The Overreaction Hypothesis on Emerging Markets

(1)

The Overreaction Hypothesis on

Emerging Markets

Master Thesis Finance MSc Business Administration Faculty of Economics & Business

Rijksuniversiteit Groningen The Netherlands

Author: Jakub Andrzej Bork

Student number: 1842161

Contact information: Jakub Bork, ul. Witosa 7/30,

80-809 Gdańsk, Poland JakubBork@gmail.com

Supervisor/1st judge: Drs. M.M. Kramer

2nd judge: Dr. L.J. Dam

(2)

2

Preface

This thesis is based upon the studies conducted between August 2008 and August 2009 at the Faculty of Economics & Business at the Rijksuniversiteit Groningen, The Netherlands. I was studying the Master of Science in Finance specializing in Risk and Portfolio Management and this thesis is the final part of my master program. Its subject is ‘The Overreaction Hypothesis

on Emerging Markets’. With this preface, I would like to take the opportunity to express my

gratitude to several people.

In first place I would like to thank my parents – Hanna and Andrzej Bork. Without their help and financial support it would not be possible for me to study and deepen my knowledge abroad.

My special thanks also goes to my leading supervisor Drs. Marc Kramer who was always willing to advice me during the writing process.

I wish to express my greatest thanks to my family and friends from Castricum, especially Rita and Flip Markowski, Ria Horvat and Arnold Kuijs for helping me to organize my room and for their hospitality on my frequent visits to Castricum.

Last but not least, I would like to thank my Dutch friends – a group of future geographers. They made my stay in Groningen easier and more entertaining always providing a good company - thanks to them during my free time I was able to enjoy the students-life of Groningen.

(3)

3

Abstract

(4)

4

1. Introduction

This master thesis presents a closer look at the ‘overreaction hypothesis’ phenomenon by examining whether it is present in emerging markets. Overreaction hypothesis is known as the long-run mean reversion in stock prices; it means, in other words, that the stocks which performed badly over the past period of time (so called ‘losers’) are expected to outperform those stocks, that did really well (so called ‘winners’) over the subsequent and similar period of time (often one to five years)1. To put it differently, when a stock price experiences a sharp increase or decrease in its value, then afterwards it will usually reverse itself.

The first authors, who performed a research about the stock market overreaction hypothesis in finance were Werner F. M. De Bondt and Richard H. Thaler (1985 - Does the

stock market overreact?). They reported the presence of this phenomenon on the US market.

Since that time the stock market overreaction effect has been given wide attention.

Generally, the stock market overreaction effect has been well documented for the US and other well developed big markets. On the other hand, a very little research has been done for emerging markets. In view of the above, this thesis will examine whether the conclusions presented by De Bondt and Thaler are also valid for emerging markets. The research will be carried out on a group of ten “Big Emerging Markets” (BEMs), plus Russia. BEMs were identified in 1993 by the U.S. Department of Commerce, as the locations for most of the growth in international trade for the following two decades. They were believed to offer the greatest potential for market growth.

The ten countries identified as BEMs in 1993 were:

Asia: China, Indonesia, India, and South Korea

America: Mexico, Argentina, and Brazil

Africa: South Africa

Europe: Poland and Turkey

(6)

6 These ten countries presented above are still included in Morgan Stanley Emerging Markets Index published in June 2006. Adding Russia to the research is a consequence of the fact, that it belongs to BRIC countries2.

In order to perform the research monthly stock market data is being used, if possible, for a 21 year period (December 1987 to December 2008). The time period used depends on the availability of the data - not all the stocks exchanges presented in this research existed in 1987/1988. That is the consequence of the fact that emerging markets’ stock exchanges have been developing and growing rapidly during the last two decades being characterized by high rates of returns. Since they have been becoming more and more popular among investors in the last few years it is interesting to check whether overreaction hypothesis phenomenon on those markets exists.

Having as many as eleven markets to examine (12 stock exchanges 3), this thesis will focus only on long-term overreaction trying to find out whether companies which were characterized by extreme good (winners) and bad (losers) performance experience a subsequent reversal in their returns. Using the methodology presented by De Bondt and Thaler (1985) and Clare and Thomas (1995) the following research questions will be answered:

Question 1: Do the emerging stock markets overreact in the long-term ? Question 2: If so, can this be explained by risk differences ?

Additionally, the used study procedure allows to examine the influence of the January Effect and other calendar anomalies. The month of January is well known for the presence of abnormal returns. In order to find out whether the January effect on the differential portfolio is present the Clare and Thomas’ (1995) methodology will be repeated once again and a January dummy will be added in regression. It will be possible to answer the third research question:

Question 3: If emerging stock markets overreact in long-term, is it mainly present in January ?

(7)

7 To test the additional seasonal effects dummy variables for the other eleven months will be added. Afterwards, those dummies will be regressed against the differential portfolio.

It is interesting to check whether the overreaction is present on emerging markets whose global importance is increasing. From economic point of view, if the overreaction is confirmed, it would be possible to earn money using The Contrarian Investment Strategy4.

I find that the portfolios of losers managed to outperform portfolios of winners on all eleven examined markets for the two and three-year formation and test period. In case of a three-year formation and test period the results on all the markets, except Turkish one, are statistically significant at the 10% level. After controlling possible different exposures to systematic risk a positive and significant beta, meaning that losers are characterized with higher systematic risk than winners, has been found only for three markets – Brazilian, Polish and Chinese (Shenzhen) for at least one out of three test periods (12, 24 or 36-month formation and test period). Completing the second test caused that the number of significant on at least 10% return differences decreased – the winner-loser effect became less pronounced. Testing for the January effect I was able to confirm it only on three markets - Turkish (as 12-month formation and test period was used) South Korean (24 and 36-month formation and test period) and Indian (36-month formation and test period). Additionally I find that the Chinese March Effect is asymmetric being pronounced more for losers then for winners.

The organization of this paper is as follows. In section two, the relevant literature on the topic will be presented and the theoretical framework will be discussed. The third section covers the methodology used in the research. The data issues are reported in section four, followed by section five, where the overreaction hypothesis and its possible explanations are being tested using the stock return data. Concluding, the results are presented and discussed in this section. Section six concludes this paper and gives some recommendations for further research. In the last section, appendixes provide some additional information about the whole research.

4

(8)

8

2. Literature review

In this section the relevant theoretical background regarding the stock market overreaction will be presented. The first research on overreaction effect suggested that it may be explained by investor irrationality, which has been highlighted by analyses in the field of cognitive psychology. Therefore, first the overreaction phenomenon’s connection with cognitive psychology will be discussed. Thereafter, I will try to answer the research questions by using the existing literature. The evidences on long term overreaction hypothesis and its possible explanations (risk differences and seasonality) will be presented. This section concludes with the summary of the previously discussed literature.

