Journal of Banking & Finance 34 (2010) 12251236

Contents lists available at ScienceDirect

Journal of Banking & Finance

journal homepage: www.elsevier.com/locate/jbf

Aggregate insider trading: Contrarian beliefs or superior information?

Xiaoquan Jiang a, Mir A. Zaman b,*
a b

Department of Finance, College of Business Administration, Florida International University, Miami, FL 33199, United States Department of Finance, College of Business Administration, University of Northern Iowa, Cedar Falls, IA 50614, United States

a r t i c l e

i n f o

a b s t r a c t
We decompose realized market returns into expected return, unexpected cash-ow news and unexpected discount rate news to test the relation between aggregate market returns and aggregate insider trading. We nd that (1) the predictive ability of aggregate insider trading is much stronger than what was reported in earlier studies, (2) aggregate insider trading is strongly related to unexpected cash-ow news, (3) market expectations do not cause insider trading contrary to what others have documented, and (4) aggregate insider trading in rms with high information uncertainty is more likely to be associated with contrarian investment strategy. These results strongly suggest that the predictive ability of aggregate insider trading is because of insiders ability to predict future cash-ow news rather than from adopting a contrarian investment strategy. These results hold even after we control for non-informative trades and information uncertainty. 2009 Elsevier B.V. All rights reserved.

Article history: Received 3 May 2009 Accepted 14 November 2009 Available online 20 November 2009 JEL classication: G11 G12 G14 G19 Keywords: Insider trading Return decomposition Discount rate news Cash-ow news Contrarian investment strategy

1. Introduction Recent studies on aggregate insider trading have documented that insiders are able to predict future market movements and that they are able to time the market (Seyhun, 1988; Lakonishok and Lee, 2001). However, it is not clear from the evidence what the source of this predictability of market returns is. One possibility is that insiders are contrarian investors (Rozeff and Zaman, 1998; Lakonishok and Lee, 2001; Jenter, 2005). It is also possible that managers are better informed about their rms future prospects and this informational advantage explains their market timing ability (Ke et al., 2003). Finally, insiders could have informational advantage and are also contrarian investors (Piotroski and Roulstone, 2005). There is substantial evidence that corporate ofcers and directors are able to discern apparent mispricing in their rms securities based on rm related information and are able to protably

trade on this.1 If this information is related to future economy-wide activity, then aggregate insider trading should predict future market movements and the market timing ability of insiders would be based on information unanticipated by the market (see Seyhun, 1988). We differentiate this from the contrarian investment strategy of insiders and dene it as superior information hypothesis, which is related to unexpected changes in future cash ow and discount rate news. If insiders are motivated to trade because of perceived mispricing, it is also conceivable they may react to market returns. It is possible that noise traders may drive market prices away from intrinsic values even in the absence of new information. Hence, a stock that was trading roughly at its intrinsic value could decline (rise) signicantly because of such noise trading. Corporate insiders may then perceive the stock to be undervalued (overvalued) and buy (sell) it. To the extent that noise trading is a market-wide phenomenon, we would

* Corresponding author. Tel.: +1 319 273 2579; fax: +1 319 273 2922. E-mail addresses: jiangx@u.edu (X. Jiang), mir.zaman@uni.edu (M.A. Zaman). 0378-4266/$ - see front matter 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.jbankn.2009.11.016

1 Previous studies based on US data unanimously documented that insiders are better informed and earn abnormal returns (Lorie and Niederhoffer, 1968; Jaffe, 1974; Seyhun, 1986; Rozeff and Zaman, 1988; Lakonishok and Lee, 2001). Using Oslo Stock Exchange data Eckbo and Smith (1998) show that insiders do not earn abnormal returns while Jeng et al. (2003) show that abnormal returns earned by insiders are restricted only to purchases. Aktas et al. (2008) nd that even though the nancial markets do not respond strongly in terms of abnormal returns, price discovery in the market is hastened on days that insiders are trading.

1226

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236

expect market returns to predict aggregate insider transactions (see Rozeff and Zaman, 1998; Chowdhury et al., 1993; Lakonishok and Lee, 2001; Jenter, 2005). Such a relationship would be viewed as insiders following a contrarian investment strategy and aggregate insider trading should be related to market expectations about future cash ows and future returns. On the other hand, if mispricing is rm specic then insiders transactions in each rm would cancel out and aggregate insider trading should not be related to market returns. Even though under both contrarian strategy and superior information scenarios insider trading is related to market returns, the key distinction is that if insiders trade on the basis of superior information aggregate insider trading will predict future market returns while the contrarian strategy implies insider trading is a reaction to market returns. Other related studies of managerial decisions also suggest that insiders are better informed about their companies future prospects. For example, Ikenberry et al. (1995) nd positive abnormal returns earned by shareholders of companies that have announced open market share repurchases. These abnormal returns persist for some time after the announcement. One of the main motivations for repurchases seems to be that insiders perceive the companys stock as being undervalued. Chan et al. (2007) uses a comprehensive sample of US repurchase announcements to look at whether executives possess market timing skills when announcing certain corporate transactions. Their results are consistent with the notion that managers possess market timing ability in the context of share repurchases. However, Ginglinger and Hamon (2007) using repurchases made by French rms show that share repurchases largely reect contrarian trading rather than managerial timing ability. Loughran and Ritter (1995), on the other hand, observe a prolonged underperformance by companies following seasoned equity offerings. This is in line with the hypothesis that companies tend to issue seasoned equity when they perceive the market to be too optimistic about the prospects of their company. Baker and Wurgler (2000) nd that the share of equity issues in total new equity and debt issues increases right after a year of high market returns and has been a stable predictor of US stock market returns between 1928 and 1996. The paper also provides evidence of issuing rms preferring equity nance before periods of low market returns and shunning equity in favor of debt before periods of high market returns. Overall, the results add to a growing body of evidence that managerial decisions are in response to or in anticipation of market conditions (see also Baker et al., 2006; Adams et al., 2009, among others). A related line of research on insider trading has focused on whether aggregate insider trading can predict market movements and could be used as a tool to time the market. Even though Givoly and Palmon (1985) found no relation between insider trading and subsequent information events, Seyhun (1988) provides evidence suggesting that some of the mispricing observed by insiders in their own rms securities is caused by unanticipated changes in economy-wide activity. Using a single-equation regression analysis he nds aggregate insider trading is correlated to market return in the subsequent 2 months following the trading activity. In a subsequent paper, Seyhun (1992) nds that aggregate insider trading is positively related to future real activity as measured by growth rates of after-tax corporate prots, the Index of Industrial Production and the Gross National Product. However, in the paper he concludes . . .both changes in business conditions as well as movements away from the fundamentals contribute to the information content of aggregate insider trading. Chowdhury et al. (1993) nd that stock market returns Granger-cause insider transactions, while the predictive content of aggregate insider transactions for subsequent market returns is slight. Lakonishok and Lee (2001) also provide evidence in support of the predictive ability of aggregate insider trading and market

movement. They conclude that this ability is partially because insiders act as contrarian investors. Previous studies simply examine the relationship between realized market return and some metric of insider trading without explicitly considering the source of predictability. Piotroski and Roulstone (2005) is an exception; their paper attempts to differentiate the source of the predictability and nds that insider trades are related to the rms future earnings performance. However, they use the change in accounting returns as proxies for future cash ows. Cohen et al. (2002) point out that the change in accounting returns is not a good measure to proxy future cash ows. Both conclusions of contrarian strategy of investing by insiders and insiders ability to predict unanticipated future economy-wide activity rely on insider trading being positively related to subsequent realized market returns. These studies, however, make no attempt to determine whether the apparent predictability of market returns by aggregate insider trading is because insiders follow contrarian strategy or are better able to predict market-wide activities. For example, Rozeff and Zaman (1998) show that insiders predominantly buy (sell) shares in value (glamour) rms and interpret this as evidence of insiders trading against the markets over-reaction to past performance. Such trading behavior is consistent with insiders purchasing (selling) securities with high (low) expected returns or the greatest amount of undervaluation (overvaluation). In a related paper, Jenter (2005) provides further evidence that a top manager tends to have contrarian views with respect to valuation of his own companys stock. Based on a methodology similar to Rozeff and Zaman (1998), Jenter nds that insiders are more likely to be net sellers in growth rms (low book-to-market ratio) and net buyers in value rms (high book-to-market ratios) and ascribes this trading pattern to contrarian views of insiders on stock valuation. To strengthen this claim Jenter controls for non-information based motives, like diversication and portfolio rebalancing, of insider trading. Using variables to control for the effect of stock ownership and equity based compensation (measures more likely to lead to trading for diversication and portfolio rebalancing purposes) he nds that book-to-market has a strong and signicant effect on insider trading and concludes that insider trading is more likely due to contrarian beliefs. In both these papers, the reported pattern of trading across cash ow-price or book-to-market portfolios could reect insiders trading on market pricing errors (e.g., over-reaction to past performance), but it could also reect insiders superior knowledge of future earnings performance. La Porta et al. (1997) show that, on average, value (growth) rms tend to have positive (negative) future earnings announcement period returns. Because returns on earnings announcement tend to be correlated with actual changes in performance both Rozeff and Zaman (1988) and Jenter (2005) ndings cannot differentiate trading on the basis of contrarian beliefs from trading on the basis of superior information about future cash ows. The purpose of this paper is to re-examine the ability of aggregate insider trading to predict market-wide movement using return decomposition in a vector autoregressive (VAR) model framework. Such a re-examination is called for because of mixed results reported in previous papers. Moreover, it is important for the capital markets to be able to distinguish between these two sources of predictability. If insiders are trading based on contrarian strategy, then in aggregate, such trading would not provide any new information about the future economy-wide activity. Aggregate insider trading would in this case imply market overreaction (under reaction) and subsequently lead to market correction. However, if insiders are trading on the basis of information related to unanticipated (from outside investors) changes in future cash ows, then aggregate insider trading will predict future real economic activities and future market returns. In order to distinguish between these two sources of predictability we closely follow

