Investor Attention, Stock Market Performance, and Momentum Returns

Transcription

1 Investor Attention, Stock Market Performance, and Momentum Returns M.Sc. Finance Thesis Zhaneta Krasimirova Tancheva August 16, 2013

2 Investor Attention, Stock Market Performance, and Momentum Returns M.Sc. Finance Thesis Zhaneta Krasimirova Tancheva ANR: Emp.: Supervisor: Dr. Alberto Manconi Second Reader: Dr. Michel R. R. van Bremen Tilburg School of Economics and Management Department of Finance Tilburg University August 16, 2013

3 Acknowledgements I would like to express the most sincere gratitude to my supervisor Dr. Alberto Manconi for all his time, valuable guidance, enthusiasm and advice throughout the last year. I truthfully appreciate all his help, understanding and support in the challenging period of completing my Master thesis. Sharing his knowledge and experience with me gave me the opportunity to enrich my knowledge in the field of Finance and improve my research and writing skills. Additionally, I would like to thank Dr. Michel R. R. van Bremen for his time and consideration as a second reader of this thesis. I would also like to thank all faculty and staff members at the Tilburg School of Economics and Management, and especially my professors at the Department of Finance for their professionalism throughout the period of the Master program in Finance and for all the valuable knowledge and experience I acquired. I would like to thank my teammates who participated in the data collection and filtering stage of this research project for their utmost determination, valuable team work and ideas during the whole process: Stergios Axiotis, Alexander H. Gavrailov, Kevin R. Koopal, Cosmin-Ionut Mazilu, Hendrik J. Petersen, Daan W. J. Röttger, and Sergey Snitsarenko. In addition, I express my gratitude to the Thomson Reuters Datastream customer support team for the information and help they provided. Last but not of least importance, I would like to express my appreciation to my parents for all their support and encouragement during the whole Master program in Finance and throughout my education. I would like to thank all my friends and relatives who support me during the period of my studies. Zhaneta Krasimirova Tancheva 2

4 Abstract Even though momentum strategies have proven to be very profitable and widely-used by investors, recent research shows that they can also experience substantial losses. Therefore, I focus on studying determinants of momentum returns, in particular investor attention and stock market performance. Exploring five years of data ( ) for 22 developed markets this thesis shows that allocating more investor attention to firm-specific indicators than to marketwide information leads to higher overreaction-driven momentum returns, an effect amplified by higher investor protection in a specific country. Stock market performance, measured by distance to 52-week high, is associated with momentum returns and closely related to momentum crashes. 3

5 Table of Contents Title Page...1 Acknowledgements...2 Abstract...3 Table of Contents Introduction Literature Review and Hypotheses Development Data Collection and Filtering Methodology Investor Attention and Momentum Returns Return R-squared Construction Construction and Testing of the Double-sorted by Return R-squared and Momentum Returns Portfolios Investor Attention and Investor Protection Regression Setting Stock Market Performance and Momentum Returns Momentum Returns Construction Momentum Returns and Distance to 52-week High Regression Setting Results Investor Attention and Momentum Returns Summary Statistics and Characteristics of the R-squared Ranked Portfolios Relation Between Investor Attention Allocation and Momentum Returns Investor Protection, Investor Attention, and Momentum Returns Stock Market Performance and Momentum Returns Momentum Returns over Time Distance to 52-week High and Momentum Returns Conclusions References Appendices...25 Appendix A: Definitions of Variables...25 Appendix B: Tables and Figures

6 1. Introduction In their seminal paper, Jegadeesh and Titman (1993) show that investment based on buying stocks which have performed well in the near past (winners) and selling stocks which have performed poorly in the past (losers) brings profits unexplained by systematic risk. Strategies based on buying winners and selling losers, also known as momentum strategies, have proven very popular among traders and money managers around the world. About one-sixth of assets in portfolios in U.S., for example, tend to be managed using quantitative strategies, which often rely on momentum (Swaminathan, 2010). Even though momentum is overall a profitable strategy, producing positive abnormal returns, it experiences periods of decline and crashes (Daniel and Moskowitz, 2011). This is why it is important to understand the determinants of momentum returns. This thesis attempts to bring more light into the field and focuses on two potential drivers of momentum, investor attention and stock market performance, and the extent to which they can determine the magnitude of momentum returns. Studying 22 developed markets in the period provides the opportunity to find the degree to which countryspecific factors and economic conditions can influence the relation between momentum returns and their determinants. Knowing whether and how investor attention and stock market performance in different parts of the world influence momentum returns can help a wide number of investors adjust their strategies and improve their profitability. Attention is a limited cognitive resource (Kahneman, 1973). Since the human brain is restricted in its information-processing capacity, attention plays an important role in learning, selecting information to be processed, and decision making (Hou, Peng, and Xiong, 2006). Peng and Xiong (2006) develop a model which shows that as a consequence of their scarce attention investors exhibit category-learning behavior, namely they focus more on macroeconomic information than on firm-specific factors, especially during economic instability. Following an approach suggested by Durnev, Morck, and Yeung (2004) which uses return R-squared (the R- squared statistic from a regression of stock returns on a market and industry index) as a proxy for allocation of investor attention to different categories of information, Hou, Peng, and Xiong (2006) show that investor attention is of high importance to the profitability of investors since it is related to momentum returns: allocating more attention to firm-specific than to market-wide information drives momentum profits up. In countries with higher investor protection this effect 5

7 is better pronounced: more informed arbitrage is promoted and hence, higher overreaction-driven momentum profits from focusing on firm-specific factors are capitalized (Morck, Yeung, and Yu, 2000; Durnev, et al., 2003). The literature also finds that stock market performance is related to momentum returns. Looking at market returns and lagged market returns researchers have shown that momentum crashes in periods of financial turmoil and the following recovery (Cooper, Gutierez, and Hameed, 2004; Daniel and Moskowitz, 2011; Cheema and Nartea, 2013). On the other hand, using lagged market returns Cooper, Gutierrez, and Hameed, (2004) show that better market state leads to higher levels of overconfidence which results in higher momentum returns. Distance to 52-week high of stocks (the ratio of the current price of a stock to its maximum value in the past year) is an anchor that investors use in their decisions and a measure of stocks' past performance whose usage can bring higher returns from momentum strategies (George and Hwang, 2004). However, Peng and Xiong (2006) make clear that investors tend to exhibit category-learning behavior and pay more attention to market-wide information than to firmspecific factors, while Li and Yu (2011) bring evidence that Distance to 52-week market high is a good predictor of stock returns. Interestingly, Kacperczyk, Nieuwerburgh, and Veldkamp (2009, 2011) find that attention of investors gets reallocated across the different states of the business cycle. Since the positive abnormal returns in the time of recession are low while the overall risk in the market is high, it would be more beneficial for investors to pay a significant amount of attention to the more uncertain factors and therefore allocate more attention to market indexes in a period of crisis than in normal economic conditions. This thesis shows that higher attention allocation to firm-specific information brings higher overreaction-driven momentum returns in four world regions and in Europe, Japan and the World these returns are increasing monotonically with the shift of attention from marketwide to company-specific information. Moreover, I find that this effect is stronger in countries with higher efficiency of the judicial system and the government. Finally, I find that stock market performance is also related to momentum returns. Looking at momentum returns over the separate years in the sample I find that momentum crashes in the period of recovery after the recent financial crisis. Using Distance to 52-week high as a stock market performance measure I 6

8 find that it is closely related to momentum crashes and the smaller the distance to 52-week high is, the higher the momentum returns are. The literature on investor attention has shown that allocation of investor attention to different categories of information can explain momentum returns (Hou, Peng, and Xiong, 2006). I contribute to the field bringing evidence from a new, filtered database from 22 developed markets grouped in four regions. I focus on the cross-sectional specifics of the different countries and on the effect of the recent financial turmoil and recovery to see how they affect the connection between investor attention and momentum returns. Researchers have shown that higher investor protection is associated with higher development of the stock markets. Moreover, higher investor protection brings greater capitalization from overreaction to firm-specific information (Durnev, et al., 2003). Using efficiency of judicial system and efficiency of government as proxies for investor protections, I contribute to the field in that my findings are consistent with the previous studies and confirm their results for a cross-section of 22 developed countries in a period of financial turmoil. The literature has shown that momentum increases during "Up" market states and experiences crashes during financial crises, during recoveries of recessions and "Down" market states (Cooper, Gutierez, and Hameed, 2004; Daniel and Moskowitz, 2011; Cheema and Nartea, 2013). While these studies use past returns as a measure of the performance of the market, I contribute to the field by showing the relation between momentum returns and Distance to 52- week market high, an alternative measure of stock market performance that is a good predictor of stocks returns and a proxy for investor overreactions. The rest of the thesis is organized as follows: Section 2 provides a literature review and development of my hypotheses. Section 3 focuses on the data collection and filtering. Section 4 explains the methodology used to reach the results, which are presented and analyzed in Section 5. Section 6 concludes the study and provides suggestions for future research. References, definitions of variables and tables and figures can be found in Sections 7 and Literature Review and Hypotheses Development Since investors have limited cognitive capabilities they face a trade-off when they decide what to focus their limited attention on (Peng and Xiong, 2006). Studies in the field show that traders 7

