Institutional Herding in International Markets This draft: April 21, 2014 Nicole Choi * University of Wyoming Hilla Skiba University of Wyoming Abstract: This paper studies herding behavior of institutional investors in international markets. First, we document the existence of wide-spread herding in 41 countries (referred to as target countries hereafter) in the sample. We then examine the relation between contemporaneous institutional demand and future returns and find that institutional herding stabilizes prices. Next, we examine the relation between institutional investors herding behavior and the level of information asymmetry in the target countries. We measure the degree of information asymmetry in each target country along five dimensions: (1) stock market development, (2) ease of access to information, (3) corporate transparency, (4) investor rights, and (5) macroeconomic factors that relate to the information environment. We find evidence that institutional investors herd more in markets characterized by low levels of information asymmetry. The results suggest that institutional investors herding behavior is likely driven by correlated signals from fundamental information. Keywords: Institutional investor; Herding; Information asymmetry; International financial markets JEL codes: G11; G15; G23; Z10 * Contact author: Address: University of Wyoming, College of Business, Department of Economics and Finance 3985, 1000 E. University Avenue, Laramie, WY 82071, United States. Tel: 1-307-766-3637; Fax: 1-307-766-5090. E-mail Address: nchoi@uwyo.edu
1. Introduction Extant literature has documented a tendency of individual and institutional investors to herd, or to follow each other in and out of the same securities. As a result, investors demand for a security is positively correlated with demand for the same security in the previous time period. This herding behavior has been mainly documented in investors trades at the security level in a single market setting (Lakonishok et al. 1992; Sias 2004; Barber et. al. 2009). Choi and Sias (2009) document that herding also occurs at the industry level, so that institutional investors demand for securities in a certain industry is positively correlated with demand for that industry in the previous period, after adjusting for size and book-to-market characteristics of the securities. Researchers have provided several explanations for herding, but much debate still exists for the reasons why individuals and institutions follow each other s trades. Often herding is viewed as negative behavior that destabilizes prices, as the herds chase hot markets and securities, and flee them as soon as the investment environment turns cold. The herds are thus believed to increase volatility and inefficiencies of the financial markets. For example, the herding behaviors that may drive prices away from fundamentals include: reputational herding (Trueman 1994), characteristic herding (Falkenstein 1996), and fads (Friedman 1984). However, it is also entirely possible that the empirically observed herding is at least partially spurious, unintentional, or driven by fundamental information. If herding is spurious, it means that investors do not actually follow each other s trades, but they trade similarly because they investigate the same information and derive similar conclusions about securities fundamental values. For example, it may appear that a herd is buying the same security, but if fundamental information drives the buying decision, the herd will only buy the security until the 1
security s market value is equal to its fundamental value. In the case of this investigative herding that is based on signals from fundamental information, the herding behavior will lead to a faster adjustment of fundamental information into securities and more efficient markets (Bikhchandani and Sharma 2000; Froot et al. 1992). Also, even if herding is intentional, it may still have a price stabilizing effect on securities if it is based on informational cascades (Sias 2004). When investors cascade, they ignore their private signals about securities fundamental values, but instead infer information from others trades (Bikhchandani et al. 1992). The contribution of our paper is to study investors herding behavior in the global environment, in a multi-market setting, using comprehensive holdings data. We believe that the multi-market setting provides us with a unique environment to test (1) if herding in international markets occurs, (2) what drives the behavior, and (3) what are the consequences of herding behavior to market stability. We make two empirical predictions in the paper. First, we expect to observe that herding behavior stabilizes prices. Specifically, we expect that institutional investors herding is more likely to stabilize prices because it is driven by fundamental information. Second, we expect to observe that target country information asymmetry is related to herding propensity. In addition to testing whether there is a significant relation between herding propensity and information asymmetry, we believe that the direction of the result will shed light on the underlying reasons for herding. Specifically, an inverse relation between herding and information asymmetry would support investigative herding. In other words, if information about securities fundamental values is more available and easier to access and interpret, then investors are more likely to make similar decisions independently from the same information. Alternatively, if herding behavior is higher in the markets with more information asymmetry, then informational cascades are more 2