2.1. Overreaction hypothesis and cognitive psychology

Examining for the first time the stock market overreaction hypothesis in finance De Bondt and Thaler (1985) were stimulated by the results of Kahneman and Tversky (1982) in cognitive psychology, who claimed that their findings supporting the overreaction hypothesis are the consequence of ‘representativeness heuristic’. People tend to overweigh recent information, at the same time underweighing prior information. While doing their evaluation they are not considering prior probabilities, but individuals base their decision on how similar or representative an event is in comparison with their own beliefs. They act overoptimistically hearing good news and show too much pessimism after bad news. Consequently, the stock prices tend to divert from their real value. When the investors realize that they have been overreacting then the prices will mean-revert – this phenomenon is called overreaction hypothesis.

This is illustrated by Power and Lonie (1993):

(9)

9

‘anchoring’ of investor judgment is gradually eradicated by the accumulation of information inconsistent with the stereotypes, the share price moves slowly in the opposite direction, resulting in a mean-reverting pattern in share returns.”

2.2. Long term overreaction hypothesis in past research

The first research that reported the overreaction hypothesis in finance was undertaken by De Bondt and Thaler (1985). They investigated monthly return data for common stocks existing on the New York Stock Exchange (NYSE), as compiled by the Centre for Research of Share Prices (CRSP) for the period January 1926 to December 1982. A portfolio of winners (35 best performing stocks) and a portfolio of losers (35 worst performing stocks) were created based on the prior thirty-six months period, called the portfolio formation period. Then for both portfolios in each of 16 non-overlapping three-years periods cumulative average residual returns were calculated over the succeeding thirty-six months (portfolio test period). Supporting the overreaction hypothesis De Bondt and Thaler found out that during the 50-year test period (1932-1982) loser portfolios of 35 stocks outperform the market by 19,6% three years after portfolio formation, while winners portfolios earned about 5% less than the market. Thus, the cumulative average residual between extreme portfolios equals 24,6 %. However, over the following five-year test periods results are even bigger in numbers. The portfolios of losers performed better than the portfolios of winners by an average of 31,9 % (De Bondt and Thaler (1987). De Bondt and Thaler interpreted their evidence as a manifestation of an irrational behavior presented by investors, which they term as ‘overreaction’.

The paper of Power (1991) investigates whether the UK companies exhibit the same regressive tendencies as their US counterparts. As reported, Power’s results confirmed findings of De Bondt and Thaler – a contrarian strategy based on investing in the shares of previously identified excellent and non-excellent UK companies can yield abnormal returns (the excellent companies assumed to produce ‘winner’ shares and the non-excellent companies ‘loser’ shares).

(10)

10 24 months the same difference got even bigger in numbers – losers outperformed winners by 37,88%.

The overreaction was also confirmed by Clare and Thomas (1995) who investigated this phenomenon on UK market using UK 1955 to 1990 drawn from a random sample of up to 1000 stocks in any one year. They found that losers manage to outperform winner over a two year period of time by a statistically significant 1,7% per annum.

More recent studies were conducted by Wang et al. (2004), who performed a research on Chinese market (Shanghai Stock Exchange, Shenzhen Stock Exchange) and by Gunasekarage and Power (2005), who examined the stock market overreaction hypothesis on the Colombo Stock Exchange in Sri Lanka. Wang argued that many Chinese share returns exhibit patterns that are consistent with the investor’s overreaction. A shares were reported to have more pronounced mean-reversion in comparison with B shares5. Also Gunasekarage and Power’s results speak in favour of overreaction hypothesis phenomenon. Their findings supported the notion that investors overreact in this emerging market. The portfolio of losers outperformed the portfolio of winners by a statistically significant amount. The observed overreaction was asymmetric being more pronounced for losers then for winners.

Some studies were also undertaken for the Japanese market. For example, Iihara et al. (2004) examined the predictability of Japanese stock returns focusing on past performance-based trading strategies. However, they found out that return reversals are observed, but especially for short-term portfolio formation strategies. That is in line with the findings reported by Mun et al. (1999), who performed a research on German and French stock markets looking for long and short-term overreaction effects, and also confirmed that the portfolio trading strategies are successful on both markets, however, short-term contrarian portfolio worked better then long-term one giving higher profits.

2.3. Risk differences as a possible explanation of overreaction hypothesis

Subsequent papers suggest that De Bondt and Thaler’s findings are subject to various methodological problems. Authors who did the research after becoming familiar with

(11)

11 De Bondt and Thaler’s results tried to look for a possible explanation of what they simply called as an overreaction.

In his study Chan (1988) created similar portfolios as De Bondt and Thaler (1985) focusing on 3-year samples. He used data from the Centre for Research of Share Prices (CRSP) for a time period from December 1926 to December 1985. However, adopting the standard Sharpe-Lintner Capital Asset Pricing Model (CAPM) he proposed an alternative interpretation of evidence on the performance of the contrarian strategy assuming that the risks of winner and loser stocks are not constant over the time. Consequently, the estimation of abnormal returns may be sensitive to how the risks are estimated. The financial leverage of the loser firm becomes bigger as the stock price falls, increasing the risk of the stock. In addition to this, the equity beta will increase when a series of negative abnormal returns occurs, thereby increasing the expected return on the stock. However, Chan reported that using an empirical method that is free of the problems caused by risk changes, he found that the contrarian investment strategy earns a very small abnormal return (probably economically insignificant). Concluding, he found no strong evidence in support of the stock market overreaction hypothesis.

Ball and Kothari (1989) decided to extend the previous research. They found out that the negative serial correlation in relative returns is mainly due to variation in relative risks and therefore changing the expected returns. Analyzing further, the evidence is not consistent with the stock market mispricing. Even though the serial correlation in portfolios’ abnormal returns is statistically significant, the magnitude of the abnormal returns is small and, in Ball and Kothari’s opinion, economically insignificant. In addition to this, it varies over time. De Bondt and Thaler reported a comparable five-year average annual return of 9.2% on an arbitrage portfolio of winner and loser stocks. Ball and Kothari’s results were slightly lower – the average annual abnormal return over the five-year post ranking period is 3.13%. Furthermore, they reported larger changes in beta then Chan, meaning that the correlation effect between this beta and the market risk premium is small. To put this differently, the return on the loser minus winner portfolio is attributable to its beta risk and not to the covariance between systematic risk and the risk premium.

(12)

12 portfolios formed on the basis of prior five-year returns, extreme prior losers outperform extreme prior winners by 5% per year (up to 10% per year among smaller firms) during the subsequent five years. In addition to this, Newton da Costa (1994) and Mun et al. (1999) also illustrated that the differences in risk, as measured by CAPM beta cannot account for the difference between the ‘loser’ and ‘winner’ portfolio returns, which is consistent with the overreaction hypothesis.

2.4. Overreaction hypothesis and seasonality

A number of studies suggested that the observed inverse relationship between security returns is mainly confined to January. One possible explanation for the superior returns in January is based on the sale of companies which have experienced a poor share price performance in order to establish a tax loss. Evidence suggesting that part of the January return may be connected with this explanation were provided by Schultz (1985). Additional evidence for the tax-loss hypothesis were given by the studies in countries with a tax year-end other than 31 December. Therefore, in addition to the January effect, a July effect has been reported in Australia where a tax year-end occurs in June (Brown, Kleidon, Marsh 1983b) and an April effect in the UK (Reinganum and Shapiro 1987).

