Change in systematic trading behavior and the cross-section of stock returns during the global financial crisis: Fear or Greed?

Transcription

1 Change in systematic trading behavior and the cross-section of stock returns during the global financial crisis: Fear or Greed? P. Joakim Westerholm 1, Annica Rose and Henry Leung University of Sydney Business School Abstract This paper examines the association between signed small trade turnover [SSTT] and stock returns. We predict that due to the change in investor behavior caused by an increased level of stress during the global financial crisis [GFC], portfolios formed during the global financial crisis period will not experience the SSTT to return association observed during pre and post GFC periods. This prediction is confirmed and we find that the SSTT to return association reverts to the pre GFC level after the crisis. Behavioral bias by small trade participants offers the most adequate explanation to the excess return generated by the low SSTT portfolio relative to the high SSTT portfolio, after adjustment for common risk factors in the CAPM and Fama-French models. This finding holds for SSTT portfolios generated during all the three periods and for both individual investors and institutional investors respectively with stronger results for individual investors. Keywords: Systematic trading behavior; Cross-section of stock returns; Signed small trade turnover JEL Classification: G11, G12, G14 1 H69, UNIVERSITY OF SYDNEY, NSW 2006, , joakim.westerholm@sydney.edu.au 1

2 1. Introduction We examine trading behavior and the cross-sectional stock returns over the recent global financial crisis [GFC] period on the NASDAQ OMX Helsinki (OMXH). Specifically we investigate whether signed small trade turnover 2 [SSTT] predicts the cross-section of future stock returns. SSTT measures temporary uninformed buy or sell pressure initiated by small individual trades trading in the same direction (or parts of larger institutional orders). It has been hypothesized that when stocks are favored by individual investors they become overvalued and subsequently underperform stocks out of favor with individuals (Hvidkjaer, 2008). We predict that due to the change in investor behavior caused by an increased level of stress during the GFC, portfolios formed during the global financial crisis period will experience a more negative SSTT to return association compared to pre and post GFC periods. New in this paper is our ability to distinctly categorize SSTT according to the type of investor that initiates these trades, hence we analyze small trades by all investors categories and compare similar size trades of individuals to similar size trades by institutions. Our motivation for doing this is that knowing the type of investor that initiates small trades, SSTT is no longer a proxy for individual investor trades, rather it becomes a proxy for uninformed trades in general. It is expected that institutions break up large orders to avoid market impact, often using computerized algorithms. It has been concluded in the literature, e.g. Chakravarty (2001), that medium size trades tend to contain most information, hence we expect that also high SSTT stocks favoured by institutions are relatively uninformed. 2 Consistent with Hvidkjaer (2008), the SSTT is defined as small trade buyer-initiated volume minus small trade seller-initiated volume, divided by the number of shares outstanding. Small trade is defined as a trade with a trading volume of 1000 shares or less, while we also investigate a trade value-based SSTT metric. The SSTT is used to measure uninformed individual investors in the case of households, and uninformed small trades as fractions of large orders in the case of institutional investors. 2

3 There are several studies in psychology which show that people's decisions are often susceptible to extreme and stressful conditions, and that such conditions modulate risk taking, potentially exacerbating behavioral bias in the subsequent decision making. For example, Porcelli and Delgado (2009) suggest that exposure to stress leads market participants to choose riskier decisions when trying to decide between taking a minor loss or a major one and the reverse proves true with gains. Lighthall et al. (2011) support similar findings except the finding is more significant for male than female. Based on the above studies, this research proposes that the riskier decision making behavior by investors in a stressful condition such as the GFC results in a different association between investor trading behavior and stock returns compared to the pre and post GFC periods. The contribution of this study is significant in two ways. First, this study examines the impact of the GFC on a new issue within the behavioral finance framework. The hypotheses relating to SSTT and stock returns have not yet been tested in the context of the GFC in the existing literature. Second, this study examines the change in the association between SSTT and stock returns across the pre GFC period, the GFC period and the post GFC among different investor categories. This is not only a cross-sectional comparison between different time periods but also a cross comparison between different investor types. Third, this research uses the dataset provided by Euroclear Finland Ltd for the OMXH, where this comprehensive dataset allows for a more precise classification of trades based on the actual investor type of the transactions and not a proxy. Compared to prior literature such as in Coval et al. (2005), Barber et al. (2009b), Frazzini and Lamont (2005) and Hvidkjaer (2008) where a trade size or volume based proxy is used for the classification of investor types, this is a major improvement in terms of methodology. The analysis of this study consists of three parts. The first part of the analysis is performed for the general market including all investors together and it focuses on the examination of the association between SSTT and the returns of the SSTT portfolios formed during the GFC period in the short term, medium term and long term. Then we compare such association with that of the portfolios formed during the pre GFC and post 3

4 GFC periods in the short term, medium term and long term. The results show that during the crisis period, there is no significant difference in stock returns in the short term between the high SSTT portfolio and the low SSTT portfolio. However, the high SSTT stocks underperform the low SSTT stocks as soon as four months after the portfolio formation and such underperformance persists in the long run. The results in the GFC period are different from the pre and post GFC period in the following aspects. (1) The association between SSTT and stock returns is similar between the pre and post GFC periods. This indicates that after the GFC period this association reverts to the pre GFC level. (2) The short term outperformance of the high SSTT stocks during the pre and post GFC period becomes economically and statistically insignificant during the GFC period. (3) The significant underperformance of the high SSTT portfolio starts from four months after the portfolio formation period in the GFC period, which is significantly earlier than one year after the portfolio formation period in the pre and post GFC periods. The second part of the analysis is performed for the different types of investors and follows the same procedure as for the general market as outlined above. The inter investor category analysis is necessary to determine whether the findings are consistent or varied among different investor categories. Given that Leung, Rose and Westerholm (2011) shows evidence to suggest that when examining the association between SSTT and stock returns in the medium and long term the results for different investor types were mixed, this study proposes that the mixed results are also obtained in this part of the research. Consistent with this hypothesis, the results obtained in this part of the analysis show that the individual investors and institutional investors show the strongest results among the different investor categories across the three different periods. The results for the two categories are mixed in the short term, medium term and long term. For example, over the pre and post GFC periods, the significant underperformance of the high SSTT portfolio occurs in the long term for individual investors, whereas it occurs in the medium term for institutional investors. Over the GFC period, the significant underperformance of the high SSTT portfolio occurs in the medium and long term for individual investors, whereas it occurs even earlier for institutional 4