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236

1227

Campbell (1991) and Hecht and Vuolteenaho (2006) method of decomposing aggregate market return into expected return, unexpected cash-ow news and unexpected discount rate news. We argue that insider trading based on new information will result in a positive relation between aggregate insider trading and unexpected cash-ow news (aggregate insider trading will lead the positive future unexpected returns). On the other hand, if insiders are trading based on a contrarian investment strategy then we expect a negative relation between insider trading and expected return (aggregate insider trading will react on the expected return). Using this decomposition, a regression of market returns on insider trading measures is then decomposed into three component regressions. We nd the following: (1) the predictive ability of aggregate insider trading is much stronger than what was reported in earlier studies, (2) aggregate insider trading is strongly related to unexpected cash-ow news, (3) market expectations do not cause insider trading, contrary to what others have documented, and (4) aggregate insider trading in rms with high information uncertainty is more likely to be associated with contrarian investment strategy. These results strongly suggest that the predictive ability of aggregate insider trading is because of insiders ability to predict unanticipated changes in economy and market-wide factors as opposed to following a contrarian investment trading strategy. Our contribution is twofold. First, this paper provides denitive evidence into the debate of whether insider trading based on perceived mispricing is a result of contrarian investment strategy or whether it is based on insiders access to information about future cash-ow news. By decomposing realized market returns into expected returns, unexpected cash-ow news and unexpected changes in discount rate news, this paper directly tests the sources of the insider trading predictability. Second, this paper contributes to the existing literature on the importance of the relation between corporate transactions and insiders ability to time the market. When the rms securities are mispriced and insiders are able to identify this mispricing, then this ability affects the nancing, investment and other corporate transactions. The paper is organized as follows. Section 2 discusses reasons to believe that insider trading can predict future market returns and develops the framework and formulates the hypotheses, Section 3 describes the data and provides summary statistics. Results are reported and discussed in Sections 46 while the last section contains a summary and interpretation of the results.

A competing hypothesis regarding aggregate insider trading relies on the contrarian strategy of investing. If stock prices are affected by the trading of both informed and uninformed (noise) traders then prices can diverge from fundamental values (Shiller, 1984; DeLong et al., 1990). According to this view noise traders may drive market prices away from current fundamental values. However, in the long run prices would revert back to fundamental values. Wang (2010) develops a model whereby an informed investor aggressively trades on her information and takes a large, opposite position against the noise trader. If the contrarian strategy is employed by insiders at the rm specic level then there should be no relation between aggregate market returns and insider trading. On the other hand, if noise trading is a market-wide phenomenon then a relation between aggregate insider trading and market return should exist. In such a scenario, market returns would predict insider trading behavior. In order to test the relation between aggregate stock returns and insider trading we use the standard log-linear present value model developed by Campbell (1991). 2.1. Log-linear present value model framework and insider trading Campbell (1991) decomposes the realized return on equities into following three components:

Rt1 Et Rt1 Et1 Et

1 X j0

qj DDt1j Et1 Et

1 X j1

qj Rt1j
1

Et Rt1 NCF;t1 NDR;t1 ;

2. Framework and hypotheses There are a number of compelling and competing reasons to believe that aggregate insider trading can predict future market returns. Assume that company executives and directors know their businesses more intimately than analysts (investors) following their stocks. They know when demand for their goods and services is increasing, when inventories are piling up, when production costs are increasing or prot margins declining, etc. Given their knowledge about their rm, insiders should be able to predict when the rms future cash ows would increase, and would then buy stocks in their rms. If the predicted increase in cash ows by insiders is strictly the result of some rm-specic improvement (e.g. prot margin) there should be no relation between aggregate insider trading and market return. On the other hand, if the cash ows are related to economy-wide activity such as increases in aggregate demand of goods and services, then subsequently when the increase in economy-wide activity is recognized by the market, stock prices will rise. This will result in a positive relation between insider buys and market return. We call this the superior information hypothesis and test this by examining the relationship between insider trading and unanticipated cash-ow news.

where R is the log return on equities, DD is dividend growth, q is the discount factor, Et(Rt+1) is the one-period expected return, NCF,t+1 is the cash-ow news, and NDR,t+1 is the discount rate news. This equation states that the realized return must be associated with the expected return, the changes in expectations of future cash ows, and/ or the changes in the expectations of future discount rates. As emphasized by Campbell (1991), Eq. (1) is really nothing more than a dynamic accounting identity relating the current return innovation to revisions in expectations. Hecht and Vuolteenaho (2006) apply this method to measure the relative importance of these three effects in regressions of returns on cash ow proxies. Based on Eq. (1), the explanatory power of cash ow proxies may arise from the correlation of cash ow proxies (predictors) with one-period expected returns, cashow news, and/or expected return news. They argue that if expected-return variation is responsible for the high explanatory power of the aggregate regressions, these R2 should not be interpreted as evidence of cash-ow news driving the returns. Similarly, if expected-return news is highly variable and positively correlated with cash-ow news, the low R2s in regressions of rm-level returns on earnings do not necessarily imply that earnings are a noisy or delayed measure of the cash ow generating ability of the rm. Even if earnings are a clean signal of cash-ow news, expected return effects (due to variation in risk-adjusted discount rates and/or mispricing) can garble the earnings-returns relation. In a similar spirit, we apply Campbells decomposition to estimate and test the dynamic relation between market returns and aggregate insider trading. This method uniquely helps us to distinguish whether the relation between market returns and aggregate insider trading is due to contrarian strategy or superior knowledge. Consider a typical forecast regression of returns on insider trading,

Rt1 a bIT t et1 ;

where IT is a measure of insider trading. Seyhun (1988) uses a similar methodology to show a weak relationship between insider trading and market returns and concludes that insider trading predict market return. As analyzed above, it is difcult to interpret the

1228

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236

coefcient b, and more importantly, using regression (2) we cannot distinguish whether the relation between market returns and insider trading is because insiders trade based on contrarian strategy or they are better able to predict unanticipated changes in marketwide factors due to changes in future cash-ow news. Using Campbells (1991) decomposition, however, we can rewrite the regression (2) as following:

Et Rt1 a bER IT t eER;t1 ; NCF;t1 a bCF IT t eCF;t1 ; NDR;t1 a bDR IT t eDR;t1 :

3a 3b 3c

Since the sum of the left-hand-side in regression (3) is the realized return and the independent variables in regression (3) are the same, regression (2) can also be expressed as:

els and/or the incorporation of biased judgments. Based on the perceived mispricing, insiders trade against outside investors sentiment. If the contrarian strategy drives the relation between market return and aggregate insider trading, we would expect that cER, cCF, and cDR in Eq. (5) to be signicantly negative. For instance, if outside market expectation Et[Rt+1] is positive, and if inside traders perceive this expectation to be incorrect, insider traders will sell their stocks, i.e., cER is negative. It is also possible that insiders are contrarians with respect to future cash-ow news and discount rate news. If insiders perceive that the markets valuation of future cash-ow news is too high relative to cash ow fundamentals, then insiders will sell their stocks. We formulate the following hypotheses: Hypothesis 1. If insider trading is not informative (in terms of cash-ow news) then the coefcients bCF, bDR in Eq. (3) are indistinguishable from zero; otherwise they are positive. Hypothesis 2. If insider trading is not informative (in terms of contrarian strategy) then the coefcients cER, cCF, and cDR in Eq. (5) are indistinguishable from zero; otherwise they are negative. 2.2. Estimating one-period expected returns, cash-ow news and discount rate news We follow Campbell (1991) and Campbell and Vuolteenaho (2004) to estimate the one-period expected return, cash-ow news, and discount rate news series using a vector autoregressive (VAR) model. We assume that the data are generated by a rst-order VAR model