9 with explorative expectations (DeLong, et al., 1990) and higher levels of overconfidence (Daniel, Hirshleifer, and Subrahmanyam, 1998) tend to overreact more to information about past stock returns. Moreover, investors would tend to overreact more to information, if they allocate more of their limited attention to it and this is why it is expected that the more investors pay attention to company-specific information, the higher their overreaction would be, bringing higher momentum returns. Hou, Peng, and Xiong (2006) show that return R-squared (as computed by Durnev, Morck, and Yeung, 2004), which obtains higher values the more investors focus on macroeconomic factors compared to firm-specific ones, is negatively related to overreaction-driven momentum returns. I test whether this holds in an international setting of 22 developed markets during a period of turmoil in the financial system and the following recovery. H1: Higher investor attention allocation to firm-specific factors, represented by lower return R- squared, is associated with higher overreaction-driven price momentum returns. Previous studies show that higher efficiency of institutions and investor protection lead to higher development of the financial market (La Porta, Lopez-de-Silanes, Shleifer and Vishny, 2000). Morck, Yeung, and Yu (2000) and Durnev, et al. (2003) show that higher investor protection (measured by a government index, efficiency of judicial system index, and rule of law index, among others) is associated with higher capitalization from overreaction to firm-specific information. They argue that higher levels of investor protection make share prices more easily predictable by arbitrageurs who gather and process information and drive arbitrageurs to trade more intensely and overreact more to information. Based on these results I test whether investors in more protected environment (countries with higher efficiency of the judicial system and the government) allocate their attention to more firm-specific factors and receive higher momentum returns in the period of the recent financial crisis and recovery in comparison to investors in countries with lower investor protection. H2: In countries with higher investor protection focusing more on firm-specific factors than on market-wide information brings higher overreaction-driven momentum returns than in countries with lower investor protection. Using lagged market returns Cooper, Gutierrez, and Hameed, (2004) show that better market state leads to higher levels of overconfidence and higher momentum returns. On the other 8

10 hand, Daniel and Moskowitz (2011) argue that in periods of stock market turmoil and especially the following recovery, momentum returns experience crashes since losers regain their true values after they have been significantly undervalued during the period of financial downturn and risk aversion. Besides returns, investors also pay attention to the price level of stocks and indexes. Distance to 52-week high of stocks is an anchor that investors use in their decisions and momentum strategies based on 52-wek high rather than past returns as a stock performance measure are more profitable (George and Hwang, 2004). Peng and Xiong (2006), however, show that investors tend to pay more attention on market-wide information than on firm-specific factors while Kacperczyk, Nieuwerburgh, and Veldkamp (2009, 2011) bring evidence that investor attention allocation to market-wide information and stock market indexes is higher during periods of crises. Driven by these results, Li and Yu (2011) find that Distance to 52-week high of the market index is a good predictor of stock returns and a proxy for overconfidence and overreaction to news. Based on these findings I use Distance to 52-week market high as a measure of stock market performance and I test whether it can determine momentum returns and explain momentum crashes. I expect that the closer the market is to its 52-week high value, the higher the momentum returns would be due to higher overreaction of investors and vice versa. H3: Lower distance to 52-week high market index value, as a measure of stock market performance, is associated with higher overconfidence- and overreaction-driven momentum returns and vice versa. 3. Data Collection and Filtering To test my hypotheses I analyze a total of twenty-two developed countries from four regions (Europe, Japan, Asia-Pacific excluding Japan and North America excluding U.S.) in the period January December Data for all active, dead and suspended common-share stocks available at Thomson Reuters Datastream, whose primary quote is at one of the stock exchanges in a given country, are selected. As suggested by previous studies in the field I collect weekly data with weeks starting on Wednesday in order to avoid the Monday effect, the tendency for Monday stock returns to be low and negative on average, compared to other weekdays 9

11 (Pettengill, 2003). I also collect data at the end of each month necessary for the analysis of momentum returns. The main variables which are necessary for the analysis are Total Return Index (RI), Market Value (MV), and SIC (Standard Industrial Classification) code (SICCODE1) for each company which determines the industry in which it operates. Total Return Index is a measure of the theoretical growth in the value of a share over a certain period of time and is preferred to the Price Index since it assumes that dividends are re-invested to purchase additional shares of equity. Thus, I use it to compute the returns of each of the stocks and of the market indexes. Market Value for each company is defined as the number of shares outstanding times the share prices. A number of static variables are necessary in order to check the cleanness of the data: Name, ADR code, DS Code, Trading Volume, Stock type, Geography group, Coverage flag, Dates of initial listing and delisting. I substitute the missing SIC codes of different classes or types of stocks of the same parent company where possible. Afterwards, I follow the two-level approach suggested by Ince and Porter (2006) to correct inaccuracies in the data and non-common share stocks. In the first level I remove all nonlocal companies and non-equity stocks using the variables Geography group and Stock type, respectively. Subsequently, I substitute with missing values all observations of Total return index after the stock is delisted which are incorrectly repeated and equal to the last trading price of the stock. Since the data are in US dollars and repeated values are only observed in local currency, the changing exchange rate alters slightly the values of return index after delisting. Therefore, I use Trading volume and the Date of delisting in the cleaning procedure. When the Trading volume is zero or the delisting date is reached, the values of Total return index are substituted with missing. Another problem with Total return index is that when the return is calculated some extreme values of over 300% can be observed. These values are substituted with missing. In order to deal with outliers, winsorising (substituting outliers with the 1 st and 99 th percentile of returns) for each period of time is applied. In the second level of screening the data I check whether the names of the stocks contain phrases which suggest that these stocks are not common shares such as "REIT" (real estate investment trust), "PREF", "PF" (preferred stock), "RESTRICTED" (restricted stock), 10

12 "DEFERRED" (deferred stock), "CLOSE" (close-ended fund), "ADR" (American depository receipt), "GDR" (Global depository receipt), etc. and remove the stocks. In addition to these two levels of screening I remove all stocks with Coverage flag "C" which are not supported by Datastream. Furthermore, I use a variable representing the type of share (TRCSDESCRIPTION) to remove all remaining non-common share stocks. Finally, I manually check all stocks for duplicates such as stocks with the same ADR code and/or Name, but different classes of shares, types of voting rights and double entries. The stocks which are most representative for the company such as Class A (over Class B) stocks and stocks with higher Trading volume are left in the database. Thus, only one stock per company remains in the sample (unless the stocks are not simultaneously traded) in order to avoid overrepresentation of certain companies in comparison to others. To analyze the effect of efficiency of institutions and corruption perceptions across countries on momentum returns I use variables available at datasets constructed by La Porta, Lopez-de-Silanes, Shleifer and Vishny (2000) (efficiency of judicial system) and La Porta, Djankov, Lopez-de-Silanes and Shleifer (2009) (efficiency of government). I retrieve the control variables GDP growth rate and Inflation rate from Thomson Reuters Datastream and Market capitalization from World Bank. All variables are described in detail in Appendix A. 4. Methodology 4.1. Investor Attention and Momentum Returns Return R-squared Construction In order to investigate the effect of investor attention on momentum returns, I compute the return R-squared in the spirit of Durnev, Morck, and Yeung (2004). For this purpose I estimate the following OLS regressions for each stock in a specific country or region for every quarter using weekly observations in the period January, 2008 December, 2012: R it = α + β 1 R Mt + β 2 R It + ε it (1) The dependent variable in the model R it is the weekly return on each of the stocks at time t, computed using the Total Return Index (RI). The first explanatory variable R Mt captures the market index return which can be found in the same way as the return of each of the stocks (using the Total return index). For each country or region I select the corresponding MSCI index 11

