Ownership Crowded with Style: Institutional Investors, Liquidity, and Liquidity Risk

Transcription

1 Ownership Crowded with Style: Institutional Investors, Liquidity, and Liquidity Risk Alessandro Beber 1, Michael W. Brandt 2, Mathijs Cosemans 3, and Michela Verardo 4 1 Cass Business School, City University and CEPR 2 Fuqua School of Business, Duke University and NBER 3 Rotterdam School of Management, Erasmus University 4 London School of Economics This Draft: May 2014 Abstract This paper studies the effect of correlated trading strategies induced by similar investment mandates on the liquidity risk of a large cross-section of stocks. We identify the potential for correlated trades using the heterogeneity in the crowdedness of stock ownership across institutional investors with different investment styles. We find that stocks with ownership concentrated among investors that follow similar strategies exhibit a substantially higher liquidity risk. This effect is not explained by cross-sectional differences in liquidity, as crowded stocks tend to be more liquid. Overall, our empirical evidence suggests that non-fundamental factors can affect asset prices due to their effect on liquidity risk, even when they have opposite effects on liquidity levels. Keywords: liquidity risk, institutional ownership, style investing, mutual funds, hedge funds JEL classification: G12 We thank Netspar and the Duisenberg School of Finance for providing financial support. We are grateful to Dion Bongaerts, Mathijs van Dijk, Frank de Jong, David Stolin, Marta Szymanowska, Marno Verbeek and seminar participants at Cass Business School, Erasmus University, and the Netspar international pension workshop for useful comments and discussions.

2 1 Introduction Liquidity is extremely important in asset pricing. For example, the seminal paper of Amihud and Mendelson (1986) shows that illiquid assets offer higher expected returns to compensate investors for higher transaction costs. Furthermore, Pastor and Stambaugh (2003) and Acharya and Pedersen (2005) show that investors demand a premium for the risk that market liquidity may deteriorate precisely when they want to trade. Although liquidity and liquidity risk play such an important role in financial markets, we still lack a full understanding of the determinants of liquidity dynamics. This became painfully clear to investors during the recent financial crisis. Some traditionally very liquid assets saw their liquidity quickly dry up, as the need to sell these assets became highly correlated across market participants. When an institutional investor has to sell due to past losses, or due to increases in risk or risk aversion, or simply because of large fund outflows, it is highly likely that other investors following similar investment styles also have to sell the same asset at the same time. In that case, liquidity risk depends on the aggregate holdings of the group of investors with correlated trading needs. This paper investigates the effects on stock liquidity risk of different ownership compositions. More specifically, when the stock ownership is crowded with institutional investors following the same investment style, we might expect to observe a larger proportion of trades in the same direction and at similar points in time, with a natural impact on the stock liquidity risk. We analyze the ownership composition of a large cross-section of stocks on three different levels with increasing level of detail. First, we look at aggregate institutional ownership, given the dominant role of institutional investors in stock markets in recent decades (French (2008)). Previous research (e.g., Gompers and Metrick (2001) and Bennett, Sias, and Starks (2003)) has shown that institutional ownership has strong predictive power for stock returns and volatility, suggesting that trading by this category of investors has a significant impact on stock market dynamics. We find that stocks with concentrated institutional ownership tend to exhibit high liquidity risk, i.e. tend to have low returns when market illiquidity is high. At times of low market liquidity, the correlated trading strategies of institutional investors have a detrimental impact on stock returns. Interestingly, this finding is in contrast with the results for the level of liquidity, which is typically positively correlated with institutional ownership. 1

3 We then examine in more detail ownership composition and separate institutional investor by investor type. Specifically, we focus on hedge funds and mutual funds as the relatively more active categories of institutional investors, as opposed to pension funds, for example. We distinguish between mutual funds and hedge funds because their trading strategies could differ considerably in terms of trading frequency and horizon and because they are likely to be exposed to different liquidity shocks. For mutual funds, shocks typically arise from fund redemptions, while hedge funds would traditionally reduce their market exposure in response to an increase in the cost of leverage. 1 The breakdown of institutional ownership by investor type reveals that ownership concentration in both mutual funds and hedge funds determines an increase in stock liquidity risk beyond the general effect of institutional ownership. Finally, we further breakdown stock ownership and investigate the implications of investment styles for an asset s liquidity risk. We hypothesize that style investing has a strong impact on liquidity risk because it leads to ownership crowded by investors with common liquidity shocks and very similar trading strategies. 2 Specifically, mutual funds with the same investment style not only tend to hold similar stocks but are also likely to experience similar in- and outflows due to positive feedback trading by fund investors at the style level (Barberis and Shleifer (2003) and Teo and Woo (2004)). We break down mutual fund ownership along the size and value-growth dimensions, because this style classification is widely used by investors (see Froot and Teo (2008) and Kumar (2009)). For example, consider a stock for which ownership is concentrated among a few small-cap growth funds while another stock is held by a large number of funds that follow uncorrelated strategies. In that case, we expect the first stock to have higher liquidity risk because its investor base is sensitive to common outflows. Hedge fund strategies can also lead to concentrated ownership. For instance, funds that rely on specific quantitative strategies are likely to trade similar stocks when they use the same underlying models (Lo and Khandani (2011) and Stein (2009)). In addition, funds that implement certain event driven strategies such as merger arbitrage, concentrate their positions in a 1 Brunnermeier and Pedersen (2009) propose a theoretical model in which variation in the funding liquidity of levered traders such as hedge funds exerts important effects on asset market liquidity and liquidity risk. In particular, they identify a potentially self-reinforcing link between funding liquidity risk (risk that a trader cannot fund his position and is forced to unwind) and market liquidity risk (risk that market liquidity worsens precisely when the investor needs to trade). 2 Concentrated ownership can also arise among investors with unrelated strategies due to herding, as documented by Grinblatt, Titman, and Wermers (1995), Nofsinger and Sias (1999), Wermers (1999), and Sias (2004). 2

