Index Shocks, Pricing Anomalies and Long-horizon Return Predictability in the Japanese Stock Market

Transcription

1 Index Shocks, Pricing Anomalies and Long-horizon Return Predictability in the Japanese Stock Market Pyemo N. Afego November 1, 2016 Abstract We investigate the extent to which the long run return behaviour of stocks added to and deleted from Japan s Nikkei 225 index can be described as having a predictable component. Contrary to US findings for the S&P 500, we document evidence that stocks added to the Nikkei 225 do not outperform the market average in the long term. Our results further show that the subpar long term performance of new additions is attributable to strong price reversals experienced by the firms 1 month after their inclusion to the index. Finally, contrarian strategies that buy new deletions and sell the additions appear profitable in the long term even after adjusting for size and book-to-market risk factors. Key Words: Return Predictability, Nikkei 225, Abnormal returns, Contrarian strategy JEL Classification: G10, G12, G14 1 Introduction The idea that prices of tradeable financial assets should be arbitrage-free is a central principle in asset pricing. An arbitrage-free market is one in which it is assumed that asset prices reflect all publicly available information (Fama, 1970), leaving little or no room for systematically positive or negative riskadjusted returns to be earned after new information about an asset has been revealed to the market. This implies that the price response of an asset to new Graduate School of Economics, Kyushu University Hakozaki, Higashi-ku, Fukuoka JAPAN. pyemo.afego@yahoo.com 1

2 information should be immediate such that returns on the asset should not be predictable beyond the date of the announcement. The literature studying the effects of index composition changes (or index shocks) on long-run abnormal return performance emerged only recently. Chan et al., (2013), using US data, investigate the long-run performance of stocks added to and removed from the well-known S&P 500 index from 1962 to They find that both added and deleted stocks outperform the market benchmark in the long run. Chan et al., used a three- to five-year horizon to evaluate the long run post-revision price performance of the event firms within the framework of the buy-and-hold abnormal returns (BHAR) and the calendar time portfolio approach. A key assumption in the traditional BHAR approach however is that the return distributions are normal and symmetrical (non-skewed). Consequently statistical and econometric techniques that evaluate the significance of longrun abnormal performance based on the Central Limit Theorem normality assumption tend to be biased (Dionysiou, 2015). Furthermore, while errors and misspecifications in risk adjustment may not matter much in short-run event studies, they can significantly bias estimates of abnormal performance in longrun event studies. Additionally, when the events are clustered in calendar time, as index recompositions tend to be, cross-correlations in abnormal returns yield biased test statistics which become amplified in long-horizon analyses (Petersen 2009). Extant studies have so far relied on inferences which assume normality and zero skewness of the distribution of the (buy-and-hold) abnormal returns measured over long-horizons. Furthermore no study, to the best of our knowledge, has examined the long-run return dynamics of changes to a major non-us stock index. Against this background, we explore the extent to which the long-run performance of stocks added to and deleted from Japan s Nikkei 225 index can be described as having a predictable component. Specifically, we investigate the nature of the price behaviour of stocks following changes to the status of their membership of the Nikkei 225 index. Furthermore, we investigate whether strategies that systematically beat the market can be designed around the return dynamics of the affected firms. The use of Nikkei 225 index data provides a unique setting to examine the long-run post-announcement price behaviour of new additions and deletions on an index for a number of reasons. First, the Nikkei 225 index is amongst the world s most widely followed stock indices. Second, and more importantly, changes to the membership of the index are based on a set of pre-defined rules which are in the public domain. Thus, in contrast to the S&P 500 index which does not follow a rigid set of rules and where membership is largely at the 2

3 discretion of the Index Committee, revisions to the Nikkei 225 are in principle non-informational, and hence should not have a significant effect on the valuation of the affected stocks. Our approach and our findings differ from those of prior studies in several respects. First, we adopt empirical strategies and test methodologies which directly address a few of the specification and inferential issues associated with long-horizon event studies. Second, we locate our study in a non-us (Japanese) setting. This approach not only provides new insight as to whether or not institutional or country-specific differences matter in driving the long-run return behaviour of index additions and deletions, but also stimulates debate on the generalizability or otherwise of US findings. Third, we limit the horizon length for evaluating long-term performance to a maximum of 3 years. Taking this step allows us to minimize the magnitudes of biases of the abnormal return estimates caused by misspecification. In contrast to Chan et al., (2013) we find that stocks added to the Nikkei 225 do not outperform the market average in the long term. The below-average long term performance of additions is caused by strong short term return reversals. Further, we find that a strategy which goes long in deletions and shorts the additions yields statistically significant profits in the intermediate- to long-term, even after adjusting for size and book-to-market risk factors. In a broader sense, our findings are consistent with Iihara et al., (2004) who document evidence that return reversals are prevalent in the Japanese stock market especially over the 1 month horizon; we provide further evidence that the 1-month return reversals tend to have a long-lasting effect. Our findings also echo a recent paper by Fama and French (2012) who study the size, value and momentum patterns in average returns in North America, Europe, Japan and Asia-Pacific, and find no evidence whatsoever of momentum in Japanese returns. The remainder of the paper is organised as follows. Section 2 presents the institutional framework governing membership of the Nikkei 225 index. Section 3 describes the test methodologies and empirical strategies employed. Section 4 discusses the data selection process and descriptive statistics of the event firms. Section 5 presents and discusses the empirical results for the long-term abnormal performance of the event firms. The section also explores the possibility of designing simple trading strategies that exploit the observed return patterns. Section 6 concludes. 3

