Do the Large Trades of an Institutional Investor Reflect Private Information?

This study tests whether Fidelity Management and Research Company possesses private information when it makes large trades. We use a conditional performance measure inspired by Admati and Ross (1985) and by Grinblatt and Titman (1993) that does not require a benchmark used in risk-adjusted measures. It conditions the performance of each trade on public information, measured by the market price of the stock at the time of the trade. Results show that Fidelity’s trades exploit public as well as private information. Furthermore, we show that Fidelity’s initial stock position, trade execution time, and stock idiosyncratic volatility help to explain it trades. Trading Strategies page 1 Do the Large Trades of an Institutional Investor Reflect Private Information? This study tests whether Fidelity Management and Research Company (hereafter FMR), a large institutional investor, possesses private information when it makes large trades. We employ a performance measure that is based on the model of Admati and Ross (1985) (hereafter AR) and Grinblatt and Titman (1993), which, unlike traditional riskadjusted CAPM (Capital Asset Pricing Model) measures, explicitly allows for private information. The measure significantly adds to our ability to evaluate whether private information is reflected in trades because it conditions performance on publicly available information reflected in stock prices. Whenever an institution buys, or has recently owned, ten percent or more of a company’s shares, it must file a Schedule 13G to disclose significant trades in those shares at the Securities and Exchange Commission’s Edgar (Electronic Data Gathering, Analysis and Retrieval) database. We use FMR’s 13G filings because it makes five times as many large trades as the next largest institution. In addition, because the potential payoff on its information gathering cost is greatest for large trades, FMR could be better informed about them. Indeed, regulators’ rationale for requiring disclosure is that ten percent owners can receive non-public information directly from firm managers or indirectly from stock brokers. Earlier studies found general evidence that large institutional trades are sometimes informed. Badrinath, Kale and Noe (1995), Sias and Starks (1997) and Chakravarty (2001) show that aggregated institutional trades impound valuable information into stock prices, and Barclay and Warner (1993) and Chakravarty (2001) show that relatively large Trading Strategies page 2 trades (often broken into several medium-size transactions) move prices more than small ones. But these studies focused more on contemporaneous effects of large intra-day trades. We study trades completed over one to three months and measure performance over three to twelve months subsequent to the trades. The underlying question of AR’s model and our empirical work is: How can we decide to what extent an investor trades on private information? The approach of earlier performance studies employing the CAPM was to judge an institution’s overall performance relative to a market benchmark. But the results of these studies differ partly because they use different benchmarks. One solution to this problem, proposed by Grinblatt and Titman (1989, 1993) – and extended in Chen, Jegadeesh and Wermers (2000) – is a portfolio change measure that does not require benchmarks. Subsequently, Ferson and Khang (2002) show that this measure needs to be conditioned on public information, because under certain circumstances, managers can use public information to boost performance. Our performance measure is similar to Ferson and Khang’s portfolio change measure but we use AR’s model to support different conditioning variables that help to better identify privately informed trades. The fundamental idea behind their measure and ours is that, although we cannot observe FMR’s information set, we can observe the relative sizes of its trades. AR show that if an investor has private information, then his shareholding size and subsequent payoff will be positively related, after controlling for stock price. This should also be true for changes in shareholdings, that is, trade size. Our paper distinguishes between FMR’s use of private and public information by conditioning the trades on trade-related variables. Following AR, we condition on Trading Strategies page 3 contemporaneous stock price, which should include all publicly available information. We also condition on the total value of the stock holding at the time of the start of a trade. The holding size measures FMR’s potential payoff, and thus its incentive to gather private information. Furthermore, we control for idiosyncratic stock return volatility. As shown by AR, the performance of privately informed trades can appear poor on a riskadjusted basis to the uninformed observer. The reason is that a well-informed investor will tend to trade in more volatile stocks where she can better exploit her information advantage. Finally, we condition our performance measure on trade execution time. As discussed in more detail below, the amount of time FMR takes to complete its trades could indicate whether its information is shortor long-lived. Overall, we find that FMR’s trades incorporate both public information (through stock price) and private information. Trade size is negatively related to contemporaneous stock price and positively related to future payoff. As predicted by AR, FMR makes larger