Institutional trading and share returns

Journal of Banking & Finance 35 (2011) 3383–3399

Contents lists available at ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

Institutional trading and share returns F. Douglas Foster a, David R. Gallagher b,c,⇑, Adrian Looi d a

School of Finance, Actuarial Studies and Applied Statistics, The Australian National University, Canberra, ACT 0200, Australia Macquarie Graduate School of Management, Sydney, NSW 2109, Australia c Capital Markets CRC Limited, Sydney, NSW 2000, Australia d Marshall Wace, London, United Kingdom b

a r t i c l e

i n f o

Article history: Received 8 March 2010 Accepted 26 May 2011 Available online 30 May 2011 JEL classification: G23 Keywords: Trading behavior Informed trading Market impact Institutional trading

a b s t r a c t Using a unique database of daily transactions from Australian equity managers, we investigate the relation between institutional trading and share returns. The 34 institutional investors included in our sample exhibit a statistically and economically significant ability to predict large capitalization share returns for the ten days following their trades. Detailed analysis indicates that investment manager style is important in understanding the link between institutional trading and stock returns. The contemporaneous relation between institutional trading and returns depends on trade size, broker use, and investment style. We find growth-oriented managers are momentum traders, while style-neutral and value managers are contrarian. Ó 2011 Published by Elsevier B.V.

1. Introduction Professional fund managers, as significant holders of equities, have the capacity to influence share returns and trading volume. Given the enormous value of assets under their management, these professional investors not only comprise a large percentage of daily trading volume but also have access to a wide pool of resources to gather costly information and develop expertise. As such, key institutional investors have the capacity to move prices both directly through their own trading, as well as indirectly by influencing the trading decisions of other market participants who may observe their actions.1 The literature shows individual institutional trades have a permanent price impact,2 suggesting that in aggregate, researchers should expect to observe active fund managers moving prices through trading. Additionally, research examining changes in the periodic holdings of fund managers indicate that increases (decreases) in holdings are contemporaneously correlated with increasing (decreasing) stock prices.3 We use detailed data from a ⇑ Corresponding author at: Macquarie Graduate School of Management, Sydney, NSW 2109, Australia. Tel.: +61 2 9850 9975; fax: +61 2 9850 9942. E-mail addresses: [email protected] (F.D. Foster), david.gallagher@ mgsm.edu.au (D.R. Gallagher), [email protected] (A. Looi) 1 Market impact studies documenting the effect of trade activity on stock prices include Chan and Lakonishok (1995), and Chiyachantana et al. (2004). 2 See for example, Chan and Lakonishok (1993, 1995), Keim and Madhavan (1997), and Chiyachantana et al. (2004). 3 See Lakonishok et al. (1992), Nofsinger and Sias, (1999), and Wermers (1999). 0378-4266/$ - see front matter Ó 2011 Published by Elsevier B.V. doi:10.1016/j.jbankfin.2011.05.018

sample of Australian institutional investors and find that links between institutional investor trades and stock returns are more nuanced than current literature suggests. Although our sample is limited in a number of ways (time period covered, number of funds, shares universe, and size of local market), our study provides new and interesting insights, while being consistent with a significant body of prior work. Using daily institutional investor trade data, we find no contemporaneous relation between stock returns and aggregate manager trading based on either the number of managers buying (selling), or the total volume of their purchases (sales). While this result at first seems counterintuitive, it can be reconciled with prior studies in that the contemporaneous price reaction depends on a fund manager’s investment style. Value managers are contrarian and may act as price stabilizers; they provide liquidity to the market during periods of high volatility through buying on weakness and selling on strength. Hence, value manager trading yields a negative relation with contemporaneous stock returns. Conversely, growth managers tend to buy (sell) shares whose price is rising (falling), so growth managers trading is positively correlated with contemporaneous stock returns. In aggregate, the net contemporaneous effect of both value and growth manager trading can be inconsequential. We explore possible price stabilization by closely examining value manager trading activity, and find their ability to obtain a negative correlation with contemporaneous stock returns requires unstable (or highly volatile) intraday stock prices. Therefore, when measuring the overall average market impact of value

3384

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

managers we must take into account market volatility (as a proxy for the likelihood of a price stabilization trade), or we may introduce a downward bias to the average market impact estimate. We find the relation between market impact and volatility depends on manager style – the market impact incurred by growth managers has little relation with volatility, while for value managers, high volatility is associated with a negative market impact from trade. Our findings also have important implications for the study of institutional ownership and stock returns. Empirical studies document a contemporaneous relation between changes in institutional holdings and stock returns (on a monthly, quarterly or yearly basis) and imply: (1) institutional traders push prices in the direction of their trade through their permanent market impact (price pressure); (2) institutional investors are intra-period momentum traders, buying as prices rise during the month, thereby inducing a positive monthly contemporaneous relation; or (3) institutions are able to predict intra-period stock returns. Without more frequent trading data, distinguishing between these three competing hypotheses is difficult. However, with daily trading data testing each hypothesis is relatively straightforward. We show that aggregate manager trading volume is not correlated with contemporaneous stock returns, rejecting the price pressure hypothesis. We also show that momentum trading depends on investment style: growth managers are momentum traders, while value managers are not. This weakens the intra-period momentum trading hypothesis since not all managers are momentum traders. Finally, we show our sample of fund managers are able to predict future stock returns, supporting the third hypothesis that manager trading contains intra-period information. Hence, our sample suggests the documented contemporaneous relation between periodic changes in fund manager holdings and stock returns may be due to fund manager trading on intra-period information. Finally, we observe the manner in which investment managers choose to process their trades. For example, we know which broker (using an established broker identification number through the Exchange) was used to facilitate the order. Analysis of the order submission corroborates our breakdown of liquidity and information-related trades. We argue that when a single fund manager splits their order across many brokers, they are more likely doing so because they have an informed basis for their trade, and there are likely to be longer-term price consequences. Further, when a single broker manages a number of similar orders from a range of fund managers, it may be a consequence of the broker soliciting liquidity to offset a prior trade. If this is the case, we expect transitory price reactions to these trades as the liquidity need is met. Both of these interpretations are confirmed by our data. The remainder of the paper is organized as follows. Section 2 provides a brief background and outlines foundations for our study. Section 3 presents a description of the data and provides basic statistics. Section 4 outlines our research design while Section 5 reports the results. Section 6 provides a summary.

2. Background Our research is related to studies examining the link between changes in institutional holdings and stock returns; however, we examine this issue with more detailed (daily) data than previous studies.4 Prior studies document a strongly positive contemporaneous relation between changes in institutional ownership and stock returns on a monthly, quarterly or yearly basis (Grinblatt et al., 1995; Nofsinger and Sias, 1999; Wermers, 1999; Sias et al., 2006). 4 Some recent studies that examine trading and return effects with high frequency daily data for investors include Keswani and Stolin (2008) and Yan and Zhang (2009).

We explore four main explanations for this result and outline how our work adds to each literature in Sections 2.1–2.4. 2.1. Price pressure Institutions may push prices in the direction of their trades. If active institutional traders trade on the premise of superior information, this price pressure may be a result of the information revealed through trading. Alternatively, active institutional traders may induce a counter-party to trade by offering a liquidity fee, thereby shifting the counter-party away from their preferred inventory positions, which could have a liquidity impact on prices. While we expect such liquidity impacts to be short-lived, sustained aggregate institutional trading (such as when several large institutions transact large trade packages over many days) may create a contemporaneous monthly relation. The debate between the liquidity and information effects of institutional trading has a long history.5 The empirical research overwhelmingly rejects the liquidity hypothesis (Holthausen et al., 1990; Lakonishok et al., 1992). Using data similar to ours, Chan and Lakonishok (1993, 1995) document a positive open-to-trade market impact for purchases, followed by price continuation rather than reversal (even after taking trading packages into account), which supports the information rather than liquidity hypothesis. However, for sales, they document reversal rather than continuation, suggesting liquidity rather than information motivations dominate sales. Prior work has used the observed market impact to determine the relative strengths of information versus liquidity effects. We take a different approach and develop measures to track the information content of manager trades (versus the potential liquidity impact they may have on prices). Liquidity effects are likely to be related to the volume of shares traded – in inventory models, the liquidity premium demanded by liquidity suppliers is related to the total volume of demanded liquidity rather than the number of traders demanding liquidity (Stoll, 1978; Grossman and Miller, 1988). To proxy for information effects we consider the unanticipated number of fund managers buying or selling on each day. If an institutional investor is trading because of a particular view about future returns, they may be unable to defer transactions as competition from other fund managers, or public announcement of information would both serve to limit discretion. From microstructure models we expect these forces to be especially striking when information is highly correlated and when the insight is fully revealed through a public signal in the near future (Foster and Viswanathan, 1996). Accordingly, if we see a number of mutual fund managers trading in the same manner on the same day, we argue that it is more likely that the motive for trade is information-based.6 Our research shows, in aggregate, neither the number of funds trading, nor the volume of shares purchased and sold by institutions is correlated with contemporaneous stock returns. However, we show that the number of value managers purchasing has a negative contemporaneous effect, while that of growth managers is positive. Consistent with prior empirical research, our findings support the information rather than the liquidity hypothesis. However, we make one important qualification: growth managers push prices in the direction of their trade due to information; however value managers often act as price stabilizers incurring negative market impact for the service of supplying liquidity. Further, we 5 For example, the liquidity effect is explored in Stoll (1978) and Grossman and Miller (1988). 6 This method of breaking down the information and liquidity effect by volume and number of institutions trading is also consistent with the prior empirical work; see Sias et al. (2006).

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

show the intensity of this price stabilization behavior is positively related to the level of price instability (or volatility). 2.2. Momentum Institutions may engage in intra-period momentum trading. If institutions purchase (sell) as prices rise (fall) during the month, then the monthly contemporaneous relation between changes in institutional ownership and stock returns will be positive. A common view of market efficiency is that historical stock returns should not be a predictor of future returns, so historical returns should not provide systematic ways to build portfolios that outperform a passive benchmark. However, a number of empirical studies document significant abnormal returns to both contrarian and momentum trading (DeBondt and Thaler, 1985, 1987; Lo and MacKinlay, 1988; Chan et al., 1996; Jegadeesh and Titman, 2001; George and Hwang, 2004). We contribute to the literature on momentum trading by documenting the relation between institutional trading and short-term past stock returns. Overall, our sample of active Australian equity managers are short-term (over 10 days) contrarian traders; however when we partition by investment style, we find that growth oriented investment managers (growth managers and growth-at-a-reasonable-price (GARP) managers) are momentum traders, while style neutral and value managers are contrarian. 2.3. Return predictability Institutional traders may possess private information subsequently impounded in prices after their trades. However, studies find a positive correlation between institutional trading and future stock returns (Daniel et al., 1997; Wermers, 1999; Nofsinger and Sias, 1999; Sias et al., 2006; Dorn et al., 2008).7 These findings are consistent with institutional traders possessing superior information. A combination of contemporaneous price impact and momentum institutional trading may also give the appearance that current institutional trading predicts future stock returns. That is, current institutional trading may simply be predicting future institutional trading (De Long et al., 1990). Our study contributes to the literature by measuring the impact of information while controlling for institutional trading volume on a daily basis. We find that, after controlling for potential liquidity-based price impacts, institutions have significant power in predicting future stock returns. 2.4. Price destabilization Our study also contributes to the literature on price destabilization. De Long et al. (1990) build a model of positive-feedback traders and rational speculators. Positive-feedback traders purchase (sell) as prices rise (fall); however, in doing so, their market impact causes prices to rise (fall) even further, which in turn provides them with even more motivation to continue purchasing (selling). In their model, the interaction between rational speculators and positive-feedback traders increases stock price volatility. Most empirical studies confirm that institutions engage in positive-feedback trading (Grinblatt et al., 1995; Wermers, 1999). Badrinath and Wahal (2002) however, decompose changes in holdings into entry and exit positions and find momentum trading for entry positions, but contrarian trading when exiting. We document the relation between daily past returns and institutional trading (partitioned by investment style) and show that value managers are deeply contrarian, while other managers are momentum traders. This finding weakens the argument that institutions as a whole 7

For exceptions see Chan and Lakonishok (1995) and Cai and Zheng (2004).