5 investors. The third and final part of the analysis performs time-series regressions to analyze the returns of the SSTT ranked portfolio in relation to systematic risks and determines whether SSTT provides the best explanation for the cross-sectional returns of stocks. This analysis is performed for the pre GFC, the GFC and the post GFC periods and for different investor categories respectively. We expect that consistent with Hvidkjaer (2008), the regression models will show that the risk factors alone in the models cannot explain the crosssectional returns of the SSTT ranked portfolios, leaving investor systematic behavior as the best explanation. The finding in this study confirms this expectation across the three different periods and between individual investors and institutional investors. The remainder of the study is organized as follows. Section 2 provides a brief literature review on SSTT in relation to stock returns and a general overview of the impact of the GFC on the financial decision making. Section 3 describes the dataset used in this study. Section 4 provides the hypotheses and empirical methodology. Section 5 includes descriptive statistics and analyses the univariate results. Section 6 describes the regression models which examine SSTT in relation to stock returns while controlling for systematic risk factors for individual investors and institutional investors. Section 7 summarizes the findings and ends the study with concluding remarks. 2. Literature review and model This section offers a brief overview on investor trading behavior in relation to stock returns and the impact of stress on financial decision making particularly in the context of the GFC. 2.1 Investor trading behavior and stock returns There are several studies that examine investor trading behavior measured by signed volume in relation to stock returns. These studies are conducted for different time intervals. For example, at a daily and intraday level, Chordia and Subrahmanyam (2004) find that signed volume is positively related to future stock 5

6 returns. Chordia et al. (2005) conclude that order imbalances predict future returns over very short intervals. At a monthly level, Subrahmanyam (2006) finds that signed dollar volume is negatively related to stock returns over the next one to two months. At a yearly level, Barber et al. (2009a) examine the association between signed volume and future stock returns using transaction data over one year. Similar to Hvidkjaer (2006), they scale the difference between buyer- and seller-initiated small trade volume by the total small trade volume, and large difference in large trade volume by the total large trade volume. For longer periods, Kaniel et al. (2008) conduct a study using data on the New York Stock Exchange (NYSE) over a four-year period, and conclude that retail investors yield a higher short term return for providing liquidity to institutional investors. In addition, Hvidkjaer (2008) concludes that stocks with high seller-initiated small trade volume, measure over the past several months, outperform those with high buyer-initiated small trade volume, and the return differences continue to be significant in the second and third year after the portfolio formation. Several studies have been conducted to examine SSTT in relation to stock returns for different investor categories. Particular attention is paid to individual retail investors and institutional investors. For example, Coval et al. (2005) use brokerage account data and find that some individual investors are able to consistently outperform the market. Grinblatt and Keloharju (2000) conclude that individual investors in Finland yield significant lower returns than institutional investors. Barber et al. (2009b) confirm Grinblatt and Keloharju s (2000) finding in the Taiwan market. Frazzini and Lamont (2005) use mutual fund data and find that individual investor favored stocks show significant underperformance in the long term. 2.2 The impact of stressful conditions on financial decision making Several studies in psychology have found evidence to suggest that extreme and stressful conditions modulate risk taking, potentially exacerbating behavioral bias in the subsequent decision making. For example, Porcelli and Delgado (2009), analyse the financial risks investors are willing to take when calm or stressed. They find that exposure to stress led investors to choose riskier decisions to avoid making losses and more conservative decisions to realize gains. They argue that one potential explanation for the effect might be 6

7 that the human brain has two ways of looking at the world, an analytical one and an intuitive one. The analytical one is more easily disrupted by outside stimuli, such as the extreme stressful condition during the GFC, and the intuitive takes over under such conditions. In addition, Lighthall et al. (2011) conducts a study examining the gender differences in reward-related decision processing under stress. They find that although both female and male respond to stress conditions by making riskier decisions, the impact on male is significantly greater than female. The global finance crisis has undoubtedly exerted tremendous stress on investors. During such extreme condition, it is reasonable to assume that the change in investor behavior, i.e., riskier financial decision making, potentially leads to different cross-sectional stock returns. Subsequently, the association between investor trading behavior measured by SSTT and stock returns could also change during the GFC compared to the pre and post GFC periods. 2.3 A model of systematic trading behavior The literature (e.g. Grinblatt and Keloharju, 2000) unambiguously confirms that individual investors are contrarian traders in the short term, hence they end up on the opposite side of more momentum-driven traders, typically institutional and foreign investors. An alternative explanation, that individuals appear contrarian due to slow adjustment of limit orders, is provided by Linnainmaa (2010). In the short term, say one month, we predict that high SSTT stocks favored by individual investor (domestic individuals and foreign registered) outperform low SSTT stocks for the same category of investors. This compensation for liquidity provision received by contrarian individuals is expected to result in the underperformance of high SSTT stocks which experience buy pressure from institutions (domestic institutions and foreign nominees) compared to low SSTT stocks in the same category of investors. The rationale is that institutions take rather than provide liquidity in the short run, particularly if they are momentum-driven traders as concluded in previous literature. 7

