University of Amsterdam
Revisiting the return to value and its potential drivers on the ASEAN Exchanges
Master Thesis By John Vuist
10363610
Amsterdam Business School
Submitted in partial fulfillment of the requirements for the degree of
Master of Science in Business Economics, Finance Track
Date 16 June 2016 Thesis Supervisor:
DECLARATION OF OWN WORK
I, John Vuist hereby certify that the submitted thesis is the product of my own work. If information of a third party is used, cited or referred to it is acknowledged in the text and its references.
The responsibility of the Faculty of Economics and Business (FEB) is limited to the supervision and completion of this thesis, which does not include its contents. The full responsibility of the contents of this thesis remains mine and is acknowledged.
Furthermore supervision of this thesis by Dr. Liang Zou is acknowledged and hereby I would like to express my gratitude for his supervision.
John Vuist Amsterdam, 16 June 2016
ABSTRACT
By employing stock market data from five ASEAN Exchanges for the period 2004-2014, this study aims to shed new light on the ongoing debate of what could be potential drivers of the return to value. There are a large number of published studies which provided evidence that portfolios formed on value stocks systematically outperform portfolios formed on growth stocks. Previous empirical studies have shown that models such as the CAPM (Sharpe, 1964) and the Fama & French three-factor model (1993) which generally are able to capture the cross-section of average stock returns, are generally unable to capture the return to value. In light of the recent publication of Fama & French (2015), who introduced a five-factor model which is presumed to improve the explanation cross-section of average stock returns compared to the aforementioned asset pricing models, it is tested whether the five-factor model improves the explanation of the return to value. A statistical significant outperformance of value portfolios is observed on the Indonesian and the Malaysian stock exchanges, whereas evidence that the five-factor model improves its explanation is inconclusive. Moreover no convincing evidence is found that the five-factor model improves the explanatory power of the three-factor model and the augmented factors of the five-factor model are not found to have any explanatory power regarding the return to value and growth portfolios.
Keywords:
JEL classifications: G12
TABLE OF CONTENTS
Chapter Page
DECLARATION OF OWN WORK ... ii
ABSTRACT ... iii
TABLE OF CONTENTS ... iv
LIST OF TABLES ... vi
LIST OF EQUATIONS ... vii
CHAPTER I: Introduction ... 1
1.1 ASEAN Exchanges ... 2
1.2 Research Aims ... 2
1.3 Research Relevance ... 3
1.4 Data & Methodology ... 4
1.5 Outline of the Thesis ... 4
CHAPTER II: Literature Review and Background ... 5
2.1 The Value Anomaly ... 5
2.2 Explanations concerning the value anomaly... 5
2.2.1 Risk Considerations ... 6
2.2.2 Behavioural Considerations ... 7
2.2.3 Occurrence due to methodological flaws? ... 8
2.2.4. Relation with other characteristics ... 8
2.3 The return to value stocks on the ASEAN Exchanges ... 10
CHAPTER III: Methodology ... 12
3.1 Formation of Value and Growth Portfolios ... 12
3.2 Testing for a value premium ... 13
3.3 Regressions ... 13
3.4 Factors & Variables ... 14
3.4.1 Factor Calculation ... 15
3.5 Evaluation of the regressions ... 19
CHAPTER IV: Data ... 20
4.1 Data collection ... 20
4.1.1 Accounting Data and Identifiers from COMPUSTAT ... 20
4.1.2 Other Data from DataStream ... 20
4.2 Data preparation ... 21
4.3 Summary statistics ... 21
4.4 Limitations of the dataset ... 22
CHAPTER V: MAIN RESULTS ... 23
5.1 The Value Premium on ASEAN Exchanges ... 23
5.2 CAPM Regressions ... 23
5.3 Fama-French three-factor regressions ... 24
5.4 Fama-French five-factor regressions ... 25
5.5 Evaluation of the various regression models ... 27
6.1 Value-weighted value premia ... 28
6.2 Crisis & Post-Crisis Analysis... 29
6.3 Asymptotically valid GRS-test ... 31
CHAPTER VII: Conclusion ... 33
7.1 Conclusion ... 33
7.2 Limitations ... 33
7.3 Suggestions for future research ... 34
REFERENCES ... 35
LIST OF TABLES
Table Page
Table 1: Research Question ... 2
Table 2: Hypotheses ... 3
Table 3: Regression Models... 14
Table 4: Factor and variables description in the regression models ... 14
Table 5: 2X3 sorts of portfolios for the Fama & French three-factor model ... 16
Table 6: Calculation of Fama & French three factors ... 16
Table 7: 2X3 sorts of portfolios, based on size, M/B, OP and INV ... 18
Table 8: Calculation of Fama & French five factors ... 18
Table 9: Amount of firms included in sample for each exchange ... 22
Table 10: Monthly value premia based on equally-weighted portfolios ... 23
Table 11: CAPM Regressions ... 24
Table 12: Fama-French three-factor regressions ... 25
Table 13: Fama-French five-factor regressions ... 26
Table 14: Comparative statistics of asset pricing models ... 27
Table 15: Value-weighted monthly value premia ... 28
Table 16: Monthly value premia during the financial crisis ... 29
Table 17: Post-crisis monthly value premia... 30
Table 18: Fama-French five-factor regression during the financial crisis ... 31
Table 19: Comparative GRS-statistics of the regression models... 32
Table 20: Descriptive statistics of the regression factors in Indonesia ... 38
Table 21: Descriptive statistics of the regression factors in Malaysia ... 38
Table 22: Descriptive statistics of the regression factors in the Philippines... 38
Table 23: Descriptive statistics of the regression factors in Singapore ... 39
Table 24: Descriptive statistics of the regression factors in Thailand ... 39
Table 25: Descriptive statistics of the portfolios ... 40
LIST OF EQUATIONS
Equation
Page
Equation 1: Portfolio Return ... 13
Equation 2: Welch's test statistic ... 13
Equation 3: Fama & French three-factor model ... 15
Equation 4: Fama & French five-factor model ... 16
Equation 5: GRS-test test statistic... 19
CHAPTER I: Introduction
The return to value has been a major area of interest within the field of finance for the last two decades. Studies over this period have provided ubiquitous evidence for the systematic
outperformance of value portfolios across the globe. Nevertheless, hitherto there has been little agreement on what drives this outperformance. The systematic outperformance is generally known as the value anomaly or value premium, which customarily is referred to as a larger yield of value over growth (glamour) stocks. This systematic outperformance has led to a major interest by investors who aim to exploit this anomaly by pursuing value investment strategies. Most commonly value stocks are referred to as stocks with high book-to-market ratios (B/M) while growth stocks have opposite characteristics (e.g. Fama & French, 1993). The literature suggests that the explanations regarding the presence of the value anomaly can be broadly classified in four categories. Firstly, related to the renowned CAPM (Capital Asset Pricing Model) by Sharpe (1964) a value anomaly could be driven by the notion of a risk premium. This notion is supported by scholars such as Fama & French (1993) who claim that value firms are by definition riskier and therefore require a higher yield.
Secondly there is a group of scholars who argue that the value anomaly is driven by behavioural biases e.g. (DeBondt & Thaler (1985) and Lakonishok, Shleifer & Vishny (1994)). The proponents of this concept believe that due to the occurrence of various behavioural biases which leads to mispricing of value stocks, will eventually result in the occurrence of the value anomaly. The third perspective is the perspective of Kothari, Shanken & Sloan (1995) who argue that the value anomaly is a spurious finding due to survivorship bias in various databases. Last but not least are the perspectives of scholars who posit that the value anomaly could be driven by other characteristics than the aforementioned.