13 which is a good representation of the market. The second explanatory variable R It is the industry return at time t, where industry is defined as the set of companies that have the same first three digits of their SIC (Standard Industrial Classification) codes as suggested by Durnev, Morck, and Yeung (2004). To find the Industry return R It the following formula is used: R It = MV j(t 1) R jt j I,j i (2) j I,j i MV j(t 1) where MV j(t 1) is the Market value of company j in industry I at time t-1 and R jt is the return of company j which operates in industry I at time t. Essentially the Industry return R It is the sum of the market weighted returns of all companies in the industry at time t, excluding the firm (i) for which regression (1) is being estimated due to possible spurious correlations between the company and industry returns in industries with a low total number of companies. β 1 and β 2 are regression (1) coefficients, α is the intercept of the regression and ε it is the error term at time t for each company i. The estimated return R-squared which is of interest in the analysis is: R 2 = 1 2 t ε i,t t(r i,t R )² i,t Construction and Testing of the Double-sorted by Return R-squared and Momentum Returns Portfolios To obtain the double-sorted by return R-squared and momentum returns portfolios I follow the approach of Hou, Peng, and Xiong (2006). First, at the end of every month (t) I sort all stocks into quintiles based on their return R-squared values over the full sample data. Since return R- squared is computed quarterly using the weekly observations within the past quarter, while momentum returns are computed on monthly basis, the monthly observations of return R- squared within each quarter are equal and have the same rank. Second, within every return R- squared quintile and month (t) I sort the stocks into quintiles based on their past six months returns (t-6:t-2) (skipping the most recent month (t-1)). The equally-weighted returns of the generated 25 portfolios are estimated over a holding period of the following one month (t:t+1). I also perform a robustness test using a holding period of 3 months (t:t+3). Finally, I compute the average momentum returns within the five R-squared quintiles, where momentum returns are defined as the differences between the portfolios with the highest past return (momentum quintile 5) and the portfolios with the lowest past return (momentum quintile 1). The variable (3) 12

14 Difference is defined as the difference between the momentum returns in the lowest return R- squared quintile and the highest return R-squared quintile. To test whether momentum returns decrease over the R-squared quintiles (meaning that higher allocation of attention to firm-specific factors than to market-wide information bring higher overreaction-driven momentum returns) I report the Spearman's rank correlation statistic (the correlation between the R-squared number of quintile and the rank of the momentum returns) and the corresponding p-values. A correlation of shows that momentum returns are monotonically decreasing from R-squared quintile 1 to R-squared quintile 5 and that allocating more attention to company characteristics than to market-wide information yields higher momentum returns which supports Hypothesis 1. A correlation of 1.00 supports the opposite statement. p-values showing whether the coefficient is statistically significantly different from zero are reported in italics below. To find the adjusted for Fama-French three factors momentum returns for each of the quarters in the separate regions I estimate the following regression and report the intercept α: Momentum return t = α + β 1 R Mt + β 2 SMB t + β 2 HML t + ε t (4) where momentum return is the difference between the return in the highest momentum quintile and the lowest momentum quintile within each R-squared quintile. The three Fama-French factors are the excess return on the market portfolios (R M ) and two factor-mimicking portfolios which measure size and book-to-market effect (SMB and HML) Investor Attention and Investor Protection Regression Setting At the next step of the analysis I test the effect of investor protection on the difference between momentum returns in the lowest and highest return R-squared quintile (capitalization of firmspecific information). I estimate the following cross-sectional OLS regression using panel data with monthly observations for all 22 countries over the period January, December, 2012 (a total of 12 5=60 observations per country): Difference it = α + β 1 Investor protection it + β 2 X it + β 3 Y + ε it (5) where the dependent variable is Difference (the capitalization of firm-specific information, described in detail in section and in Appendix A). Investor protection is measured using 13

15 two different indexes: efficiency of judicial system and efficiency of government. I estimate two separate regressions using only one of the measures per regression as a variable of interest. I also consider a model including both variables as proxies for investor protection. Vectors X and Y are a vector of country-specific control variables, capturing the size of the economy and market of the country (GDP growth rate, Market capitalization and Inflation), and a vector of year-fixed effects, respectively. In markets with fewer stocks there is a threat of a spurious correlation between the return R-squared value and the number of stocks since this small number of firms dominates the market. To control for this effect I include a measure of the stock market, Market capitalization as a percent of GDP. To control for economic conditions that may drive the results I include GDP growth rate and Inflation in the regression Stock Market Performance and Momentum Returns Momentum Returns Construction In order to calculate the momentum returns over time in a specific country/region momentum strategies introduced by Jegadeesh and Titman (1993) are applied. At the end of each month (t) I rank all stocks in a specific region/country into 5 quintiles based on their past 6 months (t-6:t-2) returns (skipping the most recent month (t-1)). Then I compute the returns over a holding period of the following 1 month (t:t+1). The average difference between the returns in the holding periods of the stocks in the highest quintile (winners) and the stocks in the lowest quintiles (losers) yields the momentum return. To find whether stock market performance is associated with momentum returns I present the momentum returns across the five years in the sample and observe the changes of magnitude of momentum returns in all return R-squared quintiles Momentum Returns and Distance to 52-week High Regression Setting To test how Distance to 52-week high is associated with momentum returns the following OLS regression is estimated for all 22 countries over the period January, 2008 December, 2012: Momentum return it = α i + β 1 Distance52 it + β 2 X it + β 3 Y + ε it (6) where the dependent variable is Momentum return, as constructed in section Distance to 52-week high is the variable of interest measuring the nearness of the current MSCI index value 14

16 to its highest value in the past 52 weeks, which is constructed in the spirit of Li and Yu (2011) (a detailed definition can be found in Appendix A): Distance 52,t = MSCI t MSCI 52max,t (7) As in equation (5), vectors X and Y are vectors of country-specific control variables (GDP growth rate, Market capitalization, and Inflation rate) and year-fixed effects, respectively. Furthermore, I estimate a fixed effects model, controlling for country-specific characteristics which remain constant within a country over time. This results in α i intercepts of the regressions which are different for each country. Note that in (5) country-fixed effects could not be added since the variables of interest (efficiency of judicial system and efficiency of government) remain constant over time, while fixed effects models use within-country variations of the variables over time. 5. Results 5.1. Investor Attention and Momentum Returns Summary Statistics and Characteristics of the Return R-squared Ranked Portfolios Table 1, Panel A presents summary statistics of return R-squared among the different countries and regions. The values seem to be highly comparable which is not surprising considering the fact that the markets included in the sample are developed markets and globalization causes decreases in diversification among markets. The return R-squared in the regions Europe and Japan is highest (35%) while in North America Excl. U.S. it is lowest (30%) meaning that on average investors in North America Excl. U.S. allocate more attention to firm-specific factors in comparison to market-wide information than in Europe. Looking at the return R-squared across countries Germany, Ireland, Singapore and United Kingdom exhibit the lowest values of R- squared while countries such as Finland, Italy, The Netherlands, Portugal and Spain have the highest R-squared values. Overall, the median return R-squared values are lower than the mean ones meaning that the right tail of the return R-squared is heavier than the left tail. Information about the number of observations and number of stocks after cleaning the data is available in Table 1, Panel A. 15

17 Table 2 provides information about the characteristics of the companies classified in the portfolios sorted by return R-squared. In the lowest R-squared quintile the values are around 5% while in the highest R-squared quintile the values are around 72%, about 14 times larger. The firm size (measured by the market value) and firm trading volume increase monotonically from R-squared quintile 1 to R-squared 5 in all four regions. This is consistent with the results of Chan and Hameed (2006) showing that firms with lower return R-squared both in U.S. and emerging markets are smaller and are covered less by analysts Relation between Investor Attention Allocation and Momentum Returns Table 3 reports the average returns of the double-sorted by return R-squared and momentum returns portfolios in the four regions and the World. Panel E1 which reports the results for the World shows that in the lowest return R-squared quintile the average return in the highest Momentum quintile is 53 basis points, while the return in the lowest Momentum quintile is 65 basis points. This yields a spread between the two (momentum return) of -12 points. In the highest return R-squared quintile the average return in the highest Momentum quintile is -50 basis points and the average return in the lowest Momentum quintile is 50 basis points. This results in a momentum return of -100 points (significant at a 1% significance level) which is lower than the momentum return in the lowest R-squared quintile calculated above (-12 basis points). The Spearman's rho coefficient of (significant at a 1% significance level) shows that the momentum returns are monotonically decreasing from R-squared quintile 1 to R-squared quintile 5. Hence, allocation of more attention to company characteristics than to macroeconomic factors increases the overreaction of investors and results in higher momentum returns. Using a holding period of 3 months instead of 1 month for robustness check brings the same results: the Spearman's coefficient of (significant at 1% significance level) shows that momentum returns are monotonically decreasing across the R-squared quintiles. Panels A1, A2, B1 and B2, Table 3 show that similar to the findings in the World, momentum returns are monotonically decreasing across R-squared quintiles in Europe and Japan for momentum returns calculated using holding periods of both 1 and 3 months. Even though the momentum returns in the lowest R-squared quintiles are higher than the momentum returns in the highest R-squared quintiles in Asia-Pacific Excl. Japan and North America Excl. U.S., momentum returns are not monotonically decreasing across R-squared quintiles. This shows that 16