4 limited number of stocks that are affected by the event. As a result, hedge fund trades become crowded and sensitive to adverse funding liquidity shocks. For instance, when a particular strategy performs poorly, hedge funds may be forced to sell part of their holdings to reduce leverage. Because of these correlated trading needs, we expect stocks with concentrated ownership among a small number of funds with similar investment styles or strategies to have higher liquidity risk than those with diversified ownership across many funds with largely uncorrelated strategies. The empirical results strongly support our predictions. Stocks with ownership crowded with hedge funds following the same investment style turn out to be much more exposed to liquidity risk. For mutual funds, stock ownership concentrated with small cap investment style funds generates the largest increases in liquidity risk, suggesting that these funds follow very similar trading strategies. All our results are robust to the use of alternative measures of liquidity risk. Our work contributes to two strands of literature. The first stream of research examines the impact of institutional ownership on stock prices. Coval and Stafford (2007) show that mutual funds that experience large outflows decrease existing positions, which creates price pressure in the securities held in common by distressed funds. Along similar lines, Lou (2012) provides a flow-driven explanation for stock return momentum. Building on these ideas, Anton and Polk (2010) and Greenwood and Thesmar (2011) show that concentrated ownership and correlated liquidity shocks of mutual funds predict stock price volatility and comovement. We add to this literature by studying whether differences in ownership structure explain variation in liquidity and liquidity risk. We find that in general, institutional ownership improves liquidity but increases liquidity risk. However, we show that the type of institutional investor matters. In particular, we show that mutual funds and hedge funds have a heterogeneous impact on liquidity risk, because they employ diverse strategies and are subject to different liquidity shocks. Our paper also contributes to a second strand of literature that studies the implications of style investing for stock prices. Barberis and Shleifer (2003) develop a theoretical model in which investors allocate their money to different styles based on relative performance. They predict that stocks that belong to the same style class comove too much relative to their fundamentals. Barberis, Shleifer, and Wurgler (2005) and Boyer (2011) provide empirical support for this model by showing that when a stock is included in an index, its return comovement with that index increases even though its fundamentals have not changed. Froot and Teo (2008) and Kumar 3

5 (2009) find strong evidence of style rotation by both institutional and individual investors, which generates a negative correlation between net flows into distant style segments. Wahal and Yavuz (2013) show that this style chasing has a significant impact on future stock returns as it amplifies momentum and reversal effects. We extend this work by analyzing the impact of style investing on liquidity. Our results show that style investing increases the liquidity risk of stocks, because it leads to concentrated ownership among investors with correlated trading needs. We further show that investment styles also have an impact on the level of liquidity, because strategies differ in terms of trading intensity and demand for immediacy. The remainder of the paper is structured as follows. Section 2 describes the data and Section 3 presents the main empirical results. Section 4 focuses on an analysis of causality and Section 5 reports results for a variety of robustness checks. Finally, Section 6 offers concluding remarks. 2 Data 2.1 Mutual Fund Sample The empirical analysis combines different data sets and covers the period between January 1990 and December We obtain information about mutual fund performance, total net assets (TNA), and investment style from the CRSP mutual fund database. We aggregate data for funds with multiple share classes because the underlying holdings for each class are the same. In particular, we sum the TNA of the different share classes and compute weighted averages of returns and net asset values (NAV), where the weights are the lagged TNA of the share classes. We match the funds in CRSP to quarterly mutual fund holdings from the Thomson Reuters CDA/Spectrum database using the MFlinks table constructed by Wermers (2000). We focus on U.S. equity mutual funds because the holdings data primarily consists of positions in U.S. stocks. Our selection criteria are based on those used by Kacperczyk, Sialm, and Zheng (2008). First, we exclude all funds with the following self-reported investment objectives in the CDA/Spectrum database: international, municipal bonds, bonds and preferred, and balanced. Subsequently, we select funds using the Lipper investment classification, which is assigned to each fund based on its actual holdings. We retain funds that are classified by Lipper in one of the 12 standard Lipper investment style classifications for U.S. diversified equity funds, which 4