Rt1 a bER bCF bDR IT t eER;t1 eCF;t1 eDR;t1 :

Regressions (3) and (4) show that there are three sources driving the relation between market return and insider trading: oneperiod expected return, cash-ow news, and discount rate news. We also consider the following regression:

IT t1 a cRt ut1 ; IT t1 a cER Et Rt1 uER;t1 ; IT T1 a cCF NCF;t uCF;t1 ; IT t1 a cDR NDR;t uDR;t1 :

5a 5b 5c 5d

Eq. (4) shows, if expected return variation is responsible for the high explanatory power of the aggregate regressions, these R2 should not be interpreted as evidence of superior information driving the returns. Similarly, if expected return news is highly variable and positively correlated with cash-ow news, the low R2s in regressions of market returns on insider trading do not necessarily imply that insider trading is a noisy or delayed measure of the cash ow generating ability of the rm. Even if insider trading is a clean signal of cash-ow news, expected return effects (due to variation in risk-adjusted discount rates and/or mispricing) can garble the insider trading-returns relation. We use regressions (3) and (5) to estimate the relation between market return and insider trading, and distinguish whether the relation is attributed to insiders possessing superior knowledge to predict market-wide movements (as evidenced in Seyhun (1988)) or contrarian strategy as evidenced in Rozeff and Zaman (1998), Chowdhury et al. (1993), Lakonishok and Lee (2001) and Jenter (2005). Assume that insiders are better able to predict future cash-ow news of the rm than outside investors. If these cash ows are related to economy-wide activity then subsequent to aggregate insider buying (selling) in stocks of their rm the market returns should increase (decrease). It may be argued that if insiders have information about their rms future cash-ow news which is related to economy-wide activity then it is likely they may be better off trading in options or other derivative securities than trading in stocks of their rm. However, given Seyhuns (1986) evidence of passive as well as active trading by insiders around rm-specic nonpublic information, insiders would also be expected to trade in stocks of their rms. If the superior information hypothesis is true, we expect positive and signicant coefcients for bCF and bDR. In contrast, if insider trading does not reveal information about future economy-wide activity then the coefcients bCF and bDR will be insignicant. Furthermore, under this hypothesis if insiders know more about their rms cash-ow news and in the aggregate, cash-ow news do not cancel out but rather are proxies of aggregate market cash-ow news, then the coefcient bCF should dominate bDR. The contrarian strategy hypothesis states that outsiders make valuation errors through the application of inferior valuation mod-

Z t1 A0 AZ t ut1 :

where Zt+1 is a vector of excess log market returns, the term yield dened as the yield difference between 10-year constant-maturity taxable bonds and short-term taxable notes, the earnings growth rate from S&P 500 index, and small-value spread,2 describing the economy at time t + 1, A0 and A are vector and matrix of constant parameters, and ut+1 is a vector of shocks. With the VAR expressed in this form, the components of identity (1) can be obtained by

Et Rt1 e10 A0 AZ t ; NCF;t1 e3 qAI qA ut1 ; NDR;t1 e1 I qA ut1 ;

0 0

7a
1

7b 7c

where e1 1 0 0 , e3 0 0 1 0 0 , and I is an identity matrix. Eq. (7) expresses EtRt+1, the one-period expected return as tted value of Zt+1 based on VAR model in Eq. (3), NCF,t+1, the cash-ow news, and NDR,t+1, the discount rate news as linear functions of the t + 1 shock vectors. Here in Eq. (7b), we directly measure the cash-ow news based on earnings growth Ncf Et1 Et P1 j j0 q DEt1j , which is different from the Campbell and Vuolteenaho (2004) residual approach (see Chen and Zhao (2009)).3 3. Data and summary statistics 3.1. Insider trading data We collect insider trading information from the Securities Exchange Commission (SEC) Ownership Reporting System (ORS). The ORS data starts in 1975 and ends in 2000 and contains all insider transaction data that are subject to disclosure by the Securities Exchange Act of 1934. Section 16(a) of the Act requires
For details of data construction, see Campbell and Vuolteenaho (2004). We also performed all of the analyses using cash ow measure from the residual method used in Campbell and Vuolteenaho (2004). The results are qualitatively similar and are available from the authors.
3 2

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236

1229

Table 1 Summary statistics. This table summarizes the statistics of insider trading data and VAR state variables. In Panel A, insider trading data are from all open market purchases and sales of NYSE/AMEX and Nasdaq CRSP- and Compustat-listed common shares (CRSP share code 10 or 11) during 1978:Q1 to 2000:Q4. We report average quarterly number of buys and sells per rm of our sample. IT is the average of IT from Eq. (8). We exclude all option transactions and transactions less than 100 shares. We dene Management: as CEOs, CFOs, and Chairmen of the Board, Directors, Ofcers, Presidents, and Vice-Presidents. Large shareholders are those who own more than 10% of shares and are not in management. Others are all those who are required to report their trading to the SEC but neither managers nor large shareholders. Large, medium, and small rms are rms based on the sample rms quintile cutoff points at the market value in previous quarter. The log excess market return (R), term yield spread (TY), 10-year moving average earnings growth (GE), and small-stock value spread (VS). In Panel B, we report the descriptive statistics of the VAR state variables estimated from the full sample period 1978:Q1 to 2000:Q4, 92 quarterly data points. It includes the log excess market return (R), term yield spread (TY), 10-year moving average earnings growth (GE), and small-stock value spread (VS). Std Dev. denotes standard deviation and Auto. denotes the rst-order autocorrelation of the series. Management Buys Sales IT data 0.255 0.073 0.275 0.542 Large shareholders Buys 1.210 1.203 1.155 1.276 Sales 0.794 0.562 0.780 1.053 IT 0.212 0.325 0.186 0.123 Others Buys 1.121 1.182 1.064 1.113 Min 0.276 1.320 0.001 1.236 Sales 1.108 0.638 1.080 1.629 IT 0.004 0.250 0.027 0.222 Max 0.182 2.720 0.027 2.045 Total Buys 1.017 1.110 0.973 0.967 Sales 1.118 0.670 1.062 1.621 IT 0.020 0.210 0.042 0.228 Auto. 0.054 0.760 0.963 0.822

Panel A: Summary statistics for insider trading All 0.778 1.432 Small rms 0.977 0.796 Medium rms 0.748 1.332 Large rms 0.622 2.122 Variables Mean

Std Dev. 0.080 0.747 0.007 0.154

Panel B: Summary statistics for state variables R 0.019 TY 0.566 GE 0.016 VS 1.497

that open market trades by corporate insiders be reported to SEC within 10 days after the end of month in which they took place. For the purposes of this reporting requirement, corporate insiders include ofcers with decision making authorities over the operations of the company (CEOs, CFOs, Other Ofcers, Presidents, Vice-Presidents, etc.), all members of the board of directors, and benecial owners of more than 10% of the companys stock. These reports led on the SECs Form 3, 4 and 5 are the source of insider trading data. From the reported transactions we exclude all transactions that are less than 100 shares and only focus on open market purchases and sales by insiders. Using the ORS data we classify insiders into three groups. The rst group, Management, includes Chairmen of the board, CEO, CFO, Ofcers, Directors, Presidents, and Vice-Presidents and is assumed to have direct access to information about the rms future prospects. Large shareholders are those who are not management but own 10% or more of shares and are assumed to have no direct access to inside information. The third group, others are all investors who are required to report their trades to SEC but are neither managers nor large shareholders. We dene a measure of aggregate insider trading activity, IT in the following manner. For each quarter in our sample from January 1978 to December 2000 we nd the total number of insiders buying (Buys), and the total number of insiders selling (Sells), stocks in their companies. We then dene the following insider trading measure for each rm i in quarter Q:

averages 0.778 buys and 1.432 sells per rm-quarter and the average IT measure is 0.255. For the large shareholder group, there are on average 1.210 buys and 0.794 sells per rm-quarter, and the average IT measure is 0.212. When we look at the trading behavior across the size of the rms we notice a monotonic decrease in buys and a monotonic increase in sales for the management group. Buys decrease from 0.977 per rm-quarter in the small rms to 0.622 per quarter in the large rms for the management group. Sales range from 0.796 per rm-quarter in the small rms to 2.122 per quarter in the large rms. The mean of the IT measure also reects this behavior with a mean of 0.073 for the smallest size rms to 0.542 for the large rms, implying smaller rms insiders are net buyers and larger rms insiders are net sellers. These results are in line with previous evidence of insiders buying more heavily in smaller rms and selling heavily in larger rms4 (Seyhun, 1986; Rozeff and Zaman, 1988; Jenter, 2005; Aktas et al., 2008). 3.2. VAR data In order to decompose the realized return into expected return, cash-ow news and discount rate news using VAR approach, we need to specify variables to be included in the state vector. Following Campbell and Vuolteenaho (2004), we choose a model with the following four state variables. The excess market return (R) is measured as the log excess return on the Center for Research Security Prices (CRSP) value-weighted index over log risk-free rate. The risk-free-rate data are constructed by CRSP from Treasury bills with approximately 3-month maturity. The term yield spread between long-term and short-term bonds (TY) is measured as the difference between 10-year constant-maturity taxable bond yield and the yield on short-term taxable notes. The markets earnings growth (EG) is the log growth of a 10-year moving average of the S&P 500 earnings. The small-stock value spread (VS) is measured as the difference between the log book-to-market ratios of small value and small growth stocks. Based on Campbell and Vuolteenaho (2004), the small-stock value spread (VS) is constructed in the following way. The portfolios, which are constructed at the
4 For example, in Jenter (2005), going from the lowest book-to-market decile to the highest book-to-market decile the number of net sellers falls from 68% to 33% while number of net buyers increase from 24% to 54%. A similar pattern in buys and sells can be discerned in our sample as one goes from large rms to small rms.