3385

are price destabilizing, since only a portion of managers engage in positive-feedback trading. Further, we show that value managers potentially stabilize prices during periods of high volatility. Evidence of value manager price stabilization has important implications for the market impact literature. Keim and Madhavan (1997) investigate the role of investment style on market impact and find technical and index managers demand immediacy more than value managers and incur higher market impact costs. Similarly, Chan and Lakonishok (1993, 1995), show that value managers on average incur negative open-to-trade market impact costs while growth managers incur positive costs, presumably due to more patient trading. This is consistent with our findings that the contemporaneous relation between value manager trading and stock returns is negative. Our research suggests that when measuring the average market impact of value managers, it is important to acknowledge that some trades may be done with the intention of incurring negative market impact as payment for liquidity provision. Additionally, when fund managers trade, they compete for liquidity, but the degree of competition depends on the style of the manager involved. We find value managers are often supplying liquidity rather than demanding it. Therefore while expected market impact costs can be aggravated by the existence of other growth managers, other value managers do not increase market impact costs because they are less often competing for available liquidity. Other studies have considered the effect of competition for liquidity. Chiyachantana et al. (2004) examine institutional trades from 37 countries over bull and bear periods and find market impact costs are higher for purchases during bull periods, and vice versa for sells. Their explanation is that during bull periods, demand for buyer-initiated liquidity is higher, which causes market impact costs for purchases to be higher than otherwise.

3. Data We examine the daily trading of a representative sample of active Australian equity fund managers during two calendar years for which complete trading records are available (2000 and 2001). While there are obvious limitations in using a sample drawn from a few investors in one country over a relatively short period, the data allow a highly detailed glimpse into the activities of key professional investors. The particular circumstances of this market, when taken with what we know from prior studies, provide some important insights. For example, the Australian market is relatively small by global standards, and so actions of institutional investors may have greater importance. Also, unlike other markets, Australian fund managers have, as a group, consistently outperformed passive benchmark indices.8 This is important because the hypothesis that institutional traders move prices due to the information revealed through their trades supposes that they actually have superior information. Our sample comprises 34 active Australian equity managers, sourced from the Portfolio Analytics Database. These institutions are representative of the institutional investment management landscape in Australia on the basis of size of assets under management, investment style, and ownership structure (e.g., large and publicly listed firms versus smaller boutique managers). The largest ten Australian investment managers account for 58% of total assets under management (AUD$399.9 billion out of AUD$688.9 billion). 8 Evidence of superior manager ability in Australian equities is documented in Mercer Investment Consulting surveys and from academic studies examining fund performance (Gallagher, 2003).

3386

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 1 Descriptive statistics. This table reports descriptive statistics for the Portfolio Analytics Database partitioned according to trade direction. Dollar trade value is the weighted average price of the trade multiplied by trade quantity. Trade size relative to volume is the number of shares traded as a percentage of the mean number of shares traded per day over the 20 days prior. Trade size relative to shares outstanding is the number of shares traded as a percentage of the number of shares outstanding. These statistics are for the sample period 2 January 2000 to 31 December 2001. Manager trading activity is defined as the number of purchases plus the number of sales made by our sample group in the sample period.

Panel A – Manager style and number of trades Number of managers Number of trade day observations Number of purchase day observations Number of sale day observations

Growth

Value

GARP

Neutral

3 4551 2907 1644

9 7428 4435 2993

8 6142 2790 3352

14 12372 6337 6035

Mean

Stdev

25th

50th

75th

Panel B – Distribution of trade size (purchases) Dollar trade value (‘000’s) Trade size relative to volume Trade size relative to SharesOutstanding

544 4.92 0.0080

1307 20.35 0.0247

83 0.46 0.0008

219 1.52 0.0027

582 4.59 0.0081

Panel C – Distribution of trade size (sales) Dollar trade value Trade size relative to volume Trade size relative to SharesOutstanding

557 5.55 0.0101

1062 15.55 0.0319

570 5.04 0.0095

208 1.54 0.0029

72 0.46 0.0009

The Portfolio Analytics Database was constructed using an ‘invitation approach’ where 45 institutional investment managers were requested to provide data, with 34 providing information in a usable format. This database was constructed with the support of Mercer Investment Consulting, with periodic monthly portfolio holdings and daily trade information provided under strict confidentiality conditions. The database includes all transactions in stocks, futures contracts, and options for each fund; we only evaluate equity trading performance. Our study also uses a detailed record of all stock price and transaction information from the Australian Securities Exchange (ASX) (provided by the Securities Industry Research Centre of Asia–Pacific (SIRCA)). Annual report information (for the book-to-market ratio) was obtained from ASPECT Financial. The investment managers were asked to provide information about their largest pooled active Australian equity funds (where appropriate) that were open to institutional investors. These funds are benchmarked against the S&P/ASX 200 and 300 Accumulation Indices.9 The term ‘largest’ was defined as the marked-to-market valuation of assets under management as at 31 December 2001, and was used as an indicative means of identifying portfolios that were representative of the investment manager. In addition, the largest pooled institutional equity fund represents the manager’s single largest revenue source from the fund family, because fund revenue is determined as a fixed percentage of assets under management, and fee variations are relatively small within common asset classes. We investigate possible survivorship and selection bias by comparing the performance of our sample with the population of investment managers, which includes non-surviving funds. These data are sourced from the Mercer Investment Consulting Manager Performance Analytics (MPA) database. The average gross outperformance of the average manager relative to the ASX/S&P 200 index is 1.78% per annum with a standard deviation of 1.39%.10 For our sample the mean manager outperformed the average MPA manager, weighted by manager years, by 0.34% per annum. While this indicates that our sample outperforms the industry, we find 9 The correlation between these indexes is very high (approximately 0.98). The additional stocks included in the S&P/ASX 300 only increases total market capitalization by around 1 percent. 10 According to the Mercer Wholesale Investment Fee Survey, mean management expense ratios for 2000 and 2001 were 0.74 and 0.8 percent respectively, indicating that active Australian equity managers were able to outperform on both a gross and net of fees basis.

that the magnitude of the outperformance is low compared to the dispersion of performance. Selection bias, it appears, is not a significant problem. In 2001, the mean return of the entire fund population was 12.42% with a standard deviation of 3.8%, while the mean performance of our sample was 12.68% with a standard deviation of 5.5%. Table 1 provides descriptive statistics for the number of manager trades, investment style, and characteristics of the trades. Panel A shows our sample comprises predominantly style-neutral and value managers. Hence, we combine growth and GARP managers as ‘‘growth oriented’’ funds. Panels B and C of Table 1 provide the distribution of trades for the 50 largest capitalization stocks (purchases and sales, respectively) in terms of three measures of trade size: dollar value of trade, trade size relative to mean daily volume and trade size relative to the number of shares outstanding. For the purpose of this study, we restrict the sample of stocks under investigation to the largest 50 stocks, ranked by market capitalization at the start of the sample. This restriction maintains a reasonable number of manager trades per day per stock.11 As at 31 December 2001, the fifty largest stocks account for 82% of total market capitalization stocks in which institutional investors are more actively engaged. Statistics regarding manager trading in the selected stocks are presented in Table 2. The mean number of purchasing and selling managers each day in the 15 largest capitalization stocks is 1.07 and 0.82, respectively. As a percentage of mean daily trading volume, the average number of shares purchased and sold per day in the 15 largest capitalization stocks is 2.83% and 2.45%, respectively. We standardize manager-trading activity with the sample mean and standard deviation of manager trading in each stock and find that the relative size of manager trades rises for lower capitalization shares. Table 2 also provides statistics on the weight of the portfolio invested in stocks included in our sample of 50 large capitalization stocks. On average, over 65% of the portfolio weight is allocated to these 50 stocks, indicating that our sample covers much of the manager’s eligible universe, weighted by market capitalization.

11 We have also repeated all analysis using the fifty most active stocks (by manager trading) over the sample period and obtain very similar results. Manager trading activity is defined as the number of purchases plus the number of sales made in a stock by the managers in the database over the sample period

3387

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 2 Institutional trading activity. This table reports descriptive statistics for the trading activity and portfolio weights for the largest 50 stocks in our sample. We select the largest stocks based on market capitalizations on the first day of our sample period. Trade size relative to volume is the number of shares traded as a percentage of the mean number of shares traded per day over the 20 days prior. Trade size relative to shares outstanding is the number of shares traded as a percentage of the number of shares outstanding. The mean sum of weights is the sum of the portfolio weights for all stocks in the corresponding stock size bucket, averaged over all the managers in the sample. The Weights of stocks is the average weight allocated by the average manager to stocks in that stock size bucket. Managers with zero holdings are included in the calculation of these figures. All figures are in percent. 3rd largest rank 35–50

2nd largest rank 16–34

1st largest rank 1–15

Mean

Stdev

Mean

Mean

Panel A – Distribution of trades partitioned by stock rank Number of purchasing managers Number of selling managers Trade size relative to volume (purchases) Trade size relative to volume (sales) Trade size relative to SharesOutstanding (purchases) Trade size relative to SharesOutstanding (sales)

0.34 0.34 4.52 4.14 0.0073 0.0093

0.58 0.60 18.18 16.22 0.0250 0.0378

0.59 0.53 3.57 4.03 0.0059 0.0065

0.77 0.73 10.50 11.33 0.0189 0.0175

1.07 0.82 2.83 2.45 0.0047 0.0038

1.04 0.99 6.16 6.88 0.0100 0.0091

Panel B – Distribution of manager weights and overweights Sum of weights of stocks within rank bucket Weights of stocks within rank bucket

6.1534 0.3846

4.7010 0.9132

14.8020 0.7791

6.0730 1.3342

44.4080 2.9605

11.9710 2.7364

Sum of overweights of stocks within rank bucket Overweights of stocks within rank bucket