6 are based on the intersection of 3 valuation categories (value, core, growth) and 4 capitalization groups (large, multi, mid, small). Because we want to compare the behavior of these active funds to a passive benchmark, we also include funds with Lipper classification SPSP, which seek to replicate the performance of the S&P 500 index. Wahal and Yavuz (2013) point out that a classification scheme should be based on styles that are widely followed by investors to be relevant for studying the impact of style investing. The Lipper scheme satisfies this requirement because it is based on the size and value-growth dimensions that are popular among both individual and institutional investors (see Froot and Teo (2008) and Kumar (2009)). Furthermore, the classification accounts for any style drift over time because it is holdings-based. Because the Lipper classification is only available in the CRSP Mutual Fund database from 1999 onwards, we backfill the classification to previous years if possible. 3 Funds that cease to exist prior to 1999 do not have a Lipper classification at all. To avoid any survivorship bias, we do include these non-surviving funds in our sample and use the Strategic Insight (SI) and Wiesenberger codes to estimate fund styles. 4 Finally, we map the SI and Wiesenberger codes into the Lipper style classification based on the algorithm proposed by Teo and Woo (2004). 5 Imposing the above restrictions leaves us with a total of 1709 mutual funds in our sample. 2.2 Hedge Fund Sample We obtain hedge fund performance information, assets under management, and strategies from the CISDM and TASS hedge fund databases for the period We combine data from these two hedge fund databases because funds may report to only one database (see Griffin and Xu (2009) and Teo (2010)). We consolidate the two databases following the steps outlined in Patton and Ramadorai (2010). Since our focus is on U.S. equity investments, we exclude hedge funds following emerging market strategies, distressed securities, fixed income strategies, and managed futures funds, along the lines of Cao, Chen, Liang, and Lo (2012). Because our holdings data only contains information about long equity positions, we also exclude short 3 We verify that backfilling the Lipper classification does not impact our results by performing the analysis using only data from 2000 to Results for this subsample are similar to those reported for the full sample. 4 In particular, we select funds with the following SI objectives: AGG, GMC, GRI, GRO, ING, or SCG. Out of funds that neither have a Lipper nor an SI classification, we pick those that have one of the following Wiesenberger codes: Wiesenberger codes: G, GCI, LTG, MCG, and SCG. 5 Teo and Woo (2004) use the Morningstar style classification. We adapt their algorithm to the Lipper classification using the fund classification roadmap published by Lipper. 5

7 sales, dedicated short bias, and convertible arbitrage funds, which typically only take short positions in common stocks. We further exclude funds of funds, because we cannot classify them into a specific style and because of the problems in reconstructing their asset holdings. Because CISDM and TASS use different strategy classifications, we rely on the mapping scheme of Agarwal, Daniel, and Naik (2009) and Teo (2010) to classify funds into four broad investment styles: directional (DI), relative value (RV), security selection (SS), and multi-process (MP). We obtain quarterly holdings data for hedge funds from their 13F filings in the Thomson Reuters institutional holdings database. Unlike mutual funds, hedge funds do not report their holdings at the fund level but at the management firm level. We therefore compile a list of management firms from TASS and CISDM and manually look up these company names in the 13F filings. Companies must report to the SEC their long positions in equity, convertible bonds, and options if assets under management exceed $100 million holdings. For equities, firms should report long positions in U.S. common stocks that consist of more than 10,000 shares or are worth more than $200,000. Subsequently, we check whether the primary business of the management firms is in hedge funds. Following Brunnermeier and Nagel (2004), we retrieve the SEC ADV forms for all registered companies and require that at least half of the firm s clients are Other pooled investment vehicles (e.g., hedge funds) or High net worth individuals. We also retain all firms that are not registered because registration is only required if the firm wants to advise non-hedge fund clients such as mutual funds and pension funds. After applying these filters, our final sample consists of 471 different holding companies. 2.3 Stock Characteristics We retrieve stock-level information for all firms listed on NYSE, AMEX, and NASDAQ. We use daily stock return and volume data from CRSP from 1990 until 2009 to construct our liquidity measures. To correct for survivorship bias, we adjust the returns for stock delisting. As control variables in our regressions, we use firm size and stock return obtained from CRSP, the book-to-market ratio calculated using data from Compustat, and return volatility computed using quarterly windows of daily returns. The firm characteristics and the ownership variables are measured as deviations from their corresponding cross-sectional means, standardized to have unit variance, and winsorized at the 1st and 99th percentiles to mitigate the effect of 6

8 outliers. Because volume data for NASDAQ stocks includes interdealer trades, we also include a NASDAQ dummy as control variable. We further add a dummy for stocks with zero reported institutional ownership, because these stocks are either very small or institutional ownership is not accurately measured. 2.4 Liquidity Measures We investigate the impact of ownership composition on two measures of stock liquidity: a measure of illiquidity cost and a measure of liquidity risk. Our principal measure of illiquidity cost is the Amihud (2002) measure of price impact, ILLIQ j t = 1 D j t D j t d=1 R j td V j td (1) for stock j in month t, where D j t denotes the number of observations available in month t, Rj td and V j td denote the volume in millions of dollars on day d in month t, respectively. Goyenko, Holden, and Trzcinka (2009) run a horse race between a large number of liquidity proxies constructed from daily data and find that the Amihud measure is the best proxy for measuring price impact. In a robustness analysis, we also consider a number of other commonly used liquidity measures, including the quoted spread and the Gibbs estimate of effective trading costs from Hasbrouck (2009). We obtain monthly measures of illiquidity by averaging the values across all days in each month. Extreme values of ILLIQ are winsorized at the 97.5th percentile in each month to reduce the impact of outliers. We measure liquidity risk by regressing daily stock returns on innovations in market-wide liquidity. Because Watanabe and Watanabe (2008) document that liquidity risk is time varying, we estimate monthly liquidity betas using a quarterly window of daily returns: R j,t,d = α j,t + β LIQ j,t LIQ M,t,d + ɛ j,t,d, (2) where R j,t,d denotes the daily excess return on stock j. The subscript d = (1, 2,..., τ) is used to index the daily returns before the end of month t, where τ is the length of the estimation 7

