Contrarian profits in Latin American markets
Master dissertation
University of Amsterdam, Amsterdam Business School MSc Business Economics, Finance track
Levien de Kraa July, 2014
Abstract
This study examines the returns of applying a contrarian strategy to a sample of Latin American markets in the period ranging from 2004 till 2013. These relatively developed markets will serve as a proxy for emerging markets through the world, which have shown stabilization after much financial turmoil in the years between 1991 and 2001. This study further investigates whether these returns are sensitive to different risk factors such as the market beta, size, price or illiquidity. Empirical findings show that a contrarian strategy yields a positive return over the sample period of 1.68% per month on average. However, these returns are highly inconsistent over time and are heavily driven by outliers in returns in 2007 and 2009, when the global financial crisis was at its peak. In addition, the risk profile of loser portfolios is significantly higher than that of winner portfolios, indicating that while the strategy yields a positive return, this could very well be due to the additional risk contained in the portfolios. Furthermore, I find that variations in contrarian returns can best be explained by the standard CAPM model. Additional risk factors size, price and illiquidity do not significantly affect these returns. While some of these results confirm results found in previous literature, it is likely that some of these results were partially biased by the substantial spread in returns between the different markets included in this thesis.
Acknowledgement
I would like to offer my sincere gratitude to my supervisor Dr. Versijp, as he readily offered important advice on multiple occasions while writing this thesis. It has been a valuable experience and would not have been possible without his help and the support of close friends and family.
Table of content
I. Introduction 5
II. Literature Review 7
a. Efficient Market Hypothesis 7
b. Investment Strategies 8
c. Illiquidity 9
d. Emerging Markets 11
III. Data 13
IV. Methodology 15
a. Holding period returns 15
b. Risk characteristics of contrarian portfolios 16
c. Risk-based portfolio returns 17
d. Contrarian returns sensitivity 18
V. Empirical findings 19
a. Holding period returns 19
b. Risk characteristics of contrarian portfolios 25
c. Risk-based portfolio returns 26
d. Contrarian returns sensitivity 28
VI. Summary and concluding remarks 29
VII. References 31
I. Introduction
Since De Bondt and Thaler (1985) showed that using a arguably simple investment strategy, forming a portfolio with a long position in past losers and a short position in past winners, it is possible to generate substantial excess returns, researchers around the world have tried to proof or disproof what is named the contrarian investment strategy. A contrarian investment strategy could imply that the Efficient Market Hypothesis (EMH) does not hold, since investors do not accurately incorporate relevant information into stock prices leading to a mean reversal effect in stock prices. However, others have argued that excess profits could also be the result of changing risk characteristics of stocks included in different contrarian portfolios. Evidence that the EMH is unlikely to hold is abundantly shown in empirical literature, see for instance Fama and French (1997), Titman and Jegadeesh (1993, 1996). In these papers it is shown that short-term investments based on different risk factors such as size, value and momentum generate significant positive profits in developed markets. Furthermore, Lo and MacKinlay (1990) find that similar significant profits can be generated by contrarian investments in developed markets. However, the evidence on contrarian strategies shows varied results and a there is a ongoing debate on whether such strategies can be applied to consistently generate excess returns.
Much of the research on contrarian returns has focused on developed markets and predominantly on U.S. equity markets. In addition, much research was traditionally more dominantly focused on the somewhat related momentum strategy. The research on emerging markets has developed rapidly over the past years, with much empirical work focusing on global indices and spreading out to the possible use of the strategy in related markets such as global commodities and foreign exchange markets. By using the indices, research has focused on the more established and developed companies in each region, which are often also the most actively traded stocks. In this thesis, I aim to contribute to existing literature by looking at a much broader spectra of Latin American stock markets, where both volatility and liquidity are expected to be much larger issues. First and foremost, I investigate whether a contrarian investment strategy would have yielded a consistent return over the market using the stock markets of Argentina, Brazil, Chile, Colombia and Mexico as a subsample of global emerging markets. In addition, I aim to investigate whether there is an illiquidity effect in these markets. A liquidity effect would effectively mean that investors require a higher return for holding illiquid stocks, lowering excess returns over the market portfolio. Finally, I intend to examine whether the excess returns of a contrarian strategy exist due to higher risk profiles found in losing portfolios within the contrarian strategy. In existing literature, the risk profile of contrarian portfolios has been suggested as the reason for the existence of excess profits reaped using the strategy (see for instance Chan (1988), while others have criticized the risk argument (see for instance Lakonishok et Al. (1994). To my knowledge, the liquidity adjusted CAPM and Three factor models of Fama and French have not been used in previous literature to measure the sensitivity of contrarian returns to risk factors in emerging markets.
In this dissertation, I find that in the financial markets included, a contrarian strategy generates a positive return of 1.68% per month on average. However, the returns are highly inconsistent over years, generating extreme values in bearish markets and erratic returns around zero percent in other years. In
addition, the risk profiles of different contrarian portfolios differ substantially. The portfolios of past losers are highly volatile and actually provide a lower mean return per unit of risk. Furthermore, I find that contrarian returns are best explained using market excess returns. Looking at the empirical results, it is not proven that contrarian returns are partly the result of a size, price or illiquidity effect.
The structure of this thesis is as follows. The first section in this thesis presents the theoretical literature background of the topic, divided into sections covering the Efficient Market Hypothesis, different investment strategies, illiquidity and a short overview of literature on investment strategies applied in (other) emerging markets. The second section describes the dataset used in this dissertation and brief summary statistics, while the third second covers the applied methodology and an explanation of the different risk factors included in this dissertation. Finally, the fourth section covers the empirical findings of this dissertation, to be followed by a discussion on the empirical findings and concluding remarks.
II. Literature Review
There is an abundance of research on the efficiency of financial markets around the world. Since the publication of one of the key papers in the field of economics, that of Eugene Fama (1965) on stock price behavior, many researchers have tried to explain the movement of stock prices in different financial markets. One of these theories implies the hypothesis of investor overreaction to information and the resulting mean reversal in stock prices, leading to the profitability of contrarian investment strategies. In the following section an overview of existing research and literature is set out in four sections. However, the returns to different trading strategies exploiting market anomalies seem to depend substantially on the chosen markets, securities and time periods, and are a continuing topic of debate among researchers.
The first section gives an overview of the development on different views on financial market efficiency. The second section follows with the development of investment strategies based on market inefficiency in general, and on contrarian investment strategies specifically. In the third section, I focus on the Amihud measure for illiquidity, in addition to giving a brief overview of some key articles on liquidity effects in
emerging markets. Finally, the fourth section sets out a compact overview of literature on different investment strategies in emerging markets around the world.
a. Efficient Market Hypothesis
The initial key papers on capital market efficiency were written by Eugene Fama in 1965 and 1970 on stock price behavior, in which he developed his influential Efficient Market Hypothesis (“EMH”). Fama finds consistent evidence in favor of the random walk model; essentially showing that past stock prices contain no information which could have enabled investors to more accurately predict future stock price movements. This implies that thorough analysis of past stock prices has no value to investors as stock prices are then by
definition equal to their fundamental value. The EMH specifies that stocks are “information efficient”, that is to say that information is immediately and fully incorporated into stock prices, thus rendering information arbitrage impossible.