An interesting finding by Ho (1990) is the presence of a February effect in stock returns on Kuala Lumpur Stock Exchange. Ho suggested that this effect may be related to the Chinese New Year. The turn of the lunar year occurs mostly during February and represents the New-Year for ethnic Chinese, who are the dominant investors in the Malaysian market.

Chan (1985) and De Bondt and Thaler (1985) have linked the turn of the year phenomena with an overreaction effect. Findings presented by De Bondt and Thaler (1985) have two main aspects - the overreaction effect appeared asymmetric (it is much larger for loser portfolio than for winner portfolio). Secondly, most of the excess returns are realized in January - excessively big positive excess returns were earned at the start of the year (1st, 13th and 25th month when a 3-year period was applied). In addition, Gunasekarage and Power (2005) found that the month of the year seems to be affecting the overreaction findings.

(13)

13 portfolios that are formed based upon five-year returns. Chopra et al. (1992) also reported that the overreaction effect is fundamentally stronger for smaller firms in comparison with larger companies.

Eun and Huang (2002) looked for the financial anomalies on Chinese market. They found no evidence for the January effect. The tax-related selling is irrelevant in China’s case because there is no capital-gain tax in China. Therefore, as expected, their test results shown that there is no January effect or the February Chinese New Year effect, which was also confirmed by Zhang, Sun and Wang (2007). Instead, Zhang, Sun and Wang (2007) found a significant and positive March effect6. This finding reveals the political nature of financial anomalies in China. The March political window-dressing effect is plausibly caused by political maneuver by the Chinese government in March. The government tries to pop up stock prices to window dress the stock market performance with a view to preventing the possible outbreak of general resentment and to maintain social stability.

2.5. Other explanations of the overreaction hypothesis

Size effect as a possible explanation of the overreaction hypothesis has been reported in several research. Zarowin (1990), documented that the tendency for losers to outperform winners is rather due to the tendency for losers to be smaller-sized firms in comparison with winners, then due to investor overreaction. He reported that the size of the firm and the return on its stock are inversely related. Without controlling for size losers significantly outperform winners, however, when losers and winners of comparable size are adapted there is no evidence of this phenomenon. At the end Zarowin came to a conclusion that the loser vs. winner phenomenon hypothesized by De Bondt and Thaler appears to be another manifestation of the size phenomenon in finance. Also Clare and Thomas (1995) reported that such an overreaction may in fact be a manifestation of the small firm effect as losers tended to be small-sized companies.

Benou and Richi (2003) completed a study examining the long-run reversal pattern for a sample of large U.S. firms. Firms that qualified for this research experienced a significant stock price declines of more than 20 per cent during a specific month. Results presented by Benou and Richi were largely consistent with the overreaction hypothesis and significantly

(14)

14 greater in magnitude than those documented by previous studies. The stocks of large firms earned approximately 4 and 12 percent in excess of what was expected, respectively, six and twelve months after their initial price decline. Another finding was that the magnitude and trend of that reversal differs substantially across the industries - technology stocks appeared to posses the strongest reversal pattern.

Also Chiao and Hueng (2005) investigated firm size and the book-to-market ratio. After controlling these factors he reported that the overreaction effect on the Japanese market is significant and plays an important role in explaining the zero-investment returns on the loser-to-winner strategy.

Another explanation of the overreaction effect was presented by Conrad and Kaul

(1993). They argued that it might be explained by various factors such as: bid-ask errors, non-synchronous trading, or price discreteness. These factors lead to substantial spurious

returns to the long-term zero-investment contrarian strategies because the single-period returns are upwardly biased.

An interesting study was presented by Dissanaike (1997). The data (UK market) was restricted to nearly 1,000 larger and better-known listed companies, whose shares are more frequently traded. The main reason for such a restriction was to eliminate two of the three alternative explanations to the overreaction hypothesis. Firstly, this restriction minimizes the biases created by the bid-ask effects and the infrequent trading. Secondly, it reduces the possibility that reversals are primarily a small-firm phenomenon. Dissanaike also investigated the third possible explanation - that time-varying risk explains the reversal effect (losers were considerably more risky than the winners, thus accounting for their superior returns). He investigated the evidence in favour of the stock market overreaction hypothesis and reported that this evidence appeared to be largely consistent with the overreaction hypothesis.

(15)

15 the long-term price reversals that can be noticed in winner and loser stocks. This last conclusion came from the evidence that the long-term investment strategies based on size and especially price produce returns higher than those based on the past performance, and that losers tend to be low price and low market value firms.

2.6. Summary of the literature review

After De Bondt and Thaler presented their findings concerning the overreaction hypothesis, extensive number of studies were completed on this topic. However, as can easily be noticed from the review just presented, most of the studies have been conducted on the US or on other well developed stock markets. Significantly less attention has been paid to the emerging markets, despite the rapidly growing interest of major international investors. Therefore, I find it interesting to check, whether the overreaction phenomenon is present on those markets as well.

This master thesis tries to fill in the gap in literature focusing on 11 emerging markets. Analyzing research on emerging markets (Newton da Costa – Brazilian market, Zamri and Hussain – Malasian market, Wang et al. – Chinese market, Gunasekarage and Power – Sri Lankan market), with all found evidence of overreaction hypothesis, the priori expectations suggest that this phenomenon will also be confirmed in this paper.

Three important alternative explanations to the overreaction hypothesis have been advanced in the literature; two of them will be investigated in this research. Firstly, time-varying risk was asserted as an explanation (Chan, 1988 or Ball and Kothari, 1989). In this paper it will be examined if the differential risk that the losers were considerably more risky than the winners, thus accounting for their superior returns, is an explanation to the overreaction hypothesis. Secondly, it was reported that the overreaction effect is merely a reincarnation of the size effect. An explanation given was that it applies primarily to smaller and lesser-known companies (Zarowin, 1990), while market efficiency is thought to still hold for the larger companies listed on the exchanges. Due to the lack of data availability the size effect is not going to be estimated in this research7. Finally, it was argued that the overreaction effect might be explained by factors such as infrequent trading (Conrad and

(16)

16 Kaul, 1993). In this study dummies will be regressed against the differential portfolio to check for the seasonal effects.

The existing literature provides mixed findings while trying to explain the overreaction hypothesis phenomenon on different markets. Therefore, what I am likely to find testing for risk differences and seasonality remains unclear.

(17)

Table 1 Literature review

The first column names the authors who performed the research, the second column presents the market on which the research has been completed. The next two columns illustrate the time period of the research and portfolio duration. Last column presents main findings.