9 window. 6 LIQ t is the liquidity risk factor, which is constructed from the daily innovation in the value-weighted average of the Amihud (2002) illiquidity measure across all stocks in the CRSP universe. Since liquidity is persistent, we compute innovations in market illiquidity as the residual u t from an AR(6) process. Finally, we use the negative of the estimated residuals, u t, as our liquidity risk factor LIQ t in (2) so that a higher beta implies higher liquidity risk. We only include stocks with a least one month of daily returns in each quarter to increase estimation precision of the firm-level liquidity betas. In a later section we also consider two other liquidity factors, which are the aggregate liquidity measure of Pastor and Stambaugh (2003) and the permanent-variable component of Sadka (2006). 3 Empirical Results In this Section, we present empirical evidence on ownership composition and on its effects on liquidity dynamics. In the first part, we present summary statistics and describe the general patterns for mutual funds and hedge funds following different investment styles and strategies. In the second part, we use this data within a number of different regression specifications to study the role for stock liquidity and liquidity risk of the ownership composition across styles. 3.1 Ownership of Mutual Funds, Preliminaries Table 1 reports the ownership composition of mutual funds and hedge funds in our sample. Panel A shows that the mutual funds in our sample hold on average about 7% of U.S. stocks during our sample period We can see strong evidence of an upward trend in mutual fund ownership, which has more than doubled in the last 20 years. The most recent figures show that average mutual fund ownership stands at 8.6%, with stocks in the upper quartile of the ownership distribution reaching about 13%. Panel B looks at the composition of ownership considering the number of mutual fund owners. We can see that the number of mutual funds owning a given stock has steadily increased over time. The average stock in our sample is owned by 27 mutual funds. We notice that this measure is highly positively skewed, with the lower and upper quartile stock being held by 4 and 36 mutual funds, respectively. 6 We have also estimated Dimson-corrected betas by adding one lag of the innovation in market illiquidity to the regression to control for any bias in liquidity betas due to non-synchronous trading effects. We find that this correction does not affect our main findings. 8

10 Table 2 illustrates the composition of ownership across styles. Styles are defined by the Lipper investment classifications for U.S. diversified equity funds, which are based on the intersection of 3 valuation categories (value (VE), core (CE), growth (GE)) and 4 capitalization groups (large (LC), multi (ML), mid (MC), small (SC)). As a benchmark, we also include funds with Lipper classification SPSP that seek to replicate the performance of the S&P 500 index. The most popular mutual fund style in total net asset values is large capital, core. The smallest style in terms of total net asset values (TNAV) is represented by small capital, value mutual funds. Table 2, panel B, contains the proportion of TNAV across style and is a better gauge of the relative importance of the different styles. We can see that there are basically three levels of TNAV: large (around 15%) for large cap core and growth and multi-cap core; medium (around 10%) for large cap value and multi-cap growth and passive funds; low (at or below 5%) for every other style. Figure 1A provides a useful visual description of this pattern. Table 3 shows the correlation of fund flows across different styles. Panel A focuses on mutual funds. A simple visual inspection of the correlation values reveals a significant degree of rotation across different styles both in the size dimension and along the value-growth spectrum. For example, flows to large cap core funds have zero correlation with flows to small cap funds. Similarly, flows to large cap growth funds have negative correlation with flows to large cap value funds. This evidence suggests at least two stylized facts. First, the mutual fund category is far from being a uniform class of investments. In this sense, it justifies our methodology that further breaks down stock ownership in different mutual fund style categories. Second, it suggests that there are likely ownership shocks due to investor style rotation that have a potential impact on the liquidity of the stocks. In Table 4 we take one step further in the direction of understanding trading behavior of different mutual fund styles and compute measures of turnover. Following Brunnermeier and Nagel (2004), we compute fund turnover as the minimum of the absolute values of buys and sells during the quarter divided by the total net assets at the end of the previous quarter. Panel A shows that, for mutual funds, growth styles exhibit higher turnover than values styles and core styles, across different capitalization categories. Along the size dimension, funds that are managing small stocks have larger turnover, but the difference is only clearly evident in the growth category. Figure 2A illustrates this pattern, showing that average turnover varies 9