18 allocating attention to company characteristics to a greater extent than to macroeconomic factors brings higher momentum returns, but the relation is not as strong as it is in Europe, Japan, and the World overall. The intercepts α from the regressions of Momentum returns on the three Fama-French factors in the different return R-squared quintiles are insignificant and not monotonically decreasing over the return R-squared quintiles for most regions. The reason for the results can be the low number observations due to the small period of time covered (5 years times 12 months amounts to 60 observations) or the overall negative momentum returns. This results is not consistent with the findings of Hou, Peng, and Xiong (2006). Overall, the results in Table 3 confirm the findings of Hou, Peng, and Xiong (2006) and provide evidence in support of Hypothesis 1. Momentum returns in the lowest R-squared quintiles are higher than the momentum returns in the highest R-squared returns in all regions and the World and are monotonically decreasing across R-squared quintiles in Europe, Japan and the World. Hence, higher attention allocation to firm-specific information brings higher overreaction-driven momentum returns Investor Protection, Investor Attention, and Momentum Returns To find whether higher investor protection can determine the capitalization from overreaction to firm-specific information I use the model described in Section 4.1.3: Difference it = α + β 1 Investor protection it + β 2 X it + β 3 Y + ε it (8) Table 4 reports the results from the regressions. As predicted, higher efficiency of the judicial system of the country is related to higher overreaction-driven momentum returns from allocating more attention to firm-specific factors than to market-wide information (the variable Difference, explained in Appendix A). An increase of 0.5 point of the efficiency of judicial system index brings additional 0.77% ( =0.77%) difference between the momentum return in the lowest and highest return R-squared quintile (p-value<0.10) (Table 4, Column 3). This, compared to the mean of the dependent variable Difference (0.23%) (Table 1, Panel B), corresponds to approximately 335% increase which is of economic importance. Increasing the efficiency of judicial system index with 1 point leads to additional 1.54% ( =1.54%) (Table 4, Column 3) gain from overreaction-driven momentum which compared to the mean 17

19 value of the dependent variable (Difference) (0.23%) (Table 1, Panel B) is an increase of about 670% and is economically significant. Efficiency of the judicial system is an important indicator for the protection of investors in a country since higher efficiency of the judicial system implies faster and more productive judgments in court trials, better economy and business and investment climate. In countries with higher investor protection informed arbitrageurs tend to trade more intensely and overreact more to information. This results is robust to controlling for country-specific macroeconomic and financial market indicators (GDP growth rate, Inflation rate and Market capitalization). The power of efficiency of judicial system remains also after using a year-fixed model, controlling for unmeasured factors which influence all countries to the same degree over the years. Efficiency of government is also statistically significantly associated with difference between momentum returns in the lowest and highest return R-squared quintile. Higher levels of government efficiency are related to greater capitalization from firm-specific overreaction-driven momentum. This result is robust to controlling for GDP growth rate, Market capitalization and Inflation. Increasing the efficiency of government with one standard deviation (0.4355) (Table 1, Panel B) leads to 1.18% ( =1.18%) (Table 4, Column 5) increase in capitalization from overreaction to firm-specific information (Difference, explained in Appendix A). This can be translated into about 500% increase above the average value of Difference (0.0023) (Table 1, Panel B), which is of economic importance. This association loses its power when year fixed effects are included in the model (Table 4, Column 6), meaning that there are some common factors that influence all countries to the same extent over the years and drive the effect of efficiency of the government on capitalization of overreaction to firm-specific information. Table 4, Column 7 presents the results when I include both efficiency of judicial system and efficiency of government as proxies for investor protection and add control variables and year fixed effects. Efficiency of government loses its explanatory power; however, efficiency of judicial system remains significant, meaning that it is not driven by efficiency of government. This result should be analyzed with caution due to the relatively high correlation between efficiency of judicial system and efficiency of government (0.77). 18

20 The results from this section bring evidence in support of Hypothesis 2, showing that the level of investor protection affects the difference between momentum returns from allocating attention to firm-specific factors and market-wide information and leads to higher capitalization from overreaction to company-specific information Stock Market Performance and Momentum Returns Momentum Returns over Time Investigating the momentum returns over time in the different return R-squared quintiles, I observe that in all four regions and the World momentum returns are negative in 2009 for all R- squared quintiles. In Japan and North America Excl. U.S., however, negative and significant returns from the momentum strategy are also observed in 2010 which can be attributed to the different development of the financial crises in these regions. Panel E shows the development of momentum returns in the World over time across the return R-squared quintiles. Momentum strategies yield negative returns in 2009 irrespective of the holding period used one or three months (Table 5, Panel E). In some R-squared quintiles in 2010 in the World I also observe negative momentum returns which might be due to the different timing of the crisis in Japan and North America Excl. U.S. In contrast to returns in the rest of the years in the World, however, the momentum returns in 2010 in all R-squared quintiles are not statistically significantly different from 0 (p-values>0.1) (Table 5, Panel E). Figures 1-10 track the monthly development of the momentum returns across the regions and return R-squared quintiles. This brings clear evidence that momentum experiences a crash in periods of market distress and the upcoming recovery irrespective whether investors allocate more attention to firm-specific factors or market-wide information. Analyzing momentum returns in the different R-squared quintiles over time gives evidence that in the highest R-squared quintiles momentum returns are larger than in the lowest R-squared quintiles (the variable Difference experiences a decrease) during periods of low stock market performance. This means that investors pay more attention to market-wide factors in these periods, which is consistent with the findings of Kacperczyk, Nieuwerburgh, and Veldkamp (2009, 2011). This result is better pronounced in Europe, Asia-Pacific Excl. Japan and the World than in Japan and North America Excl. U.S. 19

21 Distance to 52-week High and Momentum Returns Having evidence that momentum returns are negative in 2009, after the period of financial crisis and during the period of recovery I proceed with testing whether another indicator of stock market performance and a proxy for investors' overreaction to news, which is proven to be a good predictor of future returns, Distance to 52-week high (as defined by Li and Yu (2011)) can explain momentum returns. The model that I test (explained in Section ) is presented with the following equation: Momentum return it = α i + β 1 Distance52 it + β 2 X it + β 3 Y + ε it (9) Table 6 reports the results of the analysis. As predicted, Distance to 52-week high is associated with Momentum returns. Increasing the variable Distance to 52-week high with one standard deviation (0.1828) (Table 1, Panel B), which corresponds to decreasing the distance of the current market index value to its past 52-week maximum, leads to additional 1.63% momentum returns (18.28% =1.63%) (Table 6, Column 4). Compared to the mean value of momentum returns (0.30%) (Table 1, Panel B), this corresponds to an economically significant increase of more than 543%. The results are not driven by country-specific indicators of the macroeconomic conditions and market size (GDP growth, Inflation, and Market capitalization) (Column 2). Adding country-fixed effects and thus controlling for country-specific factors which are constant over time does not influence the results (Column 3). After controlling for factors which have equal effect on all countries over time with the year-fixed effects model Distance to 52-week high remains a good indicator of momentum returns (Column 4). Figures visualize the relation between Distance to 52-week high, Momentum returns and Market returns. It can be observed that in all regions momentum crashes when losers (stocks that performed poorly in the past 6 months) perform much better than the market and the winners (stocks that performed well in the past 6 months). Daniel and Moskowitz (2011) suggest that this occurs because losers have been highly undervalued during a market downturn and experience larger gains than those of winners when the market recovers, which results in negative momentum returns. Distance to 52-week high, on the other hand, moves closely to momentum returns and reaches its minimum nearly at the same time as momentum returns for all four regions and the 20