Authors Market Period Portfolio

duration Main findings

De Bondt and Thaler

(1985) USA 1926 - 1982 3 years

- overreaction of 24,6% (8,2% per year) - large positive returns in January De Bondt and Thaler

(1987) USA 1926 - 1982 3 years

- earnings of winning and losing firms show reversal patterns that are consistent with overreaction

- the winner-loser effect is not primarily a size effect, CAPM and the size-effect can not completely explain the overreaction

Chan (1988) USA 1929-1985 3 years - no evidence of overreaction as it can be explained by the CAPM that is free of the problems

caused by risk changes

Ball and Kothari (1989) USA 1930 - 1981 5 years

- overreaction of 15,65% (3,13% per year)

- substantial event-time changes in relative risk over one and five years - overreaction after controlling for risk changes less then 2,0%

Zarowin (1990) USA 1926-1978 3 years -_- size of the firm and the return on its stock are inversely related

CAPM and the January-effect can not completely explain overreaction

Power (1991) GBR 1973 - 1987 5 years -_- contrarian strategy can yield abnormal returns, even allowing for changes in risk

excess returns show no sign of disappearing for at least 5 years

Chopra et al. (1992) USA 1926 – 1986 5 years

- results consistent with a substantial overreaction effect

- overreaction in CAPM-adjusted returns of 2.5% to 9.5%, dependent on the method

employed

- overreaction is much stronger for smaller firms than for larger companies

Conrad and Kaul (1993) USA 1926-1988 3 years -overreaction might be explained by various factors such as: bid-ask errors,

non-synchronous trading, or price discreteness Newton da Costa

(1994) BRA 1970-1989 1 year

-the magnitude of the effect is more pronounced than in the U.S.

-differences in risk, as measured by CAPM-betas cannot account for the overreaction effect Clare and Thomas

(1995) GBR 1955-1990 1, 2 and 3 years

-losers manage to outperform winner over a two year period of time by a statistically significant 1,7% per annum

-overreaction may in fact be a manifestation of the small firm effect

Dissanaike (1997) GBR 1955-1975 3 to 5 years -after adjusting for risk differences and size effect an evidence in favour of the stock market

(18)

Baytas and Cakici (1999) USA, CAN, UK, JPN, GER, FRA, ITA, 1982-1991 1,2 and 3 years

-no evidence of overreaction hypothesis in the US, but returns from long-term contrarian strategies in other countries seem to be generally significant

-price and size effects might explain some of the long-term price reversals that can be noticed in winner and loser stocks

Mun et al. (1999) GER, FRA 1991-1996 1,2 and 3 years

- the highest contrarian profits are obtained in the short run and the profits decrease over time

- higher returns are not correlated to increases in the risk coefficients, which is consistent with investor overreaction

Zamri and Hussain

(2001) MYS 1986-1996 3 years

- results were consistent with patterns that might have been generated by long-term overreaction

- varying risk can not be an explanatory factor of overreaction

Benou and Richi (2003) USA 1990-2000 1,2 and 3 years

- results largely consistent with the overreaction hypothesis and significantly greater in magnitude than those documented by previous studies

- the magnitude and trend of that reversal differs substantially across industries

Iihara et al. (2004) JPN 1975-1997 1 month - significant return reversals dominating the Japanese markets

- momentum effect is not observed

Wang et al. (2004) CHN 1994-2000 20 weeks - many Chinese share returns exhibit patterns that are consistent with investor overreaction

- more pronounced mean-reversion in the market for A shares than for B shares.

Chiao and Hueng

(2005) JPN 1975-1999 5 years

- overreaction effect is significant and plays an important role in explaining the zero-investment returns

- firm size and the book-to-market ratio cannot fully explain stock returns on prior-return-based portfolios

Gunasekarage and

Power (2005) LKA 1989-2003 3 years

- investors overreact in this emerging market

- the overreaction is asymmetric; it is more pronounced for losers than for winners.

- even though the differences in firm size do not seem to influence the findings of the study, month of the year seems to be affecting the overreaction findings.

Zhang, Sun , Wang

(2007) CHN 1992-2003 n.i.

- no January effect or the February Chinese New Year effect

- significant and positive March effect caused by political maneuver by the Chinese government

(19)

19

3. Methodology and Hypothesis

In this section the methodology used in this paper will be discussed. I start by explaining the portfolio formation method. Secondly, the statistical tests for the overreaction hypothesis will be presented. This section concludes with the hypothesis formulation.

After becoming familiar with the existing literature on the overreaction hypothesis topic I decided to focus mainly on two papers. Therefore, the methodology is derived from the methodologies used by De Bondt and Thaler (1985), who examined the stock market overreaction hypothesis in finance for the first time, and by Andrew Clare and Stephen Thomas (1995), who using the past research were able to test for possible explanation of this phenomenon.

To perform the statistical analysis properly the right data is needed. Firstly, the monthly stock prices for the investigation period of all the companies listed on a specific stock exchange have been collected8. Since I am trying to explain the overreaction phenomenon, additionally market index and risk free rate data have been gathered. Based on the stock prices the returns have been computed and adjusted by market return. Thereafter, they have been ranked from high to low to form the portfolios. After having formed the portfolios for the whole investigation period I was able to perform the statistical tests.

3.1. Forming the winner and loser portfolios

The portfolios of stocks are based on the performance in the prior test period. This include the same standard event study techniques which were used by DeBondt and Thaler (1985). In order to obtain results comparable across the time three different time periods of portfolio creation (formation period) and monitoring (test period) are applied:

one year monthly data needed for portfolio creation followed by a one year period of portfolios’ performance test,

two year monthly data needed for portfolio creation followed by a two year period of portfolios’ performance test,

(20)

20

three year monthly data needed for portfolio creation followed by a three year period of portfolio’s performance test.

Stocks in portfolios are ordered according to their performance during the specific time period. The performance is measured relatively to the market index performance. For each month the market adjusted returns are calculated. This is presented in equation 1.

Uit = Rit - Rmt t = 1 . . . n (1)

where:

n = 1 … 12, or n = 1 … 24, or n = 1 … 36

(it depends on the length of test period; 12, 24 or 36 months), Uit is the adjusted return at

period t on stock calculated as a difference between Rit (return on stock at period t) and Rmt

(market return at period t).

On the basis of the market adjusted returns on all stocks it is possible to calculate the average returns for those stocks (ܴതi) as a mean of Uit over the period of t = 1 … n. Next these

average returns are ordered from high to low. Based on this ranking it is possible to form portfolios of winners and losers. The stocks in the first quintile are grouped together to form an equally weighted portfolio of winners (stocks with the highest average market adjusted returns)9. On the other hand, these stocks in the last quintile create an equally weighted portfolio of losers (stocks with the lowest average market adjusted returns)10. Using the equal weight gives the same weight, or importance, to each stock in a portfolio. This allows all of the companies to be considered on an even playing field. Average return of portfolio (ܴതp) over

the post portfolio formation period is computed based on formed portfolio of winners and losers11. This procedure is repeated for every year of the time research period using overlapping one, two and three-year period.