4 2 Institutional Setting To aid an understanding of the data and our underlying hypotheses, we provide a brief overview of the institutional arrangements and framework governing membership of the Nikkei 225 index. The Nikkei 225 index is compiled and managed by the Nihon Keizai Shimbun (Nikkei Inc.), a leading Japanese media company, and comprises of two hundred and twenty five (225) of the most liquid blue-chip stocks listed on the first section of the Tokyo Stock Exchange (TSE). Excluded from the index are non-ordinary shares such as Real Estate Investment Trusts (REITs), Exchange Traded Funds (ETFs) and preferred stocks. As at December 2014, the Nikkei 225 index covered 6 major sectors - Technology, Financials, Consumer goods, Materials, Capital goods/others, and Transportation and Utilities - representing about 64% of the total market value of the entire first section of the TSE or approximately $4.49 trillion. 1 Changes in the Nikkei 225 index constituent stocks fall under two main category types: Periodic review and Extraordinary replacement. 2 With periodic review, which occurs once a year (at the beginning of October), changes to the index take into account liquidity of stocks and the balance of the sectors in order to avoid over- or under-representation of any one sector. On this basis therefore, stocks with low liquidity may be removed from the index and replaced with stocks with high liquidity selected from the list of the top 450 liquid stocks on the first section of the TSE. Extraordinary replacement on the other hand caters for deletions and additions in response to developments such as mergers, bankruptcy, corporate restructuring, delistings or transfer to the second section of the TSE. Also, securities under supervision, investigation or monitoring may be dropped from the index under this framework. Replacements for such deleted stocks usually come from stocks within the sector with the highest liquidity which have not been previously included in the Nikkei 225 index. However in the case of a merger between a constituent and non-constituent stock, the stock of the surviving company replaces the delisted stock. Since revisions to the Nikkei 225 index are based on the pre-defined set of rules highlighted above, all of which are public knowledge, the inclusion of removal of stocks from the index are in principle non-informational and, for the most part, reflect nothing more than jumps in liquidity. If this is true, then revisions to the index should not lead to a significant change in the valuation of new additions or deletions guidebook_en.pdf 4

5 By contrast, membership of the S&P 500 index does not follow a rigid set of rules. In a recent article published on the S&P Dow Jones Indices blog, the Chairman of the S&P 500 Index Committee, David Blitzer, describes the S&P 500 index in the following terms: Unlike many other S&P Dow Jones Indices and the majority of indices offered by other index providers, there are no rigid or absolute rules for the S&P 500; the Index Committee have some discretion in selecting stocks or responding to market events (August 7, 2014). 3 Furthermore, the pool from which replacements to the index are drawn is undisclosed (Elliott et al., 2006). In light of this, membership of the S&P 500 index appears largely subjective and may be seen to convey information about the future prospects of stocks. Thus, adding (dropping) a stock to (from) the index should be expected to have a positive (negative) valuation effect. 3 Methodology Two main methods have traditionally been used to estimate long-run performance in event studies. These are the buy-and-hold abnormal returns approach and the calendar-time portfolio approach. 3.1 Buy-and-Hold Abnormal Returns The buy-and-hold abnormal returns (BHAR) method is computed as the difference between the returns of the event firms and the return on a benchmark. Because the returns are compounded, the BHAR is adjudged to provide a more realistic estimate of returns from the perspective of the investor. Furthermore, the BHAR approach is robust to microstructure biases associated with frequent portfolio rebalancing (Kothari and Warner, 2007). In addition, Loughran and Ritter (2000) notes that the BHAR method can achieve higher statistical power to detect abnormal returns relative to the calendar-time portfolio method. A major shortcoming of the BHAR approach, however, is that it is prone to bias in the presence of cross-sectional dependence induced by the overlapping long horizon return observations. The BHAR is computed as follows: T T BHAR i (t, T ) = (1 + R it ) (1 + R bt ), (1) t=1 where BHAR i is the buy-and-hold abnormal return for event firm i, R it is the 3 Inside the S&P 500: An Active Committee : Available on the S&P Dow Jones Indices blog: inside-the-sp-500-an-active-committee/ t=1 5