In light of the recent publication by Fama & French (2015), who developed a five-factor model directed at capturing the size, value, profitability and investment patterns in average stock returns, this thesis aims to test whether these potential drivers can capture the return to value by employing their five-factor model. Previous studies on the ASEAN Exchanges have failed to consider the potential impact of flaws in their datasets. In this paper, I aim to control for the potential flaws in the dataset. Moreover there is a relative paucity of research in the
subject concerning the return to value on the ASEAN Exchanges and previous research show varied findings.
1.1 ASEAN Exchanges
The ASEAN Exchanges is a collective of stock exchanges from six countries of the Association of Southeast Asian Nations (ASEAN). The national stock exchanges of Thailand (SET), Malaysia (BM), Singapore (SGX), Indonesia (BEI), Philippines (PSE) and Vietnam (HNX & HOSE) are
member of this collective. This collaboration aims to promote financial market integration in the ASEAN countries (The Stock Exchange of Thailand, n.d.). Despite its name would suggest, the ASEAN Exchanges, it is not a pan-ASEAN stock exchange as Euronext is in Europe, but rather a cross-ASEAN trading infrastructure to which each individual ASEAN Exchange is connected (Hsu & Kien, 2015). As of 2013, the ASEAN Exchanges have a joint market capitalization of more than three trillion USD, and over 3000 listed companies. (ASEAN Exchanges, n.d.)
1.2 Research Aims
This thesis aims to revisit the return to value on the ASEAN Exchanges and to examine whether it can be explained by risk-based drivers conform conventional asset pricing models. To do so, the research question in table one below is examined.
1. Can the return to value be explained by risk-based drivers conform conventional asset pricing models?
Table 1: Research Question
The first hypothesis is formed to test whether there is a presence of a value premium on any of the ASEAN Exchanges. In order to answer the research question, the latter hypotheses are formed. First, it will be evaluated whether the return to value is driven by risk exposures to the market, which follows from the CAPM-paradigm (Sharpe, 1964). From this assertion the second hypothesis is derived. The third hypothesis is formed based on the value and size patterns in average stock returns which have been proposed by Fama & French (1993). The fourth
hypothesis is formed based on the five-factor model by Fama & French (2015). Accordingly the last hypothesis is formed in accordance with the finding of Fama & French (2015) who found that their five-factor model outperforms their three-factor model (Fama & French, 1993). On its
turn the finding of Fama & French (1993) was that the three-factor model outperforms the CAPM (Sharpe, 1964). The hypotheses can be found in table two below.
1. A value premium is existent on an ASEAN Exchange.
2. The return to value can be explained by risk exposure to the market in average stock returns.
3. The return to value can be explained by risk exposure to the market and size and value patterns in average stock returns.
4. The return to value can be explained by risk exposures to the market and size, value, profitability and investment patterns in average stock returns.
5. The return to value is best explained by risk exposures to the market and size, value, profitability and investment patterns in average stock returns.
Table 2: Hypotheses
1.3 Research Relevance
As stated in the introduction, the ASEAN Exchanges are chosen as a field of study as most of the ASEAN exchanges have not been extensively studied around the return to value. In addition to this, Brown, Rhee & Zhang (2008) noted that there are varied findings on the presence of the value anomaly in East Asia and the ASEAN region. Surprisingly, much of the literature regarding the return to value in the ASEAN region cover periods before 1997, which is a period prior to the Asian Financial Crisis. Altogether this highlights the relevance to revisit the return to value on the ASEAN Exchanges, as this study covers the period 1.1.2004-1.1.2015.
Furthermore to my best knowledge, this thesis will provide the first attempt to explain the return to value by empirical testing which considers the recent Fama & French five-factor model (2015). Phalippou (2007) emphasizes the importance to test how asset pricing models perform in explaining the return to value, as asset pricing models are generally found to be unsatisfactory in doing so. The recent development of the Fama & French five-factor model (2015) highlights the relevance of conducting such a study.
Summarized the main contributions of this thesis to the literature are as follows. First, it re-examines whether a value premium is present on the ASEAN Exchanges, which is a relatively understudied area around this theme. Secondly, this thesis provides an empirical test of the Fama & French five-factor model (2015) in explaining the return to value. Thirdly, it is relatively
unexplored how asset pricing models work on explaining the return to value on emerging markets, and therefore the ASEAN Exchanges employs an interesting field of study. 1.4 Data & Methodology
In this study data on five ASEAN Exchanges which has been retrieved from DataStream and COMPUSTAT is used. The Vietnamese exchanges have been excluded from this study due to inadequate data availability. For each of the five remaining exchanges value and growth
portfolios are composed and annually rebalanced. Subsequently each exchange is tested for the presence of a value premium. Finally regressions based on the CAPM (Sharpe, 1964), Fama & French three-factor model (1993) and the Fama & French five-factor model (2015) are run on the aforementioned portfolios and the results of the regressions are evaluated.
1.5 Outline of the Thesis
This thesis is organized as follows. This chapter, chapter one was its introduction. The next chapter, chapter two will provide a literature review and a synopsis of findings from previous studies. In chapter three the methodology of this study will be covered. Chapter four will discuss the details on the data used in this study. Chapter five will provide the results of this study. Chapter six will cover additional robustness tests. To finish with chapter seven will provide a synopsis of the results of this study and discusses its limitations this study and
CHAPTER II: Literature Review and Background
This chapter aims to present a brief synopsis of the literature around the value anomaly. 2.1 The Value Anomaly
As stated in the introduction, the value anomaly is a stock return anomaly which is commonly referred to as of a systematic outperformance of value over growth (glamour) stocks. The value anomaly is also referred to as the value premium, book-to-market anomaly and various other terms. Commonly value stocks are defined as stocks with high book-to-market (B/M) ratios while growth stocks have opposite characteristics (e.g. Fama & French, 1993). In identifying value from growth stocks the B/M-ratio alongside plentiful other multiples such as earnings-to-price (E/P), cash flow-to-earnings-to-price (C/P) or dividend-to-earnings-to-price (D/P) are being used, along with their opposites (e.g. Fama & French, 1993; Lakonishok, Shleifer & Vishny, 1994; Bauman, Conover & Miller, 1998). The breakpoint considered in identifying these stocks is generally a breakpoint of the 30th percentile which means that value stocks lie within the 70th percentile of the B/M-ratio while growth stocks are those within the 30th percentile (e.g. Fama & French, 1993). Many investors engage in stock-picking based on the aforementioned multiples as they perceive that it holds information whether a firm is under-priced or over-priced relative to its fundamentals (Brown, Rhee & Zhang, 2008). In 1970 Fama claimed that all stock prices on any occasion instantly fully reflect any information developments due to the immediate process of arbitrage following any information developments. Given this notion which he refers to as the EMH (Efficient Market Hypothesis), it would be futile to engage in stock-picking activities as the value anomaly could not be existent. Nevertheless a considerable amount of studies have confirmed its existence and countless investors pursue value investing. Yet, there has been much division between scholars on what actually drives the value anomaly as it has attracted conflicting explanations, which can be broadly classified within four main categories that will be covered in the next paragraph.
2.2 Explanations concerning the value anomaly
As aforementioned in the introduction and in the previous paragraph, the literature suggests that there are roughly four possible explanations concerning the presence of the value anomaly.
Firstly I discuss the risk consideration, which has been put forward by scholars such as Fama & French (1993) who are advocates of the EMH, and therefore believe that only a rational explanation such a risk-premium could explain this phenomenon. The risk considerations will be covered in section 2.2.1. The second most common consideration with respect to the value anomaly are the behavioural considerations which have been asserted by scholars such as Lakonishok, Shleifer & Vishny (1994). The behavioural-based considerations will be covered in section 2.2.2. The third covered explanation is the perspective of Kothari, Shanken & Sloan (1995) who claim that the value anomaly occurs due to survivorship biases in various databases which will be covered in section 2.2.3. To finish with I cover the assertion by scholars who posit that the value anomaly occurs due to other firm characteristics than the aforementioned; those will be covered in section 2.2.4.