Calculating equation 1 the average stock returns for any sample length are obtained. De Bondt and Thaler (1985 and 1987) and Zarowin (1990) computed average stock returns over non-overlapping test periods. However, Chopra et al. (1992) and Ball and Kothari (1995) used overlapping test periods. The main argument for using non-overlapping test periods is that they ensure the independence in abnormal returns across securities. On the other hand,

9_{Quintile contains one-fifth (20%) of the sample population)}

10_{Using quintiles provides “the sternest test of the portfolio and because this approach will give a well} diversified portfolio which may give results of practical use to fund managers” (Clare and Thomas 1995) 11_(ܴ_ത

(21)

21 they cause a significant reduction in sample size, which reduces the power of statistical tests. Performing their study DeBondt and Thaler (1985 and 1987) and Zarowin (1990) used time period of approximately 50 years, however for emerging markets I use time period of maximum 21 years – the sample size reduction appears to be the case, and therefore, the overlapping test period is adopted in this paper. The consequence of an overlapping post-ranking periods is that the annual cross-sectional regression coefficient estimates are not independent through time. Therefore, to deal with this fact the standard errors have to be adjusted for this dependence – as a result the Newey and West (1987) correction is applied.

3.2. Statistical tests of the overreaction hypothesis

In order to find out whether losers are capable of outperforming winners during the specific period of time a new portfolio needs to be formed. This average portfolio can be called the difference portfolio (ܴതl-w), as it is created by subtracting the average return of

winners portfolio (ܴതw) from average return of losers portfolio (ܴതl)12. Equation 2 illustrates the

difference portfolio:

ܴതl-w = ܴതl – ܴതw (2)

where: ܴതl is the average return from losers portfolio, ܴതw is the average return from

winners portfolio.

When the return on the difference portfolio is insignificantly different from zero then the overreaction hypothesis can be rejected. On the other hand, a significant positive value of difference portfolio would be a confirmation of overreaction hypothesis.

The first statistical test (Test 1) is completed in order to compare the means of the winner and loser portfolio. A simple t-test on the significance of the constant α1 will inform if

there is a difference in the means of winner and loser stocks. The difference portfolio is regressed only against the constant α1 to examine if losers tend to have higher returns in

comparison with winners. This procedure is illustrated by equation 3:

12

(22)

22

ܴത (l-w)t = ܴഥl – ܴതw = α1 + εt t = 1 … n (3)

where: α1 is a constant and εt is an error term, n is equal to 12, 24 and 36 meaning one,

two and three-year test period respectively.

To confirm the overreaction hypothesis a significant and positive value for α1 is

needed.

The first regression gives only the answer to the question if losers beat the winners during the test period. However, the second regression (Test 2) is provided to examine whether some risk differences can affect the differential returns of losers and winners. ܴത(l-w)

stays as the regressand and is regressed against the market risk premium. This test controls possible different exposures to systematic risk. That may explain the differential returns between the winner and loser portfolios. Therefore ܴത(l-w)t can now be considered to be an

arbitrage portfolio, if beta is an appropriate measure of risk. The second test is presented in equation 4:

ܴത (l-w)t = α2 + β(RMt – RFt) + εt t = 1 … n (4)

where: α2 is a constant, β stands for the difference between the market betas of losers

and winners portfolios, RM is the return on the market index, RF represents the risk free rate13.

Similarly to the first regression, the positive and statistically significant α2 would be

interpreted as a proof of overreaction hypothesis. Furthermore, the statistically significant β informs that the differences in the systematic risk explain some of the variations in the ܴത(l-w)t.

If the β is positive, losers are characterized with higher systematic risk than winners.

3.3. Testing for seasonality

The month of January is well known for the presence of abnormal returns – so called The January Effect14. Therefore, the January effect on the differential portfolio will be

(23)

23 examined. A January dummy D1 is added to the second regression test. The third test (Test 3) will be performed and this is illustrated by equation 5:

ܴത (l-w)t = α3 + β2(RMt – RFt) + γ1D1 + εt t = 1 … n (5)

where: α3 is a constant, β stands for the difference between the market betas of loser

and winner portfolios, RM is the return on the market index, RF represents the risk free rate,

γi is an OLS coefficient, D1 is a January dummy and εt stands for the error term.

Interpretation of α and β remains the same as in previous tests. A positive and statistically significant α3 would give a proof of overreaction hypothesis. Positive β2 would

mean that the losers are characterized with higher systematic risk than winners. Additionally, a positive and statistically significant γ1 would confirm the January effect.

Afterwards, to test for additional seasonal effects dummy variables for other eleven months will be added. Thereafter, those dummies will be regressed against differential portfolio ࡾഥ(l-w)t . Dummy D1 stands for January, D2 for February, respectively, while D12

represents December. The regression is shown in equation 6.

ܴത (l-w)t = γ1D1 + γ2D2 + γ3D3 + γ4D4 + γ5D5 + γ6D6 + γ7D7+

+ γ8D8 + γ9D9 + γ10D10 + γ11D11 + γ12D12 + εt (6)

where: γi is an OLS coefficient and εt stands for the error term. A regression without

the constant term is applied and the number of dummies are exactly matching the number of months.

A positive (negative) and statistically significant γ would confirm that losers (winners)

did outperform winners (losers) in a specific month.

3.4. Hypothesis formulation

Previously completed research on emerging markets confirmed the evidence of long-term overreaction hypothesis phenomenon15. Therefore, I formulate the research hypothesis:

H: There exists a long term overreaction in emerging stock markets

(24)

24 If I confirm the above hypothesis two additional sub-hypotheses will be investigated:

Hsub1: Long-term overreaction in emerging stock markets can be explained by risk differences Hsub2: Long-term overreaction in emerging stock markets can be explained by seasonality

4. Data

This research is trying to confirm the overreaction hypothesis on eleven emerging markets. Having analyzed the evidence provided in financial literature it is likely that this phenomenon will also be found in this paper. In this section the data used in the research is presented. Thereafter, descriptive statistics are provided to give the reader some additional information.

4.1. Monthly stock market data

The data has been collected from Thomson Reuters Datastream. Performing the research the monthly stock market data is being used. The desired time period spreads from December 1987 to December 2008. This results in research period of 21 years. However, not all the stock exchanges existed in 1988, so for some emerging markets the research time period was reduced.

(25)

25 The arithmetic returns are being calculated in this study16. They were used in most of the past research on overreaction topic and using them will provide a better comparison of the results. The returns are computed based on the price data at the beginning of every month. To calculate the return for January the data from December, one year before, is necessary. Therefore stocks taken into formation of portfolio must have been listed on the stock exchange, not only during the whole formation and test period, but also one month before the formation period. De Bondt and Thaler (1985) in order to minimize the bias resulting from conditioning samples on prior data availability also required the stock to be listed during the entire period. As some of the new stocks were introduced to the market and some of them were delisted, the number of stocks listed on exchanges and used in this research varies throughout the study time period.

In order to calculate market adjusted returns the market indexes are needed, as the performance is measured relatively to the market index performance. In second statistical test, that controls for possible different exposures to systematic risk, the risk free rate is being used as one of the variables. Table 2 presents the sources of market index and risk free rate.

Tabel 2 Research additional information

The first two columns present the markets and stock exchanges used in this research. Third column shows the total number of stocks for which the price data has been collected. Fourth column presents the research time period. Last two columns indicate the sources of market index and risk free rate, both needed for performing the statistical tests.