8 As for the long term systematic trading behavior, a common assumption in the literature is that a majority of individual investors are noise traders (see Kyle, 1985 and Black, 1986). We hence expect high SSTT stocks which experience buy pressure from individual investors (domestic individuals and foreign registered) to underperform in the longer term, say one year. Since we propose that small trades by institutions are parts of larger liquidity-driven orders, we predict that there should be no or a negative long-term association between high SSTT stock that experience buy pressure from institutions and the cross-section of stock returns. The category to outperform in the longer term will be mid-size trades by institutions that are expected to be the most informed trades. The above systematic trading behavioral patterns are observed in a market environment with relatively normal price volatility when investors trade strategically using their endowed information and private beliefs (see Leung, Rose and Westerholm, 2011). However, in a crisis such as the GFC, investors have shown to make riskier decisions (Porcelli and Delgado, 2009). Since we are in this paper considering trader behavior in limit order book markets, we adapt the Parlour (1998) one-tick dynamic model of a limit order market to describe this phenomenon. In this model, agents choose to submit a limit order or a market order depending on the state of the limit order book. In our version of the model time is continuous, and new traders arrive at the market according to a Poisson process with parameter λ. Each trader has a type θ = {ρ, π}, where ρ is a discount rate and π is the trading profit. The payoff a trader earns as a result of trading is discounted back to his first arrival time in the market at ρ. Traders are risk neutral and submit orders to maximize their discounted profit. First, an investor s trading profit (loss) is derived from the purchase of shares at a low (high) price and selling it at a higher (lower) price. In a normal market with expected levels of volatility, individual investors experience situations described by Equation 1.1 and Equation 1.2. In a market with high levels of unpredictable volatility such as in a crisis, individual investors are likely to avoid the provision of liquidity in shares that have fallen in price, while they still provide liquidity on run-ups, as represented in Equation 8

9 2.1. In the scenario described by Equation 2.2, profits on the sales are offset by the opportunity cost of missing out on trades with temporary price setbacks. For a particular contrarian individual trader θ = {ρ, π}, profits are derived during normal market conditions: p t p t 1, sell uptick, sell p downtick, buy p t 1, buy,if,if the individual buyson a temporaryfall in price and the share recovers (1.1) the individual short sells on a temporary run - up in price and the share price adjusts (1.2) For the same individual trader θ, losses are derived during a crisis: p t p downtick, sell uptick, buy p p t 1, buy t 1, sell,if,if the individual fails to buy (and instead short sells) on a temporaryfall in price (2.1) the individual buyson a temporary run - up in price (2.2) where p downtick p 1 p, p t+1 is the market price at the end of the holding period, p downtick is the price at t uptick which a seller initiated trade fills a buy limit order and p uptick is the price at which a buyer initiated trade fills a sell limit order. Institutional investors who are momentum traders carry out trading activities described in Equation 3.1 and Equation 3.2. They are likely to take profits more quickly during a crisis, forfeiting any benefits from their momentum strategy under Equation 4.1 and Equation 4.2. For a particular institutional trader θ = {ρ, π}, profits are derived during normal market conditions: p t p t 1, sell p downtick, sell uptick, buy p t 1, buy,if the institution buyson a price increase and the share continues to rise (3.1),if theinstitution short sells on a price fall and the share continues to fall in price (3.2) where p uptick p 1 p. t downtick For the same institutional trader θ, trading profits are reduced during a crisis: 9

10 p t p uptick, buy t 1, buy p p t 1, sell downtick, sell,if the institution buyson a price increase but trades out toosoon (4.1),if theinstitution short sells on a price fall but trades out toosoon (4.2) where p downtick p 1 p. t uptick Hence, due to inconsistent strategy applications during crisis, a different association between the trading behavior of different investor categories and stock returns is expected compared to the pre and post GFC periods. Individual investors are expected to experience the worst deterioration in returns during the crisis period due to the incorrect timing of trades, and obtain the worst prices in the market swings. Institutional investors are still expected to follow a consistent trading strategy in a crisis, but are severely affected by behavioral biases. Both categories are expected to produce an inferior performance during crisis, compared to periods with more predictable volatility. 3. Data This study employs a dataset of all investor transactions in Finland s limit order book market operated by Nasdaq-OMX. These data are provided by Euroclear Finland Ltd coves the period January 2000 to December 2010 and has been used extensively in recent research, (see e.g. G-K, L Seru, Shumway,) The records includes information on investor identity, date, stock, transaction type, price and volume. The actual investor category provided with each transaction allows for the exact classification of transactions by investor type. Transactions are divided into six investor categories, namely domestic individuals, domestic institutions, foreign nominees, foreign registered, government and not-for-profit organizations and residuals. In addition to the Euroclear dataset, stock specific data such as the number of shares outstanding and market capitalizations are obtained from Thomson-Reuters. The NASDAQ OMX Helsinki Cap Index is also obtained from Thomson-Reuters to calculate market returns. 4. Empirical methodology and hypotheses 10

11 This study uses SSTT as a measure for investor trading behavior. For each stock, SSTT is calculated as small trade buyer-initiated volume minus small trade seller-initiated volume, divided by the number of shares outstanding. For the purpose of this paper, a small trade is a transaction with a volume of 1000 shares or less. SSTT is measured over periods of one to 24 months prior to the portfolio formation. Portfolios are then formed on the basis of SSTT, and the returns of the portfolios are measured over periods of one to 24 months following the portfolio formation. This study uses the Lee and Ready (1991) algorithm to estimate the direction of trade initiation. The trade classification is accurate as we know the actual trader on each side of a trade. Trades executed at a price exceeding the mid-point between the bid and ask price immediately before the execution are buyer-initiated trades, and those below the mid-point as seller-initiated trades. Trades occurring at the midpoint are excluded. In addition, the definition of small trades is simple, using the volume of stocks traded will directly avoid sensitivity to small price changes 3 (Lee, 1992). A trade size proxy for individual and institutional trading is also supported by Lee and Radhakrishna (2000). For robustness we run tests using a SSTT metric obtained by sorting stocks into quintiles then using dollar based small trade cut-off points for each quintile (see Hvidkjaer, 2008). 4.1 SSTT in relation to stock returns: across three periods for all investors Hypothesis (1): The association between SSTT and stock returns in the short, medium and long term during the GFC period is different from that during the pre and post GFC periods. This section examines the association between SSTT and the returns of the SSTT portfolios formed during the GFC period, then compares such association with that of the portfolios formed during the pre GFC and post GFC periods. 3 As a robustness check a dollar-based method similar to Hvidkjaer (2008) is used to sort portfolios and obtained similar results. 11