Fama specifies three measures of market efficiency; weak form, semi-strong and strong form efficiency (1970,383). Under a weak form efficiency, stock prices are assumed to contain all information available in past stock prices, rendering data analysis of past stock prices useless in predicting future prices. In a semi-strong market, prices are assumed to contain all information in past stock prices in addition to publicly available information. As publicly available information is fully and immediately incorporated into stock prices, investors cannot generate excess returns over the market by following different investment strategies based on analysis or as a response to information also available to other investors. Finally, the strong form efficiency implies that over and above the semi-strong form, private information is also incorporated into prices. This implies that no investor can generate a return over the average investor in the market. The EMH holds under all three measures of market efficiency, as all imply that prices move randomly.
The EMH assumes that investors are utility maximizing and fully rational. Effectively, under the EMH, an investor cannot achieve excess returns over the market return unless additional risk is taken, given the information available at that moment. Full rationality implies that investors observe and correctly incorporate new information into stock prices, adjusting for changes in the aggregate risk of their holdings. Under this assumption, the EMH implies that trading happens because of differences in personal utility, but always at fair value. As all trades take place at fair value, no investor is able to consistently generate a profit over the market return other than by sheer luck (Fama, 1970).
However, since the increasing popularity of the field of behavioral finance, research has proven that the EMH does, at least not fully, hold in financial market around the world. In the field of behavioral
economics, protagonists of the theory state that humans cannot be considered to be fully rational. At times investors are incapable of correctly applying new information to investment behavior, and as such, neither will stock prices accurately reflect all information.
Akarim and Sevim (2013) show that one aspect of investors not accurately integrating new
information into their investment practices is shown in what is termed the mean-reversal in stock prices. This concept lies at the core of what makes a contrarian investment strategy profitable. They describe the
phenomenon as “mean reversion model, one of the stock price behavior models, assumes that stocks have an average price in the long run and so an investor can identify a trading range for their investments by
estimating this average price level.” (2013,453). Testing different models in a sample of emerging markets, they find that emerging markets are in fact at least as inefficient as developed markets, showing that mean reversion in stock prices is valid in the long-run in all emerging markets examined. In addition, they show that in the sample period between 1995 and 2010, the contrarian investment strategy performs better than a buy and hold strategy or a momentum strategy.
b. Investment Strategies
Along with the increasing amount of proof that financial markets are not truly efficient, investors have developed trading strategies based on these inefficiencies. There is plenty of evidence that investment
strategies based on size, value and momentum generate a significant and positive return in developed markets of about 1%-2% per month, see for example Chan et al. (1996) and Fama and French (2012). Whereas the size effect states that smaller stocks tend to outperform the larger stocks in the long term, the value effect is based on the evidence showing that on average value stock outperform growth stocks. In addition, and more closely linked, the momentum and contrarian strategies, set out by Titman and Jegadeesh (1993) and De Bondt and Thaler (1985) respectively, benefit from irrational movements in stock prices.
De Bondt and Thaler (1985), in their research on stock market overreaction, find that U.S. stocks behave in line with the overreaction hypothesis. Under this hypothesis, they find that ‘portfolios of prior “losers” are found to outperform prior “winners”.’ (1985: 804). The formed zero-investment portfolios in their research have a formation and holding period of 36 months, in the sample ranging from 1933 till 1980. They
find that the overreaction and mean reversal effect in stock prices is still observed 5 years after formation. In addition, they find an increased January return for the loser portfolios, over and above a return which could be explained by the tax-loss selling theory. Chan (1988) however, adds to this research by showing that the risk of holding a loser portfolio differs from the winning portfolio and inherently changes substantially over time. He states that when adjusting for risk in the held portfolios, only a minor abnormal return is found using the dataset of De Bondt and Thaler.
Lo and MacKinlay (1990) further research the sources and components of contrarian profits. They prove that overreaction is not the only reason for the inefficient working of financial markets, pointing to a lead-lag effect between larger stocks leading smaller stocks. Concluding, they suggest that contrarian profits are partly due to investor overreactions, but that a large part is due to cross effects among securities, and that contrarian profits are not dependent on stock price mean reversals. Contrarily, Titman and Jegadeesh (1995) find that only a small part of profits can be attributed to lead-lag effects. They further find that the stock price overreaction is the main source of contrarian profits, mainly due to stock price overreaction to firm-specific news. They conclude that return reversals are economically significant and warrant further attention.
In different key articles on contrarian investment strategies, holding and formation periods vary between one week in the research of Lehman (1990), to a formation period of 36 months and holding of over 5 years in the research of De Bondt and Thaler (1985). Significant positive returns to the strategy are found in both papers, indicating that contrarian investment strategies are not dependent on the timeframe. However, Akarim and Sevin (2013) show by panel data regression that mean return reversal in a sample of emerging markets takes between 30 and 38 months; i.e. it takes about three years for stock prices to return to their average value. In addition, they find significant support that a contrarian investment strategy will generate superior returns in these markets when compared to a buy and hold or a momentum investment strategy. However, in several studies into contrarian strategies, such as Bildik and Gülay (2007), it is suggested that a formation/holding period format consisting of 6 and 12 months respectively, should yield reliable results. Ashgarian and Hansson (2009) concluded in their research on US markets, that returns from applying a very short term contrarian strategy, with formation and holding periods varying between a week and a month, are dependent on investor behavior. This behavior is described as the overreaction effect. Contrarily, long term contrarian strategy returns, with formation and holding periods ranging between one year and five years, are increasingly dependent on different risk factors, pre-dominantly being the market factor, size and the market-to-book ratios.
c. Illiquidity
In addition to size, value and momentum effects, researchers have more recently focused on the effect of stock illiquidity in stock returns. Amihud and Mendelson (1986) suggested that in addition to the Fama and French risk factors, illiquidity could be regarded as transaction costs for immediate execution and as such, as a cash outflow which diminishes investment returns. Furthermore, following reasoning set out by Glosten and Milgrom (1985), illiquidity is a result of adverse selection problems in financial markets, where uninformed
traders demand higher rates of return. While there has been a substantial amount of research on illiquidity effects in developed markets, this section will predominantly focus on different liquidity models developed in the past and on the key research papers most important for the emerging markets. It is important to note that all research below focusses on the liquidity effect on a full portfolio of diversified stocks and the market return and not on the relation between contrarian investment portfolios and liquidity (or size and price).
In 1984 Roll developed a liquidity measure based on the effective bid-ask spread for the stock. The Roll method is based on the negative auto-covariance between the opening price changes for a stock. A low liquidity stock will show a higher spread between the bid and ask prices, and can be considered as a transaction costs for traders. Following the earlier explained reasons that transaction costs lower portfolio returns, the spread can be used as a proxy for liquidity. While the general consensus is that price-based liquidity measures more accurately depict liquidity costs in markets, more often volume based measures are applied since share turnover and volume is often more reliable and readily available for emerging markets.