6 rate of return for event firm i on month t, R bt is the rate of return for the benchmark (market index) on month t and the holding period T ranges from 6 months (half a year) to 36 months (3 years) Skewness bias Skewness bias, as Kothari and Warner (2007) noted, is one of the inferential issues associated with long-horizon event studies. Because individual securities have long term buy-and-hold returns that are positively-skewed, the BHARs in turn tend to have positively skewed distributions. Consequently, the sampling distribution of the conventional t-statistic will lack symmetry, potentially leading to biased inferences regarding the long-term BHARs. 4 To address this problem, we compute the bootstrapped skewness-adjusted t-statistic proposed by Lyon et al., (1999) with statistical inference based on the empirical distribution from the bootstrapped resamples. The Lyon et al., (1999) statistic corrects for extreme skewness and is specified as follows: where T skew adj = n(s ˆγS2 + 1 ˆγ), (2) 6n S = BHAR(T 1, T 2 ) ˆσBHAR, (3) n i=1 ˆγ = 1, T 2 ) BHAR(T 1, T 2 )] 3 ˆσ 3, nbhar (4) and ns is the standard t statistic while ˆγ is the estimated skewness coefficient. From the original sample, bootstrap resamples are generated and the same t-statistic on each resample is calculated as follows: T b skew adj = n b (S b ˆγ b S b n b ˆγ b ), (5) where S b = BHARb (T 1, T 2 ) ˆσ b BHAR, (6) and ˆγ b = n b i=1 [BHARb (T 1, T 2 ) BHAR b (T 1, T 2 )] 3 ˆσ b3. (7) n b BHAR Critical values on which statistical inference is based are then calculated from the b resamples. The bootstrap method we employ in this study follows the 4 See for example Fama, 1998; Mitchell and Stafford, 2000; Kothari and Warner,

7 procedure outlined in Lyon et al., (1999). First, we draw 1000 bootstrapped resamples from the original sample. Second, we compute the skewness-adjusted t-statistic for each of the 1000 resamples. Finally, we generate the two critical values from the 1000 bootstrapped resamples, and these form the basis of our inference on whether or not the skewness-adjusted t-statistic is statistically significant The bad model problem The bad model problem refers to the situation whereby tests of abnormal return performance must involve a simultaneous test of a model of expected returns which is unlikely to be error-free particularly when the event sample consists of small firms. 5 The problem, as discussed in Fama (1998), is more serious in long-term returns because the errors in expected returns (due to the bad model) become amplified with horizon length. 6 As a remedy, Fama advocates the use of value-weight rather than equal-weight returns. To this end, we compute both equal-weight and value-weight BHARs; the mean value-weighted BHAR is computed as the weighted average of the individual BHARs where the weight on each event firm is determined by its market capitalization in the event month Cross-sectional Dependence Although the Lyon et al., (1999) bootstrapped approach has been argued to limit the severity of dependencies in the return data, the claim that it adequately addresses the dependence problem has been called into question. Furthermore inference for tests of post-event abnormal performance, assuming independence, typically become increasingly less unbiased as the horizon length extends beyond 3 years (Cowan and Sergeant, 2001; Kothari and Warner, 2007). Bearing in mind these inferential caveats, and since our goal is to reliably measure the long-term performance of stocks added to and deleted from the Nikkei 225 index, we limit the horizon length for evaluating the long-term performance to a maximum of 3 years. Still, for reference, we compute the estimates for the 5-year horizon. Furthermore, we augment the analysis with the calendar-time portfolio approach which should produce more precise and robust estimates of long-run abnormal performance in the presence of cross-dependence and additional risk factors, chief among which are the Fama and French (1993) size and value factors and Cahart s (1997) momentum factor. 5 See for example Fama, 1998; Kothari and Warner, 2007 and Dionysiou, The bad model problem is particularly severe under the BHAR framework because of the effect of compounding. 7

8 3.2 Calendar Time Portfolio Approach As mentioned earlier, controlling for cross-sectional dependence in the abnormal return series is particularly crucial in long-horizon event studies. Although both the BHAR and calendar-time portfolio approach are prone to misspecification, the latter approach has the advantage of accounting for cross-correlations in abnormal returns (Fama, 1998; Jeng, Metrick and Zeckhauser, 2003; Kothari and Warner, 2007). The calendar-time portfolio approach basically involves two steps: (i) For each month and for each post-event holding period T (where T = 6, 12, 18, 24, 36 or 60 months), we compute the excess return for each stock that was added to the Nikkei 225 index within the previous T months. We then calculate the mean excess returns, on a month-by-month basis, across all stocks that were added to or deleted from the index in the previous T months. Each month, the portfolio is rebalanced so that all stocks that are part of the portfolio at the end of T -month period are dropped and stocks whose addition or deletions are announced are subsequently included. This means that stocks of event firms are part of a portfolio only over the period corresponding to the T -month holding period of interest. (ii) The monthly portfolio excess returns computed in step (i) above are then regressed on selected risk factors to obtain an estimate of the mean monthly portfolio abnormal return based on the coefficient of the alpha. Since the number of stocks that make up a portfolio varies overtime due to monthly rebalancing, we employ the weighted least square (WLS) estimator in order to correct for any heteroscedasticity in the error term of the regression model, as suggested by Lyon et al., (1999). 7 Also, we compute the valueweighted portfolio return which, according to Fama (1998), limits the severity of the bad model problem. Apart from accounting for cross-correlation, the distribution of the estimator in the calendar-time portfolio method is approximately normal (Mitchell and Stafford, 2000). Furthermore, while risk adjustment in the BHAR method considers only the broad market (systematic) risk, the calendar-time portfolio approach we employ considers additional risk factors, namely the Fama-French size and book-to-market (B/M) factors which are designed to capture the relationship between returns and size on one hand, and returns and B/M on the other, respectively. 7 The weights in this case are based on the number of stocks in each portfolio each month 8