12 Formation periods of J=1, 3, 6, 12 and 24 are used, and holding periods of K=1 and 1-3 for the short term, K=4-6 for the medium term and K=7-12 and are used. 4.2 SSTT in relation to stock returns: across three periods for different investor categories Hypothesis (2): Significant differences exist in the association between SSTT and stock returns among different investor categories during of the pre GFC, GFC and post GFC periods. This section follows consistent methodology as outline in Section 4.1, except that before any calculation is performed, all transactions are divided into the aforementioned six different investor categories according the actual investor type associated with each transaction. A formation period of J=6 is used Time series regression models Hypothesis (3): SSTT is the key factor which explains the cross-sectional stock returns for individual investors and institutional investors across the pre GFC, GFC and post GFC periods respectively, particularly when traditional systematic risk factors cannot provide adequate explanations. Four regression models are constructed for each investor category. In the first model, portfolio excess returns are regressed against the market excess returns with β rm as the coefficient. In the second model, portfolio excess returns are regressed against the three Fama-French (1993) factors. The third model uses the three Fama-French (1993) factors as well as a UMD factor which is the past one year winner stock returns minus the loser stock returns. And the fourth model uses the three Fama-French (1993) factors plus a PIN factor which is the difference between the monthly returns of the high and the low PIN portfolios, where 4 Formation periods of J=1, 3, 12 and 24 are also tested and similar results are obtained. Only results for J=6 are reported. 12

13 PIN is a measure of the probability of informed trading in a stock. The calculations of PIN and the PIN portfolio are similar to that in Easley, Hvidkjaer and O Hara (2005). In this study, the four regression models are constructed for individual investors and institutional investors separately, because these two types of investors exhibit the strongest results in the univariate analysis section consistently across different periods. Therefore, the comparisons between these categories during each period are more meaningful. 5. Descriptive statistics and univariate analysis This section tests hypotheses (1) and (2) in Section 5.4. It includes descriptive statistics which describe the general characteristics of the SSTT portfolios for the pre GFC, GFC and post GFC periods. It also includes the results from testing SSTT in relation to stock returns. These results are organized as follows. Table 1a, b, and c report descriptive statistics for the general market consisting of all investors together on the OMXH during the GFC period, pre crisis period and post GFC period respectively. Table 2a, b, and c show SSTT in relation to stock returns (short, medium and long terms) for the general market during the GFC period, pre GFC period and post GFC period respectively. Table 3a, b and c show SSTT in relation to stock returns (short, medium and long terms) for the six investor categories during the GFC period, pre GFC period and post GFC period respectively. 13

14 1. Descriptive statistics for SSTT portfolios formed in different periods Table 1a Descriptive statistics for SSTT portfolios formed during the GFC period In Table 1a, portfolios are formed monthly from January 2008 to June 2009 based on the SSTT of the prior six months (J=6). The dataset consists of all stocks traded on the OMXH. The dataset consists of all stocks traded on the OMXH. ***, ** and * denote a 99%. 95% and 90% confidence level respectively. Put back old table! 14

15 Table 1b Descriptive statistics for SSTT portfolios formed before the GFC period In Table 1b, portfolios are formed monthly from January 2000 to December 2007 based on the SSTT of the prior six months (J=6). The dataset consists of all stocks traded on the OMXH. The dataset consists of all stocks traded on the OMXH. ***, ** and * denote a 99%. 95% and 90% confidence level respectively. Insert table! 15

16 SSTT portfolio Low 2 3 High Low-High t(low-high) SSTT (%) Mean -0.11% 0.08% 0.40% 40.83% % *** STD 0.31% 0.09% 0.30% 7.23% Low -7.14% -0.12% -0.03% 0.02% High 0.09% 0.41% 1.44% % Shares outstanding ('000 shares) Mean 240,574 58,237 43,608 58, , *** STD 712, ,320 79, ,435 Low High 4,663,760 4,095, ,262 4,663,760 Monthly volume (no. of shares) Mean 48, , ,401 10,721,286-10,672, *** STD 209, ,395 3,122,653 26,569,528 Low High 3,911,260 2,430,860 88,446, ,960,000 Small trade buyer-initiated turnover (%) Mean 0.29% 0.32% 0.86% 5.50% -5.21% *** STD 0.58% 0.31% 0.66% 9.01% Low 0.00% 0.00% 0.00% 0.04% High 11.40% 2.60% 5.28% 92.15% Small trade seller-initiated turnover (%) Mean 0.41% 0.24% 0.46% 2.12% -1.71% *** STD 0.79% 0.28% 0.48% 2.33% Low 0.00% 0.00% 0.00% 0.00% High 14.36% 2.45% 4.81% 20.70% Market size ( '000) Mean 4,061, , , ,825 3,529, *** STD 12,380,145 2,507, ,933 4,527,762 Low 3,856 1,836 3,569 2,101 High 108,581,000 66,257,800 10,415,700 57,690,700 Monthly return (%) Mean 0.16% 3.08% 3.07% 1.39% -1.23% -2.49*** STD 8.24% 44.18% 40.83% 11.20% Low % % % % High 46.03% % % 70.41% 16