22 World. Since investors experience category learning behavior, they pay attention to this indicator as an important psychological anchor and a good predictor of future returns. When the market index value deviates further away from the maximum over the past 52 weeks investors become more risk averse and avoid losers, bringing their prices down. Around the period in which the current value is furthest away from the past 52-week maximum and the market starts picking up investors' confidence starts increasing, the market for losers recovers, experiencing higher gains than winners and momentum crashes are observed. When the market index current value becomes closer to its highest value over the past 52 weeks, future aggregate market returns are expected to be positive, investors become overconfident and the aggregate overreaction to news raises, bringing higher momentum returns. Using Distance to 52-week high as a measure of stock market performance shows that better market performance is associated with higher momentum returns which supports Hypothesis Conclusions Strategies based on momentum trading have proven vary profitable and popular among investors. However, research has found that these strategies do not always bring positive abnormal returns and experience crashes. Therefore, it is important to find what factors determine the magnitude of momentum returns. This thesis provides evidence in support of two such indicators: investor attention and stock market performance. Using data from 22 developed countries from four regions of the world I show that investor attention allocation to firm-specific factors bring higher overreaction-driven momentum returns than attention allocation to marketwide information. Furthermore, I show that higher efficiency of judicial system and the government of a given country are associated with higher capitalization of overreaction to firmspecific information. Finally, this study brings evidence that stock market performance determines momentum returns. Focusing on Distance to 52-week high as a measure of stock market performance, which is proven by researchers to be a proxy for overreaction to information and a predictor of future returns, this thesis shows that the closer the current market index value is to its 52-week maximum, the higher momentum returns are. Distance to 52-week high is also closely related to momentum crashes. This field of research is of high importance for the decision making of investors and their profitability. Therefore, further research will be of high interest. This thesis focuses on 22 21

23 developed markets. However, due to globalization and high flow of information these markets are highly related. In order to make better generalizations for the whole world and find whether these results hold in more countries, emerging and frontier markets around the world should be investigated. A longer period of data could further contribute to the field and cover the effects of different natural phenomena, country-specific political events and policies on the magnitude of momentum returns. Last but not least, it is essential to explore further determinants of momentum returns, using behavioral and quantitative models. 22

24 7. References Investor Attention, Stock Market Performance, and Momentum Returns Chan, K. and A. Hameed, 2006, Stock Price Synchronicity and Analyst Coverage in Emerging Markets, Journal of Financial Economics 80, Cheema, M. A. and G. V. Nartea, 2013, Momentum Returns, Market States and the 2007 Financial Crisis, Working paper. Cooper, M. J., R. C. Gutierrez JR., and A. Hameed, 2004, Market States and Momentum, Journal of Finance 59, Daniel, K., Hirshleifer D., and A. Subrahmanyam, 1998, Investor Psychology and Security Market Under- and Overreactions, Journal of Finance 53, Daniel, K., R. Jagannathan, and S. Kim, 2012, Tail Risk in Momentum Strategy Returns, Columbia University and Northwestern University Working Paper. Daniel, K. and T. Moskowitz, 2011, Momentum Crashes, Columbia University and University of Chicago Working Paper. De Long, B. J., A. Shleifer, L. H. Summers, and R. J. Waldman (1990), Positive Feedback Investment Strategies and Destabilizing Rational Speculation, Journal of Finance 45, Djankov, S., R. La Porta, F. Lopez-de-Silanes, and A. Shleifer, 2010, Disclosure by Politicians, American Economic Journal: Applied Economics 2, Durnev, A., R. Morck, B. Yeung, and P. Zarowin (2003), Does Greater Firm-specific Return Variation Mean More or Less Informed Stock Pricing?, Journal of Accounting Research 41, Durnev, A., R. Morck, and B. Yeung, 2004, Value-enhancing Capital Budgeting and Firm-Specific Stock Return Variation, Journal of Finance 59, Fama, E. and K. R. French, 1993, Common Risk Factors in the Returns on Stocks and Bonds, Journal of Financial Economics 33, George, T. and C. Hwang, The 52-week High and Momentum Investing, Journal of Finance 59, Hou, K., L. Peng, and W. Xiong, 2006, R 2 and Price Inefficiency, Fisher College of Business Working Paper No Hou, K., L. Peng, and W. Xiong, 2008, A Tale of Two Anomalies: The Implications of Investor Attention for Price and Earnings Momentum, Working Paper, Princeton University. Ince, O. and B. Porter, 2006, Individual Equity Return Data From Thomson Datastream: Handle with Care!, Journal of Financial Research 29,

25 Jegadeesh, N. and S. Titman, 1993, Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency, Journal of Finance 48, Kacperczyk, M., S. van Nieuwerburgh, and L. Veldkamp, 2009, Rational Attention Allocation over the Business Cycle, NBER Working Paper Kacperczyk, M., S. van Nieuwerburgh, and L. Veldkamp, 2011, Time-Varying Fund Manager Skill, NBER Working Paper Kahneman, D., 1973, Attention and Effort, Englewood Cliffs, NJ: Prentice-Hall La Porta, R., F. Lopez-de-Silanes, A. Shleifer and R. Vishny, 2000, Investor Protection and Corporate Governance, Journal of Financial Economics 58, Li, J. and J. Yu, 2011, Investor Attention, Psychological Anchors, and Stock Predictability, Journal of Financial Economics 104, Li, X., R. S. Mahani, and V. Sandhya, 2011, Does Investor Attention Affect Stock Prices?, Social Science Research Network. Mondria, J. and T. Wu, 2011, Asymmetric Information and Stock Returns, Working Paper. Mondria, J., T. Wu, and Y. Zhang, 2010, The Determinants of International Investment and Attention Allocation: Using Internet Search Query Data, Journal of International Economics 82, Morck, R., B. Yeung, and W. Yu (2000), The Information Content of Stock Markets: Why Do Emerging Markets Have Synchronous Stock Price Movements? Journal of Financial Economics 58, Peng, L., 2005, Learning with Information Capacity Constraints, Journal of Financial and Quantitative Analysis 40, Peng, L. and W. Xiong, 2006, Investor Attention, Overconfidence, and Category Learning, Journal of Financial Economics 80, Peng, L., W. Xiong, and T. Bollerslev, 2007, Investor Attention and Time-varying Comovements, European Financial Management 13, Pettengill, G. N., A Survey of the Monday Effect Literature, Journal of Business and Economics 42, Swaminathan, B., 2010, Quantitative Money Management: A Practical Application of Behavioral Finance, Working Paper. 24

26 8. Appendixes Appendix A: Definitions of Variables Variable Return R-squared Momentum return Definition Return R-squared is the R-squared which is estimated for every quarter in the period January, December, 2012 after regressing the returns of each of the common share stocks in a specific region (country) on weekly market returns and value-weighted industry returns using weekly data within the quarter. For market returns the MSCI indexes of the regions (countries) are used. Companies are considered to belong to the same industry if they have the same first three digits of SIC code (Standard Industrial Classification). Return R-squared is used in the analysis of Durnev, Morck, and Yeung (2004) and Hou, Peng, and Xiong (2006). The variables necessary for constructing return R-squared (Total Return Index (RI), Market Value (MV), Volume (VO), MSCI indexes of all countries and regions, SIC codes (SIC1) of all common shares stocks in the selected countries and a number of static firm-specific variables necessary for cleaning the data) are downloaded from Thomson Reuters Datastream. At the end of each month (t) all stocks in a specific region/country are sorted into 5 quintiles based on their past 6 months returns (t-6:t-2) (skipping the most recent month (t-1)). The returns over a holding period of the following 1 month (t:t+1) are computed. The average difference between the returns in the holding periods of the stocks in the highest quintile (winners) and the stocks in the lowest quintiles (losers) yields the momentum return. Difference First, at the end of each month (t) the stocks are ranked into quintiles based on their return R- squared values over the full sample data. Afterwards, within every return R-squared quintile and month the stocks are sorted into quintiles based on past six months returns (t-6:t-2) (skipping the most recent month (t-1)). The equally-weighted returns of the generated 25 portfolios are estimated over a holding periods of the following one month (t:t+1). Finally, the average momentum returns within the five R-squared quintiles are computed, where momentum returns are defined as the differences between momentum quintile 5 and 1. Difference is the difference between the momentum returns in the lowest return R-squared 25