Market Stock Exchange # of

stocks Time period Market Index Risk Free Rate

Argentina Buenos Aires SE 84 1994 – 2008 Argentina Merval Argentina Deposit 30D Brazil São Paulo SE 579 1992 – 2008 Brazil Bovespa Brazil Financing Overnight China Shanghai SE Shenzhen SE 848 613 1993 – 2008 1993 – 2008 Shanghai SE Composite

Shenzhen SE Composite China Time Deposit Rate 3M India Bombay SE 2147 1991 – 2008 India BSE (100) National India Bank Deposit 90-180D Indonesia Jakarta SE 403 1991 – 2008 Jakarta SE Composite Indonesia Deposit 3M

Mexico Mexican SE 313 1989 – 2008 Mexico IPC (Bolsa) Mexico CETES 91D Av. Ret.

Poland Warsaw SE 208 1996 – 2008 Warsaw SWIG80 Warsaw Interbank 3M

Russia Russian Trading

System 266 1996 – 2008 Russia RTS Index Russia Interbank 31-90D

S. Africa Johannesburg SE 940 1988 – 2008 FTSE W South Africa SA T-Bill 91 Days

S. Korea Korea SE 1074 1988 – 2008 Korea SE Composite Korea NCD 91 Days

Turkey Istanbul SE 391 1989 – 2008 ISE National 100 Turkey Interbank Overnight

16_{The arithmetic return as calculated as R} j, t =

୔ౠ,౪ ି ୔ౠ,౪షభ

୔ౠ,౪షభ

(26)

26 4.2. Descriptive statistics

The descriptive statistics of winners’ and losers’ portfolios for both formation and test period are presented in Appendix I. It shows the number of observations on specific market, each portfolio’s mean, median and standard deviation, as well as the maximum (minimum) portfolio increase (decrease). Additionally, for each of the portfolios included in the sample a test for normality is performed. This normality test consists of a skewness and kurtosis, together forming the ingredients for the Jarque-Bera test.

Analyzing the descriptive statistics I can expect that losers will outperform winners on most of the markets. As the test period is concerned the mean values of losers’ portfolio are higher than the values of winners’ portfolio. This is computed for 9 stock exchanges when 12 months test period is applied, 12 stock exchanges for 24 months test period, and 12 stock exchanges for 36 months test period. Concluding, I might be expecting that the winner-loser effect will be found and that the research hypothesis (H: There exists a long term overreaction on emerging stock markets) will not be rejected.

The Jarque-Bera test (JB) does not give an unequivocal answer whether the returns on

winner and loser portfolios are normally distributed. For most of markets they are normally distributed – the distributions have either slightly negative or positive skewness, also kurtosis varies a bit from the expected value of 3 but it all does not affect the result of normally distributed residuals. However, for 12 months test period high value (above 5,9) of JB test with a probability below 0,05 is obtained on 3 markets for either portfolio of winners or portfolio of losers, leading to a conclusion that the returns on winner or loser portfolios are not normally distributed. For 24 months test period the returns on winner or loser portfolios are not normally distributed on 4 markets. For 36 months test period all the returns on winner and loser portfolios are normally distributed.

(27)

27

5. Results

This section presents the results of the research testing the overreaction hypothesis on emerging markets. All the statistic tests were performed in Eviews 6. At the beginning the results from first regression are illustrated showing whether losers managed to outperform winners over the subsequent and similar period of time. Afterwards, the results from second regression are presented. They give an answer whether some risk differences can affect the differential returns of losers and winners. Finally, the results from seasonality test that tries to answer if abnormal returns can be explained by some calendar-related anomalies. This section concludes with a table that summarizes all the findings.

5.1. Did losers overreact ?

The first statistical test has been completed comparing the means of the winner and loser portfolio. The standard errors were adjusted for the regression coefficient estimate’s dependence through time using Newey and West (1987) correction17. The results are presented in table 3, 4 and 5 for one, two and three-year test period respectively.

Tabel 3 Test 1 results for 12 months formation and test period

Results for portfolios with 12 months formation and test period provide an answer whether losers managed to outperform winners within that time. Annualized return difference is obtained by multiplying the coefficient (average monthly return on difference portfolio) by 12 (number of months in a year).

n = 12 Argentina Brazil (Shanghai) China

China

(Shenzhen) India Indonesia

Observations 168 192 180 180 204 204 Return on loser 0,0114 0,0282 0,00369 0,0056 0,0197 0,0178 Return on winner 0,0034 0,0163 0,00370 0,0001 0,0182 0,0034 Test 1 α1 Coefficient 0,0080 0,0119 -0,00001 0,0055 0,0015 0,0144** Std.Error 0,0074 0,0133 0,0049 0,0042 0,0057 0,0065 t-statistic 1,0823 0,9002 -0,0017 1,2886 0,2551 2,2238

Annualized return difference 9,56% 14,34% -0,01% 6,56% 1,75% 17,22%

n = 12 Mexico Poland Russia South

Africa South Korea Turkey Observations 228 144 144 240 240 228 Return on loser -0,0053 0,0093 0,0532 0,0180 0,0127 0,0112 Return on winner -0,0006 0,0041 0,0138 0,0233 0,0030 -0,0017 Test 1 α1 Coefficient -0,0047 0,0052 0,0394*** -0,0053 0,0097* 0,0128* Std.Error 0,0048 0,0070 0,0115 0,0072 0,0057 0,0066 t-statistic -0,9971 0,7516 3,4448 -0,7400 1,6978 1,9488

Annualized return difference -5,70% 6,28% 47,34% -6,39% 11,67% 15,40%

*Significant at 10% level, ** Significant at 5% level, ***Significant at 1% level

(28)

28 As can be observed in the table in most cases 12-month losers managed to outperform 12-month winners. Only on three markets (Chinese – Shanghai, Mexican, South African) portfolios consisting 20% best performing stocks (winner portfolios) did not perform good enough to beat portfolio of 20% worst performing stocks (loser portfolios) over one year test period. The biggest annualized overreaction for 12 months formation and test period is found on Russian market (47,34%), as the average return on losers’ portfolio is 0,0532, while the average return calculated for the winners’ portfolio reported only as 0,0138. The value of winners underperformance on Russian market really marks out; α1 is positive with a

significant at 1% level coefficient equal to 0,0394. Statistically significant on the level of 10%

α1 was also estimated for Indonesian, South Korean and Turkish market. Biggest

underperformance of losers’ portfolio is found on the South African market – with winners beating the losers by an average of 6,39% per year.