In 2002, Amihud suggested an illiquidity ratio as to measure the liquidity of a stock or index. It is based on the reasoning that stocks for which a large trading volume leads to a small price change can be considered very liquid. The measure takes the average of the ratio between the daily return of a stock divided by the days’ traded volume. The Amihud measure is a popular measure in literature on liquidity, but it must be noted that the measure often exhibits extreme values when there is very little trading volume, which is especially likely to happen when measuring liquidity by definition. Another critique is that the ratio returns non-real values for zero-trading days. In 2009, Hasbrouck developed an adjusted form of the Amihud measure to enhance the usability of the ratio in low volume trading circumstances by rescaling the measure. The Amihud measure is adjusted by taking the square root of the ratio between the stocks’ return and its volume. Both measures are considered volume based approaches to measuring liquidity.
Lesmond (2005) focuses on emerging markets since these markets are characterized by high volatility and scarce trading. Using the bid-ask spread costs as a measure for illiquidity, following the research of Lesmond et al. (1999) and Roll (1984), he finds that price-based liquidity measures are better at explaining liquidity effects cross-country than are volume based measures. However, he specifically states that for within country effects, the results are opposite. He concludes with findings of liquidity premia between 1% for the Taiwanese market to over 47% for the Russian market. Bekeart et Al. (2007) suggest that liquidity plays an important role in less sophisticated markets where a lower number of companies and investors lead to increased adverse selection effects. In their research, they focus on the effect of liquidity on local emerging markets, finding that liquidity is an important driver of expected returns in these markets, using a model based on the amount of zero-return trading days, averaged over the month.
Dey (2005) studies liquidity risk as measured by turnover as a determinant of returns in 49 global stock indices. He finds that there is a significant, positive relation between average returns and turnover. However, from his results he concludes that this relation only holds for indices for emerging countries. In addition, he finds that the age, type and size of the exchange, the competition for order flow and the growth
rate also influence portfolio liquidity. Lee (2011) studies 28 emerging markets for liquidity effects, using the Acharya and Pedersen adjusted CAPM augmented with a liquidity factor. He finds that liquidity is significantly priced in most of the studied markets.
d. Emerging markets
Most evidence found on investment strategies exploiting market inefficiencies stems from research on
developed markets. Rouwenhorst (1999) examines momentum returns for stocks in 20 emerging markets from around the world (a total of 1750 different entities), in which a small number of firms in Latin American countries are included, among which are Argentina, Brazil, Chile and Venuzuela. As a motive for his research on emerging markets, Rouwenhorst points to the relative isolation of these markets to international developed capital markets and to the historically low correlation between the different markets. He argues that if similar results are found in more locally oriented markets based on a local version of the CAPM model, these results will more firmly establish the true relationship between expected returns and momentum, value, size and betas.
Rouwenhorst finds that the factors that explain cross-sectional differences in expected stock returns are similar to those found in developed markets: small stocks outperform larger stocks over the longer term, value stocks outperform growth stocks and stocks in emerging markets exhibit momentum. Rouwenhorst concludes that the average momentum returns in his sample of emerging markets are statistically lower than in developed markets, though momentum in stock prices is still confirmed. Moreover, he finds that local return factors are only slightly correlated between the different countries, showing that the different returns in one country from investment strategies are independent from factors of other markets. Finally, he finds that there is no significant relation between share turnover and expected returns, but the relationship between share turnover and beta, size, momentum and value is positively correlated. According to Rouwenhorst, the latter shows that return premiums are not only a reflection of compensation for illiquidity.
Subadar and Hossenbaccus (2010) suggest that small capitalization markets possess substantially different characteristics from developed markets, pointing at the research of Harvey (1995). Harvey states that emerging markets are more likely to be driven by local information and as such possess a lower correlation with global equity indices.
Investigating a sample of emerging nations from South America, Asia and Europe, Cakici et Al. (2013) show that between the period 1990-2011 there are strategies which generate positive returns to momentum investing and investing based on the value effect for every region included in their research besides Eastern Europe. Furthermore, they investigate portfolios sorted by size and book-to-market ratios and size and lagged momentum, they use the three-factor model, the CAPM model and the Carhardt four-factor model to explain returns using local, global and U.S. data. They find that there are significant differences in premia based on size between developed and emerging markets. Cakici et al. conclude that returns generated by value and
momentum investing are negatively correlated. In addition they find that emerging market returns can best be
11
explained by local factors, suggesting a low correlation in factors between emerging nations and global capital markets.
Griffin et Al. (2010) examine and compare the market efficiency in 28 developed and 28 emerging markets between 1994 and 2005. They use the short-term reversal strategy set out by Jegadeesh in 1990, with a short-long portfolio position according to a contrarian theory to measure the returns to this strategy as a measure of market inefficiency. Furthermore they test market efficiency by applying a post-earnings
announcement drift test on the dataset in addition to testing for momentum according to the Jegadeesh and Titman (1993) model. Griffin et Al. find that these strategies earn similar returns in emerging markets as in developed markets, but that standard efficiency measures are likely to be less useful in these markets. They conclude that there are conflicting results about efficiency in emerging markets; while the strategies infer that emerging markets are at least as inefficient as developed markets, when analyzing transaction costs and information costs they conclude that these markets are likely to be less efficient. The latter is concluded because if it is more costly to gain private information and more costly to trade based on this information, stock prices will only slowly adjust to this new information.
In 2008, Silva and Chávez analyzed cross-listing and liquidity costs in the markets of Argentina, Brazil, Chile and Mexico. They emphasize that low liquidity deters foreign investment and that illiquidity raises the cost of capital and thus market values. This is especially problematic in emerging markets, which need investment capital to grow. They find that, in their sample period between 1992 and 2001, it is relatively expensive to trade stocks in Latin America, pointing to a sample mean round-trip costs of 8%. In addition, they find that trading costs in Brazil are significantly higher than in the other countries in the region. While
discussing market structure, they underscore the difficult financial situation the countries are in till the end of 2001. At the end of that year, all four countries had finalized changes in regulation, allowing foreign investors to enjoy higher transparency and better corporate governance. As such, aggregate market liquidity increased and trading normalized with an increasing turnover till the global financial crisis started in 2007.
Furthermore, Heng-hsing and Hodnett (2011) find that for stock listed on the Johannesburg Stock Exchange (JSE), longer formation periods generate larger returns when testing for a contrarian strategy (in their paper they refer to overreaction theory). Significant in their research is the finding that the strength of mean reversals is stronger in losing stocks, according to them “market corrections is more prompt for the losers than it is for the winners.” (2013: 127). However, they find that winning portfolios continue to outperform the market. This implies that, assuming the JSE can be used as a proxy for different emerging markets, contrarian investment strategies should generate positive returns in this research, as the return reversal of losers generate profits above the losses absorbed on the winning stocks.
Hearn et Al. (2010) find that illiquidity plays a large role in stock returns in a sample of African emerging markets. In addition, they find that company size is similarly influential on market returns. However, the results of similar studies performed on different emerging markets, show conflicting results. Furthermore,
the research focusses on the influence illiquidity could have on market returns, which is different from the expected relationship between liquidity and the returns from a contrarian investment strategy.
While there is little focus solely on Latin American markets, Da Costa (1993) focused on a subsample of the most stable stocks in the Brazilian market in period 1970-1989. Using the approach of De Bondt and Thaler, het concluded that after 24 months, a contrarian investment strategy generated returns of over 37%, with the loser portfolio outperforming the market by 17,63%.