27 quintile and the highest return R-squared quintile. It measures the capitalization of overreaction to firm-specific information. Efficiency of judicial system Efficiency of government Distance to 52-week high Market capitalization GDP growth rate Inflation rate The variable is an average of investors' assessments of efficiency of the judicial system in a specific country in the period (it remains constant over time). This measure is used by La Porta, Lopez-de-Silanes, Shleifer and Vishny (2000). The range of the variable is [1;10], where 10 is the highest efficiency score and 1 is the lowest efficiency score. The variable is a measure of the efficiency of each country's government and represents the average score of the Kaufmann government effectiveness index between 1998 and 2007 (which is constant over time). This index represents the perceptions of quality of public and civil services, level of independence from political pressures, quality of implementations of policies and credibility in commitment to policies. It is used in Djankov, La Porta, Lopez-de-Silanes and Shleifer (2010). The range of the variable is [-2.5; 2.5], where 2.5 is the highest efficiency score and -2.5 is the lowest efficiency score. The ratio of the current value of the market index (MSCI of the specific country) to its highest value during the past 12 months (52-weeks) which is considered as a proxy for the extent to which investors under- and overreact to news. It is estimated as in Li and Yu (2011): Distance 52,t=MSCI t /MSCI 52max,t Higher values correspond to lower distance to maximum. Market capitalization (market value) of listed companies (excluding investment companies, mutual funds and investment vehicles) in the specific country as a percent of GDP, where market capitalization equals the number of shares outstanding times the share prices. Retrieved from World Bank Annual percent change of the GDP of a specific country. Retrieved from Thomson Reuters Datastream Annual inflation rate of a specific country. Retrieved from Thomson Reuters Datastream 26

28 Appendix B: Tables and Figures Table 1: Summary Statistics Panel A: Summary statistics of return R-squared in Europe, Japan, Asia-Pacific Еxcl. Japan, North America Excl. U.S. and all countries in the sample estimated by regressing the returns of each of the common share stocks in the specific region/country on weekly market return and industry returns using weekly data over the period January, December, Appendix A contains a detailed definition of return R-squared. Mean Std Min 25 th percentile Median 75 th percentile Max Number of obs. Number of stocks Europe ,933 7,806 Japan ,055 4,115 Asia-Pacific Excl. Japan ,156 4,483 North America Excl. U.S ,132 4,037 World ,276 20,441 Australia ,168 2,047 Austria , Belgium , Denmark , Finland , France ,775 1,065 Germany ,337 1,897 Greece , Hong Kong ,030 1,500 Ireland , Italy , The Netherlands , New Zealand , Norway , Portugal , Singapore , Spain , Sweden , Switzerland , United Kingdom ,006 2,507 27

29 Panel B: Summary statistics of dependent and explanatory variables in the regressions Mean Std Minimum 25th percentile Median 75th percentile Maximum Momentum return Difference Efficiency of judicial system Efficiency of government Distance to 52-week high GDP growth rate Market capitalization Inflation rate Table 2: Characteristics of Return R-squared Sorted Portfolios This table presents the mean return R-squared values, Market values (MV) and Trading Volume. At the end of each month all stocks are sorted into quintiles based on their R-squared values over the full sample data (January, December, 2012). Details about the database and method used for computing return R-squared and presented in Appendix A. Panel A: Europe Mean R-squared Mean MV Mean Volume R-squared Q , R-squared Q , R-squared Q , R-squared Q , , R-squared Q , , Panel B: Japan Mean R-squared Mean MV Mean Volume R-squared Q , R-squared Q , R-squared Q , R-squared Q , , R-squared Q , ,

30 Panel C: Asia-Pacific Excl. Japan Mean R-squared Mean MV Mean Volume R-squared Q , R-squared Q , R-squared Q , R-squared Q , R-squared Q , , Panel D: North America Excl. U.S. Mean R-squared Mean MV Mean Volume R-squared Q , R-squared Q , R-squared Q , R-squared Q , R-squared Q , , Panel E: World Mean R-squared Mean MV Mean Volume R-squared Q , R-squared Q , R-squared Q , R-squared Q , R-squared Q , ,

31 Table 3: Return R-squared and Price Momentum The average monthly raw returns on portfolios double-sorted by return R-squared and price momentum are presented over the sample period January, December, First, at the end of each month (t) the stocks are ranked into quintiles based on their R-squared values over the full sample data. Afterwards, within every R-squared quintile and month the stocks are sorted into quintiles based on past six months returns (t-6;t-2) (skipping the most recent month (t- 1)). Finally, the equally-weighted returns of the generated 25 portfolios are estimated over two different holding periods: the following one month (t+1) and three months (t+3). Their average values and t-statistics are reported. Column 5-1 presents the average momentum returns within the five R-squared quintiles, where momentum returns are defined as the differences between momentum quintiles 5 and 1. The last column presents the intercepts of the regressions of 5-1 differences (momentum returns) on the three Fama-French factors (Market excess return, SMB and HML). The last rows report the Spearman's rank correlation statistic (the correlation between the R-squared number of quintile and the rank of the momentum returns) and the corresponding p-values. t- statistics and F-statistics for the 5-1 Column are reported in parenthesis. ***, ** and * refer to 1%, 5% and 10% significance levels, respectively. Panel A1:Europe, estimation period - past 6 months (skipping the most recent month), holding period-following 1 month (No obs. 335,052) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** ** *** (-8.83) (-5.70) (-2.57) (-0.12) (-0.87) (31.78) (1.15) R-squared Q *** *** *** *** (-5.90) (-7.01) (-3.48) (-0.13) (-0.22) (16.15) (1.16) R-squared Q *** *** * ** *** * (-9.15) (-6.01) (-1.89) (-0.45) (-2.57) (21.68) (1.86) R-squared Q *** *** ** *** (-7.24) (-3.53) (-0.39) (-1.41) (-2.05) (13.47) (0.60) R-squared Q *** *** (-3.35) (0.13) (-0.18) (-0.13) (-0.32) (4.61) (0.55) Spearman's rho ** ** p-value

32 Panel A2: Europe, estimation period - past 6 months (skipping the most recent month), holding period - following 3 months (No obs. 320,321) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** ** *** *** (-13.31) (-8.84) (-6.02) (-2.37) (-5.65) (29.88) (1.25) R-squared Q *** *** *** *** *** *** * (-11.14) (-10.02) (-5.62) (-2.60) (-3.60) (28.84) (1.88) R-squared Q *** *** *** *** *** ** (-8.63) (-6.68) (-3.85) (-0.87) (-3.71) (12.26) (2.32) R-squared Q *** ** ** ** (-4.17) (-1.52) (-0.13) (0.62) (-1.97) (2.43) (1.72) R-squared Q *** *** *** *** ** (2.88) (6.72) (4.91) (4.42) (2.04) (0.36) (1.59) Spearman's rho 1.00*** 0.90** 1.00*** 0.90** 0.90** -1.00*** 0.60 p-value

33 Panel B1: Japan, estimation period - past 6 months (skipping the most recent month), holding period - following 1 month (No obs. 215,098) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** ** *** (7.78) (9.57) (6.07) (4.41) (2.22) (15.39) (0.48) R-squared Q *** *** *** *** *** (7.84) (8.19) (7.66) (3.06) (1.16) (22.24) (-0.07) R-squared Q *** *** *** *** *** (6.93) (9.06) (3.49) (3.00) (0.15) (22.91) (0.13) R-squared Q *** *** *** ** * *** (7.28) (7.47) (4.89) (2.31) (-1.92) (42.23) (-1.10) R-squared Q *** *** *** *** *** (6.50) (5.26) (3.09) (-1.15) (-3.81) (53.06) (-0.99) Spearman's rho ** *** -1.00*** -1.00*** p-value Panel B2: Japan, estimation period - past 6 months (skipping the most recent month), holding period - fol. 3 months (No obs. 206,972) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** *** (12.49) (14.57) (8.29) (6.23) (3.53) (31.78) (-1.21) R-squared Q *** *** *** *** *** *** ** (14.63) (12.60) (9.75) (4.59) (2.66) (16.15) (-1.98) R-squared Q *** *** *** *** *** (16.44) (15.22) (7.45) (3.90) (0.40) (21.68) (-1.66) R-squared Q *** *** *** *** * *** *** (13.75) (13.84) (9.10) (4.11) (-1.91) (13.47) (-2.69) R-squared Q *** *** *** *** *** *** (10.46) (10.58) (7.26) (0.95) (-4.62) (4.61) (-2.75) Spearman's rho ** -1.00*** -0.90** 0.10 p-value

