered even if they ceased to exist after the selection period. In this study Copyright delisted companies were held until the end of the rebalancing UCT period, at which time the rebalancing process discarded the company. It was assumed that institutional investors are predominantly interested in large, highly liquid stocks and hence this study looks only at the largest 100 stocks based on market cap over the entire period, which on average represent over 95% of all the stocks based on market cap over the period. Thinly traded shares are those that have very low turnover and are not bought and sold regularly; exposure to these equities could lead to unpredictable results and wide spreads between buy and sell prices. By controlling for size we assume that exposure to these equities is limited. 1.5 Research Scope and limitations The scope of this study is to examine the relationship between portfolio returns and exposure to risk factors and combinations of factors by back testing historical data on the largest 100 equities by market capitalisation listed on the JSE over a 20-year period from The study has the following limitations: The study is limited to equities on the JSE over a 20-year period. This time period was chosen as it covers several time periods and is consistent with several other papers that look at factor investing on the JSE such as; Muller and Ward (2013), Auret and Sinclaire (2006). Page-4

12 The study does not consider the effects of transaction costs, which could have an effect on the results; for this study it is assumed that the transaction costs of each factor analysis are negligible or similar across different factors. The study only investigates a monthly holding and rebalancing schedule The study only covers 8 factors under the broader categories of value, momentum and quality factors The investigation only uses historical data as data on forward-looking factors were sparsely populated on the databases that we had access to for this study and hence we have not considered forward looking factors in this study. 1.6 Research Ethics The proposed study is a fully quantitative study, which uses secondary data, and is therefore not affected by the usual ethical concerns that affect qualitative research, which have significant human interaction. However, measures were taken to counter the following ethical concerns raised by Cooper and Schindler, (2003): Using legal sources to extract data: the study used INET BFA, Bloomberg and Datastream, which are all fully licensed for use by the University of Cape Town. Honesty in dealing with statistics: there was no intention to amend the results of the study in order to get desired results. Referencing and giving credit: all work used from other sources has been referenced throughout this dissertation. Following ethical codes and professional bodies: as a student of the University of Cape Town, I acknowledge that this research belongs to the University of Cape Town and I will not unduly profit from this research without their permission. The rest of this paper is structured as follows: The next section reviews the literature on factor investing with the view of understanding the origins of factor investing. The section outlines the significance of different factor strategies, the strengths and weaknesses of each strategy, and the research done on combined factor investing. The methodology section outlines the procedures followed in investigating the research questions. Thereafter, the paper discusses the findings on all the single-factor and combined factor strategies tested over the 20-year period. In the final section, conclusions are drawn that relate to each of the research questions, based on the analysis of the results, and areas of future research are recommended. Page-5

13 2 Literature Review 2.1 Background Markowitz on Portfolio Selection Modern factor theory and portfolio selection has its roots in the work done by Harry Markowitz. In 1952, Markowitz published a profound paper titled Portfolio Selection, in which, for the first time in finance, the risk and the return of an asset were defined (Miller, 1999). Markowitz used the expected outcome to define the yield on an asset while using the volatility or standard deviation from the mean return to define the risk of an asset. In his paper, Markowitz identifies that investor s aim not only to maximise returns but also to minimise risk through portfolio selection that meets the risk appetite of the investor. Using the assumption that investors are riskaverse, Markowitz defined the efficient frontier as the maximum expected return of a portfolio for a given risk profile, identifying that asset pricing is related to the risk or standard deviation of the asset. Markowitz s work laid the foundation for risk-based asset pricing, which led to one of the most widely used asset pricing models: the Capital Asset Pricing Model (CAPM) CAPM - Capital Asset Pricing Model The CAPM was one of the first asset pricing models developed independently by Treynor (1961), Sharpe (1964), Lintner (1965) and Mossin (1966), and is based on the principles of diversification and mean variance utility introduced by Markowitz (Ang, 2013). Sharpe (1964) investigated the behaviour of investors acting in accordance with the Markowitz mean variance theory. Sharpe (1964) looked at corner portfolios consisting of two different risky assets with several combinations of each asset. He showed that if investors acted according to the principles of maximising return and minimising risk, indifference curves would be upwardsloping, identifying an efficient portfolio as a unique combination of the two assets, where risk was minimised and return maximised. Sharpe (1964) plotted these indifference curves and efficient portfolios, identifying a linear relationship between risk and return for these efficient portfolios, which he termed the market line. He assumed that if investors had the same appetite for risk, they would hold the same portfolio, which he called the market portfolio. He further identified that if asset prices did not fall on the market line they would be bid up (or down) until they fell on the market line. Sharpe concluded his paper by stating: since all risk can be diversified except risk related to swings in economic activity, the price of an asset must be Page-6

14 directly related to the risk of the market (Sharpe, 1964, p. 442), thereby defining what we now know as the beta of an asset or the linear relationship of an asset s return with the market Multi-factor Pricing Models Despite the popularity of CAPM as a means of pricing assets and predicting market behaviour, many subsequent studies (Basu, 1977, Banz, 1981, Fama & French, 1993) showed that other factors in addition to market risk factors were responsible for asset pricing, creating an argument for a multi-factor model rather than a single-factor model. These papers identified that other factors besides risk could be used to identify higher-performing stocks, thus laying the foundation for factor investing. Basu (1977) empirically investigated the relationship between return of assets and price earnings ratio (PE). He showed that portfolios with low PE ratios on average earned higher risk-adjusted returns than those with higher PE ratios, suggesting that PE ratios could be used to predict stock prices. Banz (1981) investigated the size effect. By looking at stocks ranked on size, he found that smaller stocks based on market equity earned higher risk-adjusted returns than larger stocks, suggesting that a size premium existed in the market. Subsequent studies done after the release of this study on the size premium show a disappearance of the size factor. Ang (2013) suggests that the disappearance of the size factor could have been a behavioural response to the studies, which caused investors, in search of higher returns, to bid up the price of small cap companies. Black (1993, cited by Ang, 2003) on the other hand suggests that the initial research into size could have been flawed due to data mining Fama and French 3-factor Model Fama and French (1993) published a controversial paper called Common risk factors in stock and bond returns, in which they furthered research already done on CAPM. They concluded that the market risk factor in isolation does not account for return of assets, showing that cross section returns of US common stocks showed little relation to β (Fama and French 1993). In the paper, Fama and French propose a 3-factor model that included a size and book-to-market factor in addition to the market risk factor used in CAPM. Through a series of cross-section regressions they showed that the inclusion of a size and book-to-market factor has better explanatory power than the CAPM risk coefficient beta. Page-7

15 Fama and French (1993) investigated common stocks and bonds listed on the NYSE and NASDAQ from 1963 to They formed portfolios based on both size and book-to-market, which they rebalanced annually. They split the portfolios into two sized groups (small and large), and into three book-to-market groups (high, medium and low), forming six portfolios with the intersections of each group. In addition, they formed a SMB portfolio (small minus big portfolio), which represented a long position on small stocks and a short position on large stocks, as well as a HML (high minus low) portfolio, which represented a long position on high book-tomarket stocks and a short position on low book-to-market stocks. The SML portfolio is representative of the size premium, while the HML is representative of the value premium, or book-to-market premium. Stocks with relatively high book-to-price are referred to as value stocks, while those with low book-to-price are associated with growth stocks. Fama and French (1993) used a regression approach based on that of Black, Jensen and Scholes to identify relationships between returns and SMB and HML risk factors. The results on common stocks indicated that the two additional factors in addition to the CAPM market risk factor Copyright provided significantly better return prediction capability. They UCT concluded by providing a 3-factor model for predicting asset returns based on a size and value premium: = "#$ + "#$ + () Where: - Expected Return of an asset - Risk-free rate " - Return of a market cap weighted index that represents the return of the market "#- Value premium "#- Size premium s and h Sensitivity coefficients to size and value premiums respectively b- Coefficient of regression with respect to the market risk premium. Although these are similar to CAPM coefficients, they are not identical The Fama and French model challenged the CAPM model and revolutionised the financial ideology of the time. Fama and French (1992) challenged the underlying assumption of market efficiency, which was the cornerstone assumption on which CAPM was founded. The 3-factor model identified pockets of inefficiency that could be exploited to achieve higher excess returns, Page-8