likely to explain the behavior (e.g. Wermers 1999; Kim and Nofsinger 2005; Bikhchandani and Sharma 2000). We test these empirical predictions in several steps. First, we examine whether institutional trading stabilizes prices. We test a relation between contemporaneous institutional demand and current and future stock returns. Second, we measure country-level herding propensities of institutional investors in 41 target countries. We then test whether there is variation in herding propensities across target countries and if the observed variation is related to information asymmetry. We measure the degree of information asymmetry in each target country along five dimensions: (1) stock market development, (2) ease of access to information, (3) corporate transparency, (4) investor independence, and (5) macroeconomic factors that relate to information environment. Our results provide strong support for herding that is driven by fundamental information and appears unintentional. First, we find a statistically significant positive relation between current institutional demand and current quarter returns. We also find a positive, but only marginally significant to an insignificant, relation between the current institutional demand and the next quarter, 6-month, and 1- year returns in most target countries. Lack of return reversal indicates that institutional herding is a mechanism through which fundamental information is incorporated into security prices. The first result indicates that herding stabilizes prices and therefore is most likely due to information (investigative herding or informational cascade). Second, the country-specific information asymmetry measures explain variation in herding propensities across target countries. Information asymmetry measures across the five categories are statistically and economically significant in the majority of our analyses. The direction of the results indicates that herding occurs more in target countries that are 3
characterized by low levels of information asymmetry. The results provide evidence for herding behavior in international markets that is likely based on fundamental information about the underlying securities and markets, which institutions are discovering independent of each other, not because of psychological reasons or cascading. Some existing studies provide complementary results to our empirical predictions. These existing papers show that there is a positive relation between institutional demand and current returns in the US, accompanied by a lack of significant return reversals. The authors of these studies offer this to be evidence for institutional trading that is based on fundamental information that consequently moves prices toward equilibrium (e.g., Nofsinger and Sias 1999; Choi and Sias 2009; Puckett and Yan 2008; Sias 2004; Gutierrez and Kelley 2009). We use an extensive holdings dataset that allows us to investigate herding behavior on a large scale. Using the holdings data of international institutional investors, we are able to compute the herding measures directly. The few existing international herding studies apply market return data instead of holdings data (Chiang and Zheng, 2010, for example) to compute herding propensities. Most studies investigate herding at the security level and also in a single market setting, usually the US market with a few exceptions. To our knowledge, no other study has investigated herding across international markets using holdings data of the underlying investors. The international holdings data, provided by the Factset Company, are the most comprehensive holdings data of which we are aware. The data, for example, contain the US 13-F filings for institutional investors, and similar public filings from other countries regulatory agencies. After applying several filters to the data, there are 30,095 institutions that trade in 4
24,689 securities in 41 target countries. The filters require reliable market level information proxies and sufficient investment by institutions to compute the herding propensities. Our study makes several contributions to the existing literature. First, our study is the first to our knowledge to investigate herding in a multi-market setting using actual holdings data of the investors and thus allow for a more detailed investigation of investor behavior compared to prior studies. Second, we shed light on fundamental reasons for institutional investors herding and provide some evidence that herding behavior is based on correlated signals. Third, institutional investors comprise a large part of global portfolio investments, and understanding the decision-making of these institutions is important in highly integrated global equity markets. If these institutions in fact herd based on fundamental information, they are likely to stabilize security prices, especially in more informationally transparent markets. The rest of the paper is organized as following: Section 2 discusses related literature and testable hypotheses. Section 3 reviews the dataset and the methodology in detail. Section 4 presents the results, and Section 5 concludes. 2. Related literature and hypotheses 2.1 Herding in financial markets Herding is convergence of behavior, which in financial markets results in investors following each other from security to security and from market to market. Many recent works in the finance literature attempt to explain herd behavior. Some argue that herding behavior is not based on fundamental information and destabilizes security prices. Others argue that herding is driven by fundamental information and actually makes security markets more efficient because 5
prices adjust faster to new information (see, for example, Hirshleifer and Teoh (2003) for an extensive review of herding behavior). Information-based explanations for herding in financial markets most often include investigative herding and informational cascades. Investigative herding occurs when investors react similarly to market signals that are correlated. As a result, investors do not make buy and sell decisions based on other investors actions, but instead, all investors analyze the same data and use similar methods, and the trading strategy is the same purely because the investors receive correlated signals (Froot et al. 1992; Hirshleifer et al. 1994). Informational cascade occurs when investors ignore their own private information about a security or a market, and instead make decisions by observing others (Bikhchandani et al. 1992; Banerjee 1992). Wermers (1999) shows that herding by US mutual funds occurs more often in small, growth oriented stocks, which are characterized by high levels of information asymmetry. Similarly, Sias (2004) shows that US institutions herd more in small securities and concludes that institutional herding is best explained by informational cascades. Non-information based explanations for herding include: reputational reasons, characteristic herding, and fads. These non-information based explanations are more likely to destabilize prices, since information about securities fundamental values is a secondary consideration to the investors. First, reputational herding results when managers of investment portfolios ignore their private information and choose to mimic investment decisions by others because they would rather follow the crowd and make wrong decisions than make different decisions from the crowd. Scharfstein and Stein (1990) show that herding that occurs due to reputational reasons can be rational, because errors in prediction are not made alone, but the blame is rather shared. Trueman (1994) studies analysts herding due to reputational reasons, and 6