9 The Fama-French multi factor risk model we employ is specified as: R it R F t = a i + b i (R M t R F t ) + s i SMB t + h i HML t + e it. (8) In the above regression equation, R it is the return (equal- or value-weighted) in month t for the portfolio of stocks that were added to or removed from the Nikkei 225 within the previous T months (where T = 6, 12, 18, 24, 36 or 60); R F t is the return on a risk-free asset (proxied by the Bank of Japan discount rate); R M t is the return on the market portfolio constructed using the universe of stocks listed on the Tokyo Stock Exchange, JASDAQ 8 and the Osaka Stock Exchange; SMB t is the average return on a diversified portfolio of small stocks minus the average return on a diversified portfolio of big stocks; HML t is the average spread in returns between portfolios of value stocks (comprising of high B/M stocks) and portfolios of growth stocks (comprising of low B/M stocks); the intercept, a i, is the portfolio s alpha and provides an estimate of the portfolio s monthly riskadjusted return while b i, s i and h i are the factor loadings (or exposures) which indicate how much of the portfolios return are associated with the respective risk factors. 9 The zero-intercept hypothesis, according to Fama and French (2015), is rooted in the idea that if the exposures to the three risk factors capture all variation in expected returns, the intercept a i is zero for portfolio i Data We examine the long-run price behaviour following changes to the Nikkei 225 index from March 1970 to September Data on changes in the composition of the index were pulled from the Nihon Keizai Shimbun (Nikkei Inc.). Monthly data on the individual constituent stock prices, their market value and the market benchmark TOPIX were obtained from the Nikkei Economic Electronic Database Systems (NEEDS). A total of 294 stocks (firms) were added to or removed from the Nikkei 225 index during the sample period. 11 From this total, we drop all firms 8 Japan Association of Securities Dealers Automated Quotation 9 We thank Professor Stefano Marmi for providing the RM-RF, SMB, HML and WML factors for Japan; (accessed on 25th March 2016). 10 In the BHAR analysis, we employ the TOPIX as a proxy for market return. In the calendar time portfolio approach, the proxy for market return was computed in the spirit of Fama and French (1993) using the universe of stocks listed on the Tokyo Stock Exchange, Osaka Securities Exchange and the JASDAQ Securities Exchange. To check for consistency between the TOPIX as one market proxy and the Fama-French universe as another, we test for their correlation. The pairwise correlation coefficient between the two proxy variables is Prior to April 2000, the maximum number of stocks that could be added or removed from the index in a single review period was capped at 6. This partly explains why the total 9

10 that did not have sufficient return data over the test period (at least 1 month of post-change return data is required). We also exclude all firms that emerged as a result of bankruptcy and or the creation of holding companies. Based on these screens, we employ a final sample comprising of a total of 140 event firms (88 additions and 52 deletions) over the study period. 4.1 Descriptive Statistics for Event firms Descriptive statistics for the final sample of event firms, comprising of both additions and deletions, are presented in Table 1. With respect to firm size, it is evident that added firms, measured by their market value, are much larger than deleted firms. For example the mean (median) market value for added firms stands at JPY 913 billion (JPY 543 billion) compared to JPY 42 billion (JPY 20 billion) for deleted firms. [Table 1 around here] Also, the return performance of added stocks, over the 12 and 36 months prior to the event month, is positive while that of deleted stocks is negative. Specifically, added stocks earn an average of 18% and 44% over the previous 1 and 3 years respectively prior to their inclusion to the index whereas deleted stocks produce an average return of -12% and -29%, over the same period. The standard deviation for added stocks over the 1- and 3-year horizons is 0.69 and 1.25 respectively, slightly higher than the 0.26 and 0.45 recorded for deleted stocks. The estimates for the standard deviation appear to increase with horizon length for both additions and deletions, nearly doubling in magnitude in both cases. There is also evidence of positive skewness in both samples, suggesting that the distribution of the long-horizon returns for both additions and deletions are asymmetric and non-normal with a longer right tail. 5 Empirical Results 5.1 Market-adjusted Buy-and-Hold Abnormal Returns Additions This section reports the market-adjusted long-term buy-and-hold abnormal returns for added stocks over holding periods ranging from 6 to 60 months. The performance of the buy-and-hold returns are adjusted against returns on the Tokyo Stock Price Index (TOPIX) which is employed as a proxy for the market number of constituent changes, and hence our sample size, is relatively small. 10