1.2744 0.0797

4.7010 0.9132

3.2452 0.1708

6.0730 1.3342

5.4511 0.3634

11.9710 2.7364

4. Research design If institutional investors possess superior private information, their trading on information may have a contemporaneous effect on stock returns as their insights are revealed through trading. This means that a fund manager has a strong incentive to match their informed trading to available market liquidity. A skilled investor trades aggressively when the share price is unlikely to rise (fall) from their purchases (sales).12 If this is the case, then professional trades may have a relatively small share price impact, irrespective of the underlying trade motive. To gain a better understanding of these tradeoffs we need to posit a motive for institutional trades from transaction histories. This requires information about the trades of other potentially informed traders. Researchers have considered the effects of competition among informed investors where the information is identical (perfectly positively correlated) or merely related. When there is a large number of trading periods (when the information is long-lived in calendar time or trading is continuous), they predict dramatic changes to the intensity of trade by informed investors, price responses and expected profits to informed traders depending on the correlation between informed trader beliefs. Examples of this can be found in studies by Foster and Viswanathan (1996), and Back et al. (2000). With identical information and ‘‘near’’ continuous trading, for example, we expect to find very aggressive trade by the informed investors, low total expected profits from the information, and low market liquidity in response to their actions. This is markedly different from the case of a single informed investor. For cases that do not assume identical information, Foster and Viswanathan (1996) show that there is initially strong competition among informed traders when the conditional correlation between their information is positive.13 These insights are consistent with evidence in empirical studies such as Sias et al. (2006), who suggest that the impact of informed trading is related to the number of traders rather than their trading volume. Consequently, when we have a number of potentially informed traders all buying (selling) the same stock on the same day, it is more likely that (i) they have positive (negative) information about the company’s future share price, and (ii) that their information is

12 There is a basic question of whether we see any discretionary trading among fund managers. Some evidence consistent with discretionary liquidity trading is that trading volume from our sample of fund managers is significantly lower on Monday than any other trading day of the week. 13 See Figs. 5 and 6 in Foster and Viswanathan (1996).

Stdev

Stdev

positively correlated or is expected to last for a limited amount of time. Of course, there are other forms of less correlated (initially or conditionally, after some trading by a number of differentially informed traders) information flows that we will not detect with this approach. Gallagher et al. (2010) document trading among our sample of managers around earnings announcements and find behavior consistent with information motives for trade, when information is about to be made public. Bozcuk and Lasfer (2005) document information effects from only the larger institutional trades on the London Stock Exchange that can be consistent with greater information flows from trade when institutional investors choose not to exercise discretion. We use total trading volume, expressed as a percentage of mean daily trading volume, as a proxy for trades with an ambiguous motive, or those for which we cannot reject as being liquidity motivated (the fund manager is comfortable increasing trade intensity with the expectation that the market will accommodate the transaction). To be more certain that there is an information motive for a transaction, we require that such trades occur when more than one of the investment firms in our sample is making similar trades (e.g., buying) in the same stock on the same day. To explore further the motive for trade and the consequences for subsequent share returns, we also consider the style of the fund manager. Our results are clearly limited by our sample; even though we do not have the transactions of all institutional managers, incorporating the actions of some other managers allows us to generate insights that are both consistent with, and extend, the work of others. To investigate the relation between institutional trading and stock returns we examine three temporal orderings: (a) the influence of past stock returns on current institutional trading, (b) the contemporaneous impact of fund trading on stock returns, and (c) the ability of professional investors to forecast future stock returns (the influence of past trading on current returns). We develop regression tools for each of these three settings. 4.1. Past stock returns and institutional trading If the stock market is efficient, historical returns have no predictive power. Consequently, a rational stock trading strategy should not rely on historical stock returns. Accordingly, our null hypothesis is that past stock returns have no influence on institutional trading. However, a large body of empirical research (Lo and MacKinlay, 1988; Jegadeesh and Titman, 1993; Chan et al., 1996) shows that past stock returns do have predictive power. Studies

3388

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

investigating the role of historical stock returns on institutional trading appear to reject the null hypothesis (Grinblatt et al., 1995; Nofsinger and Sias, 1999; Cai and Zheng, 2004). To test our null hypothesis, we regress standardized institutional trading (purchasing and selling separately) against lagged stock returns, lagged institutional trading, and the lagged values of aggregate shares traded by the institutions in the sample. We also include a number of control variables: the market return, book-to-market ratio, size, and momentum. If past stock returns do not influence the trading decisions of institutional traders, the slope coefficients of lagged stock returns will not be statistically different from zero. We include lagged institutional trading to investigate the extent to which institutional trading is serially correlated. Serially correlated trading could occur for a number of reasons: temporally correlated information, herding, or because fund managers may trade over several days in order to reduce market impact (Chan and Lakonishok, 1995). Indeed, Fong et al. (2011) show that followers imitate the trades of leaders and that such activity leads to significant profits for the leader and early followers.14 To isolate the effect of past returns on institutional trading, lagged values of aggregate shares traded by sample institutions are included. We test our hypotheses with a panel data model, where we allow the coefficients on the control variables to vary according to each stock in the sample:

ys;t ¼

z¼10 X

bj;z Rs;tz þ bk MgrBuyt1 þ bl MgrSellt1

z¼1

þ bm SharesBuyt1 þ bn SharesSellt1 þ

S X

ba;s :ds

s¼1

þ

S X s¼1

þ

S X s¼1

bb;s :ds Markett þ

S X

bc;s :ds :SIZEt

s¼1

bd;s :ds :BMRatiot þ

S X

be;s :ds :Momentumt þ es;t

ð1Þ

s¼1

The dependent variable ys,t is one of four variables: MgrBuyt, MgrSellt, SharesBuyt, and SharesSellt. MgrBuy is the standardized number of managers purchasing stock s on day t. MgrSell is the standardized number of managers selling stock s on day t. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. SharesBuy is the total number of shares bought by managers in stock s on day t divided by the mean daily volume calculated over the prior 20 days (approximately a trading month). SharesSell is the total number of shares sold by managers in stock s on day t divided by the mean daily volume calculated over the prior 20 days.15 Rs,t is the return on stock s on day t. Explanatory variables include lags of the dependent variables as well as risk control variables. Market is the return on the valueweighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange, not to be confused with the ASX/ S&P 300 which is an index constructed by Standard and Poor’s) on day t. Size, is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in 14 Venezia et al. (2011) find that both amateur and professional investors exhibit herding behavior, yet it is less pronounced for professional investors. Chuang and Susmel (2011) find that individuals and institutions trade aggressively, and this activity depends on states such as market condition, stock size, risk, and momentum effects. 15 In unreported regressions, we re-estimate our results using alternative measures of volume traded by managers, including the number of shares purchased/sold divided by the number of shares outstanding. The results are qualitatively similar.

the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value-weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock.16 Since lagged values of the dependent variable in Eq. (1) are included as explanatory variables, ordinary least squares estimates are inefficient and inconsistent. We therefore employ a two-step procedure suggested by Hatanaka (1974) to compute our estimates. The severity of the inconsistency depends on the number of periods in the panel. Nickell (1981) shows that for a small number of time periods the bias can be quite severe. However in our sample we have over 500 daily observations compared to only 50 securities in each panel. Consequently, the degree of inconsistency is likely to be small. 4.2. Contemporaneous stock returns and institutional trading We expect the contemporaneous relation between institutional trading and share returns to be influenced by liquidity and information effects. If institutional trades are small or do not require the provision of additional liquidity, we expect no contemporaneous link between institutional trade and stock prices. This would be the case, for example, if institutional trades have a temporary liquidity impact that is dissipated by the end of the trading day. Hence, our first null hypothesis is that there is no contemporaneous association between the volume of trades and excess share price returns. It is possible, however, that liquidity effects may stretch beyond the day on which the trade was made; i.e. the trade impact has become permanent. If there were an information motive to the trade, we would expect the market price to shift as a liquidity effect that then extends beyond the current day. We use the number of institutions purchasing or selling to proxy for the information content of a trade. This gives us another null hypothesis: contemporaneous stock returns are not related to the number of institutions purchasing or selling. Hence we can decompose any price impact into liquidity and information components with the model given in Eq. (2).

Rs;t ¼

z¼10 X

bk;z MgrBuytz þ

z¼0

þ

z¼10 X

z¼10 X

bm;z SharesBuytz þ

z¼0

þ

S X s¼1

þ

S X s¼1

bl;z MgrSelltz

z¼0 z¼10 X

bn;z SharesSelltz

z¼0

ba;s :ds þ

S X

bb;s :ds Markett þ

s¼1

bd;s :ds :BMRatiot þ

S X

bc;s :ds :SIZEt

s¼1 S X

be;s :ds :Momentumt þ es;t

ð2Þ

s¼1

where the dependent variable Rs,t is the stock return on day t in stock s and other variables are defined as above. Note that Eq. (2) explicitly considers the possibility that institutional trading may be positively auto-correlated. Past studies examining the role of liquidity versus information in share returns overwhelming reject a pure liquidity explanation 16 In measuring stock returns, we use the midpoint of the closing bid and ask, rather than the last trade as at the close. We reproduce our results using the closing price and do not find any significant difference (not reported).

Panel A – R-squared R-squared Adjusted R-squared Variable

0.1371 0.1271 MgrBuy Coefficient

Panel B – Regression estimates MgrBuy(t-1) 0.5193 MgrSell(t-1) 0.0001 Ret(t-1: t-5) 2.0298 Ret(t-1: t-10) 2.6908 SharesBuy(t-1) 0.5336 SharesSell(t-1) 0.0951 * ** ***

Statistical significance at the 10% levels. Statistical significance at the 5% levels. Statistical significance at the 1% levels.

0.1392 0.1293 MgrSell

0.1097 0.0994 SharesBuy

0.1120 0.1017 SharesSell

t-stat

Coefficient

t-stat

Coefficient

t-stat

Coefficient

t-stat

35.53*** 0.01 3.15*** 2.88*** 6.58*** 1.18

0.0155 0.4753 1.3350 0.8793 0.2803 0.7748

2.16** 33.83*** 2.12*** 0.96 3.34*** 9.81***

0.0025 0.0006 0.0603 0.0978 0.3836 0.0066

7.61*** 1.82* 2.41*** 2.63*** 25.34*** 1.26

0.0000 0.0020 0.0130 0.0685 0.0057 0.4103

0.07 6.27*** 0.42 1.50 0.92 28.68***

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 3 P PS Institutional trading and past stock returns. This table reports regression estimates of the following regression equation: ys;t ¼ z¼10 z¼1 bj;z Rs;tz þ bk MgrBuyt1 þ bl MgrSellt1 þ bm SharesBuyt1 þ bn SharesSellt1 þ s¼1 ba;s ds þ PS PS PS PS s¼1 bb;s ds Market t þ s¼1 bc;s ds SIZEt þ s¼1 bd;s ds BMRatiot þ s¼1 be;s ds Momentumt þ es;t The dependent variable y is one of four variables: MgrBuy, MgrSell, SharesBuy, and SharesSell. MgrBuy is the standardized number of managers purchasing stock s on day t, although we leave out the subscripts. MgrSell is the standardized number of managers selling stock s on day t. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. R(t) is the return on stock s on day t, SharesBuy is the total number of shares bought by managers in stock on day t divided by the mean daily volume calculated over the prior 20 days. SharesSell is the total number of shares sold by managers in stock s on day t divided by the mean daily volume calculated over the prior 20 days. Market is the return on the value weighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange) on day t. Size is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock. These results are for the sample period 2 January 2000 to 31 December 2001. Only trades in the largest 50 stocks (based on market capitalizations on the first day of our sample period) are included. We report the sum of the lagged coefficients as follows, (t-1: t-10) is the sum of the lags from t-1 to t-10.