reports that analysts are more likely to release forecasts that are similar to others forecasts rather than what their private information suggests. Second, when investors are attracted to the same characteristics of the securities, it leads to characteristic herding 1 (Gompers and Metrick 2001; Bennett et al. 2003; Nofsinger and Sias 1999; Sias 2004). Third, investors may also herd due to fads (Freidman 1984; Barberis and Shleifer 2001). Most papers in finance study herding in stock markets, at the security level, in a single market setting. However, a few studies have also examined herding in other asset classes. These include Gleason s (2003) study of futures and Oehler and Chao s (2000) study of bonds. 2.2 International evidence of institutional herding Most herding studies thus far have examined investors herding behavior in the US market. However, some recent research documents herding in international environments as well, although most of the international studies are still focused on a single market setting and domestic institutions. International herding papers in a single market setting include studies of: Japan (Kim and Nofsinger 2005); Korea (Choe, et al. 1999; Kim and Wei 2002); Taiwan (Chen and Hong 2006); Israel (Venezia et al. 2011); Germany (Walter and Weber 2006; Kremer and Nautz 2013); the United Kingdom (Wylie 2005); Portugal (Lobao and Serra 2007); Poland (Voronkova and Bohl 2005); India (Lakshman 2013) and Hong Kong (Zhou and Lai 2009). In addition, Li and Yung (2004) study herding in American Depository Receipts. Most of the non- US studies above examine herding behavior in great detail by employing high frequency holdings data. Chiang and Zheng (2010), Chang et al. (2000), Blasco and Ferreruela (2008), and Borensztein and Gelos (2003) investigate institutional herding in a multi-market setting. Chang 1 Momentum and habit trading are examples of characteristic herding. 7
et al. (2000) employ countries market index returns to compute herding propensities, and show that herding is largely nonexistent in the US and most other developed markets, but strong in South Korea and Taiwan, the two emerging markets in their sample. In contrast, Chiang and Zheng (2010) find some evidence of herding in developed stock markets, when the countryspecific herding propensities are computed from market index data. Blasco and Ferreruela (2008) test herding behavior in seven countries using a measure of Cross Sectional Standard Deviation (CSSD) of security returns. The authors find that among the sample countries, only Spain displays significant herding propensity. Borensztein and Gelos (2003) study herding behavior of 400 emerging market mutual funds. The authors find significant herding that appears consistent over different market conditions from tranquil to crises periods, and is especially meaningful among open-end mutual funds. 2.3 Information asymmetry and investor behavior One important contribution of our paper is to study how country-specific information asymmetry affects the herding propensities of institutional investors across target countries. Information asymmetry and investors decision making has been a focus of a body of recent studies in finance. For example, Baik et al. (2010) show that geographic proximity of institutions to firms is related to their predictive abilities about future stock returns. The local advantage is especially beneficial in firms that are characterized by high information asymmetry. Gelos and Wei (2002) employ information transparency measures to study their effect on international investors behavior in 27 emerging markets. The authors find that transparency is positively related to foreign portfolio investment in the target markets. Similarly, Kang and Stulz (1997) show that foreign investors prefer more transparent firms in the Japanese market. Choi and Sias (2012) 8
employ nine information uncertainty metrics and show that the process in which information is incorporated into security prices is slower when information uncertainty is high. 2.4 Hypotheses The main contribution of the paper is to test if institutions herd in international markets and what drives the herding behavior. In the first part of the paper, we test if herding behavior in international markets is significant, and if herding is more likely driven by information rather than non-information based reasons. A body of literature (Chakravarty 2001; Sias et al. 2006, among others) reports that institutional demand is positively related to contemporaneous returns and weakly positively related to the future returns of securities. In addition, numerous studies (e.g., Nofsinger and Sias 1999; Choi and Sias 2009; Puckett and Yan 2008; Sias 2004; Gutierrez and Kelley 2009) argue that the fact that there is a positive relation between institutional demand and current return in securities, accompanied by lack of significant return reversal, is evidence that institutional trading is information-based and consequently moves prices toward equilibrium. Based on these arguments, we conjecture that if observed herding behavior in international markets is a process in which information is incorporated into prices, then there will be a positive relation between institutional demand for securities in the current quarter and return to the securities in the current quarter, followed by no significant return reversal. Alternatively, if institutional trading is not based on fundamental information, but rather destabilizes stock prices, then we expect to observe a positive relation between current institutional demand and current security returns, followed by a return reversal in the future periods. Formally, our first testable hypothesis states: H1. Institutional demand is positively related to current stock returns and positively or insignificantly related to future stock returns. 9