IT Q;i

BuysQ ;i SellsQ ;i : BuysQ ;i SellsQ ;i

To construct an average of the IT time series we dene

IT Q

N X i1

IT Q ;i =N;

where N is the number of rms with insiders trading in each quarter. In Table 1, we present summary statistics of the trading behavior of insiders during our sampled period. On average, for the total sample, there are 1.017 insiders buying stocks per rm-quarter and 1.118 insiders selling per rm-quarter and the average of the insider trading per rm-quarter, IT is 0.020 (this is the mean of the IT measure in Eq. (9) over the sample period from 1978 to 2000). Recall that the rm-quarter IT measure is the ratio of net buys to total trades (buys + sells) over the quarter. The management group

1230

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Table 2 VAR parameter estimates. This table shows the OLS parameter estimates for a rstorder VAR model including a constant, the log excess market return (R), term yield spread (TY), 10-year moving average earnings growth (GE), and small-stock value spread (VS). Each set of three rows corresponds to a different dependent variable. The rst ve columns report coefcients on the ve explanatory variables, and the sixth column shows adjusted R2. Panel B reports the correlation among the state variables. We report descriptive statistics of cash-ow news based on residual (Ncf1), cash-ow news based on earning growth (Ncf), and discount rate news (Ndr) in Panel C. We also report the NeweyWest adjusted t-statistics in parentheses. Sample period for the dependent variables is 1978:Q12000:Q4. Constant Rt TYt 0.006 (0.559) 0.564 (4.436) 0.000 (0.372) 0.017 (1.112) GEt 0.844 (0.546) 35.140 (4.063) 0.984 (28.197) 0.620 (0.314) VSt 0.114 (2.666) 0.262 (0.773) 0.000 (0.156) 0.805 (12.721) R2 0.012 0.630 0.925 0.669

end of each June, are the intersections of two portfolios formed on size (market equity, ME) and three portfolios formed on the ratio of book equity to market equity (BE/ME). The size breakpoint for year t is the median NYSE market equity at the end of June of year t. BE/ ME for June of year t is the book equity for the last scal year end in t 1 divided by ME for December of t 1. The BE/ME breakpoints are the 30th and 70th NYSE percentiles. At the end of June of year t, the small-stock value spread is constructed as the difference between the log(BE/ME) of the small high-book-to-market portfolio and the log(BE/ME) of the small low-book-to-market portfolio, where BE and ME are measured at the end of December of year t 1. For months from July to May, the small-stock value spread is constructed by adding the cumulative log return (from the previous June) on the small low-book-to-market portfolio to, and subtracting the cumulative log return on the small high-book-tomarket portfolio from, the end-of-June small-stock value spread.5 Asset pricing literature nds that these state variables are able to forecast and track market returns.6 In Panel B, we provide summary statistics of the state variables for the sample period 1978 2000. Our data description is a little different from the one in Campbell and Vuolteenaho (2004) due to data frequency (we use quarterly data while they use monthly data) and sample periods (our data is from 1978 to 2000, while they use data from 1928 to 2001, including in the sample period the Great Depression and World War Two). Our data show that all state variables except for earnings growth are fairly volatile, since we use 10-year moving average S&P 500 earnings. Quarterly log excess market return shows negative autocorrelation (0.054). Log earnings growth, log small-stock spread and log term yield spread are all quite persistent as expected, consistent with the result in Campbell and Vuolteenaho (2004).

Panel A: VAR estimates Rt+1 0.173 0.042 (2.465) (0.458) TYt+1 0.434 0.815 (0.781) (1.235) GEt+1 0.001 0.003 (0.249) (1.542) VSt+1 0.314 0.098 (3.213) (1.434)

Rt+1

TYt+1

GEt+1 0.012 0.662 1.000

VSt+1 0.065 0.041 0.023 1.000

Panel B: Correlation of the state variables 1.000 0.067 TYt+1 1.000 GEt+1 VSt+1 Mean Std Min Max

Autocorrelation 1 2 3

Panel C: Ncf1 Ncf Ndr

Descriptive statistics of cash-ow news and discount rate news 0.000 0.067 0.238 0.130 0.055 0.055 0.051 0.000 0.339 0.698 1.222 0.138 0.073 0.060 0.000 0.045 0.099 0.139 0.005 0.003 0.064

4. Results The evidence presented in this section uses a VAR model to examine the relationship between aggregate insider trading and market return. In Table 2 Panel A, we report parameter estimates for the VAR model. Each row of the table corresponds to a different equation of the model. The rst ve columns report coefcients on the ve explanatory variables: a constant and lags of the excess market return, term yield spread, earnings growth, and small-stock value spread. NeweyWest adjusted t-statistics are reported in parentheses below the coefcients. Finally, we report the adjusted R2 for each of the estimated equations. The rst row of Table 2 shows that only small-value stock spread (VS) in our VAR state variables signicantly predicts excess market return. The smallstock value spread negatively predicts the return, consistent with Eleswarapu and Reinganum (2004) and Brennan et al. (2001). Overall, the adjusted R2 of the return forecasting equation is about 1.2%, which is similar to Campbell and Vuolteenaho (2004) and a reasonable number for a quarterly model. The remaining rows in Panel A summarize the dynamics of the explanatory variables. All explanatory variables are approximately an AR(1) process. The earnings growth is highly persistent, with a root close to unity. It is well known that estimates of persistent AR(1) coefcients are biased downwards in nite samples, and that this causes bias in the estimates of predictive regressions for returns if return innovations are highly correlated with innovations in predictor variables (Stambaugh, 1999). There is an active debate about the effect of this on the strength of the evidence for return predictability (Ang and Bekaert, 2007; Campbell and Yogo, 2006; Lewellen, 2004;
All data for VAR are kindly provided by Tuomo Vuolteenaho. 6 We do not incorporate insider trading into the VAR on purpose, because our null hypothesis is that inside trading is not informative.
5

Torous et al., 2004). Panel B provides correlation among the state variables. It seems that excess return is positively correlated with other state variables. In Panel C, we report the descriptive statistics for the discount rate news and the cash-ow news from return decomposition. Consistent with Campbell and Vuolteenaho (2004), the volatility of cash-ow news is larger than the volatility of discount rate news. We regress realized market excess returns (dened as the CRSP value-weighted return minus 3-month T-Bill rates) and its three estimated components (one-period expected market excess return, cash-ow news and negative of discount rate news) individually on lagged values of aggregate insider trading measure, IT. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c and Granger-causality test in Table 3. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged insider trading are zero.7 The p-value is listed below the F-test in bracket parenthesis. In Panel A, we report results for all insiders. The rst row of Panel A shows that trading by all insiders has no explanatory power in explaining the variation in realized market returns. The F-statistic is 5.377 with a p-value of 0.146. Note that the F-statistic is used for the Granger-causality test of whether the coefcients of lagged IT explain the variation in Rt. Furthermore, none of the individual coefcients of lagged IT are signicant at the 5% condence level. This suggests that there is no relation between realized market return and insider trading in the sample period. A possible reason
7 For the sake of brevity, we only report estimates of coefcients of variables which are of interest. For example, in Table 2, we only report the estimates of lagged IT, the cis. The adjusted R2 reported is for the full model which includes the lagged X variables.

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Table 3 Regression of market excess return and its components on insider trading. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. ITt denotes the insider trading measure dened in Eq. (9), Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, tstatistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged IT. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged insider trading are zero. The p-value is listed below the F-test in bracket.