3389

3390

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

(Holthausen et al., 1990; Lakonishok et al., 1992). If information effects dominate, we expect contemporaneous stock returns to be positively (negatively) correlated to the contemporaneous number of institutions purchasing (selling), rather than their contemporaneous aggregate volume of shares purchased (sold). 4.3. Past institutional trading and stock returns Eq. (2) also contains information on the link between current returns and past institutional trades. Hence, we can document the influence of prior (as many as ten trading days) institutional trades on current returns.17 If past institutional trades continue to have impact, we expect that they are more likely to be information motivated; evidence fund managers are able to anticipate future returns. We expect liquidity influences not to persist, and information influences to be relatively permanent. Hence, we expect stock returns to be positively (negatively) correlated with lagged numbers of institutions purchasing (selling). Further, we expect stock returns to be unrelated to lagged institutional trading volume. Eqs. (1) and (2) form a recursive system of equations that can be estimated with OLS. Since daily stock returns are likely to be autocorrelated, we create and use as dependent and independent variables return innovations, that is, the residuals obtained from fitting an AR(1) model to the return series of each security. We also note that each stock is likely to have a different residual variance inducing heteroskedasticity. We, therefore, estimate the residual variance for each stock individually, and transform the variables in the regression accordingly.18 The inclusion of dummy variables in expressions (1) and (2) controls for stock fixed effects.19

5. Results 5.1. Influence of stock returns on institutional trading The estimated regression coefficients from Eq. (1) are reported in Table 3. To conserve space we do not report the coefficients of risk control factors since there is an intercept, a market, a size, a book-to-market, and a momentum slope estimate for each of the 50 stocks in our sample. We report lagged coefficients as sums: the sum of coefficients up to lag 5 and the sum of coefficients up to lag 10. 17 We use a lag length of 10 for included dependent variables in our tests. In unreported results, we consider lags of up to 20 (essentially a full trading month) and the results do not change significantly. We find the influence of variables with lags greater than 10 are not statistically different from zero. 18 In unreported results, we test the robustness of our standard error estimates by conducting a boot-strap experiment. We form numeric standard errors by randomly reconstructing the independent variables to build an empirical distribution for the coefficients. For example, for each stock, we divide the manager buying variable into blocks of 4 (in order to retain any autocorrelation structures present in the manager buying variable) and randomly reassign the order of the blocks. We then re-estimate the regression with the randomly assigned blocks and note the coefficient of manager buying. We conduct this trial 500,000 times to build up a density for the coefficient of manager buying. Our results show that the reported regression results using OLS give similar results to the numerically generated estimates. We also find similar results when we use heteroskedasticity consistent estimators on the ‘raw’ unweighted data. We have also re-estimated our results using a series of stock-by-stock regressions rather than a panel data model, and find similar results for the mean coefficients and t-statistics. 19 An alternative specification would be to drop the dummy variables and report clustered standard errors. Our approach is discussed in Petersen (2009) (see Section 4.2 on page 464 and simulation evidence reported in Table 5, Panel A, Column I on page 462). Further, we find estimated correlations between independent variables and correlations in residuals with our specifications are both small, ensuring their product is near zero. This means that there are no substantive differences between using dummy variables and reporting either least squares or clustered standard errors (Petersen, 2009; expressions (6) and (3)). The use of least square dummy variables in this setting is consistent with the discussion in Greene (2008).

The results show that institutions are contrarian traders in aggregate. From the table, we observe that the sum of the coefficients of lagged stock returns from t-1 to t-10 is 2.6908 and 0.8793 for purchases and sales, respectively. This implies that a fall (rise) in price over the last 10 days induces institutions to purchase (sell). As an example of how this might be interpreted, consider shares in BHP Billiton, a well-known, large capitalization stock. The mean number of managers purchasing on any given day in our sample is 1.67, with a standard deviation of 1.47. For a 20% fall in the price of BHP (over and above any autocorrelation effects) over 10 days, for example, the model predicts that on average, one manager will be induced to buy (2.6908  0.2  1.47 = 0.79 managers). Our finding that institutions are, on average, contrarian traders is consistent with Gompers and Metrick (2001) and Cohen et al. (2002). However, it is inconsistent with Grinblatt et al. (1995), Nofsinger and Sias (1999), and Cai and Zheng (2004). There are perhaps two explanations for differences with this second group of papers: frequency of data and value versus equal-weighting of the underlying share positions. The frequency of data in these other studies varies from monthly to annual, while our data is daily. Institutions may have a positive feedback trading strategy over longer-term horizons, while trading in a contrarian fashion over the short-term. Chan and Lakonishok (1995) find that for institutional purchases using daily trading data on a value-weighted basis, institutions are momentum traders. On a simple-weighted basis, however, institutions are contrarian. For sales, they find that institutions are contrarian on both a value- and simple-weighted basis. Our findings are consistent with Chan and Lakonishok (1995) as our regression framework does not value-weight observations. Finally, a striking feature of the results in Table 3 is that institutional trading is serially correlated. Lagged institutional purchasing (selling) is highly positively correlated with current institutional purchasing (selling), indicating that institutional trading is associated with similar future transactions. This may be caused by a variety of factors: institutions may have serially correlated information, they may be herding, or institutions may be purchasing or selling trade packages over several days in order to reduce market impact (Chan and Lakonishok, 1995). In unreported results, we find similar results when we repeat the analysis using trade packages as defined by Chan and Lakonishok (1995) – there is persistence in transactions beyond what we can attribute to multi-day trade packages. In practical terms, highly significant coefficients on lagged institutional trading show that an increase in the number of institutions purchasing, yields an increase in institutions purchasing in the future. In the case of BHP, if we observe three managers purchasing more than average in one day (three more managers represents roughly two standard deviations above the mean since the standard deviation is 1.47), we expect an increase of one manager purchasing (2  0.5193 = 1.04) over the next day. These results are consistent with studies that show institutions ‘‘herd’’ (Grinblatt et al., 1995; Nofsinger and Sias, 1999; Wermers, 1999; Sias et al., 2006). Our results are also consistent with Sias (2004), who finds institutional trading is related more to lagged institutional trading than past stock returns. 5.2. Contemporaneous price impact of institutional trading For an institutional investor’s transaction to have a contemporaneous price impact on stock returns through the information content of their trades, we expect stock returns on day t to be positively (negatively) correlated with the number of institutions purchasing (selling) on day t. However, if the institutional investor has a contemporaneous price impact due to the liquidity pressures

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

placed on liquidity providers, we expect that stock returns on day t should be positively (negatively) correlated with the number of shares purchased (sold) on day t by the manager. We test these hypotheses using Eq. (2) with results reported in the column headed ‘‘Basic’’ in Table 4. The results indicate that there is no significant contemporaneous effect. The coefficient of manager buying from our estimate of Eq. (2) is 0.0002, while that of selling is 0.0001, neither statistically significant at conventional levels. This result is surprising since we expect institutional trades to have some informational impact, especially in light of our findings in the next section indicating that managers are able to forecast stock returns. There are a number of reasons why we do not find a contemporaneous effect, and we explore these possibilities in Section 5.4. Turning to the liquidity hypothesis, if managers are adept at hiding the information content of their trades, the contemporaneous effect of manager trading volume should not be statistically significant. We find no evidence of a contemporaneous effect for purchases or sales. One should note, however, that while we find a zero contemporaneous effect when measured over the close-toclose period, these results do not say anything about the market impact costs incurred by each manager individually. Chan and Lakonishok (1993), for example, show that managers experience market impact costs when measured against open or closing price benchmarks. 5.3. Ability of institutional traders to forecast future stock returns Consistent with studies showing that institutional traders possess valuable private information (Daniel et al., 1997; Nofsinger and Sias, 1999; Sias et al., 2006), the lagged values of the number of institutions purchasing (selling) reported in Table 4 are significantly related to stock returns. Further, the correlation between lagged aggregate institutional trading volume and stock returns is not as statistically significant as the number of institutions purchasing (selling). So, our information measure of fund activity is able to predict future returns, while the liquidity measure of fund activity cannot. This suggests that, as a group, institutional investor trades predict future stock returns over and above any associated liquidity effects. In unreported computations, we have found these results to be robust over the specification of the liquidity effect (aggregate institutional trading relative to mean daily trading volume or the number of shares outstanding) and persist independently of trade package specification (indicating forecasting ability past the end of the package). In terms of economic significance, the predictive power of institutional trading can be significant. According to Table 4, using the ‘‘Basic’’ specification with BHP, an increase in the number of managers purchasing of three over the average (which is two standard deviations) has a total effect on returns of 0.42% over the following 10 days (0.0014  3 = 0.0042). The effect for sales is not quite as strong, with the sum of the lag coefficient of manager selling being 0.0011 (although still statistically different from zero at the 1% level) as compared to 0.0014 for purchases. While these figures may seem small in magnitude, an annual alpha of 2% is approximately 0.08% over 10 trading days. 5.4. Contemporaneous effects and the forecasting power of institutional trading In Section 5.2 we noted that transactions from our sample of institutional investors have no contemporaneous effect on stock prices, whether through the information content of their trades or through a liquidity effect. There may be a number of reasons why we do not find a contemporaneous effect. One possibility is that we may be aggregating many different types of trades in the

3391

MgrBuy variable that have very different contemporaneous effects. That is, the investment style of the fund manager may cause them to trade in very different circumstances, leading to distinct contemporaneous effects.