Thus, we make an empirical prediction that herding in international markets is price stabilizing and based on information. Once we establish this to be true, we further investigate if the information-based herding is intentional (cascading) or unintentional (investigative) behavior. Empirical evidence in the herding literature shows evidence for both informational cascades and investigative herding. Wermers (1999) and Sias (2004) show that in the US market, investors herd more in securities that are characterized by high levels of information asymmetry (small stocks with less precise future earnings). Both authors attribute the finding to informational cascades. According to Kim and Nofsinger (2005), investigative herding is more likely to be the reason for institutional herding in the Japanese market. The authors assert that the information environment in Japan is more transparent, and institutional investors make decisions about securities based on correlated signals. The authors also state that in a market with more information asymmetry (less transparent information environment), investors tend to ignore their own signals and informational cascades form. Following the logic of these studies, we conjecture that herding propensity will be related to the level of asymmetric information of the target country. We assert that if institutional herding is large in magnitude in target countries characterized by high levels of information asymmetry, then informational cascade is more likely to be the underlying reason for herding in international markets. Alternatively, if observed herding behavior is investigative and based on correlated signals, we expect to observe lower propensity to herd in target countries that are characterized by asymmetric information environment. Formally, we test a hypothesis that states that there is a relation between information asymmetry and herding propensity: 10
H2. Institutional herding propensity is related to the target country information asymmetry. To test the second hypothesis, we measure target country information asymmetry with numerous variables that have been used in the prior literature to proxy for country-specific information environment. We then test if there is a positive or a negative relation between country-specific herding propensity and the information asymmetry proxies. The sign of the information asymmetry measure will determine whether informational cascades or investigative herding is the likely explanation for the relationship. The information proxies are reviewed in detail in Section 3.1. Herding measures are reviewed in detail in Section 3.2. 3. Data and methods 3.1 Data 3.1.1 Holdings and security data We use quarterly institutional holdings data from the FactSet ownership (formerly LionShares) database. The data comprise various sources of institutional investors reporting, such as 13-F and N-Q for the US institutional investors, and similar filings from non-us investors regulatory agencies. FactSet contacts mutual fund associations and regulatory authorities to obtain data. Where regulatory filings are not mandatory, FactSet obtains portfolio reports from funds websites, from direct contacts with the companies, company reports and announcements, and/or industry directories. The dataset provides detailed information on securities that each institution holds, the domicile of the institutions, and the style and type of the institutional investors, among many other security and investor characteristics. The FactSet holdings data are available at the aggregate firm level and also for portfolios inside each firm, if applicable. For this study, we 11
focus on herding of portfolios, not aggregate holdings of the investment firms and refer to the portfolios as institutions or institutional investors. We match the holdings data to security data to compute herding propensities in each target country in the sample. The target countries are identified based on each security s country of domicile. We require a security to have at least five investors trading each quarter to be included in our sample. The final sample consists of 30,095 institutions. Institutions are from 87 countries, have holdings in 24,689 securities of 41 target countries. The data span the last quarter of 1999 to the first quarter of 2010. Table 1 shows descriptive statistics on institutional investors and securities in each of the 41 target countries in the sample. The first and second column show the distribution of the securities traded in the target countries. The United States has the largest number of securities held by institutions with 8,741 (35.4% of the total sample). Japan and the United Kingdom have the second and third largest number of securities in the sample (2,421 and 2,023 securities; 9.81% and 8.19% of the sample respectively). The third column shows the number of institutions that have invested at least once in a target country during the sample period. The highest number of institutions have traded in the US market (25,496 institutions, or 88% of the total sample of 30,095), followed by the United Kingdom and France (with 60.31% and 46.35% of the investors). [Insert Table 1 here] Table 2 shows the number and distribution of institutions by their country of domicile. There are 30,095 institutions, domiciled in 89 countries in the sample. The largest number of institutions are domiciled in the United States (11,869 institutions or 39.44% of the sample), 12
followed by United Kingdom and Germany (2,819 and 1,499 institutions; 9.37% and 4.98% of the sample respectively) [Insert Table 2 here] 3.1.2 Information asymmetry variables Next, we obtain country-specific information asymmetry measures from various sources. In the paper, we group the country-specific information asymmetry variables into five categories. These categories include factors that relate to: stock market development, access to information, corporate transparency, investor rights, and the macroeconomy. First, we proxy each target country s stock market development by the total market capitalization as a percent of GDP (Market Cap; obtained from World Bank); stock market Volume, computed as the aggregate volume of publicly traded securities, scaled by the market capitalization (raw data for the computation are obtained from Compustat Global); the fraction of investable shares, or Float, of the total shares outstanding (raw data obtained from WorldScope); and the standard deviation of the value-weighted stock market return (Volatility; raw data obtained from Compustat Global). The stock market development variables are inversely related to information asymmetry with the exception of Volatility. Second, we proxy access to information with the number of Internet users per 100 people in the population (obtained from World Bank); the number of Newspapers in circulation per 1,000 people in the population (obtained from World Bank); and overall access to Media in the country (from Bushman et al. 2004). All of these proxies are inversely related to information asymmetry. Third, we proxy corporate transparency by several accounting transparency measures that include: Disclosure Intensity, Accounting Principle, Analyst Coverage, Insider Trading, and 13