34 Panel C1: Asia-Pacific, estimation period - past 6 months (skipping the most recent month), holding period - following 1 month (No obs. 217,495) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** (7.79) (7.91) (9.36) (7.87) (8.12) (0.06) (0.16) R-squared Q *** *** *** *** *** (5.55) (6.99) (6.72) (5.51) (4.00) (1.17) (-0.94) R-squared Q *** *** *** *** *** (2.92) (4.30) (3.13) (3.52) (3.37) (0.11) (-0.66) R-squared Q ** *** ** ** (2.53) (3.34) (2.33) (2.31) (1.12) (0.97) (-0.14) R-squared Q (1.40) (1.06) (0.11) (0.17) (-0.20) (1.25) (-0.05) Spearman's rho -1.00*** -1.00*** -1.00*** -1.00*** -1.00*** p-value Panel C2: Asia-Pacific, est. period - past 6 months (skipping the most recent month), holding period - following 3 months (No obs. 217,495) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** (10.17) (9.87) (10.26) (10.86) (8.35) (1.58) (0.67) R-squared Q *** *** *** *** *** (8.90) (9.28) (9.01) (8.23) (7.20) (1.38) (0.49) R-squared Q *** *** *** *** *** (5.81) (6.95) (7.87) (6.39) (5.67) (0.00) (-0.01) R-squared Q *** *** *** *** *** * (7.58) (7.32) (6.95) (6.92) (5.00) (3.19) (1.09) R-squared Q *** *** *** *** *** ** (9.05) (8.38) (5.76) (4.85) (5.13) (7.12) (1.20) Spearman's rho *** -0.90** -0.90** p-value

35 Panel D1: North America, est. period - past 6 months (skipping the most recent month), holding period - fol. 1 month (No obs. 178,104) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** *** *** (9.30) (6.85) (5.58) (4.45) (2.77) (21.21) (-2.07) R-squared Q *** *** *** *** *** *** (7.60) (7.54) (3.80) (4.75) (2.94) (10.78) (-1.22) R-squared Q *** *** *** *** *** (5.91) (4.62) (4.40) (3.75) (0.12) (16.66) (-1.56) R-squared Q *** *** ** *** (3.59) (3.63) (1.98) (0.45) (-0.55) (8.57) (-1.07) R-squared Q *** * *** *** (3.27) (0.82) (1.21) (-1.91) (-2.84) (18.68) (-1.15) Spearman's rho -1.00*** -0.90** -0.90** -0.90** -0.90** p-value Panel D2: North America, est. period - past 6 months (skipping the most recent month), holding period - fol. 3 months (No obs. 170,324) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** ** *** (9.11) (9.38) (7.89) (6.27) (5.75) (5.59) (-2.06) R-squared Q *** *** *** *** *** *** (11.60) (7.29) (4.72) (6.92) (3.38) (33.55) (-1.30) R-squared Q *** *** *** *** *** (9.56) (7.09) (6.85) (4.57) (0.06) (44.93) (-1.27) R-squared Q *** *** *** *** *** *** (8.01) (5.37) (3.71) (3.04) (0.46) (28.39) (-2.02) R-squared Q *** *** *** *** *** *** (12.49) (8.77) (5.86) (0.68) (-2.92) (116.98) (-2.07) Spearman's rho ** -0.90** p-value

36 Panel E1: World, estimation period - past 6 months (skipping the most recent month), holding period - following 1 month (No obs. 945,749) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** (6.49) (6.41) (8.25) (8.55) (5.31) (0.70) (-0.51) R-squared Q *** *** *** *** *** ** (7.29) (4.95) (5.31) (6.45) (3.57) (6.94) (-0.72) R-squared Q *** *** *** *** ** (3.15) (4.08) (3.09) (5.16) (-1.12) (9.13) (-0.52) R-squared Q *** *** *** *** *** (2.87) (4.87) (2.60) (0.76) (-3.33) (19.25) (-0.55) R-squared Q *** ** *** *** (4.91) (2.53) (0.86) (-1.63) (-4.87) (47.78) (-0.83) Spearman's rho * -1.00*** -1.00*** -1.00*** -1.00*** p-value Panel E2: World, estimation period - past 6 months (skipping the most recent month), holding period - fol.3 months (No obs. 906,382) Equally-Weighted Raw Returns Momentum Q1 Momentum Q2 Momentum Q3 Momentum Q4 Momentum Q5 5-1 FF α R-squared Q *** *** *** *** *** (6.10) (7.26) (10.22) (11.30) (5.82) (0.05) (-0.58) R-squared Q *** *** *** *** *** ** (7.73) (7.05) (8.10) (9.69) (3.58) (8.63) (-0.69) R-squared Q *** *** *** *** *** (8.39) (8.36) (8.94) (7.57) (-0.16) (36.55) (-0.70) R-squared Q *** *** *** *** *** (10.64) (10.04) (9.75) (6.95) (-0.99) (67.58) (-0.75) R-squared Q *** *** *** *** *** (18.29) (14.84) (11.29) (5.02) (0.05) (165.23) (-1.21) Spearman's rho 1.00*** 0.90** *** *** -1.00*** p-value

37 Table 4: Momentum, Return R-squared, and Investor Protection The table presents the results of the cross-sectional regression of the difference between the momentum profits in the lowest and highest return R- squared quintiles on a set of country-specific indicators of investor protection. The regression equation is Difference it = α + β 1 Investor protection it + β 2 X it + β 3 Y + ε it. In columns 1, 2 and 3 the Efficiency of judicial system is used as a proxy for investor protection. In columns 4, 5 and 6 Efficiency of government is used as a measure of investor protection. Columns 2, 3, 5, 6 and 7 report the results after adding the control variables GDP growth rate, Market capitalization and Inflation rate (vector X). In Column 3, 6 and 7 Year fixed effects (vector Y) are included. In Column 7 both Efficiency of judicial system and Efficiency of government are of interest. All variables are defined in Appendix A. All standard errors are robust. t-statistics are reported in parenthesis. ***, ** and * refer to 1%, 5% and 10% significance levels, respectively. (1) (2) (3) (4) (5) (6) (7) Difference Difference Difference Difference Difference Difference Difference Efficiency of judicial system * * * * (1.77) (1.90) (1.77) (1.74) Efficiency of government * * (1.62) (1.77) (1.49) (-0.60) GDP growth rate (-1.02) (-0.45) (-0.88) (-0.27) (-0.29) Market capitalization (0.01) (-0.27) (1.28) (0.83) (-0.44) Inflation rate (-0.04) (0.43) (0.04) (0.48) (0.47) Constant * * * * (-1.65) (-1.80) (-1.71) (-1.37) (-1.58) (-1.42) (-1.71) Year fixed effects No No Yes No No Yes Yes Observations 1,212 1,194 1,138 1,212 1,194 1,138 1,138 R-squared

38 Table 5: Momentum Returns by Year The average momentum profits within the five return R-squared quintiles are reported for each of the years in the sample, where momentum profits are defined as the differences between the highest and lowest momentum quintiles. F-statistics are reported in parenthesis. ***, ** and * refer to 1%, 5% and 10% significance levels, respectively. Panel A: Europe, estimation period - past 6 months (skipping the most recent month), holding periods - fol. 1 and 3 months Holding period 1 month Holding period 3 months R-squared Q *** *** ** *** *** *** *** (38.77) (1.64) (18.38) (1.04) (9.38) (87.93) (15.10) (19.85) (0.70) (25.39) R-squared Q *** *** *** ** ** *** *** *** ** *** (45.31) (28.69) (9.51) (5.74) (8.81) (78.84) (35.31) (20.47) (9.89) (28.40) R-squared Q *** ** *** ** * *** *** *** *** (44.92) (4.92) (8.22) (4.36) (3.41) (66.62) (55.11) (25.92) (2.13) (19.08) R-squared Q *** * *** ** ** *** *** *** ** (15.49) (3.53) (9.09) (5.28) (1.18) (29.04) (60.09) (27.53) (0.43) (5.04) R-squared Q *** *** *** ** *** *** *** *** (0.26) (22.00) (15.17) (11.47) (10.18) (14.63) (159.61) (35.46) (0.73) (21.52) Observations ,546 64,936 62,617 55,321 69,684 67,654 64,062 61,692 44,834 Df 70,692 68,521 64,911 62,592 55, ,629 64,037 61,667 44,809 37