11 average. Both equal-weighted and value-weighted BHARs are computed and their estimates tested for significance using the bootstrapped skewness-adjusted t-statistic of Lyon et al., (1999). [Table 2 around here] Table 2 reports the results of the of the long-run post-event performance of added stocks. Panel A of the table shows that the equal-weighted BHARs are negative over all holding periods except the 60 month, where the mean difference in returns between the added stocks and the market average is positive but statistically indistinguishable from zero. The underperformance of added stocks relative to the benchmark is significant over the 18- and 24-month investment horizon. The significance of the underperformance of the added stocks over the 18- and 24-month horizons remain robust with value weighting (Panel B). Results of the BHARs based on equal-weighting show that while added stocks earn significant positive BHARs of 9.8% (or 6.7% with value weighting) in the announcement month [0], the gains are reversed in the following month [1] as they deliver significant negative BHARs of -2.7% (or -3.1% with value weighting). The significant return reversal that occurs between the announcement month [0] and month [1] is consistent with the finding in Iihara et al., (2004) that return reversals are prevalent in Japanese stock markets especially over the 1 month horizon. Our results further suggest that the 1-month return reversals tend have a long-lasting effect, persisting beyond 36 months. On the whole, our results, based on value-weighting (Table 2 Panel B) indicate that stocks added to the Nikkei 225 significantly underperform the Tokyo Stock Price Index (market average) over a 36-month holding period by -9.2% (tstat -2.04). This is contrary to U.S findings, documented in Chan et al., (2013), that stocks added to the S&P 500 outperform the CRSP value-weighted market index by 6% (t-stat 1.59) and the Dow Jones Industrial Average by 19%, over the same (36-month) holding period Deletions Table 3 reports the BHARs for the sample of deleted stocks over the 6-month to 5-year horizon. [Table 3 around here] Panel A of the table shows that deleted firms significantly outperform the market over all horizons under the equal weighting scheme. Panel B on the other 11

12 hand shows that value-weighting substantially reduces both the magnitude and significance of the BHARs for deleted firms. The 3-year BHAR for example shrinks from a highly statistically significant 52.5% with equal weighting, to an insignificant 17.9% with value weighting. On the whole the results for deletions, which as Table 1 reveals are mainly small stocks, show that value-weighting substantially reduces the magnitude and statistical significance of abnormal performance, consistent with the argument by Fama (1998) and Dionysiou (2015) regarding the amplified misspecification bias inherent in equal weighted schemes involving small stocks. Also evident from Table 3 is the fact that deleted stocks also experience short-term reversals. The average BHAR in the announcement month is statistically significant at -16.0% (or -12.1% with value weighting) while in the following month the estimates are positive, averaging a statistically significant 9.2% (or 7.3% with value weighting). Overall, the short-term return reversals appear to be stronger for deletions than for additions, and judging from the size characteristics of the event firms (see Table 1), this finding could be interpreted to mean that the reversal effect is stronger for small stocks relative to large stocks. To demonstrate the behaviour of the BHARs for the newly added and deleted stocks, we plot in Fig 1. the BHARs for both additions and deletions from the announcement month (month 0) to the subsequent 60 months. [Figure 1 around here] The strong reversals in the 1st post-announcement month (i.e. month 1) can be clearly observed; the reversals in that month impact the behaviour and performance of both additions and deletions up to the 60th post-announcement month. 5.2 Results of the Calendar Time Portfolio Method - Fama and French (1993) Three-factor Model Additions Panel A of Table 4 shows that the alphas for the equal-weighted portfolios for added stocks are negative but statistically insignificant for all investment horizons; the alphas range from -0.87% to -0.24% and are generally close to the BHAR estimates which, when normalized to monthly returns (see Table 8), range from -0.60% to 0.00%. Value-weighting (Panel B), as expected, reduces the magnitude of the alphas. 12

13 [Table 4 around here] On the whole, the calendar time portfolio alphas are consistent with the BHAR estimates in suggesting that additions do not earn outperform the market over the long run. With respect to factor loadings, the market risk loadings (betas) are highly significant over all horizons and their magnitudes greater than 1. This suggests that the portfolios of added stocks are more sensitive to systematic swings than the broad market index which is traditionally considered to have a beta of 1. With value-weighting however (Panel B), the market factor loadings, while still highly significant, are less than 1 at horizons longer than 24 months, suggesting that added stocks are less volatile than the benchmark at longer horizons. Turning to the loadings on the size and value factors, the coefficients for the SMB and HML factors are positive at horizons of 36 months or shorter, and negative at the 60-month horizon. However, they are all statistically insignificant and very close to 0. The insignificant loading on both the SMB and HML factors suggests that the portfolio of added stocks is composed primarily of large capitalization stocks, and that the portfolio is somewhat tilted towards growth stocks. 12 Also, the fact that the portfolio of added stocks do not load significantly on the HML factor suggests that we are looking at a portfolio of stocks with relatively strong earnings and lower distress, cash-flow or bankruptcy risks. 13 Hence a plausible explanation for the below-average performance of the portfolio of added stocks is that they are low-risk stocks. The indication that the portfolio of additions is dominated by large capitalization stocks is consistent with Table 1 which shows that the average (median) capitalization of added stocks is approximately JPY 1 trillion (JPY 600 billion), compared to JPY 42 billion (JPY 20 billion) for deleted stocks. The factor loadings on the value-weighted portfolio returns (Panel B) are similar to those of the equal-weighted portfolios (Panel A) in that they are all insignificant and very close to Deletions Contrary to the results for added stocks, Table 5 (Panel A) shows that the alphas for the equal-weighted portfolios for deleted stocks are positive over all horizons, and statistically significant in three of the six horizons. [Table 5 around here] 12 Growth stocks are typically stocks with low B/M ratios. 13 The HML factor is interpreted as a measure of distress risk in the Fama and French (1993) three-factor model for U.S equity returns. 13