11 between 14% and 19% for growth style funds, and stays between 9% and 10% for all the other investment styles. The last dimension we study is the liquidity of stocks held by mutual funds specializing in different investment styles. More specifically, we compute both the Amihud illiquidity measure and the liquidity beta of stocks held by mutual funds with different investment styles during our sample period. Panel A of Table 6 shows a large variation of illiquidity across mutual fund styles. For example, the small cap value style (SCVE) holds stocks with an average illiquidity that is about 30 times larger than the illiquidity of the large cap value style (LCVE) and more than 100 times larger than the illiquidity of S&P500 stocks. Figure 3A illustrates the large heterogeneity in illiquidity levels across mutual fund styles. Panel A of Table 6 also shows average liquidity betas for stocks owned by mutual funds that follow different investment styles. The cross-style differences in liquidity betas are less strong that those observed for illiquidity levels, with average liquidity betas ranging between 1.68 and Ownership of Hedge Funds, Preliminaries In this Section we describe our sample of hedge funds in relation to their ownership of U.S. equities. We also analyze the movements of capital within the hedge fund industry by investigating the correlation of flows across funds that are grouped by investment strategy. Finally, we present some descriptive statistics to highlight the large degree of heterogeneity in the liquidity characteristics of stocks that are owned and traded by hedge funds following different investment strategies. There are a few cautionary notes to consider in the empirical measurement of hedge fund ownership. First, we necessarily rely on a hedge fund sample that is only a subset of the universe of hedge funds, so we may miss a substantial part of hedge fund equity holdings. Second, it is not clear to what extent the hedge fund coverage in the data is a random sample from this universe and to what extent the coverage of the sample has changed over time. Finally, we only observe long positions in stocks and not short positions, since only the former need to be disclosed to the regulator. In general, all these issues with the measurement of hedge fund holdings are biasing against finding any relationship with liquidity and liquidity risk. In specific cases, some of our results could be driven by better measurement because of either better coverage or disclosure. 10

12 We will be careful in these cases to acknowledge this potential alternative explanation. We present summary statistics on hedge fund ownership in Table 1, Panels C and D. The hedge funds in our sample hold on average about 3% of US equities during the sample period Over the sample period, hedge fund ownership remains stable around 2.5% during the 1990s, shows a substantial increase in the second half of the sample period, reaching 3.8%, and declines in the more recent years after the latest financial crisis. The number of hedge funds holding US equities in their portfolios shows a similar pattern, with a sharp increase in the second decade of our sample period and a decline in the most recent years. Panel D shows that the average stock in our sample is owned by 5 hedge funds. Compared to mutual funds, the distribution of the number of hedge fund owners is less skewed and increases from about 2 to 6.5 between the bottom and top quartiles. Panel C of Table 2 shows that long/short hedge funds represent the largest fraction of the total assets under management (AUM) of the funds in our sample, with 23% of AUM. They are followed by event-driven funds (17.7%). With just below 10% of the total AUM we find multiple strategy funds, global macro funds, and funds that engage in convertible arbitrage strategies. The rest of the funds represent a smaller fraction of the total AUM of the hedge funds in our sample. Figure 1B plots the AUM distribution across different hedge fund strategies. We next analyze the flows of capital in and out of hedge funds across the set of ten investment strategies that we identify for these funds. Panel B of Table 3 reveals very heterogeneous patterns of correlation among strategies. For example, the flows to Long/Short strategies are highly correlated with the flows to long only strategies, event driven strategies, and market neutral funds, but exhibit very low correlation with global macro and merger arbitrage funds, and negative correlation with multiple strategy funds. On the other hand, merger arbitrage funds have flows that do not correlate with the flows of any other strategy, with the exception of event driven strategy funds. This evidence further motivates our investigation of the link between stock liquidity and ownership composition, providing a specific channel based on correlated flows of capital in and out of specific strategies. Panel B of Table 4 reports estimates of average turnover across hedge fund strategies. The table shows that hedge funds tend to have considerably higher portfolio turnover than mutual funds, on average. Long-only strategies stand out with a turnover of about 37%; most other 11

13 strategies exhibit portfolio turnover varying between 20% to 26%, while convertible arbitrage funds have the lowest turnover, 7% on average. Figure 2B presents an illustration of the patterns of average portfolio turnover for hedge funds. We start exploring the link between liquidity and hedge fund ownership by illustrating the liquidity characteristics of stocks conditional on their hedge fund ownership levels. As for mutual funds, we compute the Ahmihud illiquidity measure and the liquidity beta of stocks held by hedge funds that follow different investment strategies over our sample period. Panel B of Table 6 shows that illiquidity varies greatly across investment strategies. Illiquidity is greatest among stocks held by hedge funds that follow long/short and event driven strategies, reaching average values around 67% to 80%. This level of illiquidity is four times as large as the average illiquidity of stocks held by sector funds and long only funds. It is interesting to note that stocks held by hedge funds are on average substantially more illiquid that stocks held by mutual funds. These findings provide further evidence that analyzing stock liquidity in relation to aggregate institutional ownership masks two important sources of cross-sectional variations in liquidity: at one level, it masks fundamental differences between different types of owners, such as hedge funds and mutual funds; at another level, it masks important differences across investment styles within a given type of investor and across investor types. Figure 3B illustrates the patterns in illiquidity levels and liquidity betas across hedge fund strategies. 3.3 Institutional Ownership, Stock Liquidity, and Liquidity Risk Stock Liquidity and Firm Characteristics In this Section we investigate the characteristics of stocks that differ in their illiquidity levels and in their liquidity betas. Each month we sort all stocks into ten portfolios based on their illiquidity measure, c j t, or based on their liquidity beta, and compute equally weighted averages for a wide variety of stock characteristics. We find that cross-sectional differences in illiquidity are associated with systematic patterns in the cross-sectional distribution of a number of stock characteristics. Table 7, Panel A, shows that illiquidity is negatively associated with market capitalization, as expected. As we move from the portfolio of the most liquid stocks (Decile 1) to the one that contains the most illiquid stocks (Decile 10), the average market capitalization decreases 12