17 Table 1c Descriptive statistics for SSTT portfolios formed after GFC period In Table 1c, portfolios are formed monthly from July 2009 to December 2010 based on the SSTT of the prior six months (J=6). The dataset consists of all stocks traded on the OMXH. The dataset consists of all stocks traded on the OMXH. ***, ** and * denote a 99%. 95% and 90% confidence level respectively. SSTT portfolio Low 2 3 High Low-High t(low-high) SSTT (%) Mean -0.04% 12.45% 0.36% 3.79% -3.83% *** STD 0.12% 0.06% 0.25% 6.39% Low -1.33% 0.01% 0.06% 0.23% High 0.09% 0.41% 1.44% % Shares outstanding ('000 shares) Mean 282,298 53,766 40,537 30, , *** STD 706, ,297 54,701 68,729 Low 1, High 4,095,040 4,095, , ,926 Monthly volume (no. of shares) Mean 36,684 77, ,368 16,054,689-16,018, *** STD 150,292 95,382 2,791,975 44,256,731 Low ,900 10,295 High 3,611, ,549 88,446, ,943,000 Small trade buyer-initiated turnover (%) Mean 0.19% 0.26% 0.71% 6.58% -6.39% *** STD 0.46% 0.25% 0.53% 8.37% Low 0.00% 0.01% 0.07% 0.37% High 11.40% 2.60% 3.79% 92.15% Small trade seller-initiated turnover (%) Mean 0.24% 0.17% 0.35% 2.79% -2.55% *** STD 0.52% 0.22% 0.36% 2.83% Low 0.00% 0.00% 0.00% 0.04% High 12.30% 2.45% 2.67% 27.91% Market size ( '000) Mean 4,500, , , ,823 4,373, *** STD 12,727,640 2,101, , ,346 Low 1,226 1,442 1,082 1,658 High 108,581,000 66,257,800 2,548,650 2,440,010 Monthly return (%) Mean -0.68% -0.30% -0.12% 0.75% -1.43% -2.62*** STD 12.45% 14.27% 12.01% 13.51% Low % % % % High % 73.88% 90.00% % Table 1a reports the general characteristics of the SSTT portfolios formed over the GFC period. For each SSTT portfolio, the table lists the mean, standard deviation, low and high for each variable includes SSTT, 17

18 number of shares outstanding, monthly volume, small trade buyer-initiated turnover, small trade sellerinitiated turnover, market size and monthly return. The differences between the high and the low SSTT portfolios and the levels of significance are reported in the far right column. Table 1a shows that SSTT ranks are significantly positively correlated to all variables, expect for the monthly return where the correlation is negative and insignificant during the GFC period. This means that stocks with high SSTT are associated with higher number of shares outstanding, larger monthly trading volume, higher small trade buyer-initiated turnover, higher small trade seller-initiated turnover and lower market size. There is no significant association between SSTT and the historical monthly returns of the portfolios. In comparison with Table 1b and Table 1c, the following differences between the GFC, pre GFC and post GFC periods spring to attention. First, comparing the correlation between SSTT and the number of shares outstanding and market size, this is significantly positive during the GFC period and significantly negative during the pre and post GFC period. This indicates that the investors exhibit a favorable behavioral bias towards relatively larger companies during the GFC compared to the other two periods. These larger companies such as blue chip companies often provide better yield than smaller companies during market downturn. This behavior is consistent with the hypothesis that the investors behavioral bias is exacerbated by stress conditions such as the GFC where they make riskier decisions towards avoiding losses but make more conservative decisions to preserve gains. Second, in comparison to the significant positive correlation between SSTT and historical stock returns of the SSTT portfolios during the pre and the post GFC period, this correlation becomes economically and statistically insignificantly during the GFC period, meaning investors do not have a preference towards historically outperformed stocks compared to the pre and the post GFC period where they are clearly more favored. 18

19 2. Cross-sectional returns for SSTT portfolios formed in different periods for all investors Table 2a Cross-sectional returns for SSTT portfolios formed during the GFC period for all investors In Table 2a, portfolios are formed monthly from January 2008 to June 2009 based on the SSTT of the prior J months. The dataset consists of all stocks traded on the OMXH by all investors. It reports monthly % returns to SSTT portfolios for the formation periods J=1, 3, 6, 12 and 24, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high along with the t-tests for significance. *** denotes 99% confidence level, ** denotes 95% confidence level and * denotes 90% confidence level. Monthly % returns, holding period K= Formation period Portfolio J=1 Low -1.17% -1.14% -1.63% -2.78% -1.63% % -0.52% -1.81% -3.38% -2.36% % -0.77% -2.09% -3.55% -2.59% High -0.77% -1.76% -2.67% -4.97% -3.02% Low-High -0.40% 0.63% 1.04% 2.19% 1.40% *** 4.96*** 4.77*** J=3 Low -1.23% -1.35% -1.22% -2.41% -2.18% % -1.16% -1.75% -3.69% -3.14% % -0.93% -2.08% -3.56% -3.22% High -0.77% -0.85% -2.67% -4.96% -3.91% Low-High -0.46% -0.50% 1.45% 2.55% 1.73% *** 4.58*** 5.29*** J=6 Low -1.07% -0.62% -0.98% -2.42% -3.42% % -0.53% -1.41% -3.67% -4.23% % -0.41% -1.63% -3.55% -4.80% High -0.66% -0.38% -2.27% -4.96% -5.69% Low-High -0.40% -0.24% 1.29% 2.54% 2.27% ** 4.56*** 5.93*** J=12 Low -1.01% -0.77% -1.02% -3.11% -5.53% % -0.95% -1.55% -3.35% -6.22% % -0.55% -1.86% -3.10% -7.13% High -0.63% -0.45% -1.80% -4.69% -8.75% Low-High -0.38% -0.32% 0.78% 1.58% 3.22% * 4.82*** 7.34*** J=24 Low 3.19% 1.96% 5.06% 6.73% 2.39% % 1.89% 1.53% 3.68% 2.52% % 1.81% 2.15% 0.79% 1.80% High 2.27% 1.43% 1.68% 0.86% 1.79% Low-High 0.93% 0.52% 3.39% 5.87% 0.60% * 1.98** 19