Table 4 presents first regression test results for portfolios with two years (n=24) formation and test period.

n = 24 Argentina Brazil _(Shanghai)China _(Shenzhen)China India Indonesia

Observations 288 336 312 312 360 360 Return on loser 0,0139 0,0245 0,0058 0,0064 0,0225 0,0204 Return on winner 0,0031 0,0080 -0,0021 -0,0008 0,0136 0,0022 Test 1 α1 Coefficient 0,0108* 0,0165** 0,0079*** 0,0072*** 0,0089** 0,0183*** Std.Error 0,0056 0,0067 0,0023 0,0025 0,0041 0,0055 t-statistic 1,9191 2,4796 3,4130 2,8301 2,1989 3,2850 Annualized return difference 12,95% 19,85% 9,51% 8,61% 10,69% 21,93%

Africa South Korea Turkey Observations 408 240 240 432 432 408 Return on loser 0,0016 0,0202 0,0372 0,0236 0,0152 0,0175 Return on winner -0,0044 -0,0061 0,0180 0,0167 0,0031 -0,0034 Test 1 α1 Coefficient 0,0060 0,0263*** 0,0192* 0,0068 0,0121*** 0,0209*** Std.Error 0,0047 0,0063 0,0107 0,0067 0,0039 0,0065 t-statistic 1,2749 4,2053 1,7862 1,0130 3,0807 3,2099 Annualized return difference 7,18% 31,55% 22,99% 8,19% 14,55% 25,07%

(29)

29 As can be seen in a comparison with test 1 results for portfolios with one year formation and test period the results’ significance increased when two year formation and test period is used. A significant on at least 10% level winner-loser effect is found on 10 markets (compared with 4 markets for 12 months formation and test period). In addition, the winners’ outperformance of losers became greater in numbers. Only for the Russian market this outperformance decreased 47,34% for 12 months formation and test period in relation with 22,99% for 24 months formation and test period (results on annual basis). The findings are consisted with those obtained by De Bondt and Thaler (1985) for the US market and Clare and Thomas (1995) for the UK market.

Losers managed to outperform winners on all twelve stock exchanges. The average return on winners’ portfolio is negative on five markets, meaning the previously best performing stock did not match the average market return during the test period. All the winners’ portfolios achieved a positive average return, of which 10 are significant.

Next table shows the same regression test results. However, the formation and test period increased to three years (n=36).

n = 36 Argentina Brazil China

(Shanghai)

China

Observations 360 432 396 396 468 468 Return on loser 0,0142 0,0264 0,0048 0,0048 0,0229 0,0236 Return on winner -0,0004 0,0118 -0,0007 0,0001 0,0112 0,004 Test 1 α1 Coefficient 0,0146*** 0,0147** 0,0055*** 0,0047** 0,0117*** 0,0196*** Std.Error 0,0056 0,0071 0,0021 0,0024 0,0039 0,0061 t-statistic 2,6189 2,0580 2,6485 1,9678 3,0084 3,2396 Annualized return difference 17,54% 17,61% 6,62% 5,58% 14,05% 23,54%

Africa South Korea Turkey Observations 540 288 288 576 576 540 Return on loser 0,0032 0,0282 0,0345 0,0216 0,0147 0,0075 Return on winner -0,0044 -0,0057 0,0141 0,0127 0,0057 0,0039 Test 1 α1 Coefficient 0,0076** 0,0339*** 0,0205** 0,0088** 0,0090** 0,0036 Std.Error 0,0032 0,0098 0,0092 0,0043 0,0036 0,0049 t-statistic 2,4100 3,4656 2,2256 2,0421 2,5318 0,7293 Annualized return difference 9,17% 40,68% 24,57% 10,62% 10,84% 4,30%

(30)

30 As expected, losers outperformed winners on all the emerging markets analyzed in this study when three years were used for portfolio formation and test period. It is interesting to observe that the α1 is statistically significant on the level of 10% for all the markets except for

Turkish one (prob. 0,4661). Not surprising as the result obtained for Turkish market stands out compared with others - the overreaction decreased from 25,07% for 24 months formation and test period to just 4,30% for 36 months formation and test period. At seven stock exchanges losers’ outperformance became bigger compared with 24 months formation and test period. However, on both Chinese stock exchanges this outperformance decreased as the test period increased from two to three years.

Summarizing the first statistic test results it can be concluded the losers managed to outperform winners on emerging stock markets. Additionally, it can be stated that the three year formation and test period gives the best evidence for winner-loser effect with α1 being

statistically significant on the level of 10% on eleven out of twelve stock exchanges and the annualized return differences being the greatest in numbers from all other presented in this research. This finding is consistent with other papers that analyzed this topic. Next step is to try to explain what might be the reasons of the existence of this winner-loser effect.

5.2. Different exposures to systematic risk

Second regression test for possible different exposures to systematic risk has been completed. Once again, the standard errors were adjusted for the regression coefficient estimate’s dependence through time using Newey and West (1987) correction. The results are presented in table 6, 7 and 8 for one, two and three-year test period respectively.

The results from the second regression (test 2) for one year formation and test period (n=12) indicate that risk differences between the portfolios of losers and winners affect the differential portfolio Rl-w in most of the cases. A positive beta coefficient implying that losers

(31)

31 Tabel 6 Test 2 results for 12 months formation and test period

Test 2 checks for possible different exposures to the systematic risk as that may explain the differential returns between the winner and loser portfolios. Hsub1 - Long-term overreaction in emerging stock markets can be explained by risk differences, is being investigated. Hsub1 is accepted if the value of β coefficient is positive, and rejected if negative. Table presents test 2 results for portfolios with 12 months formation and test period.

(Shanghai)

China

Observations 168 192 180 180 204 204 Test 2 α2 Coefficient 0,0135 0,0127 0,0022 0,0069 -0,0113 0,0246** Std.Error 0,0086 0,0176 0,0057 0,0050 0,0079 0,0112 t-statistic 1,5767 0,7198 0,3964 1,3800 -1,4283 2,2030 β Coefficient 0,0581 0,0039 0,1087 0,0729 -0,1946** 0,0803 Std.Error 0,0540 0,1059 0,0991 0,0631 0,0807 0,0638 t-statistic 1,0765 0,0372 1,0967 1,1544 -2,4121 1,2599 R-squared 0,0125 0,00001 0,0373 0,0189 0,0617 0,0147

Africa South Korea Turkey Observations 228 144 144 240 240 228 Test 2 α2 Coefficient 0,0038 0,0057 0,0317** -0,0083 0,0153* 0,0012 Std.Error 0,0075 0,0143 0,0142 0,0129 0,0087 0,0098 t-statistic 0,5153 0,3993 2,2407 -0,6424 1,7680 0,1205 β Coefficient 0,0553 0,0045 -0,0466 -0,0271 0,0781 -0,0208 Std.Error 0,0410 0,0918 0,0462 0,0917 0,0644 0,0158 t-statistic 1,3515 0,0492 -1,0082 -0,2958 1,2121 -1,3163 R-squared 0,0116 0,00004 0,0075 0,0003 0,0071 0,0289

(32)

Test 2 checks for possible different exposures to systematic risk as that may explain the differential returns between the winner and loser portfolios. Hsub1 - Long-term overreaction in emerging stock markets can be explained by risk differences, is being investigated. Hsub1 is accepted if the value of β coefficient is positive, and rejected if negative. Table presents test 2 results for portfolios with 24 months formation and test period.