1231

Table 4 Regression of insider trading on market excess return and its components. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. ITt denotes the insider trading measure dened in Eq. (9), Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged X. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged return are zero. The p-value is listed below the F-test in bracket.

Xt a

3 X i1

bi X ti

3 X i1

ci IT ti
ITt2

4 X k1

/k Dk et
P3

Tt a

3 X i1

bi X ti

3 X i1

ci IT ti

4 X k1

/k Dk et
P3

ITt1 Panel A: All insiders 0.042 Rt (1.251) 0.086 Et1[Rt] (0.557) 0.005 NDRt (0.201) NCFt 0.030 (0.198) 0.038 NEWSt (0.217)

ITt3

R2

i1 i

F-test

Xt1 Panel A: All insiders 0.003 X = Rt (0.007) 0.015 X = Et1[Rt] (0.205) 0.323 X = NDRt (0.612) X = NCFt 0.032 (0.468) 0.021 X = NEWSt (0.343) Panel B: Management 0.124 X = Rt (0.282) X = Et1[Rt] 0.037 (0.524) 0.288 X = NDRt (0.560) 0.052 X = NCFt (0.800) X = NEWSt 0.038 (0.650)

Xt2

Xt3

R2

i1 bi

F-test

0.004 (0.100) 0.315 (1.877) 0.031 (1.281) 0.282 (1.699) 0.313 (1.659)

0.036 (0.926) 0.094 (0.625) 0.016 (0.965) 0.109 (0.787) 0.132 (0.857) 0.010 (0.339) 0.021 (0.137) 0.002 (0.121) 0.025 (0.182) 0.034 (0.222) 0.056 (1.162) 0.106 (0.684) 0.022 (1.193) 0.138 (0.987) 0.168 (1.075)

0.003 0.033 0.003 0.039 0.033

0.081 0.324 0.043 0.361 0.407

5.377 (0.146) 8.289 (0.040) 5.298 (0.151) 8.983 (0.030) 8.800 (0.032) 6.287 (0.098) 10.436 (0.015) 8.718 (0.033) 11.514 (0.009) 11.456 (0.010) 3.504 (0.320) 8.721 (0.033) 5.091 (0.165) 9.462 (0.024) 9.284 (0.026)

0.244 (0.392) 0.047 (0.611) 0.249 (0.372) 0.032 (0.376) 0.027 (0.360) 0.369 (0.605) 0.052 (0.601) 0.241 (0.340) 0.029 (0.311) 0.025 (0.297)

0.520 (1.945) 0.068 (1.384) 0.012 (0.038) 0.034 (0.721) 0.027 (0.669) 0.702 (2.217) 0.056 (0.980) 0.199 (0.580) 0.014 (0.281) 0.009 (0.202) 0.463 (1.930) 0.075 (1.857) 0.121 (0.350) 0.046 (1.021) 0.038 (0.961)

0.355 0.343 0.338 0.338 0.337

0.273 0.036 0.559 0.034 0.021

5.830 (0.120) 1.955 (0.582) 0.923 (0.820) 0.739 (0.864) 0.540 (0.910) 8.151 (0.043) 1.154 (0.764) 0.974 (0.808) 0.825 (0.843) 0.491 (0.921) 5.555 (0.135) 3.765 (0.288) 1.445 (0.695) 1.531 (0.675) 1.317 (0.725)

Panel B: Management 0.036 0.017 Rt (1.130) (0.576) Et1[Rt] 0.045 0.331 (0.298) (2.120) 0.001 0.037 NDRt (0.050) (1.682) 0.002 0.309 NCFt (0.013) (2.028) NEWSt 0.004 0.345 (0.023) (2.007) Panel C: Large shareholders 0.035 0.023 Rt (0.997) (0.464) 0.127 0.372 Et1[Rt] (0.651) (1.734) 0.013 0.033 NDRt (0.422) (1.086) NCFt 0.067 0.311 (0.348) (1.449) 0.083 0.344 NEWSt (0.379) (1.412)

0.005 0.041 0.007 0.049 0.043

0.063 0.307 0.040 0.332 0.375

0.430 0.402 0.397 0.398 0.396

0.458 0.041 0.728 0.038 0.022

0.014 0.038 0.006 0.039 0.032

0.068 0.351 0.042 0.383 0.429

Panel C: Large shareholders 0.067 0.003 X = Rt (0.202) (0.005) 0.035 0.064 X = Et1[Rt] (0.495) (1.032) X = NDRt 0.107 0.470 (0.207) (0.935) 0.053 0.060 X = NCFt (0.795) (0.920) 0.041 0.052 X = NEWSt (0.678) (0.907)

0.401 0.406 0.395 0.401 0.400

0.399 0.046 0.456 0.039 0.026

could be the inclusion of large shareholders in our sample as this groups trading may be less informative. An alternate explanation, discussed in Section 2, is that such a lack of relationship does not necessarily imply insider trading is not informative. The second, third and fourth rows report results when the realized excess return is decomposed into one-period expected return, cash-ow news and the discount rate news. The F-statistic for the expected return, cash-ow news and discount rate news are 8.289, 8.983, and 5.298, with p-values of 0.040, 0.030, and 0.151, respectively. Our results suggest that when we consider all insiders their trading has little effect on realized market excess returns. However, if we decompose the realized return into three components, insider trading is signicantly related to expected market excess return and future aggregate cash-ow news. This means insider trading can explain the variations in realized excess market return which is due to future unexpected cash-ow news. Also, the F-statistic for the NEWS regression (sum of cash-ow news and discount rate news) is 8.80 with a p-value of 0.03 suggesting that the sum of unexpected cash-ow news and unexpected discount rate news

(unexpected returns) experiences signicant positive shocks subsequent to insiders buying stocks in their rms. As dened before, the Management group comprises of insiders who are assumed to have direct access to information about the rms future prospects. If this is true, the superior information hypothesis would predict a stronger relation between insider trading and future market returns. Panel B reports results for the Management group. Here the effect of insider trading is more pronounced as predicted by the superior knowledge hypothesis. The F-statistics (p-value) for realized return, expected return, discount rate news, cash-ow news, and NEWS are 6.287 (0.098), 10.436 (0.015), 8.718 (0.033), 11.514 (0.009), and 11.456 (0.01). The two quarter lagged coefcients of IT for the realized return, expected return, cash-ow news, discount rate news and NEWS regressions are 0.017, 0.331, 0.037, 0.039, and 0.345 with t-statistics of 0.576, 2.120, 1.682, 2.208, and 2.007, respectively. This provides strong evidence that insiders who are directly related to the day-to-day activities of the rms are better able to predict market return. Furthermore, our results show that trading by this

1232

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236

group of insiders is more likely to be related to unexpected future cash-ow news two quarters later. Panel C reports results for the large shareholders group. F-statistics for expected return and cash-ow news are marginally signicant and the lagged coefcient for IT is only signicant in the NEWS regression. For this group there is marginal evidence that trading is positively related to future unexpected news. Also, the 2 quarter and 3 quarter lagged coefcients for IT are all insignicant. In Table 4, we report results of regressions between insider trading and realized market excess return and the components of realized market excess return. Here the insider trading variable, IT is the dependent variable. The motivation here is to investigate whether the markets expectation of return drives insider trading. This would provide evidence in support of Chowdhury et al. (1993), Rozeff and Zaman (1998) and Jenter (2005) assertion that insiders follow a contrarian investment strategy. Our interest is in the relation between insider trading and the lagged values of one-period expected market excess returns. If the assertion of contrarian strategy is true then we expect a negative relation between insiders trading and lagged expected return. In Panel A, we report again results for the overall group of insiders. The F-statistics are 5.830, 1.955, 0.923, 0.739, and 0.540 when the insider trading variable, IT is regressed on lagged values of realized market excess returns, expected market excess returns, discount rate news, cash-ow news, and NEWS, respectively. The results in Panel A suggest that we cannot reject the null hypothesis that realized market excess return and lagged values of expected market excess return, discount rate news, cash-ow news, and NEWS do not Granger-cause insider trading. Furthermore, none of the coefcients of the lagged values of expected market excess return is statistically signicant. The evidence from Panel A does not support the assertion of insiders following a contrarian investment strategy. Recall that this group includes all insiders and results maybe misleading as the group consist of insiders who may not have access to information relating to a rms future prospects. In Panel B and Panel C, we report results for the Management group and the large shareholders group. In Panel B, the F-statistics (p-value) is 8.151 (0.043) for the regression of realized market excess return on IT. This suggests that aggregate insider trading is related to prior periods realized return which is consistent with the evidence provided by Chowdhury et al. (1993), Rozeff and Zaman (1998) and Jenter (2005). Their sample excludes large shareholders and is comparable to the Management group in this study. However, the F-statistics for the regression of IT on the lagged values of expected market excess returns, cash-ow news, discount rate news, and NEWS are not signicant. These results do not support the contrarian strategy hypothesis as we cannot reject the null hypothesis of lagged values of expected market excess return do not Granger-cause insider trading. Evidence presented in this table clearly demonstrates that market excess return do not cause insider trading hence insider trading is not a manifestation of the contrarian strategy. These results are quite different from what Chowdhury et al. (1993),8 Lakonishok and Lee (2001) and Jenter (2005) conjecture in their papers. These papers suggest that insider trades are more likely to be a function of contrarian strategy. These studies did not
8