16 and in many ways paved the way for research into other factor premiums that exist in modern markets. As with most revolutionary papers, the 3-factor model came up against harsh criticism. Black (1993) published a paper directly attacking the 3-factor model, stating that the empirical results had little theory to back them up. In addition, he suggested that the success of the model was due to data mining, and criticised the time period in the study, stating that the results would not be able to be replicated in other periods. In another attack on the 3-factor model, Kothari, Shanken, and Sloan (1995) published a paper that re-examined the results obtained by Fama and French (1992). They criticised the COMPUSTAT database used by Fama and French, stating that the database was prone to survivorship bias, which could have influenced the results. Kothari et al. (1995) provided evidence using the Standard and Poor s (S&P) database from 1947 to They found a weak relationship between book-to-market values and stock return. In addition, Kothari et al. (1995) stated that the monthly beta calculation used by Fama and French was flawed, and suggested that an annual beta calculation was a more appropriate method for calculating beta, as most investors have investment horizons that are closer to a year than Copyright to a month. UCT Despite the wave of negative publicity around the findings of Fama and French (1992), several independent studies following the 3-factor model confirmed the findings of Fama and French (1992). Davis (1994) studied the effects of book-to-market, earnings yield, and cash flow-toprice on a database constructed using accounting information listed on Moody s Industrial Manuals for the period 1940 to Davis (1992) confirmed the absence of survivorship bias in the dataset, explaining that all companies listed were included in the manual regardless of whether they existed or not. In the study Davis (1992) confirmed the findings of Fama and French, showing a strong positive relationship between book-to-market and asset returns. In addition, Davis s (1992) study showed little relationship between beta and asset returns using annualised data to calculate beta, thus disproving the argument raised by Kothari et al. (1995). Further support of the 3-factor model came from Fama and French (1996) in a study on 13 international markets. In this study, Fama and French set out to investigate the value premium in international markets by investigating book-to-market, earnings yield, dividend yield, and cash flow-to-price. They used electronic data from Morgan Stanley s International perspective. The study identified that high book-to-market stocks achieved significantly higher returns than low Page-9

17 book-to-market stocks in all 13 developed markets, thereby confirming the results of the Fama and French 3-factor model. 2.2 Single-factor Portfolio Selection Strategies Value Investing Value investing is a method of investing or company selection that looks at companies with a high intrinsic value relative to their market value. Value investing became famous after the work done by Benjamin Graham and David Dodd in their book called Security Analysis. Benjamin Graham, often credited with being the father of value investing, focused on valuing companies by examining the companies intrinsic value, developing tools and screens to help investors identify companies that had high intrinsic values (Scott, 1996). In his book The Intelligent Investor, Graham outlined several rules that identify value stocks based on asset value, earnings, dividends and financial position. These tools suggested by Graham (1949) are listed below: PE ratio should be less than the inverse of the yield on corporate bonds PE ratio should be less than 40% of the average PE over the last 5 years Copyright Dividend yield should be greater than 2/3 yield on corporate UCT bonds Price should be less than 2/3 book value Price should be less than 2/3 net current assets Debt to Equity ratio should be less than 1 Current assets should be greater than twice current liabilities Debt should be less than twice net current assets Historical growth in EPS should be greater than 7% per annum over the last 10 years The company should have no more than 2 years of negative earnings over the last 10 years Although Benjamin Graham was an extremely successful portfolio manager, he never empirically tested his rules or screens. Shortly after Graham s death, Henry Oppenheimer (1984) tested Graham s stock selection criteria empirically. He reviewed stocks listed on the NYSE for the period 1973 to Oppenheimer (1984) found that by using Graham s screens an investor could generate significant excess risk-adjusted returns. Following Benjamin Graham s work and pre-fama and French (1992), many studies were done that confirmed the success of value factors on US equities; earnings yield (Basu, 1983), leverage (Bhandari, 1988) and book-to-market equity (Rosenberg, Reid, & Lanstein, 1985). Page-10

18 Value investing has been around since the 1920s and still is one of the major investment strategies. There has been a wealth of research done both internationally and in South Africa that confirms the existence of a value premium based on different styles or factors that represent value. This review focuses on work done on the Johannesburg Stock exchange (JSE). Van Rensburg (2001) reviewed industrial stocks listed on the JSE over the period 1983 to1999, using a portfolio-based approach to investigate style-based effects on the sample of stocks. Van Rensburg investigated 20 styles (factors), including the value factors, earnings yield, dividend yield, net asset value-to-price and turnover-to-price, adjusting for look-ahead bias, thinly traded shares and survivorship bias in the sample of industrial shares. After risk adjustment using CAPM he found evidence of a statistically significant value premium in earnings yield, dividend yield and book-toprice. Van Rensburg and Robertson (2003) extended the work done by Van Rensburg (2001) by investigating the entire population of stocks listed on the JSE between 1990 and They reexamined Copyright the 20 styles (factors) identified by Van Rensburg (2001), UCT using a cross-sectional regression approach to identify the relationship between each style and dividend adjusted returns. The study identified the presence two dominant styles: size and earnings yield, both of which have a significant relationship with returns, providing motivation for a 2-factor model. Using the same dataset as Van Rensburg and Robertson (2003), Auret and Sinclaire (2006) investigated six styles: book-to-market, price-to-net asset value, cash-flow-to-price, earnings yield, dividend yield, and size. Their results from univariate regressions indicate significant risk-adjusted returns for all of the styles, confirming the presence of a value and quality premium. After looking at a multiple regression analysis they concluded that book-to-market is the most significant value premium and completely subsumes the effect of size and earnings yield (Auret & Sinclaire, 2006). Muller and Ward (2013) performed an excellent study, looking at the longest period of any of the other studies. They performed a portfolio-based study on stocks listed on the JSE over a 27- year period from 1985 to They investigated 30 styles under the broader categories of value, momentum and quality. They adjusted the data set for survivorship bias, look-ahead bias and thinly traded shares. In addition, they only focused on the largest 160 companies according to market value, which accounted for 99% of the total market value. They argued that companies Page-11

19 that fall outside this range are too small and illiquid to be considered by institutional investors. After adjusting for market value they ranked the data based on each style, forming five quintile portfolios. Unlike other studies done that look at regression analysis of the styles, Muller and Ward (2013) performed a graphical study that looked at the compounded value of each of the five portfolios, as well as the J203 All Share Index (ALSI). Their results indicated that portfolios based on earnings yield, dividend yield and market-to-book value provided better performance than the market portfolio, indicating the presence of a value premium. They also showed the superior performance of portfolios formed based on quality and momentum. Despite the popularity of Value investing as a portfolio section strategy many value funds, particularly in South Africa, have performed badly over the past two years. The philosophy of purchasing companies that are trading at a discount to intrinsic value could be the reason why many of the funds are not performing well. Some argue that these cheap companies have dropped earnings due to fatal setbacks which appear attractive to value investors but have little chance of emerging from what is commonly known as the value trap (Lawri, 2013). Using value screens Copyright to select portfolios runs the risk of allocating funds to companies UCT that will never recover, or industries that are on the decline. Value investing can be risky, which is why Ang (2013) and others argue that there is a premium associated with it. In South Africa, funds such as the Investec Value Fund and RECM Equity Fund have suffered severe losses, primarily due to their high exposure to resources which have provided attractive valuations but have struggled to recover from severe losses due to slowing economic conditions and labour unrest (Cairns, 2015) Momentum Investing Momentum investing is a behavioural investment strategy that capitalises on the over-reaction or under-reaction by the market to past information i.e., prior returns (price momentum) or prior earnings (earnings momentum). Momentum investing aims at capitalising on firms with strong prior returns or earnings, taking a long position on winners (buying stocks with high past returns or earnings) and shorting losers (selling stocks with low prior returns or earnings) (Ang, 2013). Momentum investing became popular during the 90s following work done by Jegadeesh and Titman (1993) on stocks listed on the NYSE and the NASDAQ, from 1965 to In their study they examined the behaviour of portfolios formed based on prior returns ranging from 3 to 12 months with varying holding periods of 3 to 6 months. They found that adopting a long/short Page-12

20 strategy for all of the variations yielded significant excess returns, with 12-month prior returns and a 3-month holding period being the best performing strategy, yielding 1.49% per month. Although Jegadeesh and Titman (1993) cited under-reaction by the market as the possible reason for the success of the strategy, they however, did not provide any convincing evidence proving such. Further work by Louis, Chan K.C, Jagadeesh, Narasimhan, and Lakonishok (1996) confirmed the success of both price momentum and earnings momentum strategies, identifying that price momentum yields higher returns and is longer-lived than earnings momentum. In addition, they linked momentum strategies to earnings revisions, showing that analysts tend to under-react to prior price and earnings information, thereby confirming that the success of momentum strategies is caused by an under-reaction of the market. Van der Hart et al. (2003) performed a comparative study of value and momentum strategies, investigating stocks listed on 32 emerging markets over the period 1985 to They confirmed Copyright the presence of a momentum premium over an equally weighted UCT index, and found that a value strategy using earnings yield marginally outperforms a 12-month momentum strategy. These results contradict results on US securities, which indicate that momentum strategies have significantly outperformed value strategies over the period 1965 to 2011 (Ang, 2013). The superior performance of momentum strategies over value strategies has also been shown in several studies done on the JSE. Van Rensburg (2001) showed that a 12-month price momentum strategy with monthly rebalancing significantly (at a 95% confidence level) outperformed a similar portfolio formed using earnings yield, which was the highest performing value strategy in the study. After risk adjustment using CAPM the momentum strategy delivered a risk-adjusted return of 1.52% compared to 1.32% of the earnings yield portfolio. In a comparative study over a 27 period on the JSE, Muller and Ward (2013) showed that a 12-month momentum strategy with quarterly rebalancing generated an overall annual return of 26.1% compared to 24.1% of the highest value portfolio formed using price-to-book. The results from these studies are summarised in Table 1. Page-13