Security Disclosure (all accounting transparency variables are from Bushman et al. 2004); and Anti-self-dealing index (from Djankov et al. 2008). All of these proxies are inversely related to information asymmetry. Fourth, we measure investor rights by the index of Investor Protection (from La Porta et al. 2006); and by indices of Investment Freedom, Financial Freedom, and Economic Freedom (constructed by the Heritage Foundation). We assume that all of these proxies are inversely related to information asymmetry. Lastly, the macroeconomic factors that relate to information asymmetry include: GDP per Capita in USD and the average annual Inflation. GDP per Capita is inversely related to information asymmetry and Inflation is positively related to information asymmetry. In addition to the information asymmetry proxies reviewed above, we include the indicator variable for developed markets as an additional control. Developed takes a value of one if a country is classified as a developed country by World Bank and zero otherwise. The indicator variable is more likely to equal zero when information asymmetry is high. Table 3 reports the cross-sectional average statistics of the information asymmetry measures of the 41 target countries from 2000 to 2009. Each panel A through E shows the summary statistics for a category of information asymmetry variables along with the data source. Time varying variables are averaged and all the variables are standardized when used in the regressions as explanatory variables. Panel F shows the mean and standard deviation of the indicator variable Developed and also the number of target countries for which the indicator variable takes on value of one. [Insert Table 3 here] 14
3.2 Methodology: Institutional herding In this section, we construct the dependent variable used in the analyses. The dependent variable is the herding propensity of institutions in each target country of the sample with sufficient trading data. First, we define institutional investors as buyers of security i if their holdings in the security in the current quarter increase compared to the prior quarter, and as sellers of security i if their holdings decrease. % Buyeri j, t for security i that belongs to a target country j in quarter t is then the fraction of institutional buyers as a share of the total traders, and is defined as: % Buyer i j, t # Buyersi j, t # Buyers # Sellers i j, t i j, t (1) For example, a value of 0.60 from Eq. (1) for %Buyer i,t, indicates that 60% of institutions trading stock i in quarter t are buyers and 40% of the traders are sellers. Next, to measure the herding propensity in each target country j, we use the standardized form of Eq. (1) 2 to compute the correlation between buying intensities in securities in contiguous quarters. Formally, we measure the correlation between the standardized percentage of institutional buyers in security i in quarter t and t-1, by running cross-sectional regressions each quarter by target countries as: % Buyer % Buyer (2) i j, t j. t i j, t 1 i j, t where %Buyer i,t and %Buyer i,t-1, (both standardized) are fractions of buyers in quarter t and t-1, respectively. 2 Standardizing the herding measure allows for more intuitive interpretation of the regression coefficients and easier comparison across the target countries. Standardized fraction of buyers is calculated as: % Buyeri,t % Buyer t S.% Buyeri j, t, where % Buyer t and (%Buyer t ) are the cross-sectional average and (%Buyer t ) standard deviation across securities for each country. 15
The time-series average of coefficient β j,t for each target country j from Eq. (2) is the main measure of institutional investors herding propensity in target country j. Since the coefficient β from Eq. (2) is the correlation between the standardized fraction of buyers this quarter and the standardized fraction of buyers next quarter, it takes values between -1 and 1. Section 4.1 discusses the magnitude of this measure in more detail. Numerous papers in the finance literature document institutional investors momentum trading behavior (Sias 2004; Grinblatt et al. 1995; Nofsinger and Sias 1999; Sias et al. 2003, among others). To test if the correlation between institutional investors trading during the current quarter and the previous quarter is mainly due to institutions preference for positive momentum securities, we add a quarterly lagged return to Eq. (2) as an additional control variable: % Buyer i j, t j, t % Buyer i j, t 1 i, treti j, t 1 i j, t (3) 1 where %Buyer i,t and %Buyer i,t-1, are defined in Eq. (2), and Ret i,t-1 is a quarterly return of security i in quarter t-1. If institutional investors herding propensity is due to their momentum chasing behavior, the coefficient on %Buyer i,t will lose its significance after adding Ret i,t compared to the coefficient from Eq. (2). If observed herding behavior is not entirely due to institutions preference for past winners, %Buyer i,t will retain its statistical and economic significance. 4. Results 4.1 Institutional herding in target countries First, we compute herding propensities in each target country of the sample. Previous literature on herding across international markets has mainly used aggregate market return data to test if investors across international markets herd. Most papers conclude that little herding takes place. 16