20 Table 2b Cross-sectional returns for SSTT portfolios formed before GFC period for all investors In Table 2b, portfolios are formed monthly from January 2000 to December 2007 based on the SSTT of the prior J months. The dataset consists of all stocks traded on the OMXH by all investors. It reports monthly % returns to SSTT portfolios for the formation periods J=1, 3, 6, 12 and 24, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high along with the t-tests for significance. *** denotes 99% confidence level, ** denotes 95% confidence level and * denotes 90% confidence level. Monthly % returns, holding period K= Formation period Portfolio J=1 Low 0.36% 0.56% 0.81% 0.65% 2.22% % 1.37% 0.46% 0.75% 2.13% % 1.64% 0.60% 0.14% 1.67% High 2.08% 1.27% 0.41% 0.37% 1.08% Low-High -1.72% -0.70% 0.40% 0.28% 1.14% -1.34* -1.32* * J=3 Low -0.53% 0.35% 1.36% 1.10% 2.47% % 1.72% 0.49% 0.19% 2.12% % 2.14% 0.59% 0.06% 2.06% High 3.07% 1.67% 0.41% 0.38% 1.19% Low-High -3.59% -1.32% 0.95% 0.72% 1.28% -5.60*** -3.80*** * J=6 Low -0.61% 0.65% 1.68% 1.16% 2.21% % 1.46% 0.99% 0.24% 2.18% % 3.08% 1.15% 0.03% 1.96% High 3.23% 2.05% 0.91% 0.29% 1.40% Low-High -3.85% -1.40% 0.78% 0.87% 0.81% -4.77*** -3.24*** * J=12 Low -3.47% -0.45% 1.55% 2.92% 2.44% % 0.06% 1.16% 1.94% 0.64% % 2.59% 0.92% 1.67% 0.41% High 2.82% 2.00% 0.69% 1.94% 1.65% Low-High -6.28% -2.44% 0.86% 0.97% 0.79% -4.25*** -2.78*** ** 1.04 J=24 Low 1.19% 1.58% 5.06% 6.73% 2.39% % 3.89% 1.53% 3.68% 2.52% % 2.81% 2.15% 0.79% 1.80% High 2.27% 1.43% 1.68% 0.86% 1.79% Low-High -1.07% 0.15% 3.39% 5.87% 0.60% -1.25* 1.10* ** 20

21 Table 2c Cross-sectional returns for SSTT portfolios formed after GFC period for all investors In Table 2c, portfolios are formed monthly from July 2009 to December 2010 based on the SSTT of the prior J months. The dataset consists of all stocks traded on the OMXH by all investors. It reports monthly % returns to SSTT portfolios for the formation periods J=1, 3, 6, 12 and 24, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high along with the t-tests for significance. *** denotes 99% confidence level, ** denotes 95% confidence level and * denotes 90% confidence level. Monthly % returns, holding period K= Formation period Portfolio J=1 Low -1.07% -0.78% -0.62% -1.08% 2.28% % -0.62% -0.94% -0.95% -0.23% % -0.43% -0.93% -0.88% 0.27% High -0.04% -0.44% -0.91% -1.24% 0.37% Low-High -1.03% -0.34% 0.29% 0.16% 1.90% -1.51* ** J=3 Low -1.17% -0.83% -0.60% -1.05% 2.32% % -0.62% -0.94% -0.98% -0.24% % -0.42% -0.94% -0.87% 0.29% High -0.25% -0.58% -0.91% -1.24% 0.39% Low-High -0.91% -0.25% 0.30% 0.19% 1.93% -1.28* ** J=6 Low -1.69% -1.20% -0.70% -1.06% 2.31% % -1.24% -0.89% -0.91% -0.23% % -0.84% -0.99% -0.81% 0.27% High -0.23% -1.20% -1.02% -1.26% 0.39% Low-High -1.46% -0.01% 0.32% 0.20% 1.92% -1.64* ** J=12 Low -2.65% -1.49% -1.93% -2.38% 2.24% % -1.29% -2.21% -2.27% -0.22% % -1.22% -2.51% -2.59% 0.31% High -2.08% -1.83% -2.89% -3.20% 0.42% Low-High -0.58% 0.34% 0.96% 0.82% 1.82% ** 2.26** 2.22** J=24 Low -1.25% -0.86% -1.96% -0.98% -3.17% % -0.46% -1.94% -1.47% -3.28% % -0.97% -1.28% -1.80% -3.13% High 0.09% -0.62% -1.85% -1.93% -4.80% Low-High -1.34% -0.25% -0.11% 0.95% 1.63% -1.56* ** 4.96*** Table 2a reports the cross-sectional returns of the SSTT portfolios formed during the GFC period for all investors. In the short run, the results show an insignificant correlation between SSTT ranks and stock returns. For example, for J=6, the difference between the low and the high SSTT portfolio is -0.40% 21

22 and -0.24% for K=1 and K=1-3 respectively, and both insignificant. It shows that high SSTT portfolio yields similar returns in the first three months after portfolio formation. In the medium and long run, the results show that there is a significant negative correlation between SSTT ranks and stock returns. For example, the difference between the low and the high SSTT portfolio is 1.29% with t=2.27 at a 95% confidence level for holding period K=4-6, 2.54% with t=4.56 at a 99% confidence level for K=7-12, and 2.27% with t=5.93 for K= This shows that in the medium and long term, the high SSTT stocks significantly underperform the low SSTT stocks. Comparing Table 2a with Table 2b and Table 2c, the differences are as follows. First, the results in Table 2b and Table 2c exhibit similar patterns, and both are different from the results in Table 2a. This means that after the GFC period, the association between SSTT and stock returns revert to a similar level to the pre GFC period. This is consistent with the hypothesis that in a more stressful environment, investors trade differently and are more prone to behavioral bias. However, as soon as the condition becomes less stressful again, their trading behavior returns to be just like before. Second, the short term positive correlation between SSTT and stock returns during the pre and post GFC periods becomes insignificant during the GFC period. This indicates that the short term outperformance of high SSTT stocks during the pre and post GFC period no longer exists. Third, the significant underperformance of the high SSTT portfolio starts to appear as early as the fourth month in the holding period during the GFC. Comparing to the pre and post GFC periods where the significant underperformance only starts from the 13 th month in the holding period, this is much earlier. The above differences confirm Hypothesis (1) that the association between SSTT and cross-sectional stock returns during the GFC period is significantly different from the pre and post GFC periods. It is also consistent with the theory that when experiencing more stress, investors trading behavior changes and subsequently results in a different association between the behavior and future stock returns. 3. Cross-sectional returns for SSTT portfolios formed in different periods for different 22