14 The alphas for the 12- and 36-month horizons are the largest in magnitude, 1.4% and 1.3% per month respectively, and are relatively smaller than the equalweighted BHARs (normalized to average monthly returns) of 3.9% and 1.4% per month over the same time frames, as reported in Table 8. Clearly, adjusting the abnormal returns to the size and book-to-market risk factors diminishes both their magnitude and statistical significance. On the whole, the portfolios of deleted stocks, based on the equal-weight scheme, earn statistically significant positive alphas over a number of horizons. Value-weighting however considerably weakens the significance of the alphas so much so that only the 36-month alpha remains significantly positive, albeit at the 10% level. As for the RM-RF (beta), SMB and HML factors, we find that all three factors load positively and significantly over all investment horizons. The betas are greater (less) than 1 at horizons shorter (longer) than 24 months. Similar to added stocks, deleted stocks also are less sensitive to systematic risks than the specified market benchmark at longer horizons. In contrast to added stocks, the SMB factor for deleted stocks loads positively and significantly in every time frame, the only exception being the 6- month horizon where the coefficient is positive but insignificant. This finding suggests that the portfolio of deleted stocks is heavy on small capitalization stocks (consistent with Table 1). This in turn suggests that the higher alphas for deleted stocks can be attributed to the size effect. The significant positive loading of the HML factor on the portfolio of deleted stocks indicates that the portfolio is dominated by value (or high B/M) stocks. As Fama and French (1998) noted, value stocks tend to have weaker earnings and higher (cash-flow) risks than growth stocks. Thus, the higher risk-adjusted performance of the portfolio of deleted stocks may reflect compensation for investing in financially distressed or less profitable firms. One other plausible implication of our result is that the value effect is stronger among small firms in Japan Comparison of results with US findings Unreported results show that the value-weighted three-factor alphas for added stocks in month 0 and month 1 are 10.47% (t-value 3.86) and -4.42% (t-value -2.36) respectively. 14 Thus contrary to the findings of Chan et al., (2013) for the S&P 500, we find no evidence of short-term price momentum in the Nikkei 225 event firms. Instead, we observe strong price reversals 1 month after the initial (event month) price run up. The short-term reversal can be attributed to 14 These are comparable with the BHAR estimates for the announcement month (month 0) and the month after (month 1) reported in Table 2. 14

15 investor overreaction or cognitive biases. 15 Furthermore, the negative long term BHARs and alphas for new additions suggest that the 1-month reversals tend to have a long-lasting effect, persisting over the intermediate- (6-18 months) and long-term (2-5 years). Our finding that stocks added to Nikkei 225 index do not outperform the market on a risk-adjusted basis in the subsequent 5-year period is contrary to findings in US studies (Chan et al., 2013) that stocks added to the S&P 500 index significantly outperform the market average over the same period. On the other hand, our finding that stocks deleted from the Nikkei 225 significantly outperform the market average is consistent with findings for the S&P 500 deletions documented in Chan et al., (2013). On the whole, while Chan et al., attribute the strong post-event performance of the S&P 500 additions and deletions to short term momentum, the weak (strong) post-event performance of the Nikkei 225 additions (deletions) documented in this study can be attributed to anomalous short-term reversals experienced by the event firms 1 month after their inclusion (removal) from the index. To confirm the absence of the momentum effect in the return behaviour of the portfolios, we employ the local Japan factors associated with Cahart s (1997) four-factor model which adds a momentum (WML) factor to the three factors of Fama and French (1993). If momentum is priced, we would expect to see a positive and significant loading for the factor. On the other hand, a negative or insignificant loading on the WML factor would suggest that momentum is not relevant in explaining the risk-adjusted return of the portfolios. Table 6 shows the results of the four-factor model regressions on the portfolio of additions. [Table 6 around here] The four-factor alphas and the market, SMB and HML coefficients are similar to the estimates in the three-factor model. The WML factor loadings are insignificant and close to zero across all horizons. This suggests that the momentum factor is irrelevant for explaining returns to the sample of added stocks. On a broader note, this finding is consistent with the findings in Fama and French (2012) who document evidence of momentum patterns in stock returns in North American, European and Asia-Pacific markets but none whatsoever in the Japanese market. 15 Debondt and Thaler (2008) argue that investors lack the capacity to process available information rationally because they exhibit erroneous beliefs and cognitive biases in the way they process or act on new information. 15

16 Table 7 presents the four-factor estimates for the sample of deletions. The WML factor has a highly statistically significant negative loading over the 6- month holding period, suggesting the presence of a strong short-term reversal effect and confirming our earlier observation that the short-term reversals are stronger for deletions than for additions. [Table 7 around here] We detect an increase in deleted stocks portfolio alphas from 0.5% to 0.8% (0.3% to 0.6% with value-weighting) over the 6-month horizon. One of our more interesting observations is that including the WML factor in the model renders the hitherto relevant HML factor redundant over the 6-month horizon. Specifically, the magnitude and statistical significance of the value premium over the 6-month horizon appears to be wiped away when we include the WML factor in the regression model. This observation suggests that the HML factor is subsumed in the WML factor over short horizons for the sample of deleted firms. 5.3 Profitability of Long-Short (Contrarian) Portfolio Strategies Results for Long-Short Portfolio BHARs Having documented evidence of strong and persistent price reversals for both added and deleted firms, the question then arises whether it is possible to implement profitable investment strategies that can exploit these anomalous return patterns. To address this question, we construct a long-short (zero-cost) arbitrage portfolio which takes long positions in stocks deleted from the Nikkei 225 index and short positions in stocks added to the index. The profitability of the resulting long-short portfolio is evaluated on the basis of the statistical significance of the difference in the buy-and hold abnormal performance between the two portfolios. Before presenting the results, it is worth pointing out that investors in Japan can, subject to certain conditions, short sell stocks through either the Margin trading system or the Stock lending market. In the margin trading system, investors can engage in short selling using either standardized or negotiable margin transactions. Under the standardized margin arrangement, eligible stocks for short selling (known as taisyaku stocks) are selected from amongst TSE-listed stocks based on their number of shareholders, trading volume etc. In 2009, the number of taisyaku stocks listed on the TSE was 1,591, representing 16