2.2.1 Risk Considerations
To begin with this thesis will review the risk considerations around the theme of the value anomaly. As stated in the introduction, the advocates of the risk considerations are generally also scholars who are advocates of the EMH. As earlier covered in the introduction Fama & French (1993) put forward that it is not an anomaly, but a rational occurrence of a risk premium as they claim that value stocks are in essence riskier. Their perspective is derived from the assertion that a high book-to-market (B/M) ratio is associated with relative financial distress. As a value stock is defined as a stock which lies within the 70th percentile of the B/M ratio, it instinctively follows that a value stock has a higher perceived risk compared to a growth stock. Findings of Chen & Zhang (1998) do suggest that value firms are supposedly more likely to be under financial distress, nevertheless they dispute that the B/M ratio is at all time a reliable proxy for financial distress, especially in emerging markets where during high-growth periods the dispersion of risk amidst value and growth firms is small. Hence they contest the
consideration that the risk-pricing argument will hold for emerging high-growth markets. Other scholars who dispute the risk-pricing argument by Fama & French (1993) are Gomes, Kogan & Zhang (2003) and Zhang (2005). They claim that growth stocks derive their market price from growth opportunities. Likewise, they suggest that the market price of value stocks is inferred from assets in place. Zhang (2005) posits that due to this consideration value firms are riskier but he disputes that this would lead to a risk-pricing argument in favour of the value anomaly. In lieu of a risk-pricing argument, Zhang (2005) predicts that the higher risk of value firms will be
exposed during periods of economic decline, as he posits that during such periods a negative value premium will be realized.
2.2.2 Behavioural Considerations
In addition to the risk considerations of the value anomaly as discussed in the previous section, the behavioural considerations are also omnipresent in the literature. Behavioural
considerations on this subject can be defined as the occurrence of various behavioural biases which lead to mispricing. This mispricing will eventually lead to the occurrence of the value anomaly. Scholars who are advocates of the behavioural considerations are generally no proponents of Fama’s EMH (1970) as they believe in a mispricing argument which contests the concept of the EMH.
Early work by DeBondt & Thaler (1985) supports the association between the value anomaly and the concept of overreaction bias. Overreaction bias refers to the notion that both investors and traders overstate information developments which leads to mispricing. In the setting of the value anomaly, it occurs due to the perception that multiples such as the B/M hold information on the fundamental value of a stock to which investors overreact which in the end leads to mispricing. Given the notion of the aforementioned EMH by Fama (1970) in section 2.1, such mispricing is ought not to be long-lasting as it should immediately vanish through the arbitrage mechanism and should not result in the occurrence of the value anomaly.
Nevertheless empirical findings from e.g. LaPorta (1996), Yen, Sun & Yan (2004), show that such mispricing lasts for several years which are found to be inconsistent with the EMH. Therefore Phalippou (2008) asserts that it is inevitable that arbitrage is costly as otherwise the mispricing would vanish by means of arbitrage following the notion of the EMH.
Another common behavioural consideration is the one by Lakonishok, Shleifer & Vishny (1994) who posit that investors by some means have a natural incline towards investing in growth firms. Based on the aforementioned behavioural bias it follows that value stocks are neglected and as a result undervalued. Lakonishok, Shleifer & Vishny (1994) argue that if the stock prices revert back to their fundamentals, a value premium is ought to occur. Furthermore they mention that the preference for growth firms is more likely to be stronger among
individual investors rather than by institutional investors. Phalippou (2008) had confirmed their reasoning by finding evidence that the value anomaly is driven by the manifestation of
to this, findings of Ackert, Athanassakos & Church (2015) suggest that the personality of an individual investor significantly affect its investment style. They found that an individual’s personality significantly determines whether it prefers to invest in value stocks or in growth stocks. Given that the value anomaly is driven by the mispricing argument, it seems that the position of Lakonishok, Shleifer & Vishny (1994) holds and that the value anomaly is caused by means of a behavioural bias which causes investors to incline towards investing in value firms. 2.2.3 Occurrence due to methodological flaws?
Kothari, Shanken & Sloan (1995) argue that the results of e.g. Fama & French (1993) on the occurrence of the value anomaly are a manifestation of survivorship bias in the COMPUSTAT database. Therefore they contest the value anomaly is a stylized fact; rather they posit that it is sample specific due to the consideration that many databases may have been affected by backfilling. Kothari, Shanken & Sloan (1995) argue that firms who survived a period of financial distress are more likely to be backfilled than firms that did not. However findings of Fama & French (1998) who studied the return of value stocks in an international setting contradict the claims by Kothari, Shanken & Sloan (1995). Fama & French (1998) who conducted their study using data from MSCI (Morgan Stanley’s Capital International Perspectives), which is free from backfilling and thus less prone to survivorship bias. Hence it could be concluded that Fama & French (1998) their results suggest that the value anomaly is not occurring due to survivorship bias. Furthermore, e.g. De Groot & Verschoor (2002) and Kouwenberg & Salomons (2005) remarked that as numerous international studies have confirmed the presence of a value anomaly, its existence is well-established. Nevertheless, it is vital to keep the ideas of Kothari, Shanken & Sloan (1995) in mind for this study in order to recognize potential survivorship bias. 2.2.4. Relation with other characteristics
Fama & French (2015) postulate that there is abundant evidence that characteristics such as size, investments and profitability are related with stock returns. They posit that the
aforementioned characteristics are related with risk. Fama & French (2015) claim that their five-factor model aimed at capturing size, value, profitability and investment patterns in average stock returns performs better than their well-known three factor model (Fama & French, 1993). Fama & French (2015), ascribe their findings to the consideration that their three-factor model
is missing some explanatory factors. Considering Fama & French (2015) their recent work, I revisit findings by other scholars regarding to stock returns and the value anomaly.
First I concisely revisit the relationship of firm size and the value anomaly. To my knowledge, Banz (1981) was the first to confirm the existence of the size anomaly. The size anomaly is defined as an outperformance of stocks of firms with small market capitalization (small stocks) over firms with big market capitalization (big stocks). Generally it is presumed that stocks with small market capitalization are riskier due to the higher cost of capital that such firms face. Hence I posit that the value anomaly could be inversely related with firm size. Athanassakos (2011) agrees that the value anomaly is inversely related with firm size; however he disputes that it is a risk consideration. Instead he posits that the value anomaly is related with the mispricing argument and that firm size affects its magnitude. According to
Athanassakos (2011) value firms with small market capitalizations tend to be more under-priced than value firms with large market capitalizations which will lead that the value premium will be higher among small firms compared with large firms. A claim related to this assertion is one of Loughran (1997), who claims that the presence of the value anomaly is restricted to small stocks. Nevertheless findings of Fama & French (2006) confirm that a value premium is evident in both small and big stocks. Moreover, Phalippou (2007) disputes that size has a significant explanatory power in explaining the value anomaly as it is pervasive among both small and large stocks.
Secondly I briefly revisit the relationship of investment patterns and the value anomaly. Findings of Titman, Wei & Xie (2004) suggest that there is a negative relationship between investments and stock returns. In other words this means that firms with lower investments generally have higher stock returns and vice versa. Considering the claim by Zhang (2005), who claims that value firms tend to be riskier and have a tendency to invest less than growth firms, it is conceivable that the value anomaly can be (partially) ascribed to firms’ investment patterns. Findings of Fama & French (2015) seem to support this assertion.