trades in more volatile stocks. Furthermore, our other conditioning variables besides stock price help to explain FMR’s trading. For example, the larger FMR’s stockholding at the start of a trade, the poorer its performance. This could reflect the fact that its large holdings are public knowledge (because of Edgar filings), making large trades more difficult to execute profitably. Even though FMR’s trades incorporate private information, only 56 percent of these trades earn positive returns even though market returns are consistently positive during our sample period (1994-1998). Of course, 40 percent of trades are sells, but this doesn’t account for the seemingly low percent of positive returns. To resolve this anomaly, we estimate the binomial probability that FMR’s large trades will earn positive Trading Strategies page 4 returns. We find that when FMR’s holding is large and the stock is volatile, then the probability that FMR will earn a positive return is greater. That is, FMR tends to have its largest holdings in the volatile stocks that subsequently earn positive returns. Furthermore, FMR also makes its larger trades in relatively volatile stocks. The rest of the paper is organized as follows. Section I describes the performance measure and empirical model, and compares these to the models of Grinblatt and Titman (1993) and Ferson and Khang (2002). Section II provides some background on the disclosure rules for large trades and their implications for trading. Section III describes the data and sample. Test results are reported in Section IV and Section V concludes the paper. I. Performance Measures Our performance measure builds on Grinblatt and Titman (1993) who measure performance as the covariance between lagged portfolio weight changes and returns, and Ferson and Khang (2002), who then condition this measure on public information. Like Ferson and Khang, our approach allows tests of two non-nested models; 1) all information is public, implying that lagged portfolio weight changes are uncorrelated with returns; and 2) some significant amount of information is private, implying that lagged portfolio weight changes should be positively related to returns. A. Earlier Performance Models and Our Empirical Model Grinblatt and Titman (1993) measure performance with the covariance between lagged portfolio weight changes and returns. Their Portfolio Change Measure (PCM) is: Trading Strategies page 5 (1) , , , 1 0 ( j N t T j t j t j t k j t PCM R w w T = = − = = = − ∑ ∑ )] / Here, jt R is the return of stock j over time period t and is the portfolio weight for that stock at the beginning of time t, and jt w jt k w − is the portfolio weight for that stock, at time tk. Performance is measured over a period of length T for N stocks. There are many other candidates for the second weight, for example, the weight of the asset in some benchmark portfolio like the Standard and Poor’s 500 (S&P 500). In that case, the number in parentheses is the deviation of the investor’s portfolio weight from the index weight. Performance (the covariance) is positive when the institution places more (less) weight on S&P 500 companies that subsequently generate relatively large (small) returns. The PCM simply uses the institution’s lagged weight for comparison instead of the weight of some fixed index. If the change in weight from the previous period is positive (negative), it implies that the institution’s information signal for stock j is positive (negative). To decide if the signal is valuable (the institution is informed), one tests to see if the weight change and the subsequent stock return are significantly positively correlated. Under the null hypothesis of no information, the correlation is zero. Using the lagged weight avoids some problems associated with risk-adjusted measures, for example, the need to determine an efficient benchmark. The asset weights of an efficient benchmark represent expected return implied by publicly available information. They represent the naïve strategy of simply holding a portfolio whose weights are market determined. The PCM tests the results of an active trading strategy of changing portfolio weights, against the results of a passive strategy of no trades. Using a benchmark for comparison sets an improper conditional standard because we do not know if the benchmark precisely reflects public information (is efficient). Trading Strategies page 6 More important, it will not properly reflect conditional risk. AR point out that an informed trader might appear to do very poorly after benchmark risk adjustment: “although better information implies higher expected returns (corollary 2), it also leads to a larger variance (unconditional and also given only prices) (p. 15).” Intuitively, a trader with information on several stocks, all with the same unconditional expected return, will prefer the stock with the higher variance. The high-variance stock offers him more opportunity to exploit his information, and hence, offers him better conditional expected return. Grinblatt and Titman (1989 and 1993) and Ferson and Khang (2002) point out certain weaknesses in the PCM. They discuss several situations where a manager could use public information to forecast future returns and earn a positive PCM. For example, return serial correlation could be exploited even though it is public knowledge. In this case, the PCM can be biased because, , which is part of ( where = shares), could be correlated with jt P jt w /[ ], jt jt jt it it i w S P S P = ∑