5.4.1. Trade characteristics Perhaps the most obvious trade characteristic is trade size. Large trades relative to funds under management (FUM) should be indicative of high information content. Large trades are more likely to be information motivated since a trader with more valuable information can profit more from the information by making a larger trade. Easley and O’Hara (1987) suggest that larger trades have the capacity for greater market impact than small trades, and hence we expect the contemporaneous effect of large trades to exceed that of other trades. Small trades are more likely to be liquidity motivated, perhaps motivated by redemptions or applications. Edelen (1999) shows mutual funds engage in significant uninformed or liquidity-motivated trading. To account for trade size, we divide our sample of institutional trades into two groups: (large sized) the top quartile of trades ranked by standardized relative trade size, and (others) all other trades. We define trade size relative to both the market capitalization of the stock and the funds under management:

Tradesize ¼

price  quantity bmkweight  FUM

ð3Þ

where, bmkweight is the market capitalization of the stock as at the time of trade divided by the total value of the largest 300 stocks on the exchange. FUM is the total dollar value of holdings under management in the fund. The relative trade size is then standardized across all the trades made by our sample of institutions in each particular stock. We use the standardized relative trade size of each transaction to split the sample according to the two groups outlined above. We use this measure of relative trade size, rather than a measure relative to mean daily trading volume, because it scales according to manager size as well as share capitalization. We form two variables (and their lags) of institutional trading corresponding to the number of manager purchases and sales made on day t in each stock s and the number of large sized purchases and sales made on day t in each stock s. These variables are standardized according to the mean and standard deviation relevant for each stock in the sample. Another factor that may be contributing to a zero contemporaneous effect for manager trading is that managers may be masking their trades. One such tactic is to use multiple brokers to trade the same stock on the same day. The converse, multiple managers using the same broker in the same stock on the same day may indicate a broker providing price and time sensitive information to many managers at the same time.20 Hence we expect days where many managers use the same broker to have a positive contemporaneous relation between trading and returns. We regress the midpoint close-to-close return on the standardized number of institutions purchasing and selling in the large trade size category and the non-large trade size category as well as the number of managers trading multiplied by two sets of indicator variables: MultiBKR, set to one if a manager has used more than one broker to trade, and MultiMgr, set to one if the same broker has purchased or sold for many managers. We also include the risk control variables for market, stock size, book-to-market and momentum. Results from this analysis are presented in the column headed ‘‘TradeCharacteristics’’ in Table 4. 20 This has empirical support from Fong et al. (2011), who find multiple broker trades generate significantly higher returns over the subsequent twelve months.

Panel A – R-squared R-squared Adjusted R-squared

Panel B – Regression estimates MgrBuy(t) MgrBuy(t-1: t-5) MgrBuy(t-1: t-10) MgrSell(t) MgrSell(t-1: t-5) MgrSell(t-1: t-10) MgrBuyLRG(t) MgrBuyLRG(t-1: t-5) MgrBuyLRG(t-1: t-10) MgrSellLRG(t) MgrSellLRG(t-1: t-5) MgrSellLRG(t-1: t-10) MgrBuyMultiBroker(t) MgrBuyMultiBroker(t-1: t-5) MgrBuyMultiBroker(t-1: t-10) MgrSellMultiBroker(t) MgrSellMultiBroker(t-1: t-5) MgrSellMultiBroker(t-1: t-10) MgrBuyMultiMgr(t) MgrBuyMultiMgr(t-1: t-5) MgrBuyMultiMgr(t-1: t-10) MgrSellMultiMgr(t) MgrSellMultiMgr(t-1: t-5) MgrSellMultiMgr(t-1: t-10) SharesBuy(t) SharesBuy(t-1: t-5) SharesBuy(t-1: t-10) SharesSell(t) SharesSell(t-1: t-5) SharesSell(t-1: t-10) * ** ***

Statistical significance at the 10% levels. Statistical significance at the 5% levels. Statistical significance at the 1% levels.

Basic

0.1505 0.1364 TradeCharacteristics

Coefficient

t-stat

Coefficient

t-stat

0.0002 0.0011 0.0014 0.0001 0.0011 0.0011

1.13 4.60*** 4.89*** 1.04 4.30*** 3.73***

0.0022 0.0008 0.0031 0.0025 0.0036 0.0046

1.42 0.29 0.88 1.48 1.13 1.19

0.0003 0.0010 0.0013 0.0003 0.0013 0.0013 0.0002 0.0001 0.0002 0.0003 0.0003 0.0004 0.0008 0.0025 0.0017 0.0005 0.0004 0.0021 0.0014 0.0016 0.0002 0.0007 0.0002 0.0004 0.0027 0.0006 0.0035 0.0030 0.0034 0.0040

2.33*** 3.69*** 3.84*** 2.09*** 4.66*** 4.16*** 1.79* 0.56 0.45 2.44*** 1.21 1.11 1.28 1.97** 0.98 0.86 0.27 1.12 2.24*** 1.14 0.12 1.12 0.17 0.20 1.62 0.20 0.99 1.75* 1.03 0.99

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Variable

0.1455 0.1340

3392

Table 4 P P P Institutional trading, contemporaneous effects, forecasting, and trade characteristics. This table reports regression estimates of the following regression equation: Rs;t ¼ z¼10 b MgrBuytz þ z¼10 b MgrSelltz þ z¼10 bm;z SharesBuytz z¼1 P Pz¼10 Pz¼10 Pz¼10 Pz¼10 PS PS PS k;z PSz¼1 l;z Pz¼1   S þ z¼10 b SharesSell þ b LRGMgrBuy þ b LRGMgrSell þ b MultiBKR MgrBuy þ b MultiMGR MgrSell þ b d þ b d Market þ b d SIZE þ b d BMRatio þ tz s s t s t s t n;z o;z p;z q;z r;z a;s b;s c;s d;s tz tz tq tz z¼1 z¼1 z¼1 z¼1 z¼1 s¼1 s¼1 s¼1 s¼1 s¼1 be;s ds Momentumt þes;t The dependent variable Rs,t is the return stock s on day t. MgrBuy is the standardized number of managers purchasing stock s on day t of the mid-sized trade size. MgrSell is the standardized number of managers selling stock s on day t of the mid-sized trade size. LRGMgrBuy is the standardized number of managers purchasing stock s on day t of the large sized trade size. LRGMgrSell is the standardized number of managers selling stock s on day t of the large sized trade size. Trade size is defined by the dollar value of the trade divided by the weight of the stock in the ASX300, further divided by funds under management. Large sized trades are those larger than the 75th percentile of trade ranked by relative trade size. MultiBKR is an indicator variable set to unity when a manager on day t uses more than one broker to purchase or sell stock s. MultiMgr is an indicator variable set to unity when a broker on day t services more than one manager to purchase or sell stock s. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. R(t) is the return on stock s on day t. SharesBuy is the total number of shares bought by managers in stock s on day t divided by the mean share trading volume calculated over the prior 20 days. SharesSell is the total number of shares sold by managers in stock s on day t divided by the mean share trading volume calculated over the prior 20 days. Market is the return on the value weighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange) on day t. Size is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock. These results are for the sample period 2 January 2000 to 31 December 2001. Only trades in the largest 50 stocks (based on market capitalizations on the first day of our sample period) are included. We report the sum of the lagged coefficients as follows, (t-1: t-10) is the sum of the lags from t-1 to t-10.

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

We find that the contemporaneous return effect of the number of managers for large trades is significantly larger than that for other trades. The coefficient for large purchases is 0.0002, and is significant at the 10% level. This result lends support to the information hypothesis of price impact. A significant permanent effect is non-existent for large sized trades, however. The additional contemporaneous return effect (through the number of managers) of a large sale is larger in magnitude and more significant than that found for a large purchase. These results are consistent with Chan and Lakonishok (1993, 1995), who find market impact costs are positively correlated with trade size (relative to mean daily trading volume). Care must be taken in comparing our results, however, since our definition of trade size is relative to both funds under management, and stock capitalization. We expect that our measure is more related to information while that of Chan and Lakonishok (1993, 1995) is more related to liquidity (since their definition of trade size is relative to mean daily trading volume). There is evidence of an information masking effect through the use of multiple brokers. When managers use many brokers to purchase shares, we find that over the next five days autocorrelationadjusted stock returns rise by 0.25% (significant at the 5% level) per standard deviation of manager buying. This incremental information advantage, however, appears to be short-lived, dissipating to 0.17% (not significant) over ten days. Interestingly, this incremental information effect does not come at an increased contemporaneous cost. We find no evidence of a significant difference between the contemporaneous effects of manager buying (selling) on days when a manager has used multiple brokers. This suggests fund managers benefit from the higher information content of their purchases if they mask their trades using multiple brokers, without causing adverse changes to stock prices on the day they trade.21 When multiple managers use the same broker we find that there is a positive incremental contemporaneous effect (statistically significant at the 1% level) for purchases. This suggests that when brokers provide information to many of their institutional clients, prices adjust on the same day. We find no evidence of an incremental information effect over the following 5 or 10 days; perhaps information revealed by brokers is fully impounded in stock prices on the same day and may be about temporary liquidity needs, rather than long-term share valuations.

5.4.2. Investment style Another possible factor influencing the contemporaneous impact of institutional trading is investment manager style. Value managers, for example, aim to purchase (sell) stocks at a cheap (an expensive) price relative to fundamentals. Consequently, they behave as price stabilizers, purchasing or selling when prices deviate from fundamentals. Alternatively, momentum managers may purchase on strength to continue to boost share prices to match perceived valuations. To investigate the influence of investment style on the contemporaneous effects of institutional trading, we first examine the influence of past stock returns on institutional trading, partitioned by investment style. We do so by regressing the standardized number of purchases and sales made on day t in stock s by managers of the same style, on past stock returns, and lagged institutional trading according to investment style. We present separate results for purchases and sales in Table 5. 21 There may be other reasons for using multiple brokers; for example, a manager with a large time critical order may employ several brokers to ensure the order is filled quickly, however we control for the impact of large trades by including the trade size variable. Therefore, the coefficient of the broker variables should capture the incremental effect of using multiple brokers over and above any effect due to time critical large trades.

3393

The results in Table 5 show that investment style has a strong effect on the relation between past stock returns and institutional trading. Style-neutral purchases (sells) are significantly negatively (positively) related to past stock returns. We find growth managers are momentum traders, with past stock returns positively correlated (although not statistically significant) with purchases and negatively correlated with sells (significant at the 5% level). Value managers are strongly contrarian, with past stock returns negatively correlated with value manager purchases and positively correlated with value manager sells (both are statistically significant at the 1% level). All investment styles are highly positively serially correlated with trading activity of their own style, although not necessarily with the trading activity of other styles. For example, value manager purchases are negatively correlated with lagged growth manager purchases. Most institutional trading is uncorrelated with lagged values of aggregate shares purchased or sold by our managers. We investigate the influence of investment style on the contemporaneous effects and on the forecasting ability of institutional trading by regressing stock returns on the standardized number of style-neutral, growth, and value managers purchasing and selling on day t in stock s and their lagged values. We also include the risk control variables as discussed above. The results of the style regressions are reported in Table 6. We find that lagged values of the number of institutions purchasing are positively related to stock returns (there is a negative relation for sales), and are robust to value and growth styles. Consistent with prior studies, we find that value managers have superior forecasting ability relative to growth managers. The results also show that the contemporaneous effect of institutional trading depends on investment style. Style-neutral purchasing and selling has a statistically significant impact, but growth manager selling does not. The contemporaneous effect of value managers is much stronger than style-neutral or growth managers and is of the opposite sign. This result seems counterintuitive; according to both the liquidity and information hypotheses, we expect that as more managers purchase, stock returns would rise on that day. However, if value managers are stabilizing prices (e.g. selling when they perceive prices have risen above fundamental levels) the contemporaneous effect of their trades on stock prices will be negative. This follows the traditional notion of profitable stabilization as in Friedman (1953). During a supply shock (many investors wishing to sell for liquidity reasons), value managers may provide liquidity to the market, stabilize prices, and require a discount for the service they provide as counterparty purchasers. Put differently, value managers are likely to ‘buy on weakness and sell on strength’, which is consistent with the contemporaneous coefficients reported in Tables 4 (Trade Characteristics column) and 6. Value manager speculation (price stabilization) should be more common during times of relative uncertainty in the share market. To test this, we proxy for instability with lagged intra-day volatility and investigate the interaction between lagged volatility and value manager trading. We use lagged volatility because value managers cannot observe price instability directly, instead they must infer instability from historical data. In unreported results, we confirm that volatility is highly serially correlated, suggesting that information about lagged volatility is useful in determining current instability.22 If value managers profit from volatile prices, we would observe more aggressive buying (selling) on weakness (strength) during 22 Ideally we would use the volatility of prices immediately prior to value manager trades, however we do not know the exact time of day when each manager trades, so we use the one-day lag of volatility as a proxy.