21 2.3 Combined factor strategies The wealth of research on the topic of factor or style investing suggests that active asset managers can capitalise on pockets of inefficiency that exist in markets by using either value, momentum or quality factors to select portfolios. The success of each strategy depends on the factor and the market in which the factor is applied. Many modern portfolio managers motivate for a multifactor type strategy in which combinations of value, momentum and quality variables are used to select portfolios. Credit Suisse HOLT for instance, uses a combination of variables such as cash flow return on investment (quality), earnings sentiment (momentum) and earnings yield (value) to analyse portfolios in different markets (Rones, Curry, & Williamson, 2015). The question that must be asked is: can adopting a combined factor approach to portfolio selection outperform a single-factor approach? Van der Hart et al. (2003) studied a broad range of selection strategies across 32 emerging markets; these included single-factor momentum and value strategies as well as combinations of value and momentum, testing portfolios that ranked highly on both the momentum scale as well as on Copyright the value scale. They used a portfolio-based approach with UCT monthly rebalancing to sort portfolios based on 17 different single-factor variables and four different combinations of momentum, value and earnings revisions. Their results indicated that combined factor models outperform an equally weighted index of stocks as well as the single-factor constituents. In particular, they found superior performance of the portfolios formed using 12-month price momentum, earnings yield and market-to-book values, concluding that the performance of univariate strategies can be enhanced by combining momentum, value and earnings revisions into a multifactor strategy (Van der Hart et al., 2003 pp 130). In their investigation titled Style-based effects on the JSE, Muller and Ward (2013) investigated the effects of 30 factors using a Visual Basic style engine to manipulate data. The study covered the effects of value momentum and quality variables as well as the effects of combining variables. The study used descriptive statistics to compare the annual returns of portfolios formed using each factor. They showed that portfolios formed using a combination of factors consisting of a 12-month momentum, return on capital, cash flow-to-price and earnings yield, outperformed all other factors tested, achieving an excess return of 14% per annum compared to the J203 (All Share Index). Page-14

22 2.4 Conclusion Table 1 summarises the some of the research papers that covered the JSE as well as a paper by Van der Hart et al. (2003) that investigated several emerging markets. The papers discussed in this review and summarised in Table 1 all show signs of a strong value, momentum and quality premium in the JSE, evidenced by higher overall and risk-adjusted returns. A few papers such as those by Van der Hart et al. (2003) and Muller and Ward (2013) suggest that these single-factor strategies can be enhanced by the use of a combined factor approach, which uses several factors or styles as a means of selecting portfolios. Despite the wealth of research into the use of combined factors in developed markets and emerging markets, very few papers that investigated the JSE have shown any statistical evidence in support of a combined factor approach to portfolio selection. This study fills the gap by investigating single-factor portfolio selection strategies based on value, momentum, and quality as well as a combined factor approach, with the aim of identifying whether a combined factor strategy can achieve higher risk-adjusted returns. Table 1: Summary of research done on the JSE and emerging markets Hodnett,Hsieh, FraserandPage VanderHartet Authors Muller&Ward(2013) &VanRensburg, (2000) al.,(2003) (2012) Emerging Market JSE JSE JSE Markets Periodofstudy 1983J J J J1989 Method Descriptive/CAPM adjustment Descriptive Fama&French regression Descriptive Factor Annual Return Annual Alpha (CAPM) AnnualReturn AnnualAlphaα (Fama&French) AnnualReturn Size 19,50% NA 9,60% EarningsYield(EY) 22,60% 5,80% 2,40% 22,60% Returnoncapital(ROC) 20,30% 4,00% / / ROE 22,40% 1,70% / / BookJtoJmarket 24,90% 2,50% 29,20% 14,40% 21,90% DividendYield 22,10% 5,50% 28,50% 4,80% 21,10% Interestcover 21,10% 3,30% / / Assetgrowth 23,10% /4,70% / / CashflowJtoJprice 24,50% 7,50% / 10,80% / Liquidity 22,90% /2,40% 2,4%(P>5%) 12,70% 12Jmonthpricemomentum 26,10% 8,90% 36,00% 8,40% 20,60% Value+Momentum / / / / 24,20% Value+Quality+Momentum 32,30% 14,20% / 25,00% Page-15

23 3 Methodology 3.1 Research Approach and Strategy Figure 1: Research Onion Source, Saunders, Lewis, & Thornhill (2012) Figure 1 illustrates the research onion as defined by Saunders, Lewis, & Thornhill (2012), which identifies the philosophy, approach and strategy of this study on the effects of multifactor portfolio selection strategies. The overarching philosophy of this research can be described as being positivistic and objective by nature, as the study uses existing theory on factor investing to develop a hypothesis related to multifactor selection strategies. Furthermore, the research builds on work done by Van der Hart et al. (2003), Van Rensburg and Robertson (2003) and Muller and Ward (2013) to further investigate the credibility of multifactor selection models, focusing on equities listed on the JSE over the last 20 years from January 1995 to January 2015, thereby Page-16

24 following a typical positivistic philosophy as described by Saunders et al. (2012). The study can also be described as objective, as it views reality from an external and independent viewpoint, which is ensured through the use of mathematical models in selecting portfolios. An objective stance is critical to this study to ensure that there is no bias to stock selection besides the factors under investigation. This is a quantitative study, using returns listed on the JSE over the last 20 years as its basis. The study follows a deductive approach as identified by Saunders et al. (2012), in which the hypothesis statements that have been developed from existing theory were tested using statistical tests to establish if there are any differences between the average monthly returns of each strategy. Figure 2 outlines the approach applied to the study, modified from the deductive theory defined by Saunders et al. (2012). 1 State hypothesis There is no difference between average returns of factor-formed portfolios and an equally weighted benchmark Is there a difference between the average 2 Express hypothesis in operational terms monthly returns of portfolios selected using factor strategies and an equally weighted benchmark? 3 Test hypothesis Use t-test or Wilcoxon rank sum test of the mean to establish if there is any difference between the average monthly returns 4 Examine outcomes Is the test mean significant at the 95% confidence level? Is there a significant difference between the average monthly returns? 5 Modify theory if necessary Do the results conform to the literature? Figure 2 : Diagram Indicating Deductive Approach Source, Saunders, Lewis, & Thornhill (2012) Page-17

25 3.2 Research Design We investigated the hypothesis statements by back testing data listed on the JSE over the last 20 years. The data was collected using monthly intervals from several databases, namely: Bloomberg, Datastream and INET BFA. Most of the data however, was sourced from Datastream, as we found this database to be the most populated data source. We examined the data using a time series approach that looked at returns on a constant 1-month interval rather than using a cross section in time. Several studies, Fama & French (1992) and Van Rensburg & Robertson (2003) for example, have used cross-sectional studies successfully to investigate the relationship between factors; however, because this investigation compares the returns of portfolios that are rebalanced monthly, we opted for time series analysis, examining the average returns of portfolios formed using various factors on a monthly basis over the entire 20-year period Data Collection Monthly data was collected using Bloomberg, INET BFA, and the Datastream databases. Twenty years of data was collected on a monthly basis, from January 1995 to January 2015, for all of the equities listed on the JSE. The Total Return Index (TRI) was used to calculate the return of each stock as TRI includes return due to dividends paid and is therefore a more accurate measure of return than using price data in isolation. In addition, data related to size, value, momentum and quality factors were extracted for all companies listed on the JSE between A list of data that was collected and used in this study follows: Size Factor Market Value (MV) Value Factors Earnings Yield (EY) Dividend Yield (DY) Book-to-Market (BTM) Price Momentum Calculated using closing price data 6 month price momentum (6MPM) 12-month price momentum (12MPM) Quality factors Return on Equity (ROE) Page-18

26 Return on Invested Capital (ROIC) b Net Profit Margin (NM) Combined Factors were calculated by using a weighted sum of each of the individual factors, as listed below: Value and momentum Earnings yield and price momentum (50% EY+ 50% 12PM) Value and quality Earnings yield and ROE (50% EY+ 50% ROE) Quality and momentum ROE and price momentum (50% ROE + 50% 12PM) Value, Quality and Momentum EY, ROE and price momentum (33% ROE + 33% EY + 33% 12PM) Survivorship bias Survivorship bias is an error that can occurs when companies that have been delisted are excluded from studies. Since investors will never know when companies will delist these companies must be included to avoid survivorship bias. This study used data of all companies listed on the JSE over the 20-year period regardless if the companies delisted and therefore we assume that this study is free from survivorship bias Look ahead bias Look ahead bias is common in historical portfolio analysis such as this study. The bias occurs when historical data used in a particular study or analysis may have not been known during the analysis period. We ensured that the data used in this study is free from look ahead bias by lagging factor data and return data by a one month. This ensured that any data used to construct portfolios was available for a minimum period of one month before portfolio construction. 3.3 Sampling The investigation was confined to equities listed on the JSE, which has over 800 securities traded on two main boards: the Main board and the Alt X (JSE, n.d.). This investigation was limited to equities, and excludes all bonds, preference shares and commodities. Currently the JSE has 315 b Return on Invested Capital = NOPAT/invested capital. Since financials have extremely high-invested capital they tend to have very low ROIC which would skew results. Hence for this study, all financials from factor models that use ROIC have been omitted. ROE is a better measure of performance for financials. Page-19