(Shanghai)

China

Observations 288 336 312 312 360 360 Test 2 α2 Coefficient 0,0117 0,0285*** 0,0082*** 0,0083*** -0,0020 0,0197** Std.Error 0,0079 0,0110 0,0024 0,0027 0,0061 0,0089 t-statistic 1,4864 2,5978 3,3439 3,1116 -0,3332 2,2045 β Coefficient 0,0098 0,0600** 0,0211 0,1271*** -0,1737*** 0,0114 Std.Error 0,0542 0,0293 0,0388 0,0380 0,0646 0,0488 t-statistic 0,1809 2,0492 0,5443 3,3466 -2,6892 0,2344 R-squared 0,0003 0,0052 0,0015 0,0598 0,0415 0,0003

Africa South Korea Turkey Observations 408 240 240 432 432 408 Test 2 α2 Coefficient 0,0009 0,0460*** 0,0162 0,0061 0,0163** 0,0186** Std.Error 0,0071 0,0091 0,0124 0,0120 0,0065 0,0072 t-statistic 0,1308 5,0805 1,3022 0,5057 2,4920 2,5819 β Coefficient -0,0342 0,2448*** -0,0224 -0,0071 0,0606 -0,0040 Std.Error 0,0289 0,0708 0,0413 0,0744 0,0496 0,0076 t-statistic -1,1842 3,4552 -0,5415 -0,0954 1,2222 -0,5295 R-squared 0,0029 0,0793 0,0009 0,00002 0,0053 0,0006

(33)

Test 2 checks for possible different exposures to systematic risk as that may explain the differential returns between the winner and loser portfolios. Hsub1 - Long-term overreaction in emerging stock markets can be explained by risk differences. Hsub1 is accepted if the value of β coefficient is positive, and rejected if negative. Table presents test 2 results for portfolios with 36 months formation and test period.

(Shanghai)

China

Observations 360 432 396 396 468 468 Test 2 α2 Coefficient 0,0164* 0,0359*** 0,0054** 0,0050* 0,0071 0,0159 Std.Error 0,0088 0,0128 0,0023 0,0026 0,0068 0,0097 t-statistic 1,8566 2,7977 2,3818 1,9401 1,0500 1,6404 β Coefficient 0,0183 0,1142*** -0,0144 0,0372 -0,0742 -0,0286 Std.Error 0,0593 0,0375 0,0374 0,0398 0,0656 0,0438 t-statistic 0,3087 3,0441 -0,3843 0,9345 -1,1316 -0,6527 R-squared 0,0010 0,0090 0,0006 0,0044 0,0080 0,0013

Africa South Korea Turkey Observations 540 288 288 576 576 540 Test 2 α2 Coefficient -0,0021 0,0669*** 0,0162 0,0106 0,0138** 0,0038 Std.Error 0,0052 0,0098 0,0124 0,0092 0,0057 0,0066 t-statistic -0,4065 6,8205 1,3022 1,1539 2,4333 0,5750 β Coefficient -0,0644** 0,5203*** -0,0224 0,0171 0,0724 0,0003 Std.Error 0,0270 0,0587 0,0413 0,0633 0,0451 0,0051 t-statistic -2,3883 8,8697 -0,5415 0,2711 1,6051 0,0643 R-squared 0,0131 0,1463 0,0009 0,0002 0,0069 0,00001

I now try to answer the question, whether winner-loser effect exist after implying the second regression. Performing test 1 significant on at least 10% return differences were reported on 4 markets as 12 months formation and test period were used. However, after including risk differences significant results are found only for Indonesian, Russian and South Korean markets. The winner-loser effect on Turkish market is not significant anymore. As 24 months formation and test period are considered, 10 stock exchanges are characterized with significant winner-loser effect, however after including risk factor significant results are obtained only for 7 stock exchanges: Sao Paulo SE, Shanghai SE, Shenzhen SE, Jakarta SE, Warsaw SE, Korea SE and Istanbul SE. For 36-month formation and test period the number of stock exchanges, on which the significant return differences were reported decrease from 11 after test 1 is performed to 6 after risk differences are controlled. Significant results are computed for Argentinean, Brazilian, Chinese (Shanghai and Shenzhen), Polish and South Korean markets.

(34)

34 analyzed in this paper. However, positive and statistically significant beta values implying that losers may embody more systematic risk than winners were found for Polish market for 24 and 36 months formation and test period.

Besides Polish market also on Argentinean and Chinese (Shenzhen SE) markets positive beta coefficient values for 12, 24 and 36-month formation and test period are reported, though not statistically significant. On the other hand, Indian and Russian markets are characterized with a negative beta coefficient value in all three cases implying that losers does embody less systematic risk then winners. Therefore, it is hard to unequivocally say whether the differences in risk account for the overreaction effect the emerging markets.

5.3. Testing for seasonality

(35)

Test 3 checks for the different exposures to systematic risk and for the presence of The January Effect trying to explain the differential returns between the winner and loser portfolios. Hsub2: - Long-term overreaction in emerging stock markets explained by seasonality, is being investigated. A positive and statistically significant

γ1 would confirm January effect. Table consist of three panels – Panel A presents the results for portfolios with 12-month formation and test period, Panel B presents the results for portfolios with 24 months formation and test period and Panel C presents the results for portfolios with 36 months formation and test period.

Panel A

n=12 Argentina Brazil China

(Shanghai)

China

Observations 168 192 180 180 204 204 Test 3 α3 Coefficient 0,0119 0,0143 0,0053 0,0094* -0,0155* 0,0233** Std.Error 0,0089 0,0180 0,0057 0,0051 0,0080 0,0117 t-statistic 1,3341 0,7917 0,9331 1,8342 -1,9485 1,9954 β2 Coefficient 0,0556 0,0043 0,1012 0,0647 -0,2089** 0,0785 Std.Error 0,0567 0,1058 0,0989 0,0619 0,0805 0,0644 t-statistic 0,9802 0,0405 1,0234 1,0442 -2,5945 1,2192 γ1 Coefficient 0,0162 -0,0185 -0,0385** -0,0324** 0,0388* 0,0123 Std.Error 0,0390 0,0216 0,0187 0,0140 0,0228 0,0163 t-statistic 0,4140 -0,8561 -2,0616 -2,3106 1,7040 0,7505 R-squared 0,0151 0,0007 0,0659 0,0415 0,0802 0,0164

Africa South Korea Turkey Observations 228 144 144 240 240 228 Test 3 α3 Coefficient 0,0036 0,0095 0,0334** -0,0048 0,0149* -0,0022 Std.Error 0,0074 0,0150 0,0153 0,0132 0,0090 0,0098 t-statistic 0,4834 0,6377 2,1833 -0,3661 1,6599 -0,2258 β2 Coefficient 0,0553 0,0095 -0,0462 -0,0177 0,0780 -0,0212 Std.Error 0,0411 0,0922 0,0471 0,0926 0,0644 0,0159 t-statistic 1,3465 0,1029 -0,9815 -0,1908 1,2114 -1,3333 γ1 Coefficient 0,0030 -0,0401 -0,0188 -0,0290 0,0047 0,0377* Std.Error 0,0136 0,0260 0,0229 0,0179 0,0126 0,0221 t-statistic 0,2177 -1,5434 -0,8183 -1,6260 0,3743 1,7065 R-squared 0,0117 0,0153 0,0087 0,0052 0,0073 0,0417 Panel B

(Shanghai)

China

The Overreaction Hypothesis on Emerging Markets