Thirdly I shortly revisit the relationship of profitability and the value anomaly. Novy-Marx (2013) found that gross profitability has a similar power as value patterns in capturing average stock returns. Similarly as with the value patterns, gross profitability could be a proxy for relative distress, as it naturally follows that a firm with a lower profitability are ought more susceptible to distress. Furthermore findings from Novy-Marx (2013) suggest that the value anomaly is related to profitability as he puts forward that value firms on average generate
higher profits than growth firms. From this assertion it is followed that profitability is related to the value anomaly. Likewise to the relationship with investments, findings of Fama & French (2015) seem to support the aforementioned assertion.
2.3 The return to value stocks on the ASEAN Exchanges
In this section previous research around the return to value in the ASEAN exchanges is covered. Compared with other stock markets in the world, the ASEAN exchanges are relatively
understudied in this subject.
Following alphabetical order of the exchanges, I start with reviewing previous research concerning the return to value in Indonesia. Roll (1995) who studied the period 1985-1992 found a value premium. The study of Amanda & Husodo (2014) covered the years 2003-2013, whereas the study of Utomo and Tjandra (2015) covered 1994-2014, both studies found a value premium. Ding, Chua & Fetherston (2005) also observed a value premium for the period 1976-1997 but their findings were not statistically significant.
Secondly I cover previous research around the return to value in Malaysia. Fama & French (1998) did not observe a statistically significant value premium on the Malaysian stock exchange (BM) in the period 1987-1995. Ding, Chua & Fetherston (2005) who studied the period 1976-1997 did find a statistically significant value premium. Likewise, Shah et al. (2012)
observed value premia on the BM. According to Shah et al. (2012) on the BM it does not seem that small firms generate higher returns as generally is ought to be found. Their findings show that big firms generate higher returns over small firms. When controlling for the size-effect they found that big value firms have a higher return over small value firms. The explanation they put forward is that the big and successful firms in Malaysia are either state-owned enterprises or they have a political connection. They argue that such firms get special privileges from the government. These special privileges could entail monopolies, special licenses and other favourable conditions (Shah et al., 2012). In addition to this, findings of Faccio (2007), who studied the characteristics of political connected firms, suggest that in Malaysia value firms are more likely to be either state-owned firms or political connected firms. Therefore it is
conceivable that the value premium in Malaysia is related with state-owned firms or political connected firms. Furthermore it should be noted that the many of the big constituents of the
BM are partially owned by Khazanah Nasional1, which is the state-owned investment fund of the Malaysian government.
Thirdly I cover the Philippine stock exchange. On the Philippine stock exchange (PSE) statistically significant value premia have been observed by Fama & French (1998) and Drew (2003) who respectively studied the periods 1987-1995 and 1993-1999 on the PSE. However the aforementioned scholars did not provide any interpretation on its occurrence.
Fourth I cover the Singaporean stock exchange (SGX). Contrasting other ASEAN Exchanges, the SGX has been extensively studied by a lot of scholars around the theme of the value anomaly. Fama & French (1998), Yen, Sun & Yan (2004), Ding, Chua & Fetherston (2005) and Brown, Rhee & Zhang (2008) have confirmed the occurrence of a value premium on the SGX. Fama & French (1998) studied the period 1975-1995, whereas Yen, Sun & Yan (2004), studied the period 1975-1997. Ding, Chua & Fetherston (2005) covered a similar period as the two aforementioned scholars, namely the period 1976-1997. The study of Brown, Rhee & Zhang (2008) covered a more recent period; to be precise the years 1996-2005. Yen, Sun & Yan (2004) posit that the value anomaly on the SGX is driven by the concept of overreaction bias, which has been covered in section 2.2.2. Similarly as in Malaysia, findings of Brown, Rhee & Zhang (2008) suggest that the value anomaly on the SGX is mainly concentrated among stocks with large market capitalizations.
Last I cover the Thai stock Exchange, which unlike the Singaporean stock exchange has not been extensively documented. De Groot & Verschoor (2002) confirmed the presence of a value premium on the Thai stock exchange for the period 1984-1999. Findings from Ding, Chua & Fetherston (2005) who studied the years 1976-1997 contrast the aforementioned finding as they found a negative value premium or growth premium. Chen & Zhang (1998) did not observe a statistically significant value premium, but they posit that this finding occurs due to different economic features of high growth emerging markets. Therefore they posit that the value anomaly would not be existent in high growth emerging markets. For a more recent period, Sareewiwatthana (2011) who studied the period 1996-2010 found a value premium.
1 Khazanah Nasional (KN) is the Malaysian state-owned investment fund which is active in various sectors in Malaysia. For many big Malaysian firms the KN either has a substantial interest or is a majority shareholder.
CHAPTER III: Methodology
This chapter covers the methodology of this study. 3.1 Formation of Value and Growth Portfolios
In this section it will be discussed how the stocks are sorted to form value and growth portfolios. This study follows an approach adapted from Fama & French (1998) to construct the respective portfolios. In order to sort the stocks to form the two portfolios, the M/B (market-to-book) ratio is used. Employing other multiples would be troublesome as it would lead to a much smaller sample, which could lead to a severe manifestation of sampling bias and impairs the reliability of this study. Furthermore due to the extremely limited availability of data on other multiples, results based on those multiples would not provide any sensible interpretation. Because in this study the M/B ratio is used instead of the B/M ratio as used in the Fama & French studies, it implies that value portfolios consists of stocks within the 30th percentile, while growth portfolios consists of stocks which lie within the 70th percentile. After sorting the stocks, the portfolios are formed on equal weighting which are annually rebalanced. In this study, to form the portfolios all stocks in the dataset are included in the sample, with the exception of
financials2 and firms without available information on either M/B ratio or returns and firms with a negative M/B ratio3. In calculating the returns a half year time-lag is implemented to limit the occurrence of backfill bias4. This means that the portfolios are formed begin July; whereas the returns are calculated begin January. The returns are calculated on a monthly basis from the Return Index (RI)5 which is available on DataStream. In the equation one on the next page the calculation of the portfolio return can be found. After formation, the portfolios will be used in several analyses as described in upcoming sections.
2 This is done in accordance with academic convention, e.g. Fama & French (1998), as some scholars argue that the inclusion of financials would lead to a bias. Other scholars however argue that it would have an effect.
3 This is done as negative M/B ratios do not yield with a sensible interpretation; moreover in many cases negative ratios are not reported, hence to prevent potential biases they are excluded from the sample. This is in line with Lakonishok, Shleifer & Vishny (1994).
4 In studies which consist of back-testing such as this study, it is of great importance that data available at the time of the analysed period is used. Otherwise this could lead to an inaccurate result which is referred to as backfill or look-ahead bias. E.g. Fama & French (1998) recommends a six-month time-lag.
Portfolio Return =RIRIt-RIt-1
t-1
Equation 1: Portfolio Return
3.2 Testing for a value premium
To test for a structural value premium Welch’s t-test is used. This section aims to concisely explain the Welch t-test. The Welch t-test is a parametric statistical test which can be used to test whether there is a difference of means among two populations. The Welch t-test is preferred over the customarily used Student’s t-test because unequal variances are assumed in Welch’s t-test. Hence Welch’s t-test is more reliable. The downside of this test is that normality is assumed, and it could be that the returns of the portfolios follow a non-normal distribution. Nevertheless, the central limit theorem (CLT) posits that when the sample size is at least 30, the means of a sample will approach a normal distribution. In other words, this means that if the sample size is larger than 30, it can be assumed that the population would follow a normal distribution. Hence Welch’s t-test can still be used. In equation two below the calculation of the Welch’s test statistic can be found. X represents the average portfolio return, s² represents the variance of the portfolio returns, and N represents the number of observations, whereas the subscript v or g is an indicator for either a value or growth portfolio.
Welch's test statistic = 𝑋𝑋𝑣𝑣− 𝑋𝑋𝑔𝑔 ��sv2
Nv� + �
sg2
Ng�
Equation 2: Welch's test statistic
3.3 Regressions
In order to evaluate whether the return to value can be explained by risk-based drivers conform conventional asset pricing models three types of regressions are run and evaluated. Following Fama & French (1998) I run time-series regressions in hierarchical order. The regressions are run in three stages. The first stage consists of the single factor CAPM-regression. The second stage will employ the three-factor Fama & French (1993) model. The third stage will employ Fama & French their five-factor model (2015). To err on the side of caution heteroscedastic (HC) standard errors are employed for all the regressions. Table three on the next page provides the
equations of the aforementioned regression models. In table four the variables descriptions of the factors in the regression models are given.
This table shows the equations of the various regression models. A description of the various factors and variables as depicted in this table can be found in table 4.
Regression Model Equation
CAPM �rp�=α𝑖𝑖+bi[(Rm)-rf]
Three-factor model �rp�=α𝑖𝑖+bi[(Rm)-rf]+si[SMBt]+hi[HMLt]+eit
Five-factor model �rp�=α𝑖𝑖+bi[(Rm)-rf]+si[SMBt]+hi[HMLt]+ri[RMWt]+ci[CMAt]+eit
Table 3: Regression Models
This table shows the description of the various factors and variables of the regression models as depicted in table 3.
Factor/Variable Description
�rp� return of a portfolio
rf risk-free return
Rm return on the market portfolio
SMBt size loading factor, the difference between
returns of portfolios with small and big market capitalization
HMLt value loading factor, the difference between
returns of portfolios with high and low B/M ratios RMWt Profitability loading factor, the difference
between returns of portfolios with robust and weak profitability ratios
CMAt Investment loading factor, the difference
between returns of portfolios with conservative and aggressive investments.
𝛼𝛼𝑖𝑖 intercept
bi, si, hi, ri, ci gradient
eit residual
Table 4: Factor and variables description in the regression models
3.4 Factors & Variables
In defining and calculating the factors, Fama & French (1993, 2015) their methodology is employed. Each year, begin July, stocks are sorted and allocated to their respective portfolios. Allocation is done based on four variables, respectively market capitalization (Size), Market-to-book ratio (M/B), Profitability (OP) and Investment (INV).
Similarly as the procedure described in section 3.1 the returns are calculated a half year later, namely in the beginning of January to reduce the potential of backfill bias. Size is defined as the market capitalization. M/B is defined a ratio of the market value of equity to the book value of equity of a firm, which has been readily obtained from DataStream. OP is defined as the annual revenue minus the annual cost of goods sold, interest expenses and selling, general and administrative expenses normalized to the book value of equity. INV is denoted as a change in total assets of year t-1 to year t which is normalized by the total assets at year t-1.
3.4.1 Factor Calculation
As previously stated, Fama & French (1993, 2015) their methodology is employed in calculating the factors. First the calculation of the market portfolio return and the risk-free rate are discussed. For the various exchanges returns on composite market indices are employed as a proxy for market returns. Similarly the returns of three-month commercial papers or treasury bills are taken as a proxy for risk-free returns. This procedure is the same for all three regression models. Further detail on the data is covered in chapter four.
Secondly the construction of the factors of the Fama & French three-factor model (1993) is discussed. The Fama & French three-factor model is described in equation three below.
�rp�=α𝑖𝑖+bi[(Rm)-rf]+si[SMBt]+hi[HMLt]+eit
Equation 3: Fama & French three-factor model
In order to calculate the SMB and HML factors of the Fama & French three-factor model (1993) bivariate sorting based on size and M/B is conducted. Stocks are sorted on size by taking the 50th percentile as a breakpoint; small stocks (S) are stocks which lie within the bottom 50th percentile and big stocks (B) are the opposite. In addition to sorting on size, stocks are likewise sorted based on the M/B ratio. However in this case, the 30th percentile is used as a breakpoint. Stock within the 30th percentile of the M/B-ratio are considered value stocks (V), whereas growth stocks (G) are those within the 70th percentile, the stocks which lie in between these percentiles are considered neutral stocks (N). The 2X3 sort as aforementioned creates six types of
This table shows the six portfolios created by the 2X3 sort on size and the M/B ratio.
M/B Ratio
Value (V) Neutral (N) Growth (G)
Size Small (S) SV SN SG
Big (B) BV BN BG
Table 5: 2X3 sorts of portfolios for the Fama & French three-factor model
Based on the aforementioned 2X3 sorts, the SMB and HML factors are created based on the six value-weighted portfolios as which have been created beforehand. The SMB factor is defined as the average return of the three small portfolios (SV, SN & SG) minus the average return of the three big portfolios (BV, BN & BG). A similar method is employed in creating the HML factors, this factor is defined as the average return of the two value portfolios (SV & BV) minus the average return of the two growth portfolios (SG+BG). Table six below provides formulas how the factors are calculated.
Factor Formula SMB 1 3(𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆) − 1 3 (𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆) HML 1 2 (𝑆𝑆𝑆𝑆 + 𝐵𝐵𝑆𝑆) − 1 2 (𝑆𝑆𝑆𝑆 − 𝐵𝐵𝑆𝑆) Table 6: Calculation of Fama & French three factors
Finally, the construction of the Fama & French five-factor model (2015) is discussed. The Fama & French five-factor model is described in equation five below.
�rp�=α𝑖𝑖+bi[(Rm)-rf]+si[SMBt]+hi[HMLt]+ri[RMWt]+ci[CMAt]+eit
Equation 4: Fama & French five-factor model
The factors of the Fama & French three-factor model (2015) are formed following the 2x3 sort as mentioned in their paper. Here SMB is the average return on the nine small stock portfolios minus the average return on the nine big stock portfolios, sorted on market-to-book ratio (M/B), size, operating profitability and investments. The HML factor is similarly defined as in the three-factor model. The RMW three-factor (Robust Minus Weak) is the average return on the two robust operating profitability portfolios minus the average return on the two weak operating profitability portfolios. Last the CMA (Conservative Minus Aggressive) factor is formed; this is the average return on the two conservative investment portfolios minus the average return on
the two aggressive investment portfolios. To construct the abovementioned factors, bivariate sorting is conducted based on Size, B/M, OP and INV. Similarly as in creating the factors from the Fama & French three-factor model (1993), stocks are sorted on size by taking the 50th percentile as a breakpoint; small stocks (S) are stocks which lie within the bottom 50th percentile and big stocks (B) are the opposite. Moreover stocks are sorted on the M/B ratio. Again, the 30th percentile is used as a breakpoint. Stock within the 30th percentile of the M/B-ratio are
considered value stocks (V), whereas growth stocks (G) are those within the 70th percentile, the stocks which lie in between these percentiles are considered neutral stocks (N). Likewise stocks are sorted based on INV, in this case; again the 30th percentile is used as a breakpoint. Stock within the top 30th percentile of INV are considered aggressive stocks (A), whereas conservative (C) are the ones within the bottom 30th percentile, the stocks which lie in between these percentiles are considered neutral stocks (N). Finally stocks are sorted based on OP, again the 30th percentile is used as a breakpoint. Stock within the top 30th percentile of OP are considered robust stocks (R), whereas weak (C) are the ones within the bottom 30th percentile, the stocks which lie in between these percentiles are considered neutral stocks (N). The aforementioned sorts are then used to create eighteen blocks of portfolios as shown in table seven on the next page.
Based on the aforementioned eighteen blocks of portfolios, respectively the factors SMB, HML, CMA and RMW are created. First I start with discussing the calculation of the factor HML, this factor is created in the same way as in the Fama & French three-factor model, which is defined as the average return of the two value portfolios (SV & BV) minus the average return of the two growth portfolios (SG+BG). Subsequently the computation of the factor CMA is
discussed, this factor is defined as the average return of two conservative portfolios (SC & BC) minus the average return of the two aggressive portfolios (SA+BA). Next the creation of the RMW factor is discussed; it is defined as the average return of two robust portfolios (SR & BR) minus the average return of the two weak portfolios (SW+BW). Finally the construction of the SMB factor is discussed, which is a to some extent more complicated to calculate. The SMB factor is defined as the average returns of the following three SMB portfolios, namely 𝑆𝑆𝑆𝑆𝐵𝐵𝐵𝐵
𝑀𝑀 , 𝑆𝑆𝑆𝑆𝐵𝐵𝑂𝑂𝑂𝑂, 𝑆𝑆𝑆𝑆𝐵𝐵𝐼𝐼𝐼𝐼𝐼𝐼. For example the 𝑆𝑆𝑆𝑆𝐵𝐵𝑀𝑀𝐵𝐵 portfolio is calculated by taking the average
of the returns of the small bivariate sorts on size and B/M portfolios (SV, SN and SL), minus the average of the returns of the big bivariate sorts on size and B/M portfolios (BV, BN and BL). The
other two SMB portfolios are calculated similarly. Table eight provides the formulas of the calculation of the respective factors.
M/B RATIO
Value (V) Neutral (N) Growth (G)
Size Small (S) SV SN SG
Big (B) BV BN BG
OP (Profitability)
Robust (R) Neutral (N) Weak (W)
Size Small (S) SR SN SW
Big (B) BR BN BW
INV (Investment)
Conservative (C) Neutral (N) Aggressive(A)
Size Small (S) SC SN SA
Big (B) BC BN BA
Table 7: 2X3 sorts of portfolios, based on size, M/B, OP and INV
Factor Formula 𝑺𝑺𝑺𝑺𝑩𝑩𝑩𝑩 𝑺𝑺 13(𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆) −13 (𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆) 𝑺𝑺𝑺𝑺𝑩𝑩𝑶𝑶𝑶𝑶 1 3(𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆) − 1 3 (𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆) 𝑺𝑺𝑺𝑺𝑩𝑩𝑰𝑰𝑰𝑰𝑰𝑰 1 3(𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆 + 𝑆𝑆𝑆𝑆) − 1 3 (𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆 + 𝐵𝐵𝑆𝑆) SMB 1 3 �𝑆𝑆𝑆𝑆𝐵𝐵𝑀𝑀𝐵𝐵 + 𝑆𝑆𝑆𝑆𝐵𝐵𝑂𝑂𝑂𝑂+ 𝑆𝑆𝑆𝑆𝐵𝐵𝐼𝐼𝐼𝐼𝐼𝐼� HML 1 2 (𝑆𝑆𝑆𝑆 + 𝐵𝐵𝑆𝑆) − 1 2 (𝑆𝑆𝑆𝑆 − 𝐵𝐵𝑆𝑆) CMA 1 2 (𝑆𝑆𝑆𝑆 + 𝐵𝐵𝑆𝑆) − 1 2 (𝑆𝑆𝑆𝑆 − 𝐵𝐵𝑆𝑆) RMW 1 2 (𝑆𝑆𝑆𝑆 + 𝐵𝐵𝑆𝑆) − 1 2 (𝑆𝑆𝑆𝑆 − 𝐵𝐵𝑆𝑆) Table 8: Calculation of Fama & French five factors
3.5 Evaluation of the regressions
In order to statistically evaluate the performance of each regression model in capturing the returns of the portfolios, the GRS-test is conducted. This statistical test which has been
developed by Gibbons, Ross & Shanken in 1989, is generally used to evaluate the efficiency of a multi-factor asset pricing model. The test is based on estimating whether the zero-intercepts of the regressions of such models are simultaneously equal to zero. Fama & French are scholars who often use this test in their papers when considering the performance of their models. One of the main downsides of using this test is that it assumes normality on stock returns and the regression factors. Zhou (1993) put forward that this assumption is generally violated6 in the literature. Moreover Chou & Zhou (2006) posit that in practice the GRS-test is generally valid however non-normality of returns could affect its reliability. A red flag to consider is that as this model is based on testing whether the intercepts are simultaneously equal to zero, it does not automatically mean that a model which apparently improves on the GRS-test is necessarily a better model. In order to evaluate the performance of the models one should not merely rely on the GRS-test but also take other measures into account.
GRS= �NT� �T-N-LT-L-1� �α� ∑ _'�-1 α�
1+μ�Ὡ�-1μ�� ~F �N,�T-N-L��
Equation 5: GRS-test test statistic
CHAPTER IV: Data
This chapter will cover specific details of the data used in this study. 4.1 Data collection
In order to test the various hypotheses, multiple databases have been accessed in order to obtain the required data. The data on both currently active and inactive firms in this study is acquired from both DataStream and COMPUSTAT and consists of the years 2002-2015. Data on the return index, market-to-book ratio (M/B)7 ratio, market capitalization, market proxy and risk-free proxy is acquired from DataStream and are on monthly basis, whereas accounting data has been retrieved from COMPUSTAT which is on annual basis.
4.1.1 Accounting Data and Identifiers from COMPUSTAT
As earlier stated the accounting data has been retrieved from COMPUSTAT. The accounting data from COMPUSTAT consists of book equity, revenues, cost of goods sold, selling general and administrative expenses, interest expenses and data on total assets. This data has been retrieved for each ASEAN Exchange covered in this study. In addition to the aforementioned accounting data, firm identifiers ISIN (International Securities Identification Number) and sector identifier SIC (Standard Industrial Classifier) have been included as this enables further
processing of the data. The ISIN is required in order to merge this data with data retrieved from DataStream, whereas the SIC allows excluding financials from the sample.
4.1.2 Other Data from DataStream
As aforementioned in section 4.1 data on the return index, M/B ratio, market capitalization, market proxy and risk-free rate are retrieved from DataStream. The data is on monthly basis and is denominated in local currency. First I will discuss the data on return index, M/B ratio and market capitalization. The aforementioned variables have been retrieved for each individual exchange covered in this study. In addition to the aforementioned data, as a firm identifier ISIN
7 In the Fama & French studies (1993, 1998, 2015) the book-to-market (B/M) ratio is used, but for some reason when I download the B/M ratio instead of the M/B ratio it would yield me with a much smaller sample. As the M/B ratio is its inverse, I use this peculiarity.
is used, as this enables merging the data with the COMPUSTAT data. Second I discuss the market proxy used in this study. For each ASEAN Exchange covered in this study, the return index on the various composite market indexes has been obtained. Third I discuss the risk-free proxy used, for each ASEAN Exchange either three-month commercial papers or treasury bills are taken as a proxy for risk-free returns. The various market-proxies and risk-free proxies can be found in the appendix.
4.2 Data preparation
As the data in this study has been retrieved from two different sources, namely COMPUSTAT and DataStream it requires extensive preparation in order to merge them. The main problem encountered with the DataStream output is that the firm identifiers (ISIN) are not returned for each column, but for each row, therefore it requires reshaping to a panel dataset, with a column for each firm identifier, date, year and variable. This has been done by recording macros. In addition to this, by means of the macro, firms without any information on either return index, M/B ratio and market capitalization have been deleted. Another issue revealed using the DataStream data, is that firms that are inactive (delisted) are reported into perpetuity. In detail, this means that when a firm has been delisted its last reported value is being reported into perpetuity. In some cases it is known when a firm went bankrupt or delisted; and in such cases it could be dealt with accurately; however for most of the cases it is not clear-cut. In those cases I have assumed that a firm is bankrupt8 when the same value is reported consecutively to infinity. In other cases when the same value is not being reported consecutively to infinity, I treat the firm as temporary suspended9. Having completed the extensive cleanup of the data, the datasets are ready to be merged for further processing as discussed in the methodology section.
4.3 Summary statistics
In table nine the total amount of firms included in the sample for each exchange for each year is given. From this table it follows that the sample covers generally represents over 70% of the
8 When I assume a firm to be bankrupt, the last value before removing the firm from the sample will be set to zero, which implies that the return over that month will be -1, which corresponds with the concept of limited liability. 9When a firm is treated as temporary suspended, it means that when a firm is already in the portfolio and after portfolio formation is suspended, the next returns will be equal to zero. Also temporary suspension implies that when a firm is temporary suspended, it cannot be included in a portfolio formed in the period in which it was suspended.
constituents for each stock exchange. Compared with previous studies regarding the return to value in ASEAN Exchanges (e.g. Sareewiwatthana, 2011) this is a relative high number and enhances the reliability of this study. Other descriptive statistics regarding the portfolios and factors can be found in the appendix.
Exchange: Indonesia Malaysia Philippines Singapore Thailand
Year: Firms in sample: 2004 274 717 188 425 365 2005 282 754 185 467 383 2006 290 758 183 503 409 2007 307 784 192 532 417 2008 340 780 194 570 437 2009 348 785 196 581 440 2010 367 796 199 583 439 2011 397 808 204 585 455 2012 417 802 215 591 477 2013 446 807 212 608 501 2014 471 810 219 614 527
Table 9: Amount of firms included in sample for each exchange 4.4 Limitations of the dataset
As stated in section 4.2, the data from DataStream reveal an issue. Although I have aimed to clean the dataset for this flaw, it could be the case that minor flaws remained in the dataset. Another limitation to the data from DataStream is that it is unclear whether it is affected by backfilling. Considering more recent periods of data than covered in this study gave me reason to believe that it is indeed affected by backfilling10 but its extent remains unknown. However as covered in the methodology section, my research design is such that it does take backfilling into account, however it remains unanswered whether the measure has been sufficient. As Kothari, Shanken & Sloan (1995) do argue that backfilling will lead to potential survivorship bias; this stipulation should be kept in mind. Another caveat to consider is that stock exchanges in emerging markets are dominated by penny stocks (e.g. Brown, Rhee & Zhang, 2008; Charitou & Panayides, 2009; Fama & French, 1998) and the ASEAN Exchanges are no exception. Therefore it could be that calculated returns are illusionary as pointed out by Brown, Rhee & Zhang (2008).
CHAPTER V: MAIN RESULTS
This chapter will cover the main results of this study. 5.1 The Value Premium on ASEAN Exchanges
Exchange: Indonesia Malaysia Philippines Singapore Thailand Monthly value premium: 0.0162* 0.0130** 0.0028 0.0114 0.0073
p-value 0.0347 0.0209 0.4324 0.1160 0.1514
Std. error 0.0089 0.0064 0.0161 0.0095 0.0072
*;**;*** denotes a significance level on respectively 5%,2.5% and 1% Table 10: Monthly value premia based on equally-weighted portfolios
In table 10 above it can be seen that a value premium based on equally-weighted portfolios has been observed in Indonesia and Malaysia. In Indonesia a monthly value premium of 1.62% (p-value 0.0347) is observed whereas in Malaysia a monthly (p-value premium of 1.30% (p-(p-value 0.0209) is observed. For the remainder of the stock exchanges statistically insignificant value premia are observed. Previous studies which covered the Philippine, Singaporean and Thai exchanges (e.g. Fama & French 1998, Drew 2003 and Sareewiwatthana 2011) did observe statistically significant value premia on these exchanges. Unlike Ding, Chua & Fetherston (2005), no growth premium is found on the Thai stock exchange, nor on any other exchange. For the purpose of robustness of my findings, in section 6.1 similar tests will be conducted on value-weighted portfolios as Elze (2010) remarked that employing value-value-weighted portfolios will enhance the robustness as those are less likely to be affected by sampling bias. Furthermore as the study period includes the recent Global Financial Crisis, 2007-2008, I will re-examine the return to value for the period of the Global Financial Crisis and for a post-crisis period, 2009-2014, with the aim of confirming the robustness of my findings in section 6.2.
5.2 CAPM Regressions
In table 11 on the next page the results of the CAPM regressions in each country on value and growth portfolios is depicted. The results suggest that in none of the cases it seems that the CAPM can capture the return to value portfolios. From the alphas it can be seen that in all cases the CAPM seems to capture the return to growth portfolios. However further evaluation which will be conducted in section 5.5 is required in order to do a comparative judgement of the results. Moreover in all cases the regression coefficient beta is statistically significant. Another
interesting finding is that in the majority of the studied countries, apart from the Philippines, the explanatory power of the CAPM on growth portfolios is higher than the explanatory power of value portfolios. This can be seen from the adjusted R² and the standard error of the regression, which is respectively higher and lower for growth portfolios.
This table shows the results of the CAPM regression on value and growth portfolios in each country studied. The CAPM is given by the following equation:�rp�=α𝒊𝒊+bi[(Rm)-rf]
Portfolio a b Adj-R² s(𝛆𝛆) Indonesia Value 0.0127** (2.40) 0.8700*** (22.09) 0.4803 0.0036 Growth -0.0035 (-1.37) 0.8660*** (11.05) 0.7968 0.0009 Malaysia Value 0.0099*** (10.08) 1.0726*** (2.60) 0.4302 0.0019 Growth -0.0027 (-1.10) 0.9823*** (14.56) 0.6170 0.0008 Philippines Value 0.1670*** (2.76) 0.9200*** (8.72) 0.3640 0.0046 Growth 0.0152 (1.08) 0.8098*** (3.30) 0.0702 0.0251 Singapore Value 0.0094* (1.98) 1.0982*** (12.53) 0.5436 0.0029 Growth -0.0026 (-0.77) 1.1891*** (19.49) 0.7430 0.0014 Thailand Value 0.0117*** (3.33) 0.6289*** (11.56) 0.5031 0.0016 Growth 0.0032 (1.22) 0.7904*** (19.17) 0.7366 0.0009
Note: t-values in parentheses;*;**;*** denotes a significance level on respectively 5%,2,5% and 1% Table 11: CAPM Regressions
5.3 Fama-French three-factor regressions
In table 12 on the next page the results of the Fama-French three-factor regressions in each country on value and growth portfolios is shown. The results show that in none of the countries the Fama-French three-factor model seems to capture the return to value if we consider the alphas. In all cases, with the exception of growth portfolios in the Philippines, all regression coefficients are significantly different from zero. The regression coefficients indicate that all
portfolios, with the exception of growth portfolios in the Philippines, are positively related with the HML factor and negatively related with the SMB factor.
This table shows the results of the Fama-French three-factor regression on value and growth portfolios in each country studied which is given by the following equation:
�rp�=α𝒊𝒊+bi�(Rm)-rf�+si[SMBt]+hi[HMLt]+eit Portfolio a b s h Adj-R² s(𝛆𝛆) Indonesia Value 0.0151*** (3.93) (14.23) 0.9562*** -0.6842*** (-8.00) 0.5878*** (8.64) 0.7260 0.0019 Growth -0.0030 (-1.21) 0.8851*** (20.66) -0.1525*** (-2.80) 0.1311*** (3.02) 0.8147 0.0008 Malaysia Value 0.0199*** (10.09) 1.0968*** (18.74) -1.1489*** (-12.88) 0.9097*** (14.38) 0.8736 0.0004 Growth 0.0039** (2.45) 1.0381*** (21.64) -0.7515*** (-10.43) 0.4579*** (8.96) 0.8610 0.0003 Philippines Value 0.0130*** (2.70) 0.9912*** (11.35) -0.6317*** (-8.11) 0.3792*** (5.42) 0.6062 0.0029 Growth 0.0155 (1.09) 0.8730*** (3.36) -0.1730 (-0.75) -0.0477 (-0.23) 0.0608 0.0254 Singapore Value 0.0102*** (4.56) 1.0714*** (27.02) -1.0575*** (-19.75) 0.7959*** (10.13) 0.9193 0.0005 Growth -0.0012 (-0.55) 1.1853*** (29.74) -0.6846*** (-12.51) 0.4248*** (5.38) 0.9053 0.0005 Thailand Value 0.0148*** (8.23) 0.9219*** (24.34) -0.8562*** (-14.05) 0.6934*** (12.66) 0.8718 0.0004 Growth 0.0053*** (2.93) 1.0502*** (27.55) -0.6879*** (-11.22) 0.2811*** (5.10) 0.8802 0.0004 Note: t-values in parentheses;*;**;*** denotes a significance level on respectively 5%,2,5% and 1% Table 12: Fama-French three-factor regressions
5.4 Fama-French five-factor regressions
In table 13 on the next page the results of the Fama-French five-factor regressions in each country on value and growth portfolios are depicted. The results show that in all countries, with the exception of Singapore the Fama-French five-factor model seems to capture the return to value if solely the alphas are considered. The most interesting observation from the Fama-French five-factor regression is that the regression coefficients of the factors RMW and CMA are never significant. Chen & Zhang (1998) have argued that in high growth markets generally stock
returns of value and growth stocks do not follow a clear pattern with respect to profitability and investments. Considering that during the vast majority of time during this study, the ASEAN exchanges generally were high-growth markets, which could be an explanation why the factors RMW and CMA are never significant. However it could be that profitability and investment patterns simply do not have explanatory power on the ASEAN exchanges. On the other hand Cakici (2015) found that in the cross-section of average stock returns the RMW and CMA factors do not add explanatory power for Asia-Pacific11 stock markets. For that purpose the regression is rerun during the crisis of 2007-2008, with the aim of confirming the aforementioned
assertion. The findings related to this robustness test can be found in section 6.2.
This table shows the results of the Fama-French five-factor regression on value and growth portfolios in each country studied. The Fama-French five-factor model is given by the following equation:
�rp�=α𝒊𝒊+bi�(Rm)-rf�+si[SMBt]+hi[HMLt]+ri[RMWt]+ci[CMAt]+eit a b s h r c Adj-R² s(𝛆𝛆) Indonesia Value -0.0029 (-0.40) 0.0197 (0.10) -0.5972*** (-3.73) 0.5407*** (6.63) -0.0085 (0.13) 0.0125 (0.08) 0.6241 0.0026 Growth -0.0034 (0.99) 0.8274*** (7.28) -0.0033 (-0.04) 0.1096*** (2.38) 0.0461 (0.24) -0.0318 (-0.61) 0.8008 0.0008 Malaysia Value 0.0003 (0.12) -0.5225*** (-2.77) -1.1490*** (-7.33) 0.7058*** (7.57) -0.0116 (-0.93) 0.0930 (0.69) 0.8021 0.0007 Growth -0.0120 *** (-7.86) -0.4572*** (-4.15) -1.1401*** (-12.43) 0.1978*** (3.63) 0.0046 (0.06) 0.0083 (0.11) 0.8859 0.0002 Philippines Value -0.0007 (-0.11) -0.1221 (-0.56) -0.8424*** (-4.82) 0.2794 (3.54) -0.0532 (-0.68) 0.0614 (0.79) 0.4928 0.0037 Growth -0.0031 (0.19) -0.0186 (-0.03) -0.7460 (-1.63) -0.0663 (-0.32) 0.0130 (0.06) 0.09739 (0.48) 0.0647 0.0253 Singapore Value -0.066** (-2.59) -1.3979*** (-8.63) -1.8053*** (-14.96) 0.4623*** (4.58) -0.1351 (-1.49) -0.0993 (-1.03) 0.8862 0.0007 Growth -0.01239*** (-6.55) -0.6703*** (-5.60) -1.3693 (-15.36) 0.1345 (1.80) -0.0434 (-0.65) -0.0849 (-1.19) 0.9279 0.0000 Thailand Value 0.0015 (0.5) -0.5361*** (-3.82) -0.9527*** (-8.29) 0.6246*** (8.87) -0.0654 (-0.74) -0.0957 (-1.06) 0.7951 0.0007 Growth -0.0046* (-1.85) -0.0640 (-0.49) -0.7242*** (-6.76) 0.2233*** (3.41) -0.1636 (-1.30) -0.0718 (-0.85) 0.8360 0.0006
Note: t-values in parentheses;*;**;*** denotes a significance level on respectively 5%,2,5% and 1% Table 13: Fama-French five-factor regressions
11 In Cakici (2015) Asia-Pacific refers to the developed markets of the Pacific Rim, such as Singapore, Australia, New Zealand and others. This is similar to Fama & French their definition.
5.5 Evaluation of the various regression models
This table depicts comparative statistics of the regression models.
Model: Mean Adj-R² Mean s(𝛆𝛆) Mean absolute alpha GRS p-value
Indonesia CAPM 0.6386 0.0025 0.0012 5.7874*** 0.0039 FFTM 0.7703 0.0014 0.0061 9.8499*** 0.0001 FFFM 0.7124 0.0017 0.0032 0.4918 0.6127 Malaysia CAPM 0.0526 0.0046 0.0072 21.4216*** 0.0000 FFTM 0.8673 0.0004 0.0119 54.0796*** 0.0000 FFFM 0.8440 0.0005 0.0006 47.6157*** 0.0000 Philippines CAPM 0.2171 0.0101 0.0160 4.0675** 0.0194 FFTM 0.3335 0.0095 0.0142 3.9356** 0.0220 FFFM 0.2788 0.0115 0.0019 0.0248 0.9754 Singapore CAPM 0.6433 0.0040 0.0060 8.9961*** 0.0002 FFTM 0.9123 0.0022 0.0057 11.8213*** 0.0000 FFFM 0.9071 0.0023 0.0095 21.9201*** 0.0000 Thailand CAPM 0.6199 0.0031 0.0075 6.3005** 0.0025 FFTM 0.8760 0.0018 0.0100 35.1460*** 0.0000 FFFM 0.8156 0.0026 0.0031 2.7077 0.0706
Note:*;**;*** denotes a significance level on respectively 5%,2,5% and 1% Table 14: Comparative statistics of asset pricing models
Table 14 above provides a comparison of all the asset pricing models used in this study. When the GRS-test-statistic is considered, it is found that the Fama-French five-factor is able to explain the return to value in Indonesia, Malaysia and Thailand. However there is more than meets the eye, the GRS-test-statistic purely enables to make a judgement on the statistical performance of the model, without regarding its economic significance, as the augmented factors RMW and CMA are never significant. When considering the mean adjusted R² and the mean standard error of the regression show that in all studied countries the Fama-French three-factor model
outperforms both the CAPM and the Fama-French five-factor model in explaining the return to value and growth.