address the issue of whether the observed relationship between insider trading and market return could be due to insiders ability to predict unanticipated cash-ow news or discount rate news. In this paper, we directly test both hypotheses by using return decomposition methods. Our results show that insider trading is more likely to be based on managers ability to time the market based on superior information. Results reported in Table 3 provide evidence that insider trading is related to future cash-ow news and hence is more likely to be based on the managers ability to predict market-wide activities while results reported in Table 4 shows a lack of relation between insider trading and lagged expected market excess return thereby providing no evidence in support of the contrarian strategy. 5. Intensive insider trading criterion Insiders may trade in stocks of their companies for reasons other than information about future cash ow realizations.9 Many insiders may trade to diversify and rebalance their portfolios. They may also trade for liquidity purposes (see Rozeff and Zaman, 1988; Jenter, 2005; Aktas et al., 2008). To reduce the potential noise from such trades and to isolate trades which are information driven we develop an intensive trading criterion similar to the one in Rozeff and Zaman (1988) and apply it to the insider trading metric in Eq. (8).10 In a given month, we require that at least three insiders take the same action and that no insiders take an opposing action. Thus, in a given month, if there are at least three insiders buying stocks in their company and no insiders selling then the rm is classied as a buy. For a rm to be classied as a sell three or more insiders must sell in a given month and no insiders buy it. We then use Eq. (8) to estimate the IT measure for each quarter across all rms. In Table 5, we report results of regressions using this new measure of insider trading metric for the Management group. Results are qualitatively the same as before. Panel A reports regression results when realized market excess returns and its components are regressed on lagged values of insider trading measure IT. The F-statistics (p-value) for realized return, expected return, discount rate news, cash-ow news, and NEWS are 3.857 (0.277), 10.339 (0.016), 9.664 (0.022), 9.922 (0.019), and 10.12 (0.018). In Panel B, we report results of regressions of insider trading on realized market excess return and the components of realized market excess return. Here the insider trading variable, IT is the dependent variable. The Fstatistics (p-values) are 12.122 (0.007), 3.627 (0.305), 6.087 (0.107), 4.716 (0.194), and 4.804 (0.187). Panel A and Panel B results clearly indicate that after controlling for insider trading that may be due to portfolio diversication and rebalancing purposes, insiders are still able to predict aggregate cash-ow news and discount rate news. Trading by insiders is signicantly related to unexpected future cash-ow news two quarters later. On the other hand we nd no evidence that insider trading is Granger caused by lagged values of expected market excess return. These results add robustness to our conclusion that trading by insiders is more likely to be related to unexpected future cash-ow news and not due to contrarian beliefs. 6. Firm size, information uncertainty and aggregate insider trading Jiang et al. (2004) dene information uncertainty as the degree to which a rms value can be estimated by the most knowledge9 We would like to thank an anonymous referee for suggesting to control for noninformation motivated trades. 10 We did not follow Jenters (2005) methodology for a couple of reasons. First, Jenters methodology for controlling trades due to portfolio diversication and rebalancing motives rely on annual measures of executive compensation, etc., whereas our study looks a quarterly trading. Second, our study is focused on aggregate trading while Jenters study looks at insider trading at the individual level.

In the context of Chowdury et al. (1993) ndings of market returns having a stronger effect on insider trading (i.e. insiders are contrarians) we offer the following explanation. Even though methodologically our research is similar there are three major differences of which any or all could explain the different conclusions. First, they look at insider buys and sells separately; second, they analyze weekly data and nally their sample period is 19751986. The use of weekly data imposes serious restriction on making long-term inferences about aggregate insider trading and market return. As many studies suggest, long-horizon returns differ in many ways from short horizon returns. For example, Fama and French (1988) suggest that long horizon returns are more predictable.

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Table 5 Insider trading and market returns: intensive insider trading measure. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. ITt denotes the intensive insider trading measure, Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged IT(X). t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Grangercausality test that the coefcients of all lagged insider trading are zero. The p-value is listed below the F-test in bracket. ITt1 P3 ITt2 P3 ITt3 P4 R2 P3
i1 i

1233

F-test

Table 6 Regression of market excess return and its components on insider trading for different size portfolios. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. Small rms are the lowest quintile of the sample rms market capitalization and large rms is the highest quintile. ITt denotes the insider trading measure dened in Eq. (9), Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged IT. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged insider trading are zero. The p-value is listed below the F-test in bracket.

Panel A: X t a i1 bi X ti i1 ci IT ti k1 /k Dk et Rt 0.006 0.007 0.039 0.011 0.038 (0.331) (0.186) (0.997) 0.204 0.046 0.054 0.043 0.196 Et1[Rt] (2.299) (0.313) (0.487) 0.027 0.002 0.003 0.001 0.026 NDRt (2.345) (0.087) (0.152) 0.211 0.016 0.023 0.050 0.204 NCFt (2.342) (0.100) (0.188) NEWSt 0.234 0.019 0.022 0.041 0.231 (2.319) (0.104) (0.157) P3 Xt1 Xt2 Xt3 R2 i1 bi P P P Panel B: IT t a 3 bi X ti 3 ci IT ti 4 /k Dk et i1 i1 k1 X = Rt 0.518 0.801 1.281 0.554 (1.327) (1.793) (3.126) 0.149 0.125 0.020 0.503 X = Et1[Rt] (1.353) (1.594) (0.233) X = NCFt 0.175 1.236 1.028 0.510 (0.246) (2.049) (1.685) 0.158 0.144 0.067 0.506 X = NDRt (1.377) (1.796) (0.758) 0.127 0.130 0.068 0.506 X = NEWSt (1.279) (1.810) (0.872) 2.600 0.004 2.089 0.054 0.070

3.857 (0.277) 10.339 (0.016) 9.664 (0.022) 9.922 (0.019) 10.120 (0.018) F-test

Xt a

3 X i1

bi X ti

3 X i1

ci IT ti
ITt2

4 X k1

/k Dk et
P3

ITt1 Panel A: Small rms 0.031 Rt (0.788) 0.186 Et1[Rt] (1.788) NCFt 0.015 (1.002) 0.119 NDRt (1.184) 0.133 NEWSt (1.183) Panel B: Large rms 0.044 Rt (2.350) 0.001 Et1[Rt] (0.011) 0.007 NCFt (0.412) NDRt 0.044 (0.358) 0.049 NEWSt (0.351)

ITt3

R2

i1 i

F-test

12.122 (0.007) 3.627 (0.305) 6.087 (0.107) 4.716 (0.194) 4.804 (0.187)

0.010 (0.230) 0.394 (2.166) 0.030 (1.145) 0.332 (1.906) 0.360 (1.827) 0.005 (0.239) 0.131 (1.290) 0.015 (0.970) 0.113 (1.058) 0.127 (1.051)

0.072 (2.079) 0.198 (1.225) 0.048 (2.444) 0.238 (1.554) 0.294 (1.725) 0.023 (0.926) 0.031 (0.249) 0.003 (0.165) 0.042 (0.352) 0.047 (0.350)

0.015 0.074 0.061 0.076 0.074

0.093 0.407 0.063 0.451 0.520

7.240 (0.065) 20.027 (0.000) 15.296 (0.002) 24.188 (0.000) 24.464 (0.000) 7.324 (0.062) 3.191 (0.363) 3.315 (0.346) 4.462 (0.216) 4.396 (0.222)