14 sharply and monotonically, while average book-to-market increases monotonically, with value stocks being the most illiquid. Higher illiquidity is also associated with lower average returns. We also compute the volatility of returns in a given month using daily returns, and find that the volatility of illiquidity portfolios increases monotonically from 2% to almost 8% moving from the most liquid to the most illiquid stocks. Portfolios of stocks sorted on liquidity beta exhibit on average weaker variations in stock characteristics such as size or book-to-market, or return volatility. We then turn to examining whether variations in illiquidity identify systematic variations in the cross sectional distribution of institutional ownership. We find that they do. The average level of aggregate institutional ownership is 65% for the most liquid stocks, and decreases monotonically to 10% for the most illiquid stocks. For portfolios formed on liquidity betas we observe an inverted U-shaped pattern in institutional ownership, varying from 24% for stocks with low liquidity betas to 32% for stocks with the highest liquidity beta. When we explore mutual fund and hedge fund ownership separately, we find that mutual fund ownership exhibits again a strong monotonic pattern in relation to illiquidity: on average, about 11% of the shares of very liquid stocks are held by mutual funds, a fraction that decreases monotonically to 2.5% for the most illiquid stocks. Variations in mutual fund ownership based on liquidity betas increase from 5% for low-beta stocks to 8% to medium beta stocks, and decrease again to about 6% for high beta stocks. Interestingly, hedge fund ownership does not exhibit strong variations across liquidity deciles or across liquidity risk deciles, suggesting that hedge funds do not show a particular preference for stock-specific liquidity, on average. The aggregate result of ownership variation across illiquidity deciles may potentially mask heterogeneity across different mutual fund investment style and hedge fund strategies. We thus investigate the association between illiquidity and ownership at a finer level of disaggregation, analyzing ownership by style and strategies. The cross-sectional dispersion in stock illiquidity shows considerable variation across institutional ownership disaggregated by mutual fund investment styles and hedge fund strategies. Panel B of Table 7 shows that mutual funds that follow value strategies, across all capitalization, tend to hold stocks that are more illiquid. Mutual fund ownership increases monotonically with illiquidity for funds in the LCVE, MCGE, MCVE, and SCGE style categories, and more than 13

15 doubles moving from the lowest to the highest illiquidity deciles. In general, core strategies do not exhibit great variations in ownership across liquidity. Interestingly, mutual fund ownership does not vary considerably across portfolios of stocks sorted on liquidity betas. We also find substantial heterogeneity in the concentration of hedge fund strategies across stocks with different levels of illiquidity. As Panel C of Table 7 shows, for example, hedge fund ownership is more concentrated among illiquid stocks for global macro and event driven strategies, and in general increases with illiquidity across all different strategies. Stocks with different liquidity betas, however, do not exhibit great variation in hedge fund ownership. In summary, the univariate analysis in this section reveals that, while stock liquidity is positively associated with aggregate institutional ownership, the association between liquidity and ownership disaggregated by institutional type and investment styles is far more complex. In the next section, we investigate the impact of ownership composition on liquidity in a multivariate regression framework Ownership Composition and Stock Liquidity In this section we examine the link between ownership composition and stock liquidity in a more structured way. We ask whether ownership composition can predict cross-sectional variations in stock liquidity. We do so in a regression framework, to be able to control for a set of stock characteristics that have been shown to predict liquidity in the literature. We estimate predictive cross-sectional regressions of liquidity, measured in month t + 1, on aggregate institutional ownership and stock characteristics. Then we analyze how the impact of aggregate ownership changes with a more disaggregated investigation of ownership composition. We first consider mutual fund and hedge fund ownership, and then examine the link between ownership and future liquidity at a more disaggregated level, considering different investment styles. Our baseline regression is specified as follows: Y j,t+1 = α + β 1 IO j,t + β 2 IO MF j,t + β 3 IO HF j,t + γ X j,t + e j,t+1, (3) where Y j,t+1 is either ILLIQ, the illiquidity of stock j in month t measured as in Amihud (2002), as described in the data section, or β LIQ, the liquidity risk of stock j in month t estimated in 14