23 investor categories Table 3a Cross-sectional returns for SSTT portfolios formed during the GFC period for different investor categories In Table 3a, portfolios are formed monthly from January 2008 to June 2009 based on the SSTT of the prior six months (J=6). The dataset consists of all the stocks traded on the OMXH, and the transactions are divided into six categories by the actual investor type. For each investor category, the table reports monthly % returns to SSTT portfolios for the formation periods J= 6, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high and reported along with the t-tests for significance. *** denotes 99%, ** denotes 95% and * denotes 90% confidence level. No. of Monthly % returns, holding period K= Investor category SSTT portfolio stocks Domestic individuals Low % -0.71% -0.93% -3.05% -3.61% % -0.02% -1.80% -3.73% -4.53% % 0.22% -1.44% -3.61% -4.51% High % -0.51% -2.08% -4.64% -5.70% Low-High 0.03% -0.20% 1.15% 1.59% 2.08% ** 3.41*** 5.16*** Domestic institutions Low % -0.44% -1.56% -3.55% -4.74% % 0.06% -1.65% -3.78% -5.06% % -0.48% -1.76% -3.97% -5.16% High % -1.25% -2.54% -5.07% -5.83% Low-High 0.21% 0.81% 0.99% 1.51% 1.08% ** 2.52*** 4.49*** 2.36*** Foreign nominees Low % 0.73% -0.60% -3.97% 0.42% % 1.70% 0.37% -3.10% 1.00% % -0.74% -2.05% -4.06% 0.78% High % -0.24% -1.35% -4.06% 0.25% Low-High 0.67% 0.97% 0.74% 0.09% 0.17% Foreign registered Low % -0.62% -1.91% -4.15% 1.22% % 0.25% -1.18% -3.76% 0.60% % 0.23% -1.43% -3.90% 1.11% High % -0.74% -2.33% -4.52% 0.98% Low-High 0.01% 0.12% 0.42% 0.37% 0.24% Government and Low % 0.00% -4.15% -5.30% -7.48% not-for-profit % -0.01% -2.77% -4.45% -5.69% organisations % -0.69% -1.81% -4.83% -6.29% High % -0.59% -3.47% -6.13% -7.32% Low-High 1.14% 0.59% -0.68% 0.83% -0.17% Residuals Low % -2.93% -3.92% -6.13% -5.96% % -0.89% -2.94% -5.07% -6.78% % -0.96% -3.49% -5.00% -6.60% High % -1.89% -2.45% -5.80% -6.18% Low-High -0.16% -1.04% -1.48% -0.34% 0.22% *

24 Table 3b Cross-sectional returns for SSTT portfolios formed before GFC period for different investor categories In Table 3b, portfolios are formed monthly from January 2000 to December 2007 based on the SSTT of the prior six months (J=6). The dataset consists of all the stocks traded on the OMXH, and the transactions are divided into six categories by the actual investor type. For each investor category, the table reports monthly % returns to SSTT portfolios for the formation periods J= 6, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high and reported along with the t-tests for significance. *** denotes 99% confidence level, ** denotes 95% confidence level and * denotes 90% confidence level. No. of Monthly % returns, holding period K= Investor category SSTT portfolio stocks Domestic individuals Low % 0.17% 1.56% 0.75% 2.06% % 2.14% 0.64% 0.27% 2.46% % 2.48% 1.54% 0.06% 2.10% High % 2.30% 1.01% 0.08% 0.86% Low-High -3.59% -2.13% 0.55% 0.67% 1.20% -4.27*** -4.05*** ** Domestic institutions Low % 0.11% 0.83% 0.22% 0.54% % 0.55% 0.76% 0.66% 1.12% % 0.28% -0.18% 0.04% 0.73% High % 4.69% -0.35% -0.46% 0.48% Low-High -2.36% -4.58% 1.18% 0.68% 0.06% -3.40*** -1.74* 3.10*** 2.48** 0.30 Foreign nominees Low % 1.98% 1.38% 0.94% 2.32% % 3.42% 2.70% 0.37% 2.37% % 1.35% 0.75% -0.11% 2.88% High % 1.98% 0.93% 0.11% 1.61% Low-High -1.73% 0.00% 0.44% 0.83% 0.71% -1.68* * 1.68* Foreign registered Low % 0.82% 0.61% -0.03% 0.70% % 2.45% 1.80% 0.09% 0.88% % 3.01% 1.35% 0.02% 0.99% High % 1.74% 0.75% -0.44% 0.46% Low-High -0.49% -0.92% -0.13% 0.41% 0.23% * Government and Low % 0.77% -0.05% -1.58% 3.02% not-for-profit % 0.89% 1.41% 0.56% 3.21% organisations % 2.06% 0.28% -0.20% 2.96% High % 2.05% 0.99% -1.64% 2.72% Low-High 1.50% -1.29% -1.04% 0.06% 0.30% Residuals Low % -0.32% -0.45% -0.79% 0.46% % 1.80% 1.27% 0.18% 0.47% % 1.54% 0.07% -0.42% 0.02% High % 0.84% -0.29% -1.78% 0.00% Low-High -1.99% -1.16% -0.15% 0.99% 0.47% -1.83*