3394 Table 5 Pz¼10 Institutional trading, past stock returns and investment style. This table reports regression estimates of the following regression equation: ys;t ¼ z¼1 bj Rs;tz þ bk NeutralBuyt1 þ bl NeutralBuyt1 þ bm GrowthBuyt1 þ P P P P P bn GrowthBuyt1 þ bo ValueBuyt1 þ bp ValueBuyt1 þ Ss¼1 ba;s ds þ Ss¼1 bb;s ds Market t þ Ss¼1 bc;s ds SIZEt þ Ss¼1 bd;s ds BMRatiot þ Ss¼1 be;s ds Momentumt þ es;t The dependent variable y is one of six variables: NeutralBuy, NeutralSell, GrowthBuy, GrowthSell, ValueBuy and ValueSell. NeutralBuy is the standardized number of neutral managers purchasing stock s on day t, although we leave out the subscripts. NeutralSell is the standardized number of neutral managers selling stock s on day t. GrowthBuy is the standardized number of growth managers purchasing stock s on day t. GrowthSell is the standardized number of growth managers selling stock s on day t. ValueBuy is the standardized number of value managers purchasing stock s on day t. ValueSell is the standardized number of value managers selling stock ‘s’ on day ‘t’. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. R(t) is the return on stock s on day t, SharesBuy is the total number of shares bought by managers in stock s on day t divided by the mean daily volume calculated over the prior 20 days. SharesSell is the total number of shares sold by managers in stock s on day t divided by the mean daily volume calculated over the prior 20 days. Market is the return on the value weighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange) on day t. Size is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock. These results are for the sample period 2 January 2000 to 31 December 2001. Only trades in the largest 50 stocks (based on market capitalizations on the first day of our sample period) are included. We report the sum of the lagged coefficients as follows, (t-1: t-10) is the sum of the lags from t-1 to t-10.

Variable

* ** ***

0.1286 0.1184 0.1062 0.0957 Neutral

0.0608 0.0498 0.0693 0.0584 Value

0.1313 0.1211 0.1488 0.1388 Growth

Coefficient

t-stat

Coefficient

t-stat

Coefficient

t-stat

Panel B – Regression estimates (BUYS) NeutralMgrBuy(t-1) NeutralMgrSell(t-1) ValueMgrBuy(t-1) ValueMgrSell(t-1) GrowthMgrBuy(t-1) GrowthMgrSell(t-1) Ret(t-1: t-5) Ret(t-1: t-10) SharesBuy(t-1) SharesSell(t-1)

0.5087 0.0066 0.0034 0.0169 0.0013 0.0070 0.0930 0.2437 0.3254 0.0256

37.25*** 1.02 0.52 2.55** 0.19 0.98 0.14 0.25 3.98*** 0.32

0.0026 0.0007 0.3914 0.0247 0.0061 0.0059 4.9891 6.4709 0.2711 0.0840

0.39 0.11 21.52*** 3.61*** 0.85 0.81 7.12*** 6.33*** 3.10*** 1.00

0.0039 0.0130 0.0006 0.0158 0.5620 0.0100 0.5412 0.8226 0.4646 0.1147

0.59 2.00** 0.10 2.37*** 39.07*** 1.44 0.81 0.85 5.50*** 1.43

Panel C – Regression estimates (SELLS) NeutralMgrBuy(t-1) NeutralMgrSell(t-1) ValueMgrBuy(t-1) ValueMgrSell(t-1) GrowthMgrBuy(t-1) GrowthMgrSell(t-1) Ret(t-1: t-5) Ret(t-1: t-10) SharesBuy(t-1) SharesSell(t-1)

0.0029 0.4292 0.0007 0.0006 0.0212 0.0189 1.7002 2.0555 0.1833 0.3843

0.44 29.35*** 0.11 0.08 3.03*** 2.65*** 2.58*** 2.14*** 2.24*** 4.70***

0.0122 0.0071 0.0228 0.4015 0.0281 0.0095 3.8728 5.0408 0.1425 0.0464

1.86** 1.09 3.43*** 24.17*** 4.00*** 1.33 5.70*** 5.10*** 1.67* 0.57

0.0175 0.0000 0.0112 0.0103 0.0116 0.5236 2.1673 3.6719 0.0324 0.7794

2.72*** 0.00 1.73* 1.59 1.71* 39.09*** 3.31*** 3.84*** 0.39 9.85***

Statistical significance at the 10% levels. Statistical significance at the 5% levels. Statistical significance at the 1% levels.

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Panel A – R-squared R-squared (BUYS) Adjusted R-squared (BUYS) R-squared (SELL) Adjusted R-squared (SELLS)

3395

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 6 Institutional trading, contemporaneous effects, forecasting, and investment style. This table reports regression estimates of the following regression equation: P P P P Pz¼10 Pz¼10 Pz¼10 bl;z NeutralSelltz þ z¼10 þ z¼10 R ¼ z¼10 b NeutralBuytz þ z¼10 tz þ z¼1P z¼1 bm;z SharesBuytz z¼1 bn;z SharesSellP z¼1 bo;z GrowthBuytz þ z¼1 bp;z GrowthSelltz þ z¼1 bq;z ValueBuytz þ Ps;tz¼10 z¼1 k;z PS P PS S S S b ValueSell þ b :d þ b :d Market þ b :d :SIZE þ b :d :BMRatio þ b :d :Momentum þ e The dependent variable R is the return stock s tz s s t s t s t s t s;t s,t r;z a;s b;s c;s d;s e;s z¼1 s¼1 s¼1 s¼1 s¼1 s¼1 on day t. NeutralBuy and NeutralSell is the standardized number of style-neutral managers purchasing and selling stock s on day t respectively. GrowthBuy and GrowthSell is the standardized number of growth managers purchasing and selling stock s on day t respectively. ValueBuy and ValueSell is the standardized number of value managers purchasing and selling stock s on day t respectively. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. R(t) is the return on stock s on day t. SharesBuy is the total number of shares bought by managers in stock s on day t divided by the mean share trading volume calculated over the prior 20 days. SharesSell is the total number of shares sold by managers in stock ‘s’ on day ‘t’ divided by the mean share trading volume calculated over the prior 20 days. Market is the return on the value weighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange) on day t. Size is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock. These results are for the sample period 2 January 2000 to 31 December 2001. Only trades in the largest 50 stocks (based on market capitalizations on the first day of our sample period) are included. We report the sum of the lagged coefficients as follows, (t-1: t-10) is the sum of the lags from t-1 to t-10. Panel A – R-squared R-squared Adjusted R-squared

0.1555 0.1424

Variable

Basic

Panel B – Regression estimates NeutralMgrBuy(t) NeutralMgrBuy(t-1: t-5) NeutralMgrBuy(t-1: t-10) NeutralMgrSell(t) NeutralMgrSell(t-1: t-5) NeutralMgrSell(t-1: t-10) ValueMgrBuy(t) ValueMgrBuy(t-1: t-5) ValueMgrBuy(t-1: t-10) ValueMgrSell(t) ValueMgrSell(t-1: t-5) ValueMgrSell(t-1: t-10) GrowthMgrBuy(t) GrowthMgrBuy(t-1: t-5) GrowthMgrBuy(t-1: t-10) GrowthMgrSell(t) GrowthMgrSell(t-1: t-5) GrowthMgrSell(t-1: t-10) SharesBuy(t) SharesBuy(t-1: t-5) SharesBuy(t-1: t-10) SharesSell(t) SharesSell(t-1: t-5) SharesSell(t-1: t-10) *  ***

Coefficient

t-stat

0.0007 0.0003 0.0005 0.0005 0.0001 0.0001 0.0009 0.0010 0.0014 0.0010 0.0011 0.0010 0.0000 0.0004 0.0004 0.0001 0.0009 0.0009 0.0017 0.0005 0.0024 0.0021 0.0056 0.0064

5.54*** 1.16 1.90 4.19*** 0.69 0.28 8.02*** 4.54*** 5.27*** 8.86*** 4.98*** 3.88*** 0.17 1.83* 1.42 0.96 3.83*** 3.39*** 1.06 0.19 0.69 1.30 1.79* 1.70*

Statistical significance at the 10% levels. Statistical significance at the 5% levels. Statistical significance at the 1% levels.

periods of high volatility. Eq. (4) is a panel regression that allows us to measure the effects of manager style, prior volatility, and other risk measures on current return. Of particular interest are the coefficients of the interaction terms between lagged volatility and value buying and selling.

ys;t ¼

z¼10 X

bk;z NonValueBuytz þ

z¼1

þ

z¼10 X

z¼10 X

bm;z SharesBuytz þ

zX ¼10

z¼10 X

bn;z SharesSelltz

z¼1

bo;z ValueBuytz þ

z¼1

þ

bl;z NonValueSelltz

z¼1

z¼1

þ

z¼10 X

z¼10 X

bp;z ValueSelltz

z¼1

bq :Volatilitytz1

z¼0

þ

zX ¼10



br;z Volatilitytz1 ValueBuytz

z¼1

þ

z¼10 X z¼1

bu;z Volatilitytz1  ValueSelltz þ es;t

ð4Þ

Intra-day volatility is defined as the standard deviation of the tradeto-trade returns within each day. We use standardized intra-day volatility to ensure our results do not reflect cross-sectional variation in volatility. The results for the value manager regressions are reported in the columns of Table 7 headed ‘Value’. We also compute results using growth-oriented managers to determine if there is any symmetry between value- and growth-oriented managers (these results are reported in the columns headed ‘Growth’). The variable Style in the column of variable names refers to either value or growth corresponding to the ‘Value’ and ‘Growth’ column headings. The results show that the value manager contemporaneous effect of purchasing is less negative during periods of low intra-day volatility, as evidenced by the statistically significant negative coefficient of StyleMgrBUYLagVolatility and positive coefficient of StyleMgrSELLLagVolatility. Thus, when prices are relatively stable (when LagVolatility is small), value manager trading is less likely to be due to price stabilizing behavior and therefore yields a positive (or less negative) contemporaneous effect. Conversely, when prices are relatively unstable (when LagVolatiltiy is high), a portion of value manager trading is related to their price stabilizing