16 equation (2) of the data section. IO j,t is the stock s aggregate institutional ownership measured at the end of quarter t; IO MF j,t and IO HF j,t is the ownership of stock j across mutual funds and hedge funds, respectively; and X j,t is a vector of stock characteristics including the log of market capitalization, the book-to-market ratio, return volatility, a dummy for NASDAQ stocks, a dummy for stocks with zero reported institutional ownership, and lagged illiquidity. All these control variables are measured at the end of month t. Given the potential cross-sectional and time-series correlation that is present in the data, we compute t-statistics from standard errors that are clustered along both dimensions following the two-way clustering procedure illustrated in Petersen (2009) and Thompson (2011). Table 8, panel A, presents the results from this baseline regression. As expected, aggregate institutional ownership is negatively associated with future illiquidity. The coefficient on institutional ownership is -0.65, with a t-statistic of Stocks with a higher level of institutional ownership, on average, are significantly more liquid. The predictive ability of the other variables in the regression is in line with previous research. Stock returns and market capitalization predict higher liquidity, while book-to-market and volatility are associated with higher illiquidity in the future. Moreover, since liquidity is persistent, lagged liquidity is a strong predictor of future liquidity. We next ask whether mutual fund ownership and hedge fund ownership have any further predictive ability for stock illiquidity, after controlling for the stock s aggregate level of institutional ownership that is orthogonal to mutual fund and hedge fund ownership, in each case. We find that mutual fund ownership still predicts lower illiquidity, after accounting for aggregate ownership. The coefficient is -2.18, with a t-statistic of Similarly, hedge fund ownership also predicts lower illiquidity levels, with a coefficient of The marginal impact of mutual fund and hedge fund ownership, when examined simultaneously, leads to similar conclusions. In sum, these results indicate that the presence of institutional investors unconditionally helps to improve the future levels of stock liquidity, with a weaker role in this dimension for hedge funds. This is likely to be related to the role of liquidity providers for institutional investors. Table 8, panel B, repeats the same analysis using our measure of liquidity risk, β LIQ, as the dependent variable. Our findings are strikingly in contrast with the results for liquidity levels. Stocks with large institutional ownership overall, tend to predict a significantly higher liquidity 15

17 risk, i.e. lower returns when markets become more illiquid. This evidence is confirmed with highly statistically significant coefficients for stocks featuring large mutual fund and, especially, hedge fund ownerships. Furthermore, we notice in the last column of Table 8, panel B, that when mutual funds and hedge fund ownerships are included, the orthogonal part of institutional ownership is less important in determining future liquidity risk, suggesting that the presence of other heterogeneous institutional investors is less detrimental Investment Styles and Stock Liquidity We argue that the contrasting effects we observe between liquidity levels and liquidity risk are likely to depend on correlated trading strategies. When markets become more illiquid, mutual funds and hedge funds tend to trade in similar stocks and their correlated trades have an impact on stock returns. To corroborate this conjecture, we now examine whether variations in ownership concentration across different investment styles predict cross-sectional variations in liquidity and liquidity risk. We answer this question by examining the impact of ownership disaggregated by different investment styles for both mutual funds and hedge funds. Our baseline regression now includes aggregate institutional ownership as well as ownership disaggregated by mutual fund style and hedge fund investment strategy. To be able to identify the style effects on liquidity and better interpret their economic meaning, we aggregate the 12 Lipper style classes across capitalization styles and across valuation styles. In the empirical analysis, we consider the two most extreme capitalization categories (LC and SC) across all valuation styles, as well as the two extreme valuation categories (VE and GE) across all capitalization groups. 7 Similarly, to investigate the ownership effect of hedge fund strategies on illiquidity and liquidity risk, we aggregate funds based on the similarity in the equity strategies that they follow. We group funds together because Figure 1 shows that some individual hedge fund strategies are small in terms of AUM. In particular, we consider two broad categories: those that are predominantly long only (LO), which includes Global Macro funds, Multi-Strategy Funds, and Long-Only funds, and those that are predominantly long-short strategies (LS), including Long- Short funds, Merger Arbitrage funds, Market-Neutral strategies, and Event Driven funds. The 7 For example, LC consists of all funds in the style classes LCVE, LCCE, and LCGE while SC denotes all funds with style SCCE, SCGE, or SCVE. As expected, unreported regression results for intermediate categories such as mid cap and core equity are in between those of the extreme style classes. 16

18 predictive regressions are now specified as follows: L K Y j,t+1 = α + β IO j,t + βl L IO MFj,t M + βk H IO HF j,t H + γ X j,t + e j,t+1, (4) l=1 k=1 where Y j,t+1 is again either ILLIQ, the illiquidity of stock j in month t, or β LIQ, the liquidity risk of stock j in month t. IO MF M j,t measures the ownership of stock j across mutual funds grouped into M different categories based on either capitalization or valuation styles. The ownership of stock j across groups of hedge fund strategies is measured by IO HFj,t H. Table 9, panel A, presents the results for a few most important mutual fund styles and hedge fund styles and liquidity levels. Our results reveal that the concentration of ownership among funds that follow different investment styles has very different effects on liquidity. Ownership concentration in mutual fund growth, value, and small cap styles strongly predicts lower illiquidity, while large cap strategies seem to predict higher stock illiquidity. The incremental effect of an increase in ownership on future liquidity is particularly strong for value styles, with a coefficient of (t-stat of ); the estimated impact is still significant for small-cap styles, with a coefficient of (t-stat of -9.74). In contrast, ownership in large capitalization styles exacerbates illiquidity levels (coefficient of and t-statistic of 11.49). The results in Table 9, column (4) and (5), add ownership by hedge fund styles. Stocks with high ownership among hedge funds in the Long Short (LS) style category have a strong negative impact on stock illiquidity, stressing the general role of liquidity provision for these funds unconditionally. In summary, aggregate institutional ownership continues to be significantly associated with higher liquidity. However, ownership concentration in individual styles or strategies suggests a more complex relation between institutional ownership and stock liquidity. For example, stocks with ownership that is highly concentrated among long-short hedge funds experience lower illiquidity. We now turn to liquidity risk and styles in Table 9, panel B. It is apparent here that considering ownership by different investment styles is important, as the explanatory power of the regression increases across different specifications compared to the analysis without ownership style breakdown. Furthermore, the role of generalized ownership becomes insignificant once it 17