17 majority of stocks listed on the First Section of the TSE and about 67 percent of all TSE listings (Kato and Suzuki, 2012). As most of the stocks listed on the First Section of the TSE are taisyaku stocks and therefore eligible for short selling, the strategies considered in this paper are largely implementable. Table 8 reports the market-adjusted BHARs for the long-short portfolio over different horizons: normalized to monthly returns, the equal-weighted long-short BHARs range from 1.5% to 4.1% per month, all highly statistically significant. [Table 8 around here] On the whole, the 12-month strategy generates the largest zero-cost contrarian BHAR of 4.1% per month. Also, the long-short strategy, which simultaneously purchases deletions and short sells additions, appears to be more profitable than the long-only strategy, which invests only in deletions. This is true for all horizons except the 60-month Results for Three-Factor Regressions for Long-Short Portfolios To check the robustness of the profitability of the long-short contrarian strategy, we employ a calendar-time portfolio method which adjusts for the size and book-to-market premia believed to drive average returns in the Fama and French (1993) framework. 16 The results are presented in Table 9. The three-factor longshort portfolio alphas range from 0.5% to 1.9%, all statistically significant and comparable to the estimates of the long-short BHARs reported in Table 8. [Table 9 around here] The smaller magnitudes and statistical significance of the estimates reported in the calendar-time approach can be attributed to additional risk-adjustments and cross-sectional dependence. On the whole, the calendar time long-short portfolio alphas confirm the robustness of the profitability of the zero-cost contrarian strategy to cross-correlations in the return data and to adjustments for additional risk factors. 6 Concluding Remarks The practice of periodically reconstituting stock indices suggests that changes to the constituent stocks can impact the long-term returns of stocks added to or 16 The calendar portfolio method also implicitly accounts for cross-sectional dependence induced by event time clustering. 17

18 dropped from an index. Chan et al., (2013) presents evidence that stocks added to the S&P 500 index outperform the market average up to 5 years after their inclusion. However, the index selection rules for the S&P 500 are for the most part subjective, and give the impression that firms are selected on the basis of their future prospects. In this case, changes to the membership of the index may be considered value-relevant. In this paper, we extend our understanding of the long-term return behaviour of stocks following their inclusion or exclusion from an index by employing data from Japan s Nikkei 225 index. To the best of our knowledge, this is the first study to investigate the long term price effects of revisions to a major non-us stock index. The use of Nikkei 225 data provides a unique setting to examine the post-announcement price behaviour of stocks affected by revisions to the membership of the index: the transparent and objective index selection process suggests that inclusion or removal from the index should, in principle, be valueirrelevant. Our findings are as follows. First, although stocks added to the Nikkei 225 index exhibit significantly positive abnormal return behaviour in the announcement month of their inclusion, they deliver negative or, at best, zero abnormal returns in the subsequent 6 months to 3 years after inclusion. The muted long term return performance of the additions can be traced to the significant price reversals recorded by the firms 1 month after the announcement-month price appreciation. Deleted stocks also experience significant price reversals 1 month after the initial announcement-month price fall. In contrast to the additions, deletions earn significantly positive abnormal returns in the following 6 months to 3 years, thus suggesting that the 1-month reversal effect is stronger for deletions than for additions. Second, a significant price reaction in the announcement month which later reverses in the following month is indicative of investor overreaction to the news of the inclusion or deletion of stocks to the Nikkei 225 index. We further examine whether profitable investment strategies can be designed to exploit the return patterns exhibited by the additions and deletions. Specifically, we consider a contrarian strategy which shorts the additions and goes long in the deletions. We find that this strategy generates abnormal profits for the investor over all investment horizons considered even after adjusting for the size and book-to-market effects. Our results show that abnormal profits for this strategy are maximized over the 12-month horizon. A caveat concerning our results should be recognized: the relatively small number of constituent changes to the Nikkei 225, compared to the S&P 500, may hinder us from drawing very strong conclusions. As such, we are inclined to view our results as suggestive. 18