3396

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 7 P bk;z Institutional trading, investment style, and intra-day volatility. This table reports regression estimates of the following regression equation: Rs;t ¼ z¼10 Pz¼10 Pz¼10 Pz¼10 Pz¼10 Pz¼10 Pz¼10 Pz¼1 z¼10 NonStyleBuytz þ bm;z SharesBuytz þ bn;z SharesSelltz þ bo;z StyleBuytz þ bp;z StyleSelltz þ bq :Volatilitytz1 þ tz þ z¼1 bl;z NonStyleSell z¼1 z¼1 z¼1 z¼1 z¼0 z¼1 br;z Pz¼10  Volatilitytz1  StyleBuytz þ z¼1 bu;z Volatilitytz1 StyleSelltz þ es;t The dependent variable Rs,t is the return stock s on day t. NonStyleBuy is the standardized number of nonvalue or non-growth managers purchasing stock s on day t. NonStyleSell is the standardized number of non-value or non-growth managers selling stock s on day t. StyleBuy is the standardized number of value or growth managers purchasing stock s on day t. StyleSell is the standardized number of value or growth managers selling stock s on day t. We standardize by subtracting the mean and dividing by the standard deviation of the institutional trading variable particular to each stock over the sample period. R(t) is the return on stock s on day t. SharesBuy is the total number of shares bought by managers in stock s on day t divided by the mean share trading volume calculated over the prior 20 days. SharesSell is the total number of shares sold by managers in stock s on day t divided by the mean share trading volume calculated over the prior 20 days. Market is the return on the value weighted portfolio of all stocks listed on the ASX300 (the largest 300 stocks on the exchange) on day t. Size is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by market capitalization) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. BMRatio is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by book to market ratio) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. Momentum is the value weighted return on a portfolio of stocks comprised of the largest quintile of stocks (as ranked by stock return calculated over the last 130 days) in the ASX300 less the value weighted return on a portfolio of stocks comprised of the smallest quintile of stocks in the ASX300. d is an indicator variable for each stock. These results are for the sample period 2 January 2000 to 31 December 2001. Only trades in the largest 50 stocks (based on market capitalizations on the first day of our sample period) are included. We report the sum of the lagged coefficients as follows, (t1: t-10) is the sum of the lags from t-1 to t-10. Volatility is defined in Eq. (4). Panel A – R-squared R-squared Adjusted R-squared Variable

Panel B – Regression estimates NonStyleMgrBuy(t) NonStyleMgrBuy(t-1: t-5) NonStyleMgrBuy(t-1: t-10) NonStyleMgrSell(t) NonStyleMgrSell(t-1: t-5) NonStyleMgrSell(t-1: t-10) SharesBuy(t) SharesBuy(t-1: t-5) SharesBuy(t-1: t-10) SharesSell(t) SharesSell(t-1: t-5) SharesSell(t-1: t-10) StyleMgrBuy(t) StyleMgrBuy(t-1: t-5) StyleMgrBuy(t-1: t-10) StyleMgrSell(t) StyleMgrSell(t-1: t-5) StyleMgrSell(t-1: t-10) StyleMgrBuyLagVolatility(t) StyleMgrBuyLagVolatility(t-1: t-5) StyleMgrBuyLagVolatility(t-1: t-10) StyleMgrSellLagVolatility(t) StyleMgrSellLagVolatility(t-1: t-5) StyleMgrSellLagVolatility(t-1: t-10) LagVolatility(t) LagVolatility(t-1: t-5) LagVolatility(t-1: t-10) * ** ***

0.1573 0.1438

0.1499 0.1363

Value

Growth

Coefficient

t-stat

Coefficient

t-stat

0.0005 0.0004 0.0006 0.0005 0.0005 0.0005 0.0020 0.0004 0.0028 0.0026 0.0039 0.0047 0.0009 0.0010 0.0013 0.0010 0.0010 0.0010 0.0003 0.0001 0.0005 0.0003 0.0005 0.0009 0.0000 0.0003 0.0003

3.82*** 1.95** 2.36*** 4.02*** 2.36*** 1.80* 1.26 0.16 0.82 1.59 1.26 1.25 7.83*** 4.49*** 4.99*** 8.55*** 4.89*** 3.71*** 2.95*** 0.27 1.98** 2.95*** 2.60*** 3.39*** 0.26 1.14 1.62

0.0001 0.0009 0.0013 0.0002 0.0008 0.0008 0.0025 0.0009 0.0032 0.0035 0.0041 0.0049 0.0000 0.0005 0.0005 0.0002 0.0008 0.0008 0.0001 0.0002 0.0002 0.0001 0.0004 0.0004 0.0001 0.0002 0.0004

0.63 4.11*** 4.83*** 1.94* 3.34*** 2.74*** 1.59 0.30 0.92 2.17** 1.31 1.29 0.08 2.14** 1.81* 1.62 3.49*** 3.01*** 1.02 0.78 0.56 0.99 1.82 1.61 0.62 1.05 1.82*

Statistical significance at the 10% levels. Statistical significance at the 5% levels. Statistical significance at the 1% levels.

behavior and hence they are able to transact with a negative relation to contemporaneous returns.

5.4.3. Market impact, investment style and prior trading If some value manager trades are price stabilizing, then we would expect the market impact of value manager transactions to be negatively related to return volatility. Further, such price stabilizing behavior would imply that value managers do not compete for liquidity with non-value managers (i.e. value managers would provide liquidity). We explore these additional implications by considering stock returns during the day for both purchase and sale transactions with respect to differences in the investment styles of fund managers. In particular, we relate intraday returns to the prior day’s activities; namely prior day return volatility and trading from investors of different styles.

To measure the market impact of a fund manager’s trade we consider the open-to-trade return each day.23 If there is a rise in the share price between the market open and when a manager bought stock, we interpret this as a positive market impact, as it suggests that the manager used liquidity from the market to facilitate their trade. If there is a fall in the share price between the market open and when a manager bought shares we view this as a negative market impact, as it suggests that the manager provided liquidity to the market to facilitate the trades of others (they bought on weakness). Similar interpretations are used for sell transactions for managers.

23 Open-to-trade market impact is defined as the value weighted trade price obtained during the day, divided by the midpoint of the opening bid and ask, minus one.

3397

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

Table 8 Stock returns around institutional trading. This table reports the results of regressing the open-to-trade return, defined as the trade weighted transaction price, divided by the midpoint of the open bid and ask, minus one, on (i) one day lagged abnormal intra-day volatility and (ii) one day lagged abnormal net growth, value and neutral manager purchasing, as well as control variables for forecast trade size (FRTS), and dummy variables for manager identification. We also include the trade-to-close regression, where the trade-to-close is defined as the midpoint of the closing bid and ask divided by the trade weighted price, minus one. We omit stock size control variables since our sample is already limited to the largest 50 stocks on the exchange. Abnormal intra-day volatility is measured as intra-day volatility divided by mean intra-day volatility measured from t-60 to t-20. Abnormal net manager trading is measured by the number of managers of a particular style purchasing, less the number selling, divided by the mean calculated from t-60 to t-20. We use lagged values of abnormal volatility and manager trading to avoid look-ahead bias. Open-to-trade

Trade-to-close

Growth

Value

Neutral

Growth

Value

Neutral

Panel A – Purchases Coefficient FRTS (%) VOL (%) Growth (%) Value (%) Neutral (%)

0.1916 0.0842 0.0102 0.0007 0.0169

0.0065 0.5479 0.0072 0.0034 0.0035

0.9088 0.2393 0.0032 0.0047 0.0003

0.0664 0.1256 0.0028 0.0046 0.0000

0.0230 0.0811 0.0007 0.0031 0.0026

0.3216 0.0888 0.0027 0.0015 0.0016

t-stat FRTS VOL Growth Value Neutral

0.91 1.51 3.11 0.20 3.11

0.02 8.64 1.58 0.68 0.48

3.50 4.85 1.09 1.26 0.07

0.64 4.60 1.71 2.63 0.00

0.15 2.69 0.32 1.30 0.75

2.54 3.69 1.89 0.84 0.87

R-squared R-squared (%) Adjusted R-squared (%)

4.0646 3.8737

5.9762 5.7240

2.0245 1.7676

1.6200 1.4243

1.5297 1.2656

2.3238 2.0676

Panel B – Sells Coefficient FRTS (%) VOL (%) Growth (%) Value (%) Neutral (%)

0.0653 0.1645 0.0017 0.0023 0.0172

0.5326 0.2802 0.0137 0.0170 0.0116

0.8024 0.0675 0.0019 0.0034 0.0048

0.4773 0.0315 0.0012 0.0025 0.0058

0.5484 0.0159 0.0012 0.0037 0.0058

0.5531 0.0597 0.0027 0.0082 0.0008

t-stat FRTS VOL Growth Value Neutral

0.38 3.13 0.43 0.60 2.63

1.75 3.89 3.83 4.23 1.99

2.83 1.20 0.60 0.92 1.04

5.23 1.11 0.59 1.21 1.64

3.68 0.45 0.66 1.88 2.03

3.68 2.01 1.66 4.10 0.31

R-squared R-squared (%) Adjusted R-squared (%)

3.2279 2.9846

7.9546 7.6167

We regress open-to-trade returns on one-day lagged abnormal intra-day volatility, one-day lagged abnormal net purchases from growth, value and neutral managers,24 control variables for forecast trade size (FRTS),25 and manager fixed effect dummy variables.26 Abnormal intra-day volatility is measured as intra-day volatility divided by mean intra-day volatility measured from days t-60 to t-20. Abnormal net manager trading is measured by the number of managers of a particular style of purchasing, less the number selling, divided by the mean calculated from days t-60 to t-20. The results of these regressions are presented in Table 8. The market impact incurred by value managers for both purchase and sell transactions are strongly and negatively correlated with lagged intra-day volatility. For example, the estimate for value manager purchases implies that for a 10% increase in lagged abnormal intra-day volatility, the expected open-to-trade return for a purchase falls by 5.4 basis points. However, this result does not extend to managers of other styles. While the estimated market 24 We proxy competition for liquidity as lagged abnormal net purchasing, since manager trading is highly persistent with a high autocorrelation coefficient. This suggests that if net activity is high, then the expected level of purchasing over the subsequent period will also be high. 25 This is measured by dollar trade value divided by the mean daily trading value (calculated over the last 20 trading days). 26 We omit stock size control variables as our sample is already limited to the largest 50 stocks on the exchange.