27 companies listed on the two boards; however, this value has changed through time as some companies were listed and others delisted. Since institutional investors are interested in larger, more liquid equities, a sample consisting of the largest 100 companies based on market value every month for the entire 20-year period was investigated. The sample accounts (on average) for more than 95% of the market cap of all the equities listed on the JSE. This type of sampling is based on subjective judgement and can be defined as non-probability sampling (Saunders et al., 2012). All data collected for the sample of 100 companies was scrutinised to ensure that data is available for each of the factors listed in In certain instances where data was not available, these companies were left out of the study for that particular factor. 3.4 Data Analysis Portfolio Selection and Data Manipulation The data mentioned in was collected and converted into an Excel database that was used as the engine behind the analysis. A Visual Basic (VBA) program was written to manipulate data and form portfolios at monthly intervals, based on a particular factor. Initially the programme collected Copyright the largest 100 stocks every month based on market cap and UCT created a matrix of codes that represented the largest 100 stocks. Thereafter, another VBA program was used to rank the equities on a monthly basis according to the factors mentioned in For each of the 240 monthly periods, five portfolios (quintiles) were created ranked from highest to lowest, based on each variable. An equally weighted return was calculated for each portfolio and the compounded return graphed for the entire time series. The portfolios were labelled from P1 to P5, with P1 representing the highest-ranking portfolio according to a particular factor. In addition, a long/short strategy (L/S), which takes a long position on the highest-ranking portfolio (P1) and a short position on the lowest ranking portfolio (P5), was calculated. This L/S portfolio is labelled as P1-P5 in all the graphical illustrations that follow. Lastly, an equally weighted portfolio consisting of all 100 equities based on market cap was formed and used as a benchmark to compare each of the portfolios Visual inspection and overall return calculation The graphs of the compounded return versus time for the six portfolios were plotted for each factor. A logarithmic scale was used to plot the returns, in order to enhance the separation of the returns and allow for easy inspection. The overall annual geometric return and annual standard deviation was also calculated for each portfolio and is commented on. This initial inspection was used to get an idea of the separation of each portfolio (a visual account for the difference Page-20

28 between each portfolio), which assisted in identifying any correlation between the factor under investigation and the returns of the portfolio. A clear spread between the highest-ranking portfolio, P1, and lowest-ranking portfolio, P5, evidenced that the factor had a strong signal during the test period. We also observed the ordering of the portfolios and took note of any changes that might have occurred during the 20-year period. A consistent ordering of the portfolios is an indication that that particular factor performed similarly through different business cycles Analysis of the average monthly return vs the equal weighted benchmark In order to test the difference between the average returns of portfolios vs the benchmark, we first had to establish if the returns were normally distributed. This was done by testing the monthly return data of the P1 and P1-P5 portfolio using the Shapiro-Wilk test for normality. The Shapiro-Wilk test assumes a null hypothesis of normality: if the test statistic is less than the required alpha value then the data does not pass the test for normality. In our case we tested for normality at the 95% confidence level. The results, summarised in Appendix 1, indicate that the data failed the normality test and hence we progressed to testing the difference in average monthly returns with the Wilcoxon Rank Sum Test. The Wilcoxon Rank Sum test is a nonparametric test of the mean that does not assume normality. The test can be used instead of the t- test of the mean to test for the difference between two means. The hypothesis statement for the test is given below: There is no difference between the average monthly returns of portfolios formed using a single-factor strategy and an equal weighted portfolio of all stocks in the sample. : " = : " 0 Where: X = factor being investigated EW= Equal weighted portfolio Page-21

29 The results of the Wilcoxon rank sum test were used to establish whether the average monthly return of the highest-ranking portfolio (P1) is greater than the return of the equal weighted portfolio. The results of this test confirmed whether a single-factor approach to portfolio construction outperformed an equal weighted benchmark Analysis of each of the highest-ranking portfolios To investigate the second research question we analysed each of the highest-ranking portfolios against each other, testing the difference of the average return obtained from each of the highestranking portfolios against each other. The hypothesis statement that was tested is given below: There is no difference between the average monthly returns formed using one single-factor strategy when compared with every other single-factor strategy. : " " = : " " 0 Where: X1 = factor being investigated X2= different factor being investigated The results of the all the tests were summarised in a matrix and used to draw conclusions on whether any one single-factor strategy outperformed any other single-factor strategy. 3.5 Data Reliability The data used in this study was collected from reputable databases such as Bloomberg, Datastream and INET BFA. In addition, several studies (Van Rensburg & Robertson, 2003, Ward & Muller, 2012, Muller & Ward, 2013) have successfully used data from INET BFA without any reported reliability issues. Based on this we assumed that the data collected was reliable for the purpose of this study Analysis of risk-adjusted returns To investigate the risk-adjusted behaviour of portfolios we incorporated the standard deviation of the portfolio as a measure of risk. We investigated the return per unit risk of each strategy in order to establish which portfolio provided the best return per unit risk. We also looked at the change in the return per unit risk, as factors were combined to investigate whether a combined factor approach to Page-22

30 portfolio construction achieves higher risk-adjusted returns. The return per unit risk measurement was used instead of the Sharpe ratio, as the risk-free rate is the same across all portfolios and thus can be excluded. Lastly, the Information Ratio (IR) was also calculated, which aims to quantify the consistency of excess returns over a particular benchmark. The measurement gives an indication of consistency to outperform the benchmark. The equation for the Information Ratio (IR) is given below. R x = Return of portfolio " = " " REW- Return of the equally weighted benchmark σ X-EW = Standard deviation of excess returns vs the equally weighted benchmark The ratio or index looks at the excess return per unit of standard deviation of the difference between the portfolio and the benchmark, thereby establishing a measurement for the consistency in achieving excess returns over the benchmark. We used the information ratio as the primary measure of riskadjusted returns, and ranked performance of factors based on this index. 3.6 Data Validity Cooper & Schindler (2003) define three types of validity: face validity, construct validity and internal validity. They define face validity as surface assurance by a non-researcher that the research methods measure the intended goals. Can a layperson see the reasoning behind the proposed methods? In this study we used statistical methods to test the returns of portfolios using various selection strategies. This satisfies the criteria of face validity through the use of simple techniques that are easy to understand and which measure distinct variables (monthly returns). Construct validity is defined as confirmation that the study actually measures what is intended. In this study, several research papers (Van der Hart et al., 2003, Van Rensburg & Robertson, 2003, Muller & Ward, 2013) were used as guides to constructing the research methodology. These papers successfully conducted similar research using different markets or different time periods, and thus we assume that the proposed methods are valid and meet the requirements of construct validity. Lastly, internal validity is defined as confirmation of cause and effect. Does the stated variable actually cause the measured effect? Although we cannot draw conclusions directly related to cause and effect, we adopted the Wilcoxon rank sum test of the mean and the information ratio to analyse the association of factors with excess returns, thereby satisfying requirements of internal validity. Page-23

31 4 Research Findings, Analysis and Discussion 4.1 Size Effect Figure 3: Graph of portfolio returns ranked according to Market Value Figure 3 shows the portfolio returns of the largest 100 equities ranked according to market value. The P1 portfolio represents 20 of the highest-ranking companies according to market value from the sample of 100. The average market value of companies in the P1 portfolio over the 20-year period was R117 billion. The P5 portfolio on the other hand, represents the lowest 20 companies (from the sample of 100) according to market value, and had an average market value over the of R2.4 billion over the 20-year period. The results appear to be random with very little evidence of any spread between the highest and lowest raking portfolios, which is evidenced by the relatively flat P1-P5 portfolio. Although the highest-ranking portfolio according to market value (P1) did perform the worst over the 20-year period from 1995 to 2015, the results are inconclusive and suggest that there is no size premium in the JSE over the period under review. Although some papers on the JSE indicate the presence of a size premium (Van Rensburg, 2001, Van Rensburg and Robertson, 2003b), our results are consistent with those of Strugnell, Gilbert, and Kruger (2011), Muller and Ward (2012), who find no conclusive evidence of a size effect on the JSE. Page-24

32 4.2 Single-factor Strategies Value Factors Earnings Yield (EY) Figure 4: Graph of portfolio returns ranked according to Earnings Yield (EY) Figure 4 shows the compounded returns of portfolios ranked according to earnings yield c for the period 1995 to The P1 portfolio represents the companies with the highest earnings yield figures while the P5 portfolio represents companies with the lowest earnings yield figures from the sample of the 100 largest companies based on market value. The graph clearly indicates a spread between the highest and lowest ranking portfolios. In addition, the graph shows clear ordering of the five portfolios in relation to levels of earnings yield, which is consistent through the entire time series, with a reversal during the Dotcom bubble of 1997 to In general, over the 20-year period the results indicate that a long strategy on high earnings yield stocks achieved a premium of 11.4% per annum over an equal weighted index, formed with the largest 100 stocks every month. The statistical significance of these results is addressed in the analysis section. c The data for earnings yield is historical data, which applies to all other factors. Forward-looking data was difficult to source and thus not used in this study. Page-25