19 References [1] Carhart, M. M., On persistence in mutual fund performance. The Journal of Finance, 52(1), pp [2] Chan, K., Kot, H.W. and Tang, G.Y., A comprehensive long-term analysis of SP 500 index additions and deletions. Journal of Banking and Finance, 37(12), pp [3] Cowan, A.R. and Sergeant, A.M., Interacting biases, non-normal return distributions and the performance of tests for long-horizon event studies. Journal of Banking and Finance, 25(4), pp [4] Dionysiou, D., Choosing Among Alternative LongRun EventStudy Techniques. Journal of Economic Surveys, 29(1), pp [5] Elliott, W. B., Ness, B. F., Walker, M. D., Warr, R. S., What drives the S&P 500 inclusion effect? An analytical survey. Financial Management, 35(4), pp [6] Fama, E. F., Efficient capital markets: A review of theory and empirical work. The journal of Finance, 25(2), pp [7] Fama, E. F., and French, K. R., Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), pp [8] Fama, E.F., Market efficiency, long-term returns, and behavioral finance. Journal of Financial Economics, 49(3), pp [9] Fama, E. F., and French, K. R., Size, value, and momentum in international stock returns. Journal of Financial Economics, 105(3), pp [10] Hendricks, K.B. and Singhal, V.R., An empirical analysis of the effect of supply chain disruptions on longrun stock price performance and equity risk of the firm. Production and Operations Management, 14(1), pp [11] Iihara, Y., Kato, H. K., Tokunaga, T., The winner - loser effect in Japanese stock returns. Japan and the World Economy, 16(4), pp [12] Jeng, L. A., Metrick, A., Zeckhauser, R., Estimating the returns to insider trading: A performance-evaluation perspective. Review of Economics and Statistics, 85(2), pp

20 [13] Kothari, S.P. and Warner, J.B., Econometrics of Event Studies. Handbook of Corporate Finance: Empirical Corporate Finance. B. Espen Eckbo. [14] Lyon, J.D., Barber, B.M. and Tsai, C.L., Improved methods for tests of longrun abnormal stock returns. The Journal of Finance, 54(1), pp [15] Mitchell, M.L. and Stafford, E., Managerial Decisions and LongTerm Stock Price Performance. The Journal of Business, 73(3), pp [16] Petersen, M.A., Estimating standard errors in finance panel data sets: Comparing approaches. Review of Financial Studies, 22(1), pp

21 Figure 1: Post-Event Performance of BHARs (value-weighted) for Additions and Deletions Additions Deletions 40.00% 30.00% 20.00% BHARs (%) 10.00% 0.00% % % Holding period (months) relative to announcement month 21

22 Table 1: Descriptive statistics for the sample of event firms Panel A: Additions Firm Size (trillions of yen) Prior 1-year return Prior 3-year return Mean Median Std. Dev Skewness Ex. Kurtosis Minimum Maximum Doornik-Hansen a a Panel B: Deletions Mean Median Std. Dev Skewness Ex. Kurtosis Minimum Maximum Doornik-Hansen c a This table reports the firm characteristics and descriptive statistics for the sample of firms added to and deleted from the Nikkei 225 Index during the period March 1970 to September Firm size, measured in Japanese Yen, is the market value of the sample firms as of the event month 0; market value is defined as the product of a firm s month-end stock price and the number of shares outstanding. Prior 1-year (3- year) return is defined as the 1-year (3-year) compounded return earned from month t-12 (t-36) to month t-1; it is the return earned by the event firm over the 12-month (36-month) period prior to the event month. Superscripts a, b, and c denote statistical significance at the 1%, 5%, and 10% level, respectively for the Doornik-Hansen test of normality of the return distributions. 22

23 Table 2: Market-adjusted buy-and-hold abnormal returns for added stocks Holding Period (Months) [0] [1] [1,6] [1,12] [1,18] [1,24] [1,36] [1,60] BHARs Panel A: Equal- Weighted Mean Skew adj. t-stat a b c c Bootstrapped p value N Panel B: Value- Weighted Mean Skew adj. t-stat a a a a c Bootstrapped p value N Note: This table reports the long-term average buy-and-hold abnormal returns (BHARs) for the sample of firms added to the Nikkei 225 Index during the period March 1970 to September The long-term BHARs is evaluated against the Tokyo Stock Price Index, TOPIX, which serves as the benchmark for market returns. The holding periods range from 6 to 60 months. We report both equal-weighted and value-weighted BHARs and their significance are evaluated using the skewness-adjusted t-statistics of Lyon et al., (1999). The critical values are obtained by bootstrapping. Superscripts a, b, and c denote statistical significance at the 1%, 5%, and 10% level, respectively. 23

24 Table 3: Market-adjusted buy-and-hold abnormal returns for deleted stocks Holding Period (Months) BHARS Panel A: Equal- Weighted [0] [1] [1,6] [1,12] [1,18] [1,24] [1,36] [1,60] Mean Skew adj. t-stat a a a a a a a a Bootstrapped p- value N Panel B: Value- Weighted Mean Skew adj. t-stat a a a a a a c Bootstrapped p- value N Note: This table reports the long-term average buy-and-hold abnormal returns (BHARs) for the sample of firms deleted from the Nikkei 225 Index during the period March 1970 to September The long-term performance of the BHARs is evaluated with the Tokyo Stock Price Index, TOPIX, as benchmark for market return. The holding periods range from 6 months to 60 months. We report both equal-weighted and value-weighted BHARs and these are tested for significance using skewness-adjusted t-statistics of Lyon et al., (1999). The critical values are obtained by bootstrapping. Superscripts a, b, and c denote statistical significance at the 1%, 5%, and 10% level, respectively. 24