25 Table 3c Cross-sectional returns for SSTT portfolios formed after GFC period for different investor categories In Table 3c, portfolios are formed monthly from July 2008 to December 2010 based on the SSTT of the prior six months (J=6). The dataset consists of all the stocks traded on the OMXH, and the transactions are divided into six categories by the actual investor type. For each investor category, the table reports monthly % returns to SSTT portfolios for the formation periods J= 6, and for holding periods K=1, 1-3, 4-6, 7-12 and Monthly returns are calculated as the average of the current period s return for the portfolios constructed in each prior month. For example, when K=1-3, the portfolio monthly return is the average of this month s return for the portfolios constructed in each of the prior three months, i.e., the average of this month s return of three different portfolios constructed in month 1, 2 and 3. Return differences between low and high SSTT portfolios are calculated as low minus high and reported along with the t-tests for significance. *** denotes 99% confidence level, ** denotes 95% confidence level and * denotes 90% confidence level. No. of Monthly % returns, holding period K= Investor category SSTT portfolio stocks Domestic individuals Low % -0.64% -1.04% -1.18% 2.02% % 0.06% -1.11% -0.98% -0.38% % -0.04% -0.57% -0.84% 0.01% High % 0.86% -0.89% -1.06% 0.53% Low-High -3.30% -1.50% -0.15% -0.12% 1.49% -4.05*** -2.31** * Domestic institutions Low % -0.83% 0.39% -0.03% 0.95% % -0.01% -0.04% 0.50% 0.71% % 0.15% -0.34% -0.09% 0.89% High % 0.16% -1.56% -1.12% 0.87% Low-High -2.91% -0.99% 1.95% 1.09% 0.07% -3.22*** -1.70* 3.24*** 2.51** 0.10 Foreign nominees Low % -1.01% -1.06% -1.21% -0.60% % -1.38% -0.89% -1.01% -2.04% % -0.78% -0.90% -1.17% -2.35% High % -1.12% -1.32% -1.29% -1.83% Low-High -1.04% 0.11% 0.26% 0.09% 1.23% * Foreign registered Low % -1.01% -1.06% -1.21% 0.77% % -1.38% -0.89% -1.01% 0.80% % -0.78% -0.90% -1.17% 1.08% High % -1.12% -1.32% -1.29% 0.12% Low-High -0.36% 0.11% 0.26% 0.09% 0.65% Government and Low % -3.65% -3.34% 0.43% 1.22% not-for-profit % -3.96% -3.69% 0.15% 0.94% organisations % -2.35% -3.56% 0.24% 1.04% High % -3.45% -3.65% -0.37% 0.78% Low-High -0.75% -0.19% 0.31% 0.80% 0.44% Residuals Low % -0.44% -3.03% 1.06% -0.60% % -0.01% -1.18% -0.75% -2.04% % 0.04% -1.36% -0.62% -2.35% High % -0.36% -2.46% 0.67% -1.83% Low-High -3.82% -0.08% -0.57% 0.39% 1.23% -1.91* * 25

26 Table 3a shows the cross-sectional returns of the SSTT portfolios formed during the GFC period for different investor categories. Among all the investor categories, individual investors and institutional investors show the strongest results. However the results between the two investor types are varied. Consistent with the results for the general market in the same period, individual investors show a insignificant correlation between SSTT and returns in the short term, and a significant negative correlation in the medium term and long term. However, the significant negative correlation starts as early as from K=1-3 for institutional investors. This means that significant underperformance of the high SSTT stocks occurs much earlier than individual investors during the GFC period. When comparing the results in Table 3a with Table 3b and Table 3c, there are several similarities and differences as follows. Firstly, during the pre and post GFC periods, the association between SSTT and returns are different for individual investors and institutional investors. Although the association is significantly positive in the short term for both, it is different for them in the medium and long term. For example, the high SSTT stocks significantly underperform in the long term for individual investors, whereas they significantly underperform in the medium term for institutional investors. And this inter category difference is consistent for the pre and post GFC period. Secondly, the results for the pre and post GFC periods are very similar, and they differ from the results for the GFC period. In addition, the inter category differences for the pre and post GFC periods are also similar, and they also differ from the results from the GFC period. All the above differences confirm the theoretic prediction outlined by Hypothesis (2) that the association between SSTT and cross-sectional stock return for different investor categories is a mixed one across the pre GFC, GFC and post GFC periods. 4. Multivariate analysis As outlined in the methodology section, this section performs the time-series regression analysis for 26

27 individual investors and institutional investors. Given that they show the strongest results in the univariate analysis section across different periods, it is more meaningful to conduct the regression analysis for these two investor categories separately to see the similarities and differences. Table 4a, b and c report the regression results for the GFC period, pre GFC period and post GFC period respectively. Within each table, Panel A shows the results for individual investors and Panel B shows the results for institutional investors. Figure 1 summarizes these results as the difference between α for low SSTT portfolio and the α for the high SSTT portfolio, reported for the three holding periods, 1-3, 4-6 and 7-12 months and for the three periods pre GFC, GFC period and post GFC. Figure 1 Distribution of α for low α high SSTT portfolios pre, during and post GFC. Pre 1-3, 4-6 and 7-12 reports the difference in low and high SSTT alpha for the holding periods 1-3, 4-6 and 7-12 months during the pre GFC period January 2000 to December 2007, GFC 1-3, 4-6 and 7-12 reports the difference in low (large trades) and high (small trades) SSTT alpha for the holding periods 1-3, 4-6 and 7-12 months during the GFC period January 2008 to June 2009 and post 1-3, 4-6 and 7-12 reports the difference in low and high SSTT alpha for the holding periods 1-3, 4-6 and 7-12 months during the post GFC period July 2009 to December a Low SSTT -a High SSTT LARGE TRADES Individuals LARGE TRADES Institutions pre 1-3 pre 4-6 pre 7-12 GFC 1-3 GFC 4-6 GFC 7-12 post 1-3 post 4-6 post Low - High SSTT