39 Panel B: Japan, estimation period - past 6 months (skipping the most recent month), holding period - fol. 1 and 3 months Holding period 1 month Holding period 3 months R-squared Q *** *** *** *** *** *** ** (24.29) (43.85) (28.16) (0.03) (1.28) (77.67) (101.59) (52.29) (9.47) (0.02) R-squared Q ** *** *** *** *** *** *** (6.57) (35.74) (42.09) (0.24) (0.02) (19.18) (127.63) (38.13) (18.84) (2.70) R-squared Q ** *** *** * *** *** *** *** (8.55) (49.00) (24.68) (0.30) (3.03) (20.87) (237.74) (63.55) (11.20) (0.31) R-squared Q *** *** *** *** *** *** *** (6.43) (50.30) (43.84) (0.00) (20.62) (33.11) (226.03) (70.58) (19.26) (2.38) R-squared Q *** *** *** *** *** *** ** ** (0.48) (91.55) (23.78) (1.63) (16.82) (5.16) (287.79) (43.86) (4.42) (6.73) Observations 45,939 44,693 43,465 42,568 38,433 45,669 44,388 43,227 42,370 31,318 Df 45,414 44,668 43,440 42,543 38,408 45,644 44,363 43,202 42, Panel C: Asia-Pacific Excl. Japan, estimation period - past 6 months (skipping the most recent month), holding period - fol. 1 and 3 months Holding period 1 month Holding period 3 months R-squared Q ** *** *** (1.00) (2.31) (2.40) (0.00) (5.70) (1.10) (25.35) (0.29) (1.21) (11.19) R-squared Q ** ** ** *** *** (4.91) (6.62) (1.87) (0.13) (3.86) (0.02) (23.15) (0.21) (0.60) (43.17) R-squared Q *** ** *** * *** *** *** (0.23) (13.13) (2.70) (6.08) (7.62) (2.87) (32.82) (0.01) (7.79) (42.91) R-squared Q ** *** ** *** ** *** (6.32) (8.34) (2.21) (4.82) (1.08) (0.32) (53.15) (1.02) (4.53) (17.35) R-squared Q *** ** ** *** (10.73) (0.39) (0.12) (4.32) (0.14) (7.35) (20.35) (0.07) (0.89) (6.30) Observations 42,702 43,508 43,755 45,172 42,358 42,491 43,329 43,568 44,932 34,445 Df 42,677 43,483 43,730 45,147 42,333 42,466 43,304 43,543 44,907 34,420 38

40 Panel D: North America Excl. U.S., estimation period - past 6 months (skipping the most recent month), holding per. - fol. 1 and 3 months Holding period 1 month Holding period 3 months R-squared Q ** ** ** * ** (2.51) (9.71) (2.05) (10.67) (1.47) (0.93) (5.75) (2.12) (3.02) (4.85) R-squared Q ** * *** *** ** * *** (6.62) (10.45) (2.87) (0.39) (2.59) (14.81) (42.93) (9.72) (2.89) (24.79) R-squared Q *** *** ** *** ** ** * (2.67) (25.17) (0.18) (0.31) (0.33) (8.25) (52.79) (8.75) (3.31) (3.29) R-squared Q *** *** *** ** *** (1.83) (13.71) (1.23) (0.01) (0.35) (11.19) (28.68) (9.34) (2.43) (11.54) R-squared Q *** *** *** *** ** * (11.56) (24.18) (0.79) (1.92) (0.03) (52.66) (125.00) (5.69) (2.31) (3.02) Observations 36,357 36,532 35,623 36,078 33,514 35,994 36,103 35,330 35,762 27,135 Df 36,332 36,507 35,598 36,053 33,489 35,969 36,078 35,305 35,737 27,110 Panel E: World, estimation period - past 6 months (skipping the most recent month), holding period - fol.1and 3 months Holding period 1 month Holding period 3 months R-squared Q *** *** *** *** *** *** (64.57) (13.47) (0.47) (4.86) (26.26) (218.98) (30.01) (0.44) (1.16) (52.52) R-squared Q *** *** ** *** *** *** ** *** (36.84) (62.63) (1.37) (9.76) (18.15) (119.07) (50.35) (0.11) (9.89) (104.15) R-squared Q *** *** *** ** *** *** *** *** (36.45) (65.90) (0.52) (14.46) (9.07) (98.14) (113.79) (1.00) (19.71) (87.27) R-squared Q ** *** ** *** *** ** *** (5.10) (57.86) (0.16) (6.19) (0.78) (32.26) (145.04) (0.17) (2.00) (32.45) R-squared Q * *** *** *** (3.46) (57.97) (1.28) (0.37) (0.48) (0.29) (174.28) (1.19) (0.75) (20.92) Observations 159, , , , , , , , , ,597 Df 159, , , , , , , , , ,572 39

41 Figure 1: Momentum returns by return R-squared quintiles (time-series). Europe Figure 2: Momentum returns and Difference between momentum returns in the lowest and the highest return R-squared quintile (time-series). Europe 40

42 Figure 3: Momentum returns by return R-squared quintiles (time-series). Japan Figure 4: Momentum returns and Difference between momentum returns in the lowest and the highest return R-squared quintile (time-series). Japan 41

43 Figure 5: Momentum returns by return R-squared quintiles (time-series). Asia-Pacific Excl. Japan Figure 6: Momentum returns and Difference between momentum profits in the lowest and the highest return R-squared quintile (time-series). Asia-Pacific Excl. Japan 42

44 Figure 7: Momentum returns by return R-squared quintiles (time-series). North America Excl. U.S. Figure 8: Momentum returns and Difference between momentum returns in the lowest and the highest return R-squared quintile (time-series). North America Excl. U.S. 43

45 Figure 9: Momentum returns by return R-squared quintiles (time-series). World Figure 10: Momentum returns and Difference between momentum returns in the lowest and the highest return R-squared quintile (time-series). World 44

46 Table 6: Momentum Return and Distance to 52-week Market High The table presents the results of the cross-sectional regressions of Momentum returns on distance of the current MSCI of a given country to the moving past 52-week highest value of the MSCI of the specific country (Distance to 52-week high). The regression equation is Momentum return it = α i + β 1 Distance52 it + β 2 X it + β 3 Y + ε it. Monthly observations for all countries in the sample period January, December, 2012 are used. In Columns 2, 3 and 4, a set of country-specific control variables is added (GDP growth rate, Market capitalization and Inflation rate - vector X). Columns 3 and 4, include country fixed effects. Column 4 includes year fixed effects (vector Y). All variables are defined in Appendix A. All standard errors are robust and clustered by country. t-statistics are reported in parenthesis. ***, ** and * refer to 1%, 5% and 10% significance levels, respectively. (1) (2) (3) (4) Momentum return Momentum return Momentum return Momentum return Distance to 52-week high *** *** *** *** (8.74) (7.45) (7.91) (8.28) GDP growth rate *** ** (1.56) (3.12) (-2.10) Market capitalization * (-1.66) (-1.35) (-2.03) Inflation rate (0.73) (1.71) (-0.97) Constant *** *** *** *** (-7.29) (-5.94) (-5.18) (-5.21) Country fixed effects No No Yes Yes Year fixed effects No No No Yes Observations 1,297 1,274 1,274 1,215 R-squared

47 Figure 11: Averages of all countries' Scaled MSCI, MSCI return, Momentum return and Distance to 52-week high (time series). World Figure 12: Averages of all countries' MSCI return, Returns of Losers and Winners, Distance to 52-week high and Momentum return (time series). World 46

48 Figure 13: MSCI return, Returns of Losers and Winners, Distance to 52-week high and Momentum return (time series). Europe Figure 14: MSCI return, Returns of Losers and Winners, Distance to 52-week high and Momentum return (time series). Japan 47

49 Figure 15: MSCI return, Returns of Losers and Winners, Distance to 52-week high and Momentum return (time series). Asia-Pacific Excl. Japan Figure 16: MSCI return, Returns of Losers and Winners, Distance to 52-week high and Momentum return (time series). North America Excl. U.S. 48