We compute the time-series average herding propensities for each target country j based on Eq. (2) and (3). We report the results of herding tests in panels A and B of Table 4 for 41 target countries. The first column, Herding, reports the time-series average coefficient (β j,t ) from Eq. (2). Associated t-statistics are reported in the second column. In panel B, Herding is the time-series average coefficient (δ j.t ) from Eq. (3). Results of Table 4 show that institutional herding across international markets exists and is statistically and economically significant in magnitude. In panel A, all but eight target countries exhibit statistically significant herding propensities. Also, all but one herding propensity measures are positive (Kenya has a negative but insignificant herding propensity). The significant coefficients range from 0.0931 (Thailand) to 0.3122 (New Zealand). The herding measure for the United States is 0.2233, which indicates the correlation between percentage of buyers in the current quarter and percentage of buyers in the previous quarter, and compares well with the herding measure reported by Sias (2004). 3,4 In panel B, we control for the momentum in returns. The results show that in 33 out of 41 target countries, institutional investors are momentum chasers. 5 However, even after controlling for institutional investors positive feedback trading, all but two target countries (Argentina and Turkey) retain the significant coefficient on the herding measure. Therefore we conclude that institutional investors convergence in trading behavior is not entirely due to their momentum trading. Overall, the results of Table 4 show that herding is abundant across the sample s target countries. 3 Sias (2004) reports a herding measure of 0.1755 for US securities with 5 or more institutional traders. 4 Because we use standardized variables and a single independent variable, the regression coefficients are the correlation between the fraction of institutional buyers in the current quarter and the fraction in the previous quarter. Refer to the appendix on Sias (2004) for the proof. 5 Because we require a security to have the past quarter returns for Eq. (3), the number of observations for the regression decreases, as reported in the columns labeled N in each panel. 17
[Insert Table 4 here] 4.2 Evidence of herding in international markets using the LSV measure Most previous studies examining herding behavior employ the measure developed by Lakonishok et al. (1992; LSV measure hereafter). The LSV measure is calculated as following: H % Buyer % Buyer AF (4) i, t i, t t i, t where %Buyer i,t is defined in Eq. (1), and % Buyer is cross-sectional average of %Buyer i,t for each period for each country. The Adjustment Factor (AF i,t ) is calculated based on the assumption that the number of institutional traders in security i in quarter t follows a binomial distribution with the probability of buying equal to the cross-sectional average of fractional buyers ( % Buyer ). t [Insert Table 5 here] Table 5 summarizes previous research on institutional herding in international markets using the LSV measure. Table 5 shows the variation in herding propensities in different countries. For example, Kremer (2013) estimates a herding propensity of 0.0248 for the German market and Choe et al. (1997) estimate a herding propensity of 0.2628 for the Korean market. These international studies find higher levels of herding compared to the US specific studies that use the LSV measure (Lakonishok et al. (1992), Sias (2004) and Grinblatt, Titman and Wermers (1995), among others). 6 [Insert Table 6 here] Table 6 reports the average of LSV herding measure defined in Eq. (4) for each of the target countries in our sample as well as associated t-statistics. Results show that herding t 6 Lakonishok et al. (1992) report an LSV measure of 0.0270 for the US market. Sias (2004) reports an LSV measure of 0.0178. 18
propensity is positive and significant at the 1% level for all 41 of the target countries. The average herding propensity (H i.t from Eq. (4)) for the United States is 0.0552 and statistically significant at 1% level, which implies that in a given quarter and security, the number of institutions moving in the same direction (either buying or selling) is approximately 5.52% higher than it would be if institutions acted independently. The result is comparable, though slightly higher, to LSV measures found in previous US studies. Overall, results in Sections 4.1 and 4.2 provide evidence of institutional investors herding behavior measured by two different metrics. Both measures show herding that is positive (i.e., institutions move to the same direction) and statistically significant in most target countries in the sample. The results remain strong after controlling for momentum, one of the most popular reasons as to why institutions would prefer the same stocks. 4.3 Institutional demand and security returns In the previous section, we document that herding across target countries is wide-spread and large in magnitude. Next, we test whether this herding has a price stabilizing effect on the target countries. Hypothesis 1 predicts that institutional herding is information-based and is not followed by significant price reversals in the future periods. We conjecture that institutional herding is more likely to be due to informational cascades or investigative herding if we observe a positive relation between institutional demand and current return and positive to no relation between demand and future returns. Table 7 reports correlation coefficients between institutional demand (Eq. (1)) and returns to securities for the previous quarter (Q t=-1 ), the current quarter (Q t=0 ), the next quarter (Q t=1 ), the next 2 quarters (Q t=1,2 ), and the next four quarters (Q t=1-4 ). Reported numbers in Table 7 are time-series average correlation coefficients across the securities traded in each target country and their associated t-statistics. 19