0.018 0.017 0.040 0.011 0.019

0.072 0.161 0.024 0.198 0.222

able investors at reasonable costs. Using this denition, high information uncertainty rms would be those rms whose expected cash ows may be difcult to estimate due to their environment or nature of operations, etc. These rms are likely to have high information acquisition costs and their fundamental values are more likely to be unreliable and volatile. If aggregate insider trading is driven by the contrarian strategy then insiders are more likely to trade in high information uncertainty rms as these are more likely to have current market values deviating from the true fundamental values. It is also true that high information uncertainty rms have greater information asymmetry which can lead to insiders exploiting their superior knowledge. If this is true then aggregate insider trading in high information uncertainty rms would predict market return. On the other hand, low information uncertainty rms are more likely to have market values equal to the fundamental values. Insider trading in these types of rms is more likely to be a manifestation of the insiders ability to predict market return based on superior knowledge rather than contrarian strategy. To the extent that small rms have high information acquisition costs and are likely to be followed by fewer analysts we use rm size as a proxy for information uncertainty. For each quarter in our insider trading sample we form size quintiles based on the market capitalization value. The rst quintile comprises of the smallest rms while the fth quintile comprises the largest rms. We repeat our earlier analyses on smaller rms and larger rms but conne it to the Management group of insiders. In Table 6, we report regression results for the two groups small rms and large rms. Realized market excess returns and its components are regressed on lagged values of IT

for the small rm and large rm samples. Panel A reports results for small rms. The F-statistics (p-values) for realized market excess return, expected market excess return, cash-ow news, discount rate news, and NEWS regressions are 7.24 (0.065), 20.027 (0.00), 15.296 (0.002), 24.188 (0.000), and 24.464 (0.000), respectively. In Panel B results for large rms are reported. Here the F-statistics (p-values) for realized returns, expected return, cash-ow news, discount rate news, and NEWS are 7.324 (0.062), 3.191 (0.363), 3.315 (0.346), 4.462 (0.216), and 4.396 (0.222), respectively. The results suggest that for both small and large rms, insider trading positively predicts the realized market excess return. In addition, it is only for small rms we nd insider trading and cashow news are positively and signicantly related suggesting that aggregate insider trading predicts future cash-ow news. Table 7 reports results when IT is regressed on lagged values of realized returns, expected returns, cash-ow news, discount rate news, and NEWS for both the small rm group and the large rm group. In Panel A, for small rms the F-statistics (p-values) for the regressions of realized returns, expected returns, cash-ow news, discount rate news, and total news are 4.436 (0.218), 1.662 (0.645), 1.927 (0.588), 0.855 (0.836), and 0.794 (0.794), respectively. In Panel B, results are reported for large rms. The F-statistics (p-values) for the regressions of realized returns, expected returns, cash-ow news, discount rate news, and total news are 4.454 (0.216), 1.966 (0.580), 1.817 (0.611), 0.845 (0.839), and

1234

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Table 8 Regression of market excess return and its components on insider trading for number of analysts following portfolios. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. Least followed rms are the lowest quintile of the sample rms number of analysts and closely followed rms are the highest quintile, ITt denotes the insider trading measure dened in Eq. (9), Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged IT. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged insider trading are zero. The p-value is listed below the F-test in bracket.

Table 7 Regression of insider trading on market excess return and its components for different size portfolios. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. Small rms are the lowest quintile of the sample rms market capitalization and large rms are the highest quintile. ITt denotes the insider trading measure dened in Eq. (9). Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged X. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged return are zero. The p-value is listed below the F-test in bracket.

IT t a

3 X i1

bi X ti

3 X i1

ci IT ti
Xt2

4 X k1

/k Dk et
P3

Xt a

3 X i1

bi X ti

3 X i1

ci IT ti
ITt2

4 X k1

/k Dk et
P3

Xt1 Panel A: Small rms 0.453 X = Rt (1.281) 0.001 X = Et1[Rt] (0.013) X = NCFt 0.119 (0.309) 0.027 X = NDRt (0.517) 0.019 X = NEWSt (0.411) Panel B: Large rms 0.163 X = Rt (0.346) 0.048 X = Et1[Rt] (0.511) 0.822 X = NCFt (1.259) 0.037 X = NDRt (0.423) X = NEWSt 0.040 (0.523)

Xt3

R2

i1 bi

F-test

ITt1

ITt3

R2

i1 i

F-test

0.468 (0.832) 0.094 (1.219) 0.459 (0.698) 0.075 (0.868) 0.065 (0.845) 0.435 (0.538) 0.040 (0.336) 0.515 (0.484) 0.067 (0.488) 0.061 (0.500)

0.137 (0.591) 0.022 (0.539) 0.203 (0.699) 0.000 (0.005) 0.003 (0.098) 0.640 (1.890) 0.080 (1.153) 0.113 (0.235) 0.037 (0.560) 0.032 (0.548)

0.353 0.349 0.340 0.342 0.342

0.784 0.073 0.781 0.048 0.049

4.436 (0.218) 1.662 (0.645) 1.927 (0.588) 0.855 (0.836) 0.794 (0.851) 4.454 (0.216) 1.966 (0.580) 1.817 (0.611) 0.845 (0.839) 0.875 (0.831)

0.270 0.248 0.253 0.245 0.246

0.367 0.073 0.195 0.067 0.053

Panel A: Least followed 0.039 0.024 Rt (1.254) (0.639) 0.082 0.110 Et1[Rt] (0.739) (0.708) NCFt 0.004 0.002 (0.208) (0.108) 0.022 0.072 NDRt (0.212) (0.490) 0.028 0.074 NEWSt (0.239) (0.452) Panel B: Closely followed 0.039 0.016 Rt (1.852) (0.847) 0.007 0.157 Et1[Rt] (0.060) (1.676) 0.006 0.018 NCFt (0.335) (1.326) 0.035 0.147 NDRt (0.284) (1.563) NEWSt 0.038 0.165 (0.279) (1.547)

0.088 (2.195) 0.335 (1.846) 0.057 (2.688) 0.358 (2.324) 0.420 (2.448) 0.018 (0.799) 0.030 (0.231) 0.002 (0.115) 0.035 (0.290) 0.040 (0.293)

0.042 0.038 0.040 0.049 0.047

0.103 0.362 0.055 0.409 0.466

9.931 (0.019) 8.520 (0.036) 8.941 (0.030) 10.337 (0.016) 10.753 (0.013) 7.443 (0.059) 4.852 (0.183) 4.081 (0.253) 6.390 (0.094) 6.209 (0.102)

0.021 0.009 0.034 0.000 0.008

0.073 0.179 0.026 0.217 0.244

0.875 (0.831). The results in Panels A and B clearly suggest that realized market excess return, expected returns, cash-ow news, discount rate news, and total news do not Granger-cause insider trading. In summary, we nd little evidence of a contrarian investment strategy employed in aggregate insider trading. To reinforce our results we use an alternate proxy for information uncertainty. We classify rms into quintiles based on the number of analysts following a rm.11 Each year we nd the number of analysts following a rm from the I/B/E/S tapes and classify rms in one of ve quintiles based on the number of analysts following the rm. Quintile 1 comprises of rms least followed by analysts and quintile 5 has rms closely followed by analysts. In Table 8, in Panel A, for least followed rms the F-statistics (p-values) for the regressions of realized returns, expected returns, cash-ow news, discount rate news, and total news are 9.931 (0.019), 8.52 (0.036), 8.941 (0.03), 10.337 (0.016), and 10.753 (0.013), respectively. In Panel B results for closely followed rms are reported. Here the F-statistics (p-values) for realized returns, expected return, cash-ow news, discount rate news, and NEWS are 7.443 (0.059), 4.852 (0.183), 4.081 (0.253), 6.390 (0.0.094), and 6.209 (0.102) The results are qualitatively similar when size is used as a proxy for information uncertainty with one exception.
11 Our choice of number of analysts as a proxy for information uncertainty is based on earlier studies. For example, Agca and Mozumdar (2008) list analyst following as one of the factors associated with market imperfections.

For rms closely followed by analysts we nd weak evidence of aggregate insider trading being related to unexpected changes in future discount rate news. Table 9 reports results when IT is regressed on lagged values of realized returns, expected returns, cash-ow news, discount rate news, and NEWS for least followed rms and closely followed rms. In Panel A, for least followed rms the F-statistics (p-values) for the regressions of realized returns, expected returns, cash-ow news, discount rate news, and total news are 8.391 (0.039), 0.808 (0.848), 5.392 (0.145), 1.136 (0.768), and 1.312 (0.726), respectively. In Panel B, results are reported for large rms. The F-statistics (p-values) for the regressions of realized returns, expected returns, cash-ow news, discount rate news, and total news are 4.850 (0.183), 1.993 (0.574), 2.557 (0.465), 1.016 (0.797), and 1.109 (0.775). For rms which are least followed by analysts, we nd that aggregate insider trading is related to prior realized market excess returns. This can be misinterpreted as evidence of contrarian investment strategy by insiders for this group of rms. However, we nd no evidence of expected returns, cash-ow news, discount rate news, and NEWS Granger-causing aggregate insider trading. In summary, we cannot nd any evidence of a contrarian investment strategy employed in aggregate insider trading when realized market excess returns are decomposed. Our results from Tables 69 conrm what we conjectured earlier regarding information uncertainty and insider trading. We nd that for rms which are high in information uncertainty (small

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Table 9 Regression of insider trading on market excess return and its components for number of analysts following portfolios. This table shows the results of the regressions between insider trading and market return (its components) over the period 1978:Q12000:Q4. Least followed rms are the lowest quintile of the sample rms number of analysts and closely followed rms are the highest quintile. ITt denotes the insider trading measure dened in Eq. (9), Rt denotes the realized market excess return, Et1[Rt] denotes the expected market excess return, NCFt denotes the cash-ow news, NDRt denotes the discount rate news, NEWS denotes the sum of cash-ow news and discount rate news and D is a dummy variable to control for seasonality. The return decomposition is based on the VAR system in Eq. (7). We report the estimates, t-statistics, adjusted R2, sum of c(b) in Panel A (B), and Granger-causality test. Note that the adjusted R2 is for the full model, whereas in the table we only report the coefcients of lagged X. t-Statistics are computed using NeweyWest heteroskedastic-robust standard errors with 5 lags, and are list below each estimate in parentheses. F-test is the Granger-causality test that the coefcients of all lagged return are zero. The p-value is listed below the F-test in bracket.