33 Book-to-Market (BTM) Figure 5: Graph of portfolio returns ranked according to Book-to-Market A book-to-market d strategy aims at capitalising on companies that are trading low relative to their book value. Our results, shown in Figure 5, confirm the existence of a book-to-market premium, evidenced by the clear spread between high and low book-to-market portfolios as well as the clear ordering of intermediate portfolios. The ordering between portfolios is also generally consistent throughout the entire series except during the Dotcom bubble consistent with returns of portfolios ranked according to earnings yield. The top ranked portfolio based on book-tomarket values out-performed both the equally weighted portfolio and the J203, achieving a 7,4% premium over an equally weighted index. The results are consistent with several studies done on the JSE by Auret and Sinclaire (2006), Van Rensburg and Robertson (2003b) and Muller and Ward (2013), all of whom indicate the existence of a book-to-market premium on the JSE for several different time periods. d Book-to-market = (Total assets intangible assets liabilities)/market value. P1 portfolio represents companies with the highest book-to-market values while P5 portfolio represents companies with lowest book-to-market values from the sample of 100 companies each month. Page-26

34 Dividend Yield Figure 6: Graph of returns of portfolios ranked according to Dividend Yield (DY) A dividend yield e strategy invests in companies that have a high dividend yield. The strategy takes advantage of the positive signalling effect that dividend pay-out has on the market. There is much evidence that supports this strategy, both in emerging markets and in developed markets. Van Rensburg and Robertson (2003b) and Muller and Ward (2013) both give supporting evidence for the existence of a dividend yield premium in the JSE. Our results, shown in Figure 6, support these findings, seen by a clear spread between high and low dividend yield portfolios over the period under review. As with the other value factors (earnings yield and book-tomarket) the returns of portfolios ranked according to dividend yield also show a reversal during the Dotcom bubble from 1997 to Thereafter the portfolios switched to the hypothesised order in support of a dividend yield strategy. The results indicate that a long strategy on high dividend yield equities out-performed both the J203 and equal weighted index of the largest 100 equities by market cap, with a 4.5% per annum premium over the equal weighted index. The results indicate that high dividend yield companies performed well relative to the market. e Dividend yield is calculated using the last dividend paid divided by price. The P1 portfolio represents the highestranking portfolio based on dividend yield while the P5 portfolio represents the lowest ranking portfolio from the sample of 100 companies. Page-27

35 4.2.2 Momentum Factors Month Price Momentum Figure 7: Graph of portfolio returns ranked according to 6-month price momentum A price momentum strategy invests in companies with high prior returns. Figure 7 shows the results of a 6-month price momentum f strategy with a 1-month holding period on the JSE over the last 20 years. The results indicate a clear spread between the highest- and lowest-ranking portfolios, which is consistent with evidence from developed markets (Louis et al., 1996) and emerging markets (Van der Hart et al., 2003, Muller and Ward, 2013). A long strategy on 6- month price momentum outperforms both the equal weighted index and the J203 with an 11.2% premium over an equal weighted index. The results also show a clear ordering of the highest and lowest ranking portfolios over the period, which gives further support for a momentum premium on the JSE over the last 20 years. f 6-month price momentum is calculated by looking at 6-month prior returns: Price(t+6)/ Price (t)- 1. The P1 portfolio represents companies with the highest 6-month price returns every month while the P5 portfolio represents companies with the lowest 6-month price returns every month, from the sample of 100 companies. Page-28

36 Month Price Momentum Figure 8: Graph of portfolio returns for portfolios ranked according to 12-month price momentum The 12-month price momentum g strategy showed the best performance of all momentum strategies tested over the 20-year period. Figure 8 shows the results of the 12-month price momentum strategy with a 1-month holding period. The graph shows a distinct ordering between portfolios, which remains intact for the entire period. A long strategy on the highest-ranking portfolio outperforms both the J203 and an equal weighted index with a 17.7% per annum premium over the equal weighted index. In addition, a long/short strategy using 12-month price momentum also outperforms both the J203 and equal weighted index, but slightly underperforms a long strategy. Looking at both Figures 7 and 8, our results suggest the presence of a very strong momentum premium on the JSE over the last 20 years, which is consistent with studies done by Van der Hart et al. (2003) and Muller and Ward (2013). g 12-month price momentum is calculated by looking at 12-month prior returns: Price(t+12)/ Price (t)- 1. The P1 portfolio represents companies with the highest 12-month price returns every month while the P5 portfolio represents companies with the lowest 12-month price returns every month from the sample of 100 companies. Page-29

37 4.2.3 Quality Factors Net Profit Margin Figure 9: Graph of portfolio returns ranked according to net profit margin Figure 9 shows the results of portfolios ranked according to net margin h. Although our results show a premium in investing in companies with a high net margin, the ordering of the intermediate portfolios is not as distinct as either the value or momentum portfolios. The random nature of the portfolios is prevalent during the 2000 and 2008 crashes. On the whole, the net profit margin signal does not appear to be a very distinct signal. However, a long view on high net profit margin companies did outperform both the J203 and an equal weighted index with a 10.7% per annum premium over the equal weighted index. Gordan (2013) has shown evidence of a profitability premium in emerging markets and shows that gross margin produces stronger signals. Unfortunately, due to lack of data on gross margin we could not test this particular quality factor. h Net margin data was extracted from Reuters Datastream, which included data for financial companies. The net margin for financial companies is calculated using net profit after tax divided by total interest income. P1 portfolio represents companies with highest net margin while P5 portfolio represents companies with lowest net margin from the sample of 100 companies. Page-30

38 Return on Equity (ROE) Figure 10: Graph of portfolio returns ranked according to Return on Equity (ROE) Return on Equity i is an accounting ratio that is used to indicate the amount of net income generated per unit of equity invested. The results shown in Figure 10 give strong support for the existence of a ROE risk premium on the JSE, evidenced by a clear spread and distinct ordering of portfolios. The results indicate that the highest-ranking ROE portfolio P1 outperformed the equally weighted benchmark by 13,6%. Furthermore our results indicate that a long/short strategy outperforms a long strategy on high ROE stocks, which is caused by the extremely poor performance of low ROE stocks. Our results contradict the results obtained by Muller et al. (2013) who identify poor performance of companies with high ROE, suggesting that the market has over-priced these companies. Our results suggest the presence of a strong ROE risk premium, consistent with the work done by Van Rensburg, (2001) and Gordan (2013). i ROE data was extricated from Reuters Datastream and includes both financials and non-financials. The data is based on historical data rather than forward-looking data. P1 portfolio represents the highest companies ranked according to ROE every month while P5 portfolio represents the lowest ranking companies based on ROE from the sample of 100 companies. Page-31

39 Return on Invested Capital (ROIC) Figure 11: Graph of Portfolio returns ranked on Return on Invested Capital (ROIC) Return on invested capital (ROIC) j looks at the efficiency of operating earnings related to capital invested. In a study done on emerging markets Gordan (2013) shows evidence of a profitability premium using both ROIC and ROE as factors. The study shows strong support for an ROIC premium in emerging markets, suggesting that profitable companies tend to remain profitable. The results of portfolios ranked according to ROIC displayed in Figure 11 show a strong visual relationship between ROIC and portfolio returns, consistent with the work done by Gordan (2013). The graph shows a clear spread between high and low ROIC portfolios with a distinct ordering which remains consistent throughout the period. The results indicate that a long strategy on high ROIC stocks outperformed both the J203 and an equally weighted index with a 15,8% per annum premium over the latter, the highest of any of the other single factor strategies. j Return on Invested Capital (ROIC) is defined as net profit after tax (NOPAT)/Invested Capital. Data for ROIC was extracted from the Reuters Datastream database. The dataset included financials and non-financials. For this study we removed financials as these companies have low ROIC values due to high invested capital, which would skew results. Page-32

40 4.2.4 Summary of single-factor strategies Table 2: Summary of Single-factor Strategies Long Strategy (Portfolio P1) Average Return (Geometric) Average Return (Arithmetic) Average Excess Return R(P1-EW) Standard Deviation Return/ Risk Informati on ratio ROIC 33,65% 36,12% 6,69% 19,59% 1,72 1,58 ROE 31,45% 33,95% 4,24% 19,78% 1,59 1,43 12(m) Momentum 35,49% 37,62% 15,14% 21,54% 1,65 1,40 EY 29,18% 31,11% 9,43% 17,41% 1,68 1,29 NP 28,50% 30,95% 9,02% 18,23% 1,56 1,17 6 (m) Momentum 29,80% 32,53% 10,53% 20,52% 1,45 0,95 Book-to- Market 24,99% 27,64% 12,37% 21,03% 1,19 0,58 DY 22,76% 24,61% 4,24% 17,53% 1,30 0,50 Table 3: Summary of Single-factor Strategies Long/Short Strategy - Portfolio P1-P5 Copyright Average Return Average Standard UCT Return/ (Geometric) Annual Return Deviation Risk ROE 32,19% 34,18% 17,67% 1,82 12(m) Momentum 31,74% 34,29% 19,93% 1,59 ROIC 29,15% 30,86% 16,59% 1,76 NP 27,47% 28,78% 14,59% 1,88 6 (m) Momentum 19,89% 22,13% 19,58% 1,02 EY 12,11% 13,51% 15,92% 0,76 Book-to- Market 8,67% 10,44% 18,22% 0,48 DY 6,44% 8,33% 18,99% 0,34 Table 2 and Table 3 summarise the results for all single-factor strategies tested both with a long position on the highest ranking portfolio (P1) as well as a long/short strategy (P1-P5). For the long strategy 12-month price momentum achieved the highest overall return of 35,16% while portfolios ranked according to ROIC achieved the highest return per unit risk as well as the highest information ratio compared with an equal weighted index. The long/short strategy portfolios ranked according to ROE achieved the highest overall annual return, while net profit margin achieved the highest return per unit risk. The results give evidence of a strong quality premium represented by ROIC and ROE as well as a momentum premium on the JSE over the last 20 years. Page-33