1.9131 1.6409

2.2397 1.9939

3.9949 3.6425

1.9187 1.6465

impact incurred by neutral manager purchases is negatively correlated with lagged abnormal volatility, the magnitude is only half that of value managers (and the result for neutral manager sells is of the opposite sign and is not statistically significant). Further, the open-to-trade market impact incurred by growth manager purchases is not statistically significantly related to lagged abnormal intra-day volatility. These results are consistent with Table 7 and support our hypothesis that value managers are behaving as price stabilizers since they show that value managers are able to attain low or even negative market impact during periods of price instability. For sells, growth managers appear to be demanding liquidity (share prices are falling as they sell). These results suggest that growth managers are less likely to stabilize prices and are using liquidity, whereas value managers are most likely to provide liquidity and stabilize prices. The trade-to-close returns show that managers of all styles benefit from price instability while purchasing (the results for sell transactions are statistically significant only for neutral managers). For example, the growth manager VOL coefficient of 0.1256% indicates that for a 10% increase in volatility, the post trade stock return for the average growth manager increases by 1.2 basis points. Finally, we consider the extent to which managers of different styles might compete for liquidity. To document this interaction we consider the influence of prior day net demand for shares from

3398

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399

each investment style on the market impact of a manager’s trade. The regression results from Table 8 indicate that growth manager open-to-trade market impact for purchases is significantly and positively related to growth and neutral manager purchases the prior day, suggesting some competition for liquidity (although the results for growth manager sell transactions are not as significant). For example, in Panel A, the coefficient of 0.0102% for growth manager purchases indicates that for a 1 standard deviation increase in lagged growth manager demand, market impact increases by 1 basis point. If both net neutral and net growth manager purchases increase by 10%, we estimate the market impact of growth manager purchases the next day will increase by about 2.7 basis points. For value managers we find that their purchases are unrelated to the prior day’s trade by managers of all investment styles. For value manager sells there is a significant and positive relation between the open to trade return and prior day net purchases by managers of all styles. This is consistent with value managers providing liquidity when other managers demand it, and earning a small spread for their efforts.

6. Conclusions Using a unique database of daily fund manager transactions, we investigate the relation between institutional trading and stock returns for the largest 50 stocks listed on the Australian Securities Exchange. While our sample is limited in some respects (time period, shares considered, and trades of a portion of institutional managers) we are nevertheless able to gain deeper insights into drivers and consequences of institutional trading than in prior studies. For example, while our sample is not exhaustive the securities included represent about 82% of total market capitalization. Institutions in our sample are, on average, contrarian traders over short-term horizons of about ten days. When we partition according to investment style, we find that growth-oriented managers are momentum traders while style-neutral and value managers are contrarian. This is consistent with the manager’s self-reported investment styles; that is value managers appear to buy on weakness and sell on strength. We also find that institutional trading is highly persistent for managers with the same investment style. Value manager purchasing, for example, is strongly positively correlated with lagged value manager purchasing, but is weakly negatively correlated with growth-oriented manager purchasing. For all investment styles, the auto-correlation with lagged value of aggregate purchase or sell volume is of less importance than the lagged number of institutions within their own investment style purchasing or selling. From the pattern of autocorrelation within investment styles, we conclude that institutional traders may either engage in information-based herding, or may receive serially correlated signals. In either case, the evidence suggests that such behavior is rational, since we find no evidence of price reversals over the ten days following their trades. Much of the literature finds a strong correlation between changes in institutional holdings (or inferred trades) and stock returns in the same period. Many of these studies, however, use monthly, quarterly or even annual data. With daily data we find that institutional purchasing and selling is not correlated with contemporaneous stock returns. There are several factors related to this result. Trade characteristics such as size and broker use can affect the contemporaneous relation between returns and trading. Investment manager style also has a strong effect on this relation. Lagged values of the number of institutions purchasing or selling are correlated with stock returns. This suggests that institutional investors are able to predict returns. This finding is robust over trade size, broker use, and investment style. In general we

note that a deeper understanding of the details of investment manager actions and strategies provides a more nuanced view of the relation between institutional trading and share returns. Acknowledgements The authors gratefully acknowledge the assistance and support of the Securities Industry Research Centre of Asia-Pacific (SIRCA) for provision of the ASX SEATS data, and the investment managers who provided daily trading information used to construct the Portfolio Analytics Database. The identities of the participating investment managers remain strictly confidential. We are indebted to seminar participants at the Asian FMA Annual Meetings, The Australian National University, Australasian Finance and Banking Conference, China International Conference in Finance, Chinese University of Hong Kong, EFA Annual Meetings, FMA (Europe) Annual Meetings, Fudan University, Monash University, Hong Kong Polytechnic University, Journal of Banking and Finance Conference, Nanyang Technological University, National University of Singapore, The University of New South Wales, Singapore Management University, The University of Sydney and the UNSW Investment Management Conference for their comments. Detailed comments from Sohnke Bartram, Tom George, Jennifer Gippel, Sara Hartley, Petko Kalev, Ameziane Lasfer, Tom Smith, Terry Walter, Geoff Warren, and Robert Whaley improved the paper substantially. Part of this research was completed when Foster and Gallagher were at UNSW, while Foster was a visitor at HKUST, and while Gallagher was at University of Texas at Austin and UTS. References Back, K., Cao, H., Willard, G., 2000. Imperfect competition among informed traders. The Journal of Finance 55, 2117–2155. Badrinath, S., Wahal, S., 2002. Momentum trading by institutions. The Journal of Finance 57, 2449–2478. Bozcuk, A., Lasfer, M., 2005. The information content of institutional traders on the London stock exchange. Journal of Financial and Quantitative Analysis 40, 621– 644. Cai, F., Zheng, L., 2004. Institutional trading and stock returns. Finance Research Letters 1, 178–189. Chan, L., Jegadeesh, N., Lakonishok, J., 1996. Momentum strategies. The Journal of Finance 51, 1681–1713. Chan, L., Lakonishok, J., 1993. Institutional trades and intraday stock price behaviour. Journal of Financial Economics 33, 173–199. Chan, L., Lakonishok, J., 1995. The behavior of stock prices around institutional trades. The Journal of Finance 50, 1147–1174. Chiyachantana, C., Jain, P., Jiang, C., Wood, R., 2004. International evidence on institutional trading behavior and price impact. The Journal of Finance 59, 869– 898. Chuang, W.I., Susmel, R., 2011. Who is the more overconfident trader? Individual vs. institutional investors. Journal of Banking and Finance 35, 1626–1644. Cohen, R., Gompers, P., Vuolteenaho, T., 2002. Who underreacts to cash-flow news? Evidence from trading between individuals and institutions. Journal of Financial Economics 66, 409–462. Daniel, K., Grinblatt, M., Titman, S., Wermers, R., 1997. Measuring mutual fund performance with characteristic-based benchmarks. The Journal of Finance 52, 1035–1058. DeBondt, W., Thaler, R., 1985. Does the stock market overreact? The Journal of Finance 40, 793–805. DeBondt, W., Thaler, R., 1987. Further evidence on investor overreaction and stock market seasonality. The Journal of Finance 42, 557–581. Dorn, D., Huberman, G., Sengmueller, P., 2008. Correlated trading and returns. The Journal of Finance 63, 885–920. De Long, J.B., Shleifer, A., Summers, L., Waldmann, R., 1990. Positive feedback investment strategies and destabilizing rational speculation. The Journal of Finance 45, 379–395. Easley, D., O’Hara, M., 1987. Price, trade size, and information in securities markets. Journal of Financial Economics 19, 69–90. Edelen, R., 1999. Investor flows and the assessed performance of open-end mutual funds. Journal of Financial Economics 53, 439–466. Fong, K., Gallagher, D., Gardner, P., Swan, P., 2011. Follow the leader: fund managers trading in signal-strength sequence. Accounting and Finance. doi:10.1111/ j.1467-629x.2010.00367.x. Foster, F.D., Viswanathan, S., 1996. Strategic trading when agents forecast the forecasts of others. The Journal of Finance 51, 1437–1478. Friedman, M., 1953. The case for flexible exchange rates. In: Friedman, M. (Ed.), Essays in Positive Economics. The University of Chicago Press, Chicago.

F.D. Foster et al. / Journal of Banking & Finance 35 (2011) 3383–3399 Gallagher, D., 2003. Investment manager characteristics, strategy, top management changes and fund performance. Accounting and Finance 43, 283–309. Gallagher, D., Looi, A., Pinnuck, M., 2010. Are active fund managers collectors of private information or fast interpreters of public information? Accounting and Finance 50, 635–662. George, T., Hwang, C., 2004. The 52-week high and momentum investing. The Journal of Finance 59, 2145–2176. Gompers, P., Metrick, A., 2001. Institutional investors and equity prices. Quarterly Journal of Economics 116, 229–259. Greene, W., 2008. Econometric Analysis. Pearson-Prentice Hall, New Jersey. Grinblatt, M., Titman, S., Wermers, R., 1995. Momentum investment strategies, portfolio performance, and herding: a study of mutual fund behavior. American Economic Review 85, 1088–1105. Grossman, S., Miller, M., 1988. Liquidity and market structure. The Journal of Finance 43, 617–633. Hatanaka, M., 1974. An efficient estimator for the dynamic adjustment model with autocorrelated errors. Journal of Econometrics 2, 199–220. Holthausen, R., Leftwich, R., Mayers, D., 1990. Large-block transactions, the speed of response, and temporary and permanent stock-price effects. Journal of Financial Economics 26, 71–95. Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: implications for stock market efficiency. The Journal of Finance 48, 65–91. Jegadeesh, N., Titman, S., 2001. Profitability of momentum strategies: an evaluation of alternative explanations. The Journal of Finance 56, 699–720. Keim, D., Madhavan, A., 1997. Transaction costs and investment style: an interexchange analysis of institutional equity trades. Journal of Financial Economics 46, 265–292.

3399

Keswani, A., Stolin, D., 2008. Which money is smart? Mutual fund buys and sells of individual and institutional investors. The Journal of Finance 63, 85–118. Lakonishok, J., Shleifer, A., Vishny, R., 1992. The impact of institutional trading on stock prices. Journal of Financial Economics 32, 23–43. Lo, A., MacKinlay, C., 1988. Stock market prices do not follow random walks: evidence from a simple specification test. The Review of Financial Studies 1, 41– 66. Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49, 1417–1426. Nofsinger, J., Sias, R., 1999. Herding and feedback trading by institutional and individual investors. The Journal of Finance 54, 2263–2295. Petersen, M., 2009. Estimating standard errors in financial panel data sets: comparing approaches. The Review of Financial Studies 22, 435–480. Sias, R., 2004. Institutional herding. The Review of Financial Studies 17, 165–206. Sias, R., Starks, L., Titman, S., 2006. Changes in institutional ownership and stock returns: assessment and methodology. The Journal of Business 79, 2869–2910. Stoll, H., 1978. The supply of dealer services in securities markets. The Journal of Finance 33, 1133–1151. Venezia, I., Nashikka, A., Shapira, Z., 2011. Firm specific and macro herding by professional and amateur investors and their effects on market volatility. Journal of Banking and Finance 35, 1599–1609. Wermers, R., 1999. Mutual fund herding and the impact on stock prices. The Journal of Finance 54, 581–622. Yan, X., Zhang, Z., 2009. Institutional investors and equity returns: are short-term institutions better informed? The Review of Financial Studies 22, 893–924.