1235

IT t a

3 X i1

bi X ti

3 X i1

ci IT ti
Xt2

4 X k1

/k D k et
P3

Xt1 Panel A: Least followed 0.483 X = Rt (1.183) 0.002 X = Et1[Rt] (0.019) X = NCFt 0.029 (0.061) 0.033 X = NDRt (0.446) 0.026 X = NEWSt (0.393) Panel B: Closely followed 0.325 X = Rt (0.644) 0.055 X = Et1[Rt] (0.636) 0.914 X = NCFt (1.534) 0.052 X = NDRt (0.646) X = NEWSt 0.054 (0.756)

Xt3

R2

i1 bi

F-test

0.104 (0.187) 0.063 (0.787) 0.423 (0.559) 0.058 (0.628) 0.051 (0.615) 0.504 (0.663) 0.054 (0.434) 0.610 (0.562) 0.085 (0.599) 0.077 (0.606)

0.441 (1.849) 0.019 (0.325) 0.335 (0.845) 0.014 (0.206) 0.016 (0.274) 0.691 (2.087) 0.074 (0.972) 0.054 (0.114) 0.030 (0.419) 0.025 (0.408)

0.297 0.277 0.279 0.277 0.277

0.145 0.043 0.786 0.039 0.041

8.391 (0.039) 0.808 (0.848) 5.392 (0.145) 1.136 (0.768) 1.312 (0.726) 4.850 (0.183) 1.993 (0.574) 2.557 (0.465) 1.016 (0.797) 1.109 (0.775)

0.303 0.276 0.284 0.275 0.276

0.512 0.072 0.250 0.062 0.048

other hand, if these trades are a result of contrarian beliefs then insider trading should be negatively related to past expected return. We nd strong evidence that aggregate insider trading is positively related to unexpected cash-ow news for all types of insiders. When we partition our sample based on insiders who are more likely to have access to performance related information these results are much stronger and signicant. We also examine whether aggregate insider trading is in response to market expectations. We nd no evidence of aggregate insider trading being caused by market expectations. Our results strongly suggest that insiders are able to predict market returns because of having superior information about future cash-ow news. We also control for non-information based motives of insider trading. Using an intensive trading criterion we isolate trades which are more likely to be based on information and nd results to be consistent. To further substantiate our results we classify rms into high information uncertainty and low information uncertainty rm and use rm size and analyst following as proxies for information uncertainty. If aggregate insider trading is due to contrarian strategy, then insiders are more likely to trade in small rms (or rms with less analyst following), as these rms are likely to have high information uncertainty. On the other hand, if insiders are trading due to superior knowledge then the trades should be concentrated on larger rms as these trades are more likely related to economywide activity. However, small rms also have greater information asymmetry which can lead to insiders exploiting their superior knowledge. We nd that the predictive ability of aggregate insider trading in high information uncertainty rms is due to superior knowledge. We nd weak evidence that aggregate insider trading in these rms is associated with contrarian strategy of investment. The fact that insider trading is due to informational advantage has an important implication. Given that insider trades are driven by superior information, aggregate insider trading should be construed as a leading indicator of market-wide activities. Furthermore, such trading by insiders will drive prices towards fundamental values. Acknowledgements We appreciate valuable comments from an anonymous referee, Gordon Klein, Bong-Soo Lee, Ike Mathur (the editor), Paul Schultz, seminar participants at the University of Northern Iowa, Florida International University, Hohai University, and Hunan University, and presentation participants at the 2007 Financial Management Association Annual Conference. We alone are responsible for errors. Zaman acknowledges nancial support from the College of Business Administration, University of Northern Iowa. References

rms or fewer analysts following rms) insider trading is more likely due to managers exploiting their superior knowledge about future cash-ow news, whereas for rms which have low information uncertainty (large rms or larger number of analysts following rms) aggregate insider trading is not a manifestation of contrarian strategy or of informational advantage. 7. Conclusion Evidence from recent research on insider trading has shown that insiders are able to predict the market return either on the basis of contrarian beliefs (e.g., Rozeff and Zaman, 1998; Chowdhury et al., 1993; Lakonishok and Lee, 2001; Jenter, 2005) or on the basis of superior knowledge about future cash-ow news (e.g., Ke et al., 2003; Seyhun, 1988). Piotroski and Roulstone (2005) document that insiders trade on the basis of both contrarian beliefs and superior knowledge. In this study, we examine the ability of aggregate insider trading to predict market-wide movement using return decomposition in a vector autoregressive (VAR) model framework. We decompose market returns into expected return, unexpected cash-ow news, and unexpected discount rate news by closely following the methods outlined in Campbell (1991). Such decomposition enables us to identify the source of predictability of aggregate insider trading. We argue that if insiders are trading on the basis of superior information, then aggregate insider trading is more likely to be positively related to unexpected cash-ow news. On the

Adams, B., Carow, K., Perry, T., 2009. Earnings management and initial public offerings: The case of the depository industry. Journal of Banking and Finance 33, 23632372. Agca, S., Mozumdar, A., 2008. The impact of capital market imperfections on investment-cash ow sensitivity. Journal of Banking and Finance 32, 207216. Aktas, N., De Bodt, E., Van Oppens, V., 2008. Legal insider trading and market efciency. Journal of Banking and Finance 32, 13791392. Ang, A., Bekaert, G., 2007. Stock return predictability, is it there? Review of Financial Studies 20, 651707. Baker, M., Taliaferro, R., Wurgler, J., 2006. Predicting returns with managerial decision variables: Is there a small sample bias? Journal of Finance 61, 1711 1730. Baker, M., Wurgler, J., 2000. The equity share in new issues and aggregate stock returns. Journal of Finance 55, 22192257. Brennan, M.J., Wang, A.W., Xia, Y., 2001. A simple model of intertemporal capital asset pricing and its implications for the FamaFrench three-factor model. Working Paper, UCLA. Campbell, J.Y., 1991. A variance decomposition for stock returns. Economic Journal 101, 157179. Campbell, J.Y., Vuolteenaho, T., 2004. Bad beta, good beta. American Economic Review 94, 12491275.

1236

X. Jiang, M.A. Zaman / Journal of Banking & Finance 34 (2010) 12251236 Jiang, G., Lee, C., Zhang, G., 2004. Information uncertainty and expected returns. Review of Accounting Studies 10, 185221. Ke, B., Huddart, S., Petroni, K., 2003. What insiders know about future earnings and how they use it: Evidence from insider trades. Journal of Accounting and Economics 35, 315346. Lakonishok, J., Lee, I., 2001. Are insider trades informative? Review of Financial Studies 14, 79111. La Porta, R., Lakonishok, J., Shleifer, A., Vishny, R., 1997. Good news for value stock: Further evidence on market efciency. Journal of Finance 52, 859874. Lewellen, J.W., 2004. Predicting returns with nancial ratios. Journal of Financial Economics 74, 209235. Lorie, J.H., Niederhoffer, V., 1968. Predictive and statistical properties of insider trading. Journal of Law and Economics 11, 3551. Loughran, T., Ritter, J., 1995. The new issues puzzle. Journal of Finance 50, 2351. Piotroski, J.D., Roulstone, D., 2005. Do insider trades reect both contrarian beliefs and superior knowledge about future cash ow realization? Journal of Accounting and Economics 39, 5581. Rozeff, M., Zaman, M., 1988. Market efciency and insider trading: New evidence. Journal of Business 61, 2544. Rozeff, M., Zaman, M., 1998. Overreaction and insider trading: Evidence from growth and value portfolios. Journal of Finance 53, 701716. Seyhun, H.N., 1986. Insiders prots, cost of trading, and market efciency. Journal of Financial Economics 16, 189212. Seyhun, H.N., 1988. The information content of aggregate insider trading. Journal of Business 61, 124. Seyhun, H.N., 1992. Why does aggregate insider trading predict future