41 4.2.5 Investigation of Research Question 1 Is there evidence that supports the view that adopting a single-factor portfolio selection strategy (either value, momentum or quality) consistently outperforms the market, represented by an equal weighted portfolio? In the previous section we graphically illustrated the superior performance of several singlefactor strategies against an equal weighted index. In examining research question No 1 we investigated whether there is a significant difference between the average monthly returns of the highest ranking portfolio (P1) against an equally weighted benchmark consisting of the largest 100 companies based on market value Test for Normality Appendix 1 summarises the results from the Shapiro-Wilk test for normality. The results indicate that all the portfolios, except a long/short strategy using ROE and EY have a significance value of less than 5%, which indicates that the data is non-normal. We therefore proceeded to test the average returns with the Wilcoxon rank sum test, which is a non-parametric test that does not assume normality Wilcoxon Rank Sum test of the mean Appendices 2 and 3 outline the results of the Wilcoxon rank sum test for all single-factors, tested using a long strategy, compared to an equal weighted portfolio. Table 4 summarises the average annual excess returns and the p values from the test. The results indicate that a long strategy for all factors, except dividend yield, produces positive excess returns with a significance (p value) less than 5%, implying that the null hypothesis (there is no difference between the average return of a particular single-factor strategy and the average return of an equal weighted portfolio) can be rejected at the 95% confidence interval. We therefore concluded that all the single-factor strategies tested (except dividend yield) using a long strategy produce significant positive excess returns compared with an equally weighted portfolio. The results support the view that adopting a single-factor strategy (value, momentum or quality) does outperform an equal weighted benchmark. Page-34

42 Table 4: Summary of excess returns and Wilcoxon Rank Sum Test Long Strategy Average excess return R(P1-EW) Significance (p value) Book-to-Market 6,69% 0,9% EY 9,43% 0,0% DY 4,24% 11,9% 12(m) Momentum 15,14% 0,0% 6 (m) Momentum 10,53% 0,0% ROIC 13,27% 0,0% ROE 11,62% 0,0% NP 9,02% 0,0% Investigation of Research Question 2 Is there evidence that any one single-factor strategy outperforms every other single-factor strategy? Table 5 summaries the results of the Wilcoxon rank sum test of all long single-factor strategies. The Table gives the significance p-values of all long single-factor strategies tested against each other. The green shaded areas indicate tests that are significant. From the Table we can see that no one strategy significantly outperformed every other strategy at the 95% confidence interval. From the results shown in Table 5, we accept the null hypothesis (There is no difference between the average monthly returns formed using one single-factor strategy when compared with every other single-factor strategy), and conclude that there is no one dominant single-factor strategy that significantly outperforms ever other single-factor strategy. Although ROIC and 12- month price momentums were the highest performing strategies in absolute terms, from Table 5 we can see that ROIC only significantly outperformed book-to-market and dividend yield, while 12-month price momentum only significantly outperformed the value strategies. The results do however confirm the presence of a strong quality and momentum premium on the JSE. Page-35

43 Table 5: Wilcoxon Rank Sum-Significance values for all long single-factor strategies BTM DY EY 12MPM 6MPM ROE ROIC NP BTM 47,6% 13,8% 0,3% 26,5% 52,0% 4,8% 16,8% DY 47,6% 1,3% 0,0% 1,0% 0,4% 0,1% 0,5% EY 13,8% 1,3% 3,8% 87,6% 51,1% 18,8% 91,4% 12MPM 0,3% 0,0% 3,8% 1,5% 13,6% 59,4% 12,2% 6MPM 26,5% 1,0% 87,6% 1,5% 68,0% 52,6% 98,4% ROE 52,0% 0,4% 51,1% 13,6% 68,0% 61,6% 35,8% ROIC 4,8% 0,1% 18,8% 59,4% 52,6% 61,6% 22,8% NP 16,8% 0,5% 91,4% 12,2% 98,4% 35,8% 22,8% Linear regression with equally weighted benchmark 4.3 Combined factors Correlation matrix Table 6: Excess return R (P1-EW) Correlation Matrix BTM DY EY 12MPM 6MPM ROIC ROE NP BTM 1,00 0,27 0,33 /0,04 /0,02 0,10 /0,01 0,24 DY 0,27 1,00 0,23 /0,32 /0,26 /0,05 /0,21 0,03 EY 0,33 0,23 1,00 0,25 0,17 0,10 /0,01 0,21 12MPM /0,04 /0,32 0,25 1,00 0,72 0,30 0,29 0,35 6MPM /0,02 /0,26 0,17 0,72 1,00 0,17 0,21 0,26 ROIC 0,10 /0,05 0,10 0,30 0,17 1,00 0,83 0,52 ROE /0,01 /0,21 /0,01 0,29 0,21 0,83 1,00 0,40 NP 0,24 0,03 0,21 0,35 0,26 0,52 0,40 1,00 Page-36

44 Table 7: Long/Short Strategies Correlation Matrix BTM DY EY 12MPM 6MPM ROIC ROE NP BTM 1,00 0,03 0,37 0,12 0,09 /0,12 /0,12 0,15 DY 0,03 1,00 0,34 /0,24 /0,22 /0,03 /0,06 /0,08 EY 0,37 0,34 1,00 0,22 0,14 0,10 0,12 0,10 12MPM 0,12 /0,24 0,22 1,00 0,75 0,25 0,26 0,40 6MPM 0,09 /0,22 0,14 0,75 1,00 0,21 0,21 0,30 ROIC /0,12 /0,03 0,10 0,25 0,21 1,00 0,66 0,39 ROE /0,12 /0,06 0,12 0,26 0,21 0,66 1,00 0,41 NP 0,15 /0,08 0,10 0,40 0,30 0,39 0,41 1,00 Table 6 and Table 7 summarise the correlation coefficients of factors tested using a long strategy and a long/short strategy. For the long strategy the excess returns for each factor were used to calculate the correlation coefficients. This was done to remove the effect of the market, which causes a similar effect and thus increases the correlation between factors. Both strategies show a very similar correlation pattern. The results indicate that value factors have a low to negative correlation with both momentum and quality factors, with a more negative correlation with the latter. This relationship can be expected as high quality companies i.e. companies with high profitability are usually growth stocks that have high price relative to earnings and thus would be considered low value stocks. Of the value factors tested, dividend yield shows a highest negative correlation with momentum factors while book-to-market shows the highest negative correlation with quality factors. Momentum factors on the other hand have a moderate positive correlation (0,25-0,40) with quality factors which is consistent with the work done by Novy-Marx (2013), who suggests that because profitable companies tend to remain profitable and generate excess returns, they are likely to be picked up by momentum strategies, thereby explaining the high correlation between momentum and quality factors. Page-37

45 4.3.2 Value and Momentum (12MPM+ EY) Figure 12: Returns of portfolios ranked according to 12-month price momentum and earnings yield Figure 12 shows the returns of portfolios ranked according to both value and momentum, represented by a factor formed by combining earnings yield and 12-month price momentum k. To form this combined factor an equally weighted score of the two factors was used; i.e. 0,5x12MPM +0,5xEY. The new factor was then used to rank the equities in the sample. The results displayed graphically in Figure 12 show a clear spread between the highest- and lowestranking portfolio as well as a clear ordering of portfolios, which remained consistent throughout the time series. Both the highest-ranking portfolio (P1) and the long/short portfolio (P1-P5) outperformed the equal weighted benchmark. The overall performance of the P1 portfolio of 35,14% per annum is slightly lower than the P1 momentum portfolio, but considerably higher than the highest-ranking earnings yield portfolio, which earned 29,18% per annum over the same period. A long position on the highest-ranking portfolio produced a 17,7% per annum premium over the equal weighted portfolio. k The factor was formed using an equally weighted score of 12-month price momentum and earnings yield. Portfolio P1 represents the highest-ranking companies according to the new combined factor, while P5 represents the companies that ranked lowest based on this combined factor of momentum and value. Page-38

46 4.3.3 Value and Quality (ROE+EY) Figure 13: Returns of portfolios ranked according to ROE and EY Figure 13 graphically displays the returns of portfolios ranked according to a factor formed using value and quality, represented by earnings yield and return on equity l. The factor was formed by creating an equally weighted score of earnings yield and ROE for all companies in the sample, over the 20-year period. The new factor was then used to rank each of the companies. The graph shows a clear spread and ordering of the portfolios ranked according to this factor. The highestranked portfolio (P1) outperforms the lowest ranking portfolio (P5) by 31,33% and an equally weighted benchmark by 14,51%. The ordering of the portfolios is consistent through the entire period, with some reversal during 1995 to The long/short portfolio also shows strong performance over the 20-year period, with an overall return of 28,8% per annum. l A factor was formed by taking the average between ROE and EY for all companies in the sample. P1 represents companies that ranked highest based on this new factor, while P5 represents the companies that ranked the lowest based on this factor that represents quality and value. Page-39