[Insert Table 7 here] First, Table 7 shows that correlation coefficients between current institutional demand and past quarter returns are significant in all but 7 target countries in the sample (columns 1 and 2). This is consistent with the results in Table 4. Second, consistent with previous studies (albeit mainly done only in the US market), the relationship between institutional demand and current quarter returns is statistically significant and positive in all but 11 target countries (columns 3 and 4). Third, the last six columns show virtually no evidence for return reversal in the future time periods. When returns over the next quarter (Q t=1 ) and the next 2 quarters (Q t=1,2 ) are used to compute the correlation coefficients, there is no return reversal in any of the sample target countries. However, when the next 4 quarter return (Q t=1,4 ) is used to compute the correlation coefficients, the UK and the US markets show some return reversal. Overall, the results of Table 7 show that institutional demand is positively related to the securities current returns and weakly or not related to the securities future returns with the exception of two markets. This result mainly supports Hypothesis 1, and leads us to conclude that informational cascades or investigative herding are likely to be the underlying reasons for institutional herding in the target countries. In other words, it appears that institutional herding is a mechanism through which fundamental information is incorporated into security prices and does not destabilize the security prices of the target markets. 4.4 Univariate regression analysis In this section, we test Hypothesis 2 and examine the relation between institutional herding propensity and information asymmetry of the target countries. Empirical prediction of Hypothesis 2 is that we expect to observe a negative relation between herding propensity and information asymmetry if herding behavior is more likely to be unintentional or investigative in 20
nature. Alternatively, if we observe a positive relation between herding propensity and information asymmetry, then informational cascades are more likely to be the underlying cause of herding behavior. To test Hypothesis 2, we run several pooled univariate OLS regressions, where the dependent variable is the target country herding propensity from Eq. (2) and the independent variables consist of the information asymmetry proxies reviewed in Section 3.1.2. The regression equation is: Herding InfoAsym (5) j j j where Herding j indicates the time series average of herding propensity, β j,t, from Eq. (2) for each target country, and InfoAsym j represents each of the 20 target country specific information asymmetry measures. In addition to the 20 variables, we also construct one principal component for each of the five categories of information asymmetry proxies. The principle components include: Stock Market Development, Access to Information, Corporate Transparency, Investor Rights, and Macroeconomy. We also construct one principal component based on all 20 information asymmetry proxies (Information Asymmetry). All principle components are the first principle components of the underlying variables. Table 8 shows the results from regression equation (5). The table reports each explanatory variable s coefficient, an associated t-statistic, and an adjusted R 2. All the explanatory variables are standardized to mitigate the significant differences in units and ranges of the variables. 7 The principal component variables of the categories are reported in the last row of each panel and are in italic. 7 Every explanatory variable is standardized to avoid significant differences in scale. To standardize each variable, we subtract the mean of each variable and divide by standard deviation of each variable across 41 countries so that each standardized variable has a mean of 0 and standard deviation of 1. Formally, a standardized explanatory 21
[Insert Table 8 here] Panel A of Table 8 displays the regression results of four stock market development factors and the factors first principal component. Market Cap, Volume, and Float are inversely related to information asymmetry. Volatility is positively related to information asymmetry. Of the five variables, all but Volume are statistically significant. The result is consistent with the empirical prediction of Hypothesis 2. Specifically, it appears that institutions herd more, when information asymmetry, proxied by stock market development factors, is low. The finding is evidence against informational cascading and is more consistent with correlated signals. Panel B shows the regression results of the variables that relate to ease of access to information. All variables of panel B are inversely related to information asymmetry. Consistent with the results of panel A, variables related to access to information support the empirical prediction of Hypothesis 2. Namely, Media, Internet, Newspapers, and their principle component are all positive and statistically significant. The result provides strong support for herding that is likely driven by correlated signals from the fundamental information and supports the investigative herding story. Panel C reports the coefficients on the independent variables that represent the level of Corporate Transparency and the first principle component Corporate Transparency. Three variables, Accounting Principle, Analyst Coverage and Anti-self-dealing, are positive and statistically significant, and all seven variables are positive. The results in panel C suggest again that institutional investors show higher degree of herding behavior in more informationally efficient markets, and supports investigative herding behavior. variable is expressed as: S. Exp j Exp j Exp, where Exp j is a value of an explanatory variable for a country j, and Exp and σ are the cross-sectional mean and standard deviation of the explanatory variable across 41 countries. Time varying variables are averaged for each country before being standardized. 22
The results from regressions where information asymmetry is captured by variables related to investor rights are reported in panel D of Table 8. All independent variables in the panel are inversely related to information asymmetry. Investment Freedom, Financial Freedom and Economic Freedom, as well as the principal component variable Investor Rights, take on positive and significant coefficients. Consistent with all previous panels, the results from panel D indicate that institutions herd more in target countries with stronger investor freedom and rights. The last panel E shows results from regressions where information asymmetry is captured by macroeconomic variables. The variables are the logarithm of GDP per capita in US dollars, and Inflation measured as GDP deflator in percentage. Macroeconomy is the first principal component of the macroeconomic factors. Panel E of Table 8 shows that GDP per Capita, and the principal component Macroeconomy are positive and statistically significant. Inflation is also significant and has its expected sign. Again, the results provide strong support for investigative herding in the target countries. Finally, the first principal component of all 20 information asymmetry variables, Information Asymmetry, is positive and significant in the univariate regression. In summary, 20 of 26 univariate regressions of herding propensity measure on country level information asymmetry proxies yield statistically significant or marginally significant coefficients with their expected signs. In addition, all information asymmetry proxies have signs that are consistent with the idea that institutions follow each other s trades more in markets where information is easier to obtain. The results support Hypothesis 2 and specifically the idea of investigative herding rather than informational cascades. 23