Muga and Santamaría (2007) study the momentum effect in Latin American markets in order to discover the mean reversal in sample containing the markets of Argentina, Brazil, Chile and Mexico. They find that the returns to winners stochastically dominate the returns of loser portfolios for risk-averse investors. They further suggest that it is not clear beforehand whether emerging markets will show larger momentum returns or smaller. On one hand, lower analyst coverage and the inclusion of smaller stock could lead to an increased overreaction effect. On the other hand, they state that “there are arguments that suggest a lower level of momentum due to the financial turbulence and economic crises that tend to beset emerging markets.” (2007,25). These arguments hold in similar vein for contrarian returns, since they are characterized by similar profits under volatile markets, albeit in a contrary manner.
III. Data
In this paper monthly stock return data is used for several nations in Latin America in order to test the performance of contrarian investment strategies in these markets over the past decade (Jan 2004 – Jan 2013). Data will be collected for a sample of nations in which stock markets are considered stable and developed enough to lend credibility to the results of this dissertation, in addition to providing sufficient reliable data. As a criteria for stock exchange stability and development, selected countries are those included in the Morgan Stanley Capital International (MSCI) index for South America. In addition to the MSCI index, Argentina is included because sufficient information on prices and trading volume is available through the Thomson Datastream financial database. In total, the six selected countries included are Argentina, Brazil, Chile,
Colombia, Mexico and Peru. In order to further enhance the credibility of this study, only stocks which are part of the Worldscope index are entered into the study, resulting in a total of 1270 companies included in the sample. The Worldscope index is used since it offers a complete constituents list for every country. By using the constituents list rather than only the index, the financial information of a broader range of companies can be analyzed, including a larger sample of smaller stocks. Smaller stocks are generally omitted in other research due to their highly volatile nature and substantial illiquidity issues, even when key financial data is available. However, considering the scope of this research towards Illiquidity, only stocks with no information on market capitalization or prices are excluded from the sample. In addition, the constituents list contain both active and dead stocks, thereby enlarging the dataset and lowering survivorship bias.
Prices and market capitalization are converted into U.S. dollars using the integrated Thomson One exchange rates and excess returns are computed using the one-month U.S. Treasury bill rate as the risk-free proxy. Survivorship bias is not a problematic issue, as the Worldscope database contains both active and dead firms. As is typical for emerging markets, a considerable subsample of the dataset contains stocks with a relatively low value and/or price. While the average market value of all firms included is close to $1437 million, the median value is significantly lower at $149.43 million, as is shown in table 1.
Table 1
Characteristics of included companies by year. The table provides annual total market values, price and the number of firms included in the sample period. Countries included are Argentina, Brazil, Chile, Colombia, Mexico and Peru. Reported annual values represent the median value over the months in that year.
Year Firms Market Capitalization ($) Price ($)
2004 870 47.25 0.78 2005 898 71.825 0.99 2006 931 91.07 1.25 2007 998 171.08 2.22 2008 1,059 188.36 2.20 2009 1,072 141.87 1.60 2010 1,101 214.95 2.41 2011 1,138 250.45 2.60 2012 1,163 215.99 2.34 2013 1,180 207.81 2.44 2014 1,184 195.94 2.29 Total 1,041 149.43 1.87
(Average) (median) (median)
Note: 2014 contains only data for the month January. Average total firms represent the average of firms in the period between 2004 and 2013.
The sample includes stock information for the period between January 2004 and January 2014, covering 121 monthly intervals in which end of month prices and market capitalization is used. 18 months are lost in the formation of contrarian portfolios using a 6 month formation and 12 month holding period. As such, the data of the first six months of 2004 are lost in generating the cumulative monthly return used in forming the first contrarian portfolios and the last 12 months of the dataset are lost generating the holding period return of the last 12 months of 2012. The sample period from 2004 onwards is selected in order to ensure that possible lagged effects resulting from the turmoil in financial markets in the region in the period between 1991 and 2001 does not affect the results of this study.
IV. Methodology and hypotheses
The methodological section is split into four parts, in order to provide answers to different questions. In the first part, the monthly average returns of contrarian portfolios are analyzed. First using the entire sample period, followed by an examination of the returns by year, in order to increase insight into when contrarian strategies are more likely to generate excess returns. Furthermore, in an attempt to dampen the effect of a large market capitalization spread within the sample data, the sample will be split into subsamples based on market capitalization. Last, the returns to the contrarian strategy are displayed per country. In the second section, the risk characteristics of the broad contrarian portfolio are investigated in order to discover whether the different quintile portfolios are related to size, price or Illiquidity. In the third section, the returns to portfolios based the size, value and illiquidity effect are generated and displayed. Finally, in the last section, using the CAPM model and the adjusted Fama/French three factor model, both further augmented with an additional liquidity risk factor, the contrarian returns will be regressed on the market return and the different risk factor loadings in order to gain further insight into the nature of excess returns resulting from the contrarian investment strategy in these markets. What follows now is the methodological process for the different steps.
a. Holding period returns
In the first section, contrarian portfolios will be formed by ranking all companies according to their cumulative return in the 6-month period previous to the formation date, which I will refer to as the formation period. Following the research of Titman and Jegadeesh (1993), the last month of returns previous to the formation date is excluded in order to avoid possible influences due to the bid-ask spread effects, price pressure effects and lagged reaction effects. Cumulative returns are thus calculated as follows:
𝐹𝑜𝑟𝑚𝑎𝑡𝑖𝑜𝑛 𝑝𝑒𝑟𝑖𝑜𝑑 𝑟𝑒𝑡𝑢𝑟𝑛𝑡= (𝑃𝑟𝑖𝑐𝑒𝑡−1− 𝑃𝑟𝑖𝑐𝑒𝑡−6)�𝑃𝑟𝑖𝑐𝑒𝑡−6
Following this, the companies will be divided into five equally-weighted portfolios based on their formation period lagged returns. Companies with the lowest cumulative return over the six months before that date, are appointed into portfolio one ( the “losers”) and companies with the highest cumulative return are appointed into portfolio five (the “winners”).
The return in the following 12-month period is computed, referred to as the holding period, which is then averaged to get the monthly average return per portfolio. A fictional long/short position will be formed consisting of a short position in the top quintile representing the past winners (with high lagged returns) and a long position in the bottom quintile representing the past losers (with low lagged returns), following the method as specified in Titman and Jegadeesh (1993), which is the method followed in the majority of empirical work on contrarian investment strategies, such as Heng-Hsing and Hodnett (2011). In this research, the contrarian portfolios will be based on a 6 month formation period, and both the contrarian and the risk characteristic portfolios will be evaluated based on a 12-month holding period. In previous literature there is
little proof of an ideal framework, it is shown that the contrarian strategy should provide excess returns using different formation and holding periods between one week and one year as a formation period, and between 1 week and 3 years as holding periods.