47 4.3.4 Momentum and Quality (12MPM+ROE) Figure 14: Returns of portfolios ranked according to 12-month price momentum and ROE 12-month momentum and ROE were used in creating a factor that represents momentum and quality m. An equally weighted score of 12-month price momentum and ROE was used as a combined factor to rank the equities in the sample. The returns of portfolios ranked according to this factor are shown in Figure 14. As with both 12-month momentum and ROE, both the long strategy (P1) and the long/short strategy (P1-P5) considerably outperformed the equal weighted benchmark with a 15,07% premium over the benchmark. The results also show a clear spread and distinct ordering of quintile portfolios that remains consistent for most of the time series, with some re-ordering during the Dotcom bubble from 1995 to The overall performance of the P1 portfolio is only slightly (1,5% per annum) greater than the highest-ranked ROE portfolio, and 2,5% per annum less than the highest-ranking 12-month momentum portfolio. m The factor was formed using an equally weighted score of 12-month price momentum and ROE. Portfolio P1 represents the highest-ranking companies according to the new combined factor while P5 represents the companies that ranked lowest based on this combined factor of momentum and quality. Page-40

48 4.3.5 Value, Momentum and Quality (12MPM+EY+ROE) Figure 15: Returns of portfolios ranked according to value momentum and quality A combined factor was formed using value momentum and quality by creating an equal weighted factor of earnings yield, 12-month price momentum and Return on Equity n. The factor was formed by creating an equally weighted score using each of the 3 variables. This new factor was then used as a factor to rank each of the equities in the sample. The results using this 3- factor combination are shown in Figure 15. The graph clearly shows the superior performance of both a long and long/short strategy over the equally weighted benchmark, with a 14,4% and 17,5% premium respectively over the equally weighted portfolio over the 20 year period. The absolute return of the long/short strategy using this combination outperformed all constituent factors, with a 3,13% per annum premium over ROE, which was the highest performing long/short factor. n The factor was formed using an equally weighted score of 12-month price momentum, earnings yield EY and ROE. Portfolio P1 represents the highest-ranking companies according to the new combined factor while P5 represents the companies that ranked lowest based on this combined factor of momentum, value and quality. Page-41

49 4.3.6 Investigation of Research Question 3 Is there evidence that suggests that using a multifactor approach to portfolio construction offers higher risk-adjusted returns? The results of combined factors covered thus far have only looked at the absolute performance of the portfolios ranked according to each combined factor. In investigating research question 3 we analysed both the absolute return and the volatility of each of the portfolios represented by the standard deviation of each portfolio. Table 8 Table 9 summarise the results of all factors tested both as a long strategy and a long/short strategy. The results include the absolute returns (both geometric and arithmetic) as well as the standard deviations and information ratios. Figure 16Figure 17 Figure 18 depicts the results graphically. Table 8, and Figure 16 and Figure 17 summarise the results of a long strategy on all single-factors and combined factors tested. The results indicate that portfolios ranked according to 12-month price momentum achieved the highest annual return and highest excess return of 35,49% and 15,14% Copyright respectively. Despite being the highest performing factor overall UCT based on overall return, the 12-month momentum strategy was not the best performing factor based on information ratio or return per unit risk. Portfolios ranked according to ROIC had the highest return per unit risk of 1,72 as well as the highest information ratio (of all single-factors) of 1,58. These results indicate that although the highest-ranking ROIC portfolio did not achieve the highest overall return, it did produce the highest return per unit risk and was the most consistent at outperforming the equal weighted benchmark. Single-factor strategies, ROE and 12-month momentum followed closely with information ratios of 1,43 and 1,40 respectively. In Table 8 the results of overall return and information ratios are tabulated and ranked according to information ratios. The results indicate that combined factors tested using a long strategy show a marked improvement in information ratios over their constituent factors. For example, 12-month price momentum combined with earnings yield achieved an information ratio of 1,5, which is an improvement over both earnings yield and momentum, while a combined factor consisting of ROE and EY produced an information ratio of 1,50, which is a fair improvement over both ROE and EY of 1,43 and 1,29 respectively. The portfolio formed using 12-month price momentum and ROE had an information ratio of 1,57, which is a significant increase over its constituent factors, 12-month momentum and ROE, which had information ratios of 1,40 and 1,43 respectively. Lastly, the results indicate that the combination of three factors; 12-month momentum, EY and ROE Page-42

50 achieved the highest information ratio of all factors tested. Although the 3-factor combination did not achieve the highest return per unit risk, it was the most consistent at outperforming the benchmark of all factors tested using a long strategy. Table 9 tabulates the results for all long/short strategies ranked according to return per unit risk while Figure 18 graphically displays the data. The results show a similar pattern to the factors tested using a long strategy. The quality factor, ROE, was the highest performing single-factor, evidenced by the highest overall return and the highest return per unit risk. The 2 factor combinations of value + momentum, quality + value and momentum + quality produced a return per unit risk between each of their constituent factors while a 3-factor combination of value, quality and momentum long/short strategy produced the highest return per unit risk of all long/short strategies. In general the return per unit risk of 2,18 was the highest recorded return per unit risk of all factors tested over the 20-year period, both as long and long/short strategy. Table 8: Long Strategy Summary of Return vs Risk for all factors tested Average Average Standard ExcessReturn Return/ Information Return Annual Deviation R(P1JEW) Risk ratio (Geometric) Return 12MPM + ROE + EY 32,23% 34,67% 19,47% 12,18% 1,66 1,62 ROIC 33,65% 36,12% 19,59% 13,27% 1,72 1,58 12MPM + ROE 32,92% 35,37% 19,51% 12,70% 1,69 1,57 12MPM + EY 35,41% 34,80% 21,11% 14,99% 1,68 1,50 ROE + EY 32,34% 38,42% 19,60% 12,27% 1,65 1,50 ROE 31,45% 33,95% 19,78% 11,62% 1,59 1,43 12(m) Momentum 35,49% 38,62% 21,54% 15,14% 1,65 1,40 EY 29,18% 31,11% 17,41% 9,43% 1,68 1,29 NP 28,50% 30,59% 18,23% 9,02% 1,56 1,17 6 (m) Momentum 29,80% 32,53% 20,52% 10,53% 1,45 0,95 Book-to-Market 24,99% 27,64% 21,03% 6,69% 1,19 0,58 DY 22,76% 24,61% 17,53% 4,24% 1,30 0,50 Equally Weighted EW 17,83% 19,52% 16,86% - 1,06 - Page-43

51 Table 9: Long/Short Strategy Summary of Return vs Risk for all factors tested Average Return (Geometric) Average Annual Return Standard Deviation Return/Risk 12MPM + ROE + EY 35,32% 37,00% 16,17% 2,18 NP 27,47% 28,78% 14,59% 1,88 ROE 32,19% 34,18% 17,67% 1,82 12MPM + ROE 31,23% 33,15% 17,42% 1,79 ROIC 29,15% 30,86% 16,59% 1,76 ROE + EY 28,80% 30,70% 17,44% 1,65 12(m) Momentum 31,74% 34,29% 19,93% 1,59 12MPM + EY 25,06% 28,16% 22,40% 1,12 Equally Weighted EW 17,83% 19,52% 16,86% 1,06 6 (m) Momentum 19,89% 22,13% 19,58% 1,02 EY 12,11% 13,51% 15,92% 0,76 Book-to-Market 8,67% 10,44% 18,22% 0,48 DY 6,44% 8,33% 18,99% 0,34 Page-44