19 is orthogonalized on style ownership, suggesting that all the effects on liquidity risk are coming in from concentration in styles. We notice that while mutual fund style results were generally stronger for liquidity levels in Table 9, panel A, the findings are now most striking for the LS strategy group of hedge funds. More specifically, these hedge funds seem to predict a much larger liquidity risk for the stocks they own, with a coefficient of 2.55 and a t-stat of The mutual fund style that is closer in terms of significance of the results is small-cap, with a coefficient of 4.00 (t-stat equal to 3.61). There is only one style that significantly contributes to a lower future liquidity risk and that is value. If a stock is owned by these funds, it is likely to experience a relatively higher return when the stock market as a whole is more illiquid, suggesting that funds tend to hold on more to these stocks. 4 Analysis of Causality One concern in interpreting our empirical results is that stock liquidity and ownership may be endogenous. More specifically, a priori it is not clear whether institutional ownership causes liquidity or whether causality runs in the other way. For instance, it is possible that hedge funds that can place a limit on outflows have a preference for illiquid stocks to capture the liquidity premium in stock returns. In contrast, mutual funds that can experience outflows at any time may prefer more liquid stocks. 8 We acknowledge that there is no perfect solution to this endogeneity concern but seek to address it in several ways. First, the predictive regressions in the previous section relate future liquidity and liquidity risk to current ownership and control for lagged values of liquidity and liquidity risk. While current liquidity and ownership may well be endogenous, for most funds it is less likely that current ownership is endogenous with respect to future stock liquidity or liquidity risk. 9 Second, following Aghion, Van Reenen, and Zingales (2009), we use the inclusion of a firm 8 Da, Gao, and Jagannathan (2011) decompose a mutual fund s stock selection skill into various components depending on whether their trades are liquidity-supplying or liquidity-demanding. They find that income-oriented funds are more likely to generate performance by providing liquidity while for growth-oriented funds liquidity absorbing trades motivated by superior information are the most important component of skill. 9 Cao, Chen, Liang, and Lo (2012) and Cao, Simin, and Wang (2012) document that some hedge fund and mutual fund managers are able to time liquidity at the aggregate level, reducing the market beta of their portfolio when equity-market liquidity is low. However, this does not imply that hedge funds and mutual funds are also able to time liquidity at the individual stock level. 18

20 in the S&P 500 index to identify exogenous changes in ownership. The idea is that mutual fund managers tilt their portfolios towards firms that are included in a broad stock market index like the S&P 500 because their performance is evaluated relative to such an index. 10 For that reason, the inclusion restriction for the instrument is likely to be satisfied. We also expect the exclusion restriction to be satisfied because Standard and Poor s claim that their decision to include a firm in the index is only related to past liquidity and not to predictions about future liquidity or liquidity risk. In addition, we control for any relation between past and future liquidity (risk) by adding lagged liquidity (risk) to the regression. The left panel in Table 10 reports the results for the predictive regressions of stock illiquidity using index inclusion as an instrument for institutional ownership. Column 1 shows that total ownership remains a significant negative predictor of future illiquidity after controlling for possible endogeneity. In fact, the coefficient is much larger in absolute value than in the standard panel regression in Table 8, panel A. As a test of the strength of the instrument, we also report F -statistics on the excluded instrument based on the first stage estimates. These tests confirm that index inclusion is a strong instrument for ownership. 11 Our findings for mutual funds and hedge funds in columns 2 and 3 show a similar increase in the absolute value of the ownership coefficient. A potential explanation for this result is that the firms that join the S&P 500 are quite different from the rest of the sample. However, in the regressions we control for differences in characteristics across firms by including measures of firm size, book-to-market, and volatility. Thus, the evidence in Table 10 shows that even after controlling for potential endogeneity concerns, the impact of institutional ownership on future stock liquidity remains economically large and statistically significant. On the right hand side of Table 10 we present empirical results for instrumental variable regressions with liquidity betas as dependent variable. Again we find that after dealing with possible endogeneity of ownership our results become even stronger than in Table 8, panel B. In particular, both total institutional ownership and mutual fund and hedge fund ownership predict an increase in liquidity betas. These results are consistent with the hypothesis that correlated trading among institutional investors increases the liquidity risk of stocks. 10 In addition, Cremers and Petajisto (2009) show that a significant fraction of equity mutual funds that claim to be active are in fact closet indexers. 11 The coefficient on the index inclusion dummy in the first stage regression is highly significant. Index inclusion not only increases total institutional ownership, but also ownership by mutual funds and hedge funds. 19