4.5 Univariate regression analysis with additional controls In this section, we add additional controls to the regression equation (5) to further explore what explains herding propensities across the target countries. In these analyses, we add the indicator variable Developed to the basic regression equation (5). Developed takes on a value of 1 if the target country is a developed economy and zero otherwise. By inclusion of the indicator variable, we test if the information asymmetry proxies are capturing an omitted effect of the size of a target country s economy. A more developed target country is likely to have a less information asymmetric environment and many of the information asymmetry proxies will be positively related to market development in general. To disentangle the power of the main informational asymmetry variables from that of the macroeconomic factor mentioned above, we add the indicator variable to univariate analyses. [Insert Table 9 here] In Table 9, we include the Developed indicator as an additional control in the regression equation. The results show that the Developed indicator is positive and significant in most regression specifications. This indicates that institutions herd more in a more developed target country, consistent with the results from the panel E of Table 8, where we find a positive and significant sign on GDP per capita. More interestingly, the results of Table 9 show that even after adding the Developed indicator in the specifications, 11 of the 20 information asymmetry proxies retain their statistical significance and correct signs. The result provides further support for Hypothesis 2 and investigative herding. It appears that herding propensity increases as the information environment becomes more transparent, even after controlling for the general level of development of the target country. 24
In summary, the results in this section indicate that herding propensities in target countries are inversely related to information asymmetry. We find that institutions herd more when information is easier to access and interpret. The findings lead us to conclude that it is more likely that herding in international markets is driven by institutions trading based on correlated signals from fundamental information rather than inferring information from each other s trades. In other words, it is more likely that institutional herding is investigative and unintentional in nature rather than informational cascades. 4.6 Robustness checks To check whether the regression results that test Hypothesis 2 remain statistically significant after controlling for the momentum trading by institutions, we repeat the analysis of Tables 8 and 9 with an alternative herding measure. For tests in this section, the dependent variable in Eq. (5) is the time-series average of the coefficients δ j,t from Eq. (3). We omit the regression tables for brevity, but discuss the key results of the robustness checks below. The results that replicate the analysis of Table 8 with the metric from Eq. (3) as a herding measure for each of the target countries are quantitatively similar to those in Table 8. With variables related to market development, the same variables that show significance in Table 8, namely, Market Cap, Float, Volatility, and Market Development, remain significant. The rest of the robustness checks reveal that Newspapers, Accounting Principal, and Investment Freedom become insignificant, but Security Disclosure is significant with the alternate measure of herding propensity. All the variables related to macroeconomic factors retain their significance with the alternative herding measure. We also repeat the analysis of Table 9 with the herding measure defined in Eq. (3) as the dependent variable. The results show a small decline in statistical significance and economic 25
magnitude compared with the results of Table 9. Specifically, Accounting Principal, Anti-selfdealing, Economic Freedom, Investor Rights, and GDP per capita lose their statistical significance. However, overall, the results of the robustness checks indicate that the herding measure that controls for institutions momentum trading yields similar results to the univariate regression tests presented in Tables 8 and 9. 5. Conclusion This paper studies institutional herding in international markets on a large scale. We make several contributions to the finance literature. First, we use an extensive holdings data to measure herding propensities in a large set of target countries. To our knowledge, our paper is the first to measure herding in a multi country setting using actual holdings data. Second, we document that herding in international markets is wide-spread. We find, using two different herding measures, statistically significant herding propensities in 41 target countries that have significant presence of institutional investors. Third, we show that herding seems to be price stabilizing rather than irrational behavior. Consequently, institutional demand is positively related to current and adjacent periods returns, but we do not observe price reversals. Therefore we conclude that herding is more likely to be based on fundamental information and that it is a mechanism through which fundamental information is incorporated into security prices. Fourth, we show that target countries level of information asymmetry is inversely related to herding propensities. In other words, institutions herd less when information asymmetry is high. Previous papers in the literature have shown that information asymmetry can lead to informational cascades, where institutions in a presence of high information asymmetry infer fundamental information from other investors trades rather than from security level information. However, we do not find evidence for informational cascades, but rather for investigative 26