52 Figure 16: Return and Risk for all factors tested using a Long Strategy Page-45

53 Figure 17: Long Strategy Excess return and Information ratio (IR) Page-46

54 Figure 18: Long/Short Strategy Annual return vs risk Page-47

55 4.3.7 CAPM Risk adjustment using linear regression In addition to the return per unit risk and the information ratios, the study also investigated the risk-adjusted returns using a CAPM model with a linear regression against the equally weighted benchmark. Although our data does not conform to the requirements of linear regression, which assumes normality, many other papers (Fama et al. 1993, Van Rensburg et al and Basiewicz et al. 2010) perform regressions on financial data and thus we used the regression analysis as a means of comparison. Table 10 and Figure 19 summarise the results of the linear regression with the equally weighted benchmark. Table 10 tabulates the alpha and beta values of linear regression as well as the standard error and R 2 values. The results indicate that all strategies tested using a long strategy produced significant excess returns (at the 95% confidence level), evidenced by the t-stat values greater than 2,25. The beta values are comparable across all factors with no significant differences. 12-month price momentum has the highest risk-adjusted return, which is illustrated by the highest alpha value followed by portfolios formed using 12-month price momentum and earnings Copyright yield. The other combined factors show little improvement UCT in alpha from constituent factors. The P<0 value indicates the probability of excess returns being less than zero. These values highlight the benefit of diversified exposure using combined factors evidenced by the lower P<0 values of combined factors compared to single-factors. Further support for a combined factor approach is highlighted by the lower standard error values of combined factors. Table 10: Results of regression against equally weighted benchmark AverageExcess ReturnR(P13 EW) Beta β Alphaα (monthly) Alphaα (annual) Standard Error R 2 t3stat P<0 12MPM 15,14%& 1,03& 1,21%& 14,52%& 3,65%& 0,66& 3,80& 11,54%& 12MPM+EY 14,99%& 1,04& 1,19%& 14,28%& 3,38%& 0,69& 5,18& 10,04%& ROIC 13,27%& 1,00& 1,11%& 13,32%& 2,90%& 0,74& 5,65& 9,29%& 12MPM+ROE 12,70%& 1,01& 1,05%& 12,60%& 2,77%& 0,76& 5,58& 9,27%& 12MPM+ROE+EY 12,18%& 1,03& 0,97%& 11,64%& 2,56%& 0,79& 5,61& 8,49%& EY 9,43%& 0,90& 0,94%& 11,28%& 2,50%& 0,75& 4,89& 14,20%& ROE 11,62%& 1,03& 0,93%& 11,16%& 2,75%& 0,77& 4,97& 11,13%& 6MPM 10,53%& 0,96& 0,93%& 11,16%& 3,63%& 0,63& 5,65& 20,08%& NP 9,02%& 0,94& 0,84%& 10,08%& 2,61%& 0,75& 4,77& 16,05%& DY 4,24%& 0,87& 0,55%& 6,60%& 2,80%& 0,70& 2,92& 33,48%& Book3to3Market 6,69%& 1,01& 0,54%& 6,48%& 3,58%& 0,65& 2,25& 29,44%& Page-48

56 16,00% 14,00% 33,48% Alpha0and0Standard0error0for0factors0tested0using0long0strategy 1995T ,00% 35,00% 12,00% 29,44% 30,00% Annual0 Alpha 10,00% 25,00% 8,00% 20,08% 20,00% 16,05% 6,00% 15,00% 14,20% 11,54% 11,13% 10,04% 4,00% 9,27% 10,00% 3,58% 9,29% 8,49% 3,65% 3,63% 2,80% 2,50% 2,90% 2,61% 3,38% 2,75% 2,77% 2,56% 2,00% 5,00% Standard0error/0P<0 0,00% Book0To0 Market0 DY EY 12(m)0 Momentum 60(m)0 Momentum ROIC ROE NP 120(m)0 120(m)0 120(m)0 Momentum0+0 Momentum0+0 Momentum0+0 EY ROE ROE0+0EY 0,00% Alpha0 α0(annual) Standard0Error0 P<0 Figure 19: Alpha and standard error for factors tested using long strategy Page-49

57 month excess rolling returns Figure 20 graphically depicts the 12-month excess returns of the best performing single factors against the equally weighted benchmark. The graph indicates the consistency at which each factor outperformed the benchmark. Positive excess returns indicate periods of outperformance while negative returns indicate period of underperformance. Momentum factors represented by 12-month price momentum show exceptional performance during growth periods, evidenced by high positive excess returns, and poor performance during periods of poor economic growth shown by the high negative excess returns during the 2000 and 2008 crashes. The graph for 12- month momentum emphasises the high-risk nature of momentum strategies, which is consistent with the high standard deviation, reported in Figure 17. Quality factors represented by ROIC and ROE show exceptional performance during period of high growth however they also show resilience during economic crashes evidenced by lower negative excess returns compared with other single factors. The results re-iterate the superior performance of quality factors on the JSE. Value factors represented by EY also show positive excess returns during periods of growth but not as high as either momentum or quality. However, during periods of poor economic growth portfolios selected using EY crashed shown by extremely poor performance during the 2008 crash as well as during the last year. The results during the last year are consistent with findings by Cairns (2015) who highlights the poor performance of value portfolios during the last year citing investments into resource stocks as the reason for the poor performance of value asset managers. Results on combined factors shown in Figure 21. The results indicate that portfolios formed using momentum and value have similar performance to 12-month momentum with high excess returns during periods of growth as a well as high negative excess returns during periods of low economic growth. During the crash of 2000 and 2008 portfolios formed using momentum and value experienced extremely high negative excess returns, lower than any of the other combined factor portfolios. The results on the other combined factor portfolios show similar performance over the 19-year period with high positive excess returns during growth periods and negative excess returns during periods of low economic growth. Page-50

58 Figure 20: Single factors 12-month excess rolling returns vs the equally weighted benchmark Page-51

59 Figure 21: Combined factors: 12-month excess rolling retrun vs the equally weighted benchmark Page-52

60 year vs 20-year comparison Table 11: 10-Year vs 20-Year comparison Average Return (Geometric) Standard Deviation Return /Risk IR Alphaα (annual) 1995A2015(20Ayearperiod) 12(m)Momentum+ROE+EY 32,23%% 19,47%% 1,66% 1,62% 11,68%% ROIC 33,65%% 19,59%% 1,72% 1,58% 13,31%% 12(m)Momentum+ROE 32,92%% 19,51%% 1,69% 1,57% 12,57%% EY+ROE 32,34%% 19,60%% 1,65% 1,50% 12,08%% 12(m)Momentum+EY 35,41%% 21,11%% 1,68% 1,50% 14,23%% ROE 31,45%% 19,78%% 1,59% 1,43% 11,12%% 12(m)Momentum 35,49%% 21,54%% 1,65% 1,40% 14,52%% EY 29,18%% 17,41%% 1,68% 1,29% 11,29%% 6(m)Momentum 29,80%% 20,52%% 1,45% 0,95% 11,20%% BookAtoAMarket Copyright 24,99%% 21,03%% 1,19% UCT 0,58% 6,54%% DY 22,76%% 17,53%% 1,30% 0,50% 6,63%% % % % % % 2005A2015(10Ayearperiod) % % % % % ROIC 30,18%% 17,09%% 1,77% 1,54% 10,56%% 12(m)Momentum+ROE+EY 28,05%% 16,88%% 1,66% 1,43% 8,63%% 12(m)Momentum+ROE 25,78%% 16,85%% 1,53% 1,15% 6,74%% EY+ROE 25,44%% 16,97%% 1,50% 1,09% 6,37%% ROE 25,11%% 16,69%% 1,50% 1,07% 6,41%% 12(m)Momentum 27,81%% 19,90%% 1,40% 0,96% 7,08%% 12(m)Momentum+EY 25,94%% 19,06%% 1,36% 0,93% 5,47%% 6(m)Momentum 26,15%% 18,74%% 1,40% 0,80% 7,77%% DY 21,19%% 15,31%% 1,38% 0,49% 5,48%% EY 19,33%% 16,55%% 1,17% 0,34% 1,20%% BookAtoAMarket 18,92%% 15,58%% 1,21% 0,27% 2,50%% Table 11 shows the results of the 10-year period from 2005 to 2015 vs the 20-year period from 1995 to 2015 ranked according to information ratios (IR) for each period. The overall returns for the 10-year period are lower than the 20-year period and are less volatile, as indicated by the lower standard deviations. The rank according to IR, although not identical, is similar in both periods. ROIC and the 3-factor combination of momentum value and quality are the best Page-53

61 performing portfolios based on IR, followed by the 2-factor combinations of momentum + value and value + quality. Six-month momentum and value factors were the worst performing factors based on IR in both periods. Once again, this data suggests a strong momentum and quality premium on the JSE for both periods investigated. Furthermore, the results indicate the superior performance of combined factor strategies during both period based on IR values. 4.4 Limitations of the research The purpose of this study was to investigate the performance of several factors on the JSE over a 20-year period from 1995 to 2015, and to identify whether adopting a combined factor approach to portfolio construction produces superior risk-adjusted returns. The study has several limitations, which are listed below: The research is limited to the Johannesburg Securities Exchange (JSE) and cannot be used to draw conclusions on other markets. Other markets might behave differently and produce different results. Due to the availability of data this study only considered eight factors grouped under three styles: value, momentum and quality. There are several other styles that have not been covered such as mean reversion, cash flow to price, gross margin etc. In addition, factors that look at forward-looking data based on analysts reports have not been considered, as they were not available on the databases used in this study. The study only considered a one-month holding period, which may be bias toward factors that perform well with shorter holding periods. For example, momentum factors tend to work well with shorter holding periods, while value factors tend to perform better over slightly longer holding periods. Therefore this study might be considered biased toward momentum factors. Transaction costs were ignored in this study. In reality transaction costs are related to the rate of turnover of the portfolio, or the frequency at which securities are bought and sold. Portfolios with a high turnover rate generally attract higher fees while those with a low turnover have lower fees. Momentum portfolios tend to have higher turnover rates compared either to the value or quality factors, and thus attract higher fees. These fees, related to transaction costs and capital gains tax, must be considered. We performed a brief study that looked at the top-ranking portfolio of each of the factors tested, and calculated the percentage change of each monthly portfolio from the preceding month. The results, Page-54

62 shown in Figure 22, indicate that quality factors tend to have the lowest rate of change or the lowest turnover, followed by value factors. Momentum factors on the other hand have the highest turnover rate and are susceptible to higher transaction costs. Interesting to note is that adopting a combined factor strategy tends to reduce the turnover rate. An in-depth study into the effects of fees should be done if any of these trading strategies are to be implemented. Figure 22: Graph comparing the average percentage monthly change of the top-ranking portfolio Page-55