Following the method above, I form overlapping portfolios, such that in every subsequent month after July 2004, an investor applying the strategy would hold a maximum of 12 similar portfolios (the current months in addition to a maximum of 11 portfolios over the past 11 months), hereby increasing the total number of portfolios evaluated. It also serves to ensure the starting month picked in this research (July 2004) does not significantly affect the results, which in turn safeguards that seasonality in returns does not
significantly impacts the results of this thesis. In addition, I assume the portfolios are held for the full period and are not rebalanced after every month, leading to the following portfolio returns:
𝐴𝑅𝑡= 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 �
(𝑃𝑟𝑖𝑐𝑒𝑡+12− 𝑃𝑟𝑖𝑐𝑒𝑡)
(𝑃𝑟𝑖𝑐𝑒𝑡)
12 𝑚𝑜𝑛𝑡ℎ𝑠
� �
Where ARt indicates the monthly average return over the holding period of the portfolio and Price indicates the individual stock prices. Annual returns are then computed taking the average over all months. The individual returns are then average per month to get the monthly average return. The results of each portfolio is then evaluated both over the whole sample period and by year, showing whether the 6/12 formation/holding period contrarian strategy yields significant returns by assuming the six different markets can be considered as one market. Next, the sample is divided into two groups based on market capitalization. In every month, the sample is split into two using the median as a cutoff point. Both subsamples are then taken as individual markets, and contrarian portfolios are formed following the same process as above.
b. Risk characteristics of contrarian portfolios
In the second part, the specific characteristics of the selected contrarian portfolios separated. For portfolios 1 through 5, the average and median values for market capitalization, price and illiquidity are reported. This is done in order to investigate possible correlations between returns and risk characteristics. In previous
formation period holding period
Formation date (time = t) t = 12 t =-1 t = -6
16
literature (see for instance Cakici et al. 2013), it was often found that contrarian returns are correlated with size meaning that companies in portfolio 1 (past-losers) are generally the smaller stocks as measured by market capitalization. There is however, no previous literature specifying a relationship between illiquidity and contrarian returns.
c. Risk-based portfolio returns
In the third part, using similar steps as in the formation of contrarian portfolios, similar quintile portfolios will be selected in an equal holding period framework based on different risk characteristics. By using these portfolios, I am able to generate the returns relating to the risk factors size, price, and liquidity in these markets. The different risk factors are summarized as follows:
Size
The selected stocks in the combined countries are ranked in order of market capitalization. The selection is divided into quintiles sorted by size, and the variable SMB denotes the monthly return of the quintile with the smallest stocks minus the monthly return on the portfolio with the largest stocks. Following the abundance of research in this field, I hypothesize that the holding period return on smaller stocks is larger than the return on larger stocks.
𝐴𝑅𝑠𝑖𝑧𝑒= 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 �𝑅𝑒𝑡𝑢𝑟𝑛𝑝𝑓1− 𝑅𝑒𝑡𝑢𝑟𝑛𝑝𝑓5 �
Where ARsize denotes the average return per month and subscripts pf1 and pf5 denotes the top and bottom quitiles selected on size respectively. The difference between the portfolios is then annualized by taking the average over the 12 months.
Price
The selected stocks in the combined countries are ranked in order of stock price and divided into the variable P, where the difference in returns between the portfolios with the lower priced stocks and the higher priced stocks denotes the average monthly difference in returns. it is hypothesized that on average, a higher percentage of small cap stocks will be among the loser portfolio and vice versa.
𝐴𝑅𝑝𝑟𝑖𝑐𝑒= 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 �𝑅𝑒𝑡𝑢𝑟𝑛𝑝𝑓1− 𝑅𝑒𝑡𝑢𝑟𝑛𝑝𝑓5 �
Where ARprice denotes the average return per month and subscripts pf1 and pf5 denotes the top and bottom quitiles respectively. The difference between the portfolios is then annualized by taking the average over the 12 months.
Liquidity
The selected stocks are further ranked on the basis of their liquidity. Liquidity is measured by the Amihud measure (2002) as described below. A volume based measure is used because of the increased reliability and
availability of trading volume over bid-ask spreads in emerging markets. The variable ILLIQ measures the difference in average monthly returns between the portfolio with the least liquid and the most liquid stocks. Amihud measure:
𝐼𝐿𝐿𝐼𝑄 = 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 �𝑉𝑜𝑙𝑢𝑚𝑒|𝑅𝑡|
𝑡�
Where | Rt | measure the absolute stock return for the individual stock in period t and Volume measures the traded volume in dollars for the same period. It is worth noting that this measure only has a real value in positive trading days, as the formula is undefined for days with zero-volume. As such, for days where the Amihud measure returns negative or no value, the value will be set to zero. Following a similar approach as for the risk factors size and price, the returns to a long/short portfolio based on illiquidity is then annualized by taking the average over the year.
Value
The value effect, indicating the effect that companies with a higher market-to-book value generally outperform companies with a lower market-to-book value, is not included as a risk characteristic in this dissertation. While the effect has been well documented in developed markets (see for instance Fama and French, 1993), the ratio found in the sample data has an distribution which is severely influenced by extreme values, leading to exceptionally high outliers throughout the sample. These high outliers are unlikely to be reliable and could significantly bias the results of this thesis. Following the model set out by Martinez et Al. (2005), I therefor exclude the Market-to-Book ratio as a factor loading, while realizing this may lead to an positively biased effect of contrarian excess returns.
d. Contrarian return sensitivity
Finally, in the last part, the selected risk factor loadings described will be entered into the two different asset pricing models in order to control for other sources of excess returns, such as company size and liquidity premia. The returns will be measured according to the CAPM model of Sharpe (1964) augmented for liquidity as done in the research of Hearn et Al. (2010) and the similarly augmented three-factor model as set out by Fama and French (1993). In addition, a high minus low price factor is included to see whether some of the return to contrarian portfolios can be explained by a price effect.
𝐶𝐴𝑃𝑀 𝑀𝑜𝑑𝑒𝑙: 𝐶𝑅𝑖− 𝑅𝑓,𝑖= 𝛼𝑖+ 𝛽𝑖�𝑅𝑚𝑎𝑟𝑘𝑒𝑡,𝑖− 𝑅𝑓,𝑖� + 𝛾1(𝐼𝐿𝐿𝐼𝑄𝑖) + 𝜖𝑖
𝐴𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑇ℎ𝑟𝑒𝑒 − 𝐹𝑎𝑐𝑡𝑜𝑟 𝑀𝑜𝑑𝑒𝑙: 𝐶𝑅𝑖− 𝑅𝑓,𝑖
= 𝛼𝑖+ 𝛽𝑖�𝑅𝑚𝑎𝑟𝑘𝑒𝑡,𝑖− 𝑅𝑓,𝑖� + 𝛿1,𝑖(𝑆𝑀𝐵𝑖) + 𝛿2,𝑖(𝐻𝑀𝐿𝑖) + 𝛾1(𝐼𝐿𝐿𝐼𝑄𝑖) + 𝜖𝑖
Where CRI denotes the contrarian returns in month i, Rrf denotes the risk-free rate as measure by the one month US treasury rate. Rmarket denotes the return on the entire market portfolio in month i and SMB and ILLIQ denote the average returns of the size and illiquidity effect in that month respectively. HMLi denotes the high minus low price factor in timeslot i.
V. Empirical Findings
The following section covers the empirical findings. Sections are divided according to the division found in the setup of the methodology section. First, the holding period returns of the contrarian strategy will be
presented, followed by the returns per year and the returns based on different splits in subsamples. Second, the risk characteristics for size, price and illiquidity of the contrarian portfolios are set out. Following, the returns to risk based portfolios are generated, followed by the return sensitivity analyses by testing the two different asset pricing models.
a. Holding period returns
Considering the performance of different contrarian portfolios over the sample period, I find that past losers (Portfolio 1) significantly outperform past winners (Portfolio 5) with 1.68% per month on average. Using equally weighted portfolios, the average return over the entire period is 2.67% per month over the entire index. This effectively means that a more diversified investment in the market would have outperformed the contrarian strategy over the sample period. Table 2a displays the monthly average returns for all portfolios and the equally-weighted index of all portfolios. Portfolios 1 and 2, constituting the bottom two quintiles of the stock, outperform the index by 1.44% and 0.74% per month respectively. However, especially these two portfolios show rather high volatilities, which could possibly be the result of traditionally highly volatile penny stocks included in the sample. While the two portfolios generate a higher return, the mean return ‘per unit of risk’ is significantly lower for these portfolios. This is in line with findings in previous literature, where research has shown losing portfolios to generally exhibit a higher volatility, see for instance Cakici et Al. (2013). Portfolio 3, constituting of the companies which neither over nor underperformed significantly over the 6 months before formation, perform worst in the subsequent 12 months, generating a return of 1.51% per month on average, which is considerably lower than the mean return of the combined portfolios. However, the variability of this return is lower with a standard deviation of 14.50%.
When analyzing the performance of contrarian portfolios over all countries by year, it is evident that the returns are all significantly different from zero, but do not show a consistent positive or negative return. The annualized average monthly returns are summarized in table 2b and graph 1. In the first 1,5 years of the sample, the strategy generates a slight positive return. However, in 2006 the total return is wiped out by a negative average return of 1.62% per month on average. In the year 2007 the strategy returned close to a massive 11% per month, while the returns to the strategy turn negative after the year 2009. From graph 1 it is clear that the strategy produces high returns in years where there is significant turmoil in financial markets.
Table 2a
Contrarian portfolio returns over the period between July 2004 and Jan 2013 for all countries. Returns and standard deviations are the result of a 6-month formation period and a 12-month holding period. Portfolio 1 constitutes the portfolio with stocks which performed worst in the previous 6 months, while portfolio 5 contains the winning portfolios.
Monthly average return (%) Standard deviation (%)
Portfolio 1 (Past losers) 4.11 60.16
Portfolio 2 3.41 102.72
Portfolio 3 1.51 14.50
Portfolio 4 1.82 8.72
Portfolio 5 (Past winners) 2.43 15.07
P1-P5 1.68 5.58
Total 2.67 54.66
Note: Monthly average return of the total is based on an equal weighted index of the portfolios.
Table 2b
Contrarian portfolio performance by year. Displayed average is the mean difference between the portfolio of past losers (PF1) and the portfolio of past winners (PF5). Standard deviations of the difference in means are given in parenthesis.
Average monthly return
Portfolio 1 (%) Portfolio 5 (%) difference (%) Mean
2004* 4.81 4.25 0.56*** (0.43) 2005 4.65 3.92 0.73*** (1.07) 2006 7.98 9.6 -1.62*** (2.24) 2007 13.88 2.91 10.97*** (8.45) 2008 0.05 -1.19 1.24*** (5.45) 2009 8.06 3.52 4.54*** (6.36) 2010 1.67 1.72 -0.05*** (0.97) 2011 -0.78 -0.72 -0.06*** (0.96) 2012 -0.69 0.61 -1.3*** (0.68) Total 4.09 2.43 1.66*** (Standard deviation) (5.58)
Note: 2013 and 2014 results lost in generating holding period returns. *2004 results are shown for the last 6 months of the year. *** indicates significance at the 1% level.
Graph 1
This is consistent with previous findings, that show mean reversion in stock prices is stronger in bearish markets (Heng-hsing and Hodnett (2011). Graph 1 visually shows the performance of the contrarian portfolio since 2004. From the graph it is clear that returns are erratic around zero percent, with extreme spikes in returns of up to around 24% percent per month in 2007 and up to 17% in 2009. It is likely these results were driven by extreme shocks the aggregate economy of Latin America during the global financial crisis.
Furthermore, in table 2c, the data is split into two subsamples based on market capitalization. As in all stock markets around the world, there is a substantial difference between total market value of the companies included in the index. To research whether the returns to a contrarian strategy differ substantially between the smaller stocks and the larger stocks, all observations in each year have been divided into two groups according to the median value of market capitalization in that year. Next, all stocks below the median have been pooled into a subsample denoted “low” and the stocks which have a higher value into another subsample denoted “high”. From the table, it is clear that the total return from sample of smaller stocks is significantly higher than that of the larger stocks; 3.35% and 1.98% on average per month respectively. However, there is substantially more risk involved in investing in these stocks; the standard deviation on a return of 3.35% is 66.88%. Contrarily, the standard deviation on a portfolio of the larger stocks is 38.48%. It is surprising to find that a large part of the risk is driven by stocks included in portfolio 2 of the “low” subsample, and by portfolio 1 of the “high” subsample. When looking at returns using a contrarian strategy, the smaller portfolio only returns 0.98% per month on average, while the larger portfolio returns 2.28% on average per month. This effectively means that in the sample period, it would have been more profitable to invest in a subsample of larger stocks using a contrarian strategy. However, this result is unlikely to be consistent over different markets and periods, and is more likely to be driven by outliers in different years. Stocks could especially have been hit in extremes around the financial crisis.
-.1 0 .1 .2 .3 R et ur n 2004.01 2006.01 2008.01 2010.01 2012.01 2014.01 Time
Contrarian strategy returns
In table 2d the sample has been split into the different Latin American markets which together form the sample dataset. For this table, first subsamples have been formed for the different countries, after which the same methodology is applied to generate the returns to a contrarian strategy. The returns portrayed in the table offer further insight in the source of the contrarian returns found in table 2a. Especially for Peru, a
Table 2c
Contrarian portfolio performance over the entire sample period from 2004 till 2014. The entire sample has been divided into two subsamples. Division is based on median market capitalization value by year. For instance, in 2007 median market capitalization was $ 171.08m. All companies with a market value over this amount are appointed to "high". Mean and median values are in USD million, and given per subsample.
Low High
Average monthly return (%) deviation (%) Standard Average monthly return (%) deviation (%) Standard
Portfolio 1 (Past - losers) 4.09 20.67 4.15 89.19
Portfolio 2 4.96 138.54 1.54 10.19
Portfolio 3 1.93 19.59 1.06 4.77
Portfolio 4 2.18 11.94 1.55 5.17
Portfolio 5 (Past winners) 3.11 13.66 1.87 16.10
P1-P5 0.98 5.35 2.28 9.44 Total 3.35% 66.88 1.98 38.48 Observations 59.879 59.030 Mean 42.42 2944.12 Median 20.07 862.35
Note: 2013 and 2014 results lost in generating holding period returns. Total number of observations per group differs slightly due to division into groups around the median.
contrarian strategy appears to generate a somewhat consistent positive return, with returns between 2.88% and 10.07% per month on average in the period between July 2004 and December 2012, with the exception in 2006. Furthermore, the returns in these years are significantly different from zero in most years. In Brazillian markets, a contrarian strategy returns an extreme return of over 40% in 2007, and generating a sample return of 5.20% per month on average, which is significantly different from zero. From the table, it is seems highly likely that the spike in return found in the sample dataset in 2007 is at least in part driven by the returns in Brazilian markets. For the remaining four countries, the returns to a contrarian strategy are largely insignificant and lie mostly between approximately -1% and 4% per year. Moreover, the return over the sample period lies between -0.39% and 1.31%, which makes it improbable that investors would significantly outperform the market when adjusting for the risk free return. Furthermore, it can be concluded that a contrarian strategy is unlikely to generate a consistent return in most emerging markets when markets are not in significant turmoil. It is likely that the results of contrarian strategies applied to all six countries as if one market, are likely to show
22
high standard errors due to a substantial spread in returns from within the countries. These results lead to a possible conclusion that there is low correlation between the selected markets, which can lead to substantially higher standard deviations, possible affecting the statistical significance of the results.
De lta de no te s t he m ea n di ffe re nc e be tw ee n po rt fo lio 1 m inus po rt fo lio 5 . Por tfol io 1 Por tfol io 5 De lta Por tfol io 1 Por tfol io 5 De lta Por tfol io 1 Por tfol io 5 De lta Por tfol io 1 Por tfol io 5 De lta Por tfol io 1 Por tfol io 5 De lta Por tfol io 1 Por tfol io 5 De lta 20 04 3. 48 1. 46 2. 02 ** * 6. 03 5. 74 0. 29 3. 17 4. 07 -0 .9 0* * 7. 13 7. 30 -0 .1 7 1. 46 3. 02 -1 .5 6* ** 6. 98 2. 20 4. 78* ** (1 .3 9) (1 .3 6) (0 .3 6) 0. 00 0. 00 (1 .6 1) (0 .1 7) (0 .5 2) 20 05 2. 70 0. 30 2. 40 ** * 4. 90 4. 78 0. 12 0. 71 1. 37 -0 .6 5* * 5. 15 5. 55 -0 .4 0 2. 08 1. 57 0. 52 10 .7 3 5. 58 5. 15* ** (1 .8 2) (0 .4 4) (0 .3 3) 0. 00 0. 00 (0 .4 9) (0 .5 3) (1 .3 6) 20 06 2. 93 3. 57 -0 .6 4 13 .7 2 13 .7 1 0. 00 2. 92 2. 60 0. 32 4. 26 1. 93 2. 34 6. 56 6. 30 0. 26 10 .0 6 11 .5 4 -1. 48 (1 .8 2) (1 .2 0) (0 .3 7) 0. 00 0. 00 (2 .5 4) (0 .4 5) (1 .1 0) 20 07 0. 03 0. 48 -0 .4 5* 46 .8 0 5. 52 41 .2 8* * 0. 20 -0 .3 2 0. 53 ** 11 .3 0 3. 67 7. 63 -0 .2 3 3. 40 -3 .6 3* ** 10 .2 8 0. 21 10. 07* ** (1 .1 5) (1 6. 96 ) (0 .2 0) 0. 00 0. 00 (6 .2 9) (0 .5 3) (1 .3 1) 20 08 -1 .6 4 -2 .7 8 1. 14 0. 36 -0 .4 8 0. 84 0. 21 -0 .7 2 0. 93 -0 .3 9 -0 .0 6 -0 .3 3 -2 .2 4 -2 .3 9 0. 15 0. 06 -1 .1 9 1. 25 (3 .5 2) (2 .1 1) (0 .7 2) 0. 00 0. 00 (0 .4 4) (1 .0 3) (0 .9 7) 20 09 5. 02 3. 19 1. 83 9. 04 4. 40 4. 64 ** 7. 07 3. 48 3. 58 ** 5. 53 4. 28 1. 25 6. 82 2. 76 4. 06 ** 6. 16 3. 24 2. 92* ** (5 .7 5) 2. 46 (1 .6 0) 0. 00 0. 00 (1 .0 6) (2 .0 9) (0 .3 5) 20 10 1. 41 3. 35 -1 .9 3* * 0. 72 1. 21 -0 .4 9* ** 5. 32 3. 43 1. 89 2. 99 1. 78 1. 21 1. 01 2. 17 -1 .1 7* ** 4. 58 1. 70 2. 88* ** (2 .5 1) 0. 16 (2 .1 1) 0. 00 0. 00 (0 .8 4) (0 .2 3) (0 .7 9) 20 11 -2 .1 1 -2 .3 2 0. 20 -1 .6 0 -0 .8 9 -0 .7 1* ** -0 .4 2 -0 .7 1 0. 30 0. 23 0. 04 0. 19 -0 .2 1 0. 40 -0 .6 1* ** 2. 03 -0 .0 9 2. 13* ** (1 .5 9) 0. 13 (0 .2 8) 0. 00 0. 00 (0 .3 8) (0 .2 0) (0 .5 4) 20 12 0. 89 0. 04 0. 85 ** -1 .5 1 -0 .4 0 -1 .1 0* ** 0. 08 0. 30 -0 .2 2 -0 .5 6 -0 .0 3 -0 .5 2* * 1. 54 3. 54 -1 .2 0* ** -0 .5 5 0. 65 -1. 19 (1 .2 9) 0. 16 (0 .6 6) (0 .2 3) (0 .5 1) (0 .6 3) To ta l 1. 30 0. 77 0. 53 ** 8. 77 3. 57 5. 20 ** 2. 04 1. 32 0. 72 ** 3. 72 2. 42 1. 31 * 1. 86 2. 25 -0 .3 9 5. 43 2. 64 2.8 0** * (0 .2 9) (5 .2 1) (0 .3 5) (0 .8 2) (0 .3 5) (0 .4 5) COL OM BI A M EX IC O PER U No te : 2 01 3 and 2 01 4 r es ul ts lo st in ge ne rat ing ho ldi ng pe rio d r et ur ns . * 20 04 re sul ts ar e s ho w n f or the las t 6 m ont hs o f t he y ear . * ** , * *, * indi cat es si gni fic anc e at the 1 % , 5 % , 1 0% si gni fic anc e l ev el re spe ct iv el y. Ta bl e 2 d AR GE NT IN A BR AZ IL CH IL E Co nt ra ri an po rt fo lio pe rf or m anc e by c ount ry . P er ce nt ag es de no te a ve ra ge m ont hl y r et ur n o ve r e ac h y ea r. A ll r et ur ns a re g iv en i n pe rc ent ag es , w ith t he c or re spo ndi ng s ta nda rd de va tio ns in pa re nt he se s.
24
b. Risk characteristics of contrarian portfolios
In past literature, researchers have linked the return on a stock to be related to the size of the company, as measured by its total market value, the price of the stock, in addition to its market-to-book value (excluded in this dissertation). In addition, it has been suggested that liquidity of the stock is similarly priced. In table 3a the average and median values of total market capitalization and price are reported, sorted by the contrarian portfolios. It is evident from the table that companies which are constituents of the loser portfolio (portfolio 1) are the smaller stocks with a lower market capitalization. Furthermore, portfolios 4 and 5 are the large market capitalization stocks measured by both average and median value. However, it is surprising to find that the average and median size of portfolio 5 companies is lower than those found in portfolio 4. When looking at average and median prices a similar trend appears; the past losers are generally the low priced stocks, and the price increases up to the fourth quintile. However, for the fifth quintile the average and median prices are again lower. It is possible that this is due to stock splits generally performed by more advanced and established companies in order to increase the market liquidity of the stock, this is however not clear from the data and falls outside the scope of this empirical work.
Furthermore, in table 3b the illiquidity measure of the different contrarian portfolios is reported. It is important to note the illiquidity measure has been multiplied by a factor thousand for presentation purposes. The past loser portfolio, constituting of low market value and low price stocks, shows a marginally higher illiquidity than portfolios 2 through 4. However, most notable is that the portfolio consisting of past winners shows a significantly higher illiquidity. In addition, similar to the relation with size and price, there is a large difference between the median and mean in all portfolios.
Table 3a
Characteristics of contrarian portfolios over the sample period. Both average and median values are computed using monthly values.
Portfolios
Market Capitalization ($) Price ($)
Average Median Average Median
Portfolio 1 (Past - losers) 1,077.37 92.77 1,195.24 0.97
Portfolio 2 1,214.89 109.71 3,116.65 1.65
Portfolio 3 1,178.09 138.31 3,261.62 3.07
Portfolio 4 1,990.63 247.25 4,601.40 2.44
Portfolio 5 (Past winners) 1,975.23 248.18 2,739.83 2.19
Total 1,482.91 153.09 2966.62 1.92
Note: Based on equally weighted contrarian portfolios as reported in table 2.
Table 3b
Illiquidity characteristics of contrarian portfolios over the sample period. The Illiquidity measure is formed according to Amihud (2002), using monthly values rather that daily trading information. Where no trading volume is available, monthly trading volume is set to 0, indicating that the stock was not traded in this timeslot.
Illiquidity
Average P10 P25 Median P75 P90
Portfolio 1 (Past - losers) 24.22 0 0 0.03 0.5 7.4
Portfolio 2 11.14 0 0 0.01 0.21 3.31
Portfolio 3 18.48 0 0 0.02 0.44 6.02
Portfolio 4 16.54 0 0 0.02 0.32 4.65
Portfolio 5 (Past winners) 92.89 0 0.01 0.06 1.04 17.09
Total 37.86 0 0 0.03 0.48 7.58
Note: The Amihud measure is multiplied by a factor thousand for presentation purposes. P10, P25, P75 and P90 denote the 10th, 25th, 75th and 90th percentile respectively.
c. Risk-based portfolio returns
In order to investigate whether the returns of a contrarian strategy are truly significant, additional portfolios are formed based on the different risk characteristics. Following a similar methodology as in the formation of the contrarian portfolios, quintile, equally-weighted portfolios are selected based on size, price and illiquidity, in order to generate the returns on these portfolios using the same 12-month holding periods. The returns generated by the strategies based on risk factors are summarized in table 4.
Mean differences for size based portfolios are significantly different from zero in all years. Portfolio 1 contains the bottom quintile containing the smallest stocks in the markets in the selected years. Portfolio 5 contains the large cap companies in all markets. From the data it is clear that a long/short portfolio based on size generates a significantly positive return up to 2008, after which it turns negative for the remaining years, with the exception of 2010. Contrarily, price based portfolios generate a significantly positive return in all years, while the returns decrease towards the end of the sample period. In similar vein, mean difference returns for illiquidity based portfolios also significantly differ from zero. From the data, it is clear that in 2008, 2009 and 2012, a illiquidity based long/short strategy generates losses, while it generates positive returns in the other sample period years.
Table 4
Returns on portfolios selected on size (Market Capitalization), price and illiquidity. Size is measured by market capitalization and illiquidity is measured by the Amihud measure. Monthly data is annualized to present annual returns.
Size
Portfolio 1 (%) Portfolio 5 (%) Mean difference (%) Standard deviation (%)
2004 7.28 3.30 3.98 2.18 2005 6.07 2.71 3.36 1.60 2006 19.76 3.63 16.13 16.70 2007 38.31 2.67 35.65 30.46 2008 -0.08 -0.36 0.28 1.50 2009 4.19 4.22 -0.03 0.89 2010 2.42 1.08 1.34 0.52 2011 -0.56 -0.46 -0.10 0.83 2012 -0.23 0.14 -0.37 0.40 Total 8.08 1.77 6.31 16.01 Price
Portfolio 1 (%) Portfolio 5 (%) Mean difference (%) Standard deviation (%)
2004 7.37 2.96 4.40 1.95 2005 6.57 1.96 4.61 1.85 2006 15.07 3.29 11.77 4.09 2007 17.02 1.21 15.82 7.93 2008 0.85 -1.17 2.03 1.82 2009 5.53 2.88 2.64 1.19 2010 3.85 0.90 2.94 1.13 2011 -0.32 -0.74 0.42 0.33 2012 0.13 -0.60 0.73 0.64 Total 5.87 1.08 4.80 5.85 Illiquidity
Portfolio 1 (%) Portfolio 5 (%) Mean difference (%) Standard deviation (%)
2004 6.05 4.19 1.86 0.53 2005 4.62 3.67 0.96 0.52 2006 10.94 7.23 3.71 3.56 2007 10.97 1.87 9.10 20.54 2008 -1.25 -0.73 -0.51 2.02 2009 3.62 6.06 -2.44 2.09 2010 1.71 1.35 0.36 0.83 2011 -0.26 -0.58 0.33 0.50 2012 0.18 0.26 -0.08 0.58 Total 3.73 2.38 1.35 7.76
Note: Portfolio selection based on equal-weighted portfolios, in which each portfolio consists of the same number of stocks.
d. Contrarian return sensitivity
Table 5 reports the result of OLS testing the contrarian returns over the period 2004 through 2012 by using the adjusted CAPM and adjusted Fama and French three-factor Model. This is done in order to test whether the contrarian returns exist due to higher risk embedded in the different portfolios. If the models accurately predict expected returns, the intercept in de model should be close to zero. While this is true for all models tested, the results show that contrarian returns can best be modeled using a standard CAPM model,
controlling for the market return minus the risk free rate proxy. The standard CAPM model shows an adjusted R2 of 0.400, indicating that the model is able to explain approximately 40 percent of the changes in the contrarian portfolio return. Furthermore, it shows that the difference between winning and losing portfolios is approximately covered by movements in the excess market return. For every point increase in the excess market return, contrarian returns will increase roughly 0.83. When expanding the model with different risk factors however, is seems these factors do not significantly contribute to an explanation of where contrarian returns come from. While the adjusted R2 value increases from 0.400 to 0.416 when adding the different factors, the results show that this increase does not significantly change the results. Expected returns from contrarian investment strategies are therefore most completely explained by the market return.
Table 5
Various regressions showing the sensitivities of the returns of a contrarian strategy (portfolio 1 minus portfolio 5). Contrarian returns have been used as the dependent variable.
CAPM Multiple factor models
Intercept 0.002 0.002 0.002 -0.006 -(0.46) (0.41) (0.44) (-1.13) Market – Rf 0.829*** 0.807*** 0.853*** 0.795*** (4.93) (4.54) (3.75) (3.59) Illiquidity 0.043 0.051 0.018 (0.48) (0.55) (0.18) Size -0.017 -0.068 (-0.26) (-1.05) Price 0.0255 (-1.29) R 0.406 0.409 0.410 0.440 R adjusted 0.400 0.397 0.392 0.416 F 24.35 12.95 9.21 8.87 Significant F 0.00 0.00 0.00 0.00
Note:*** denotes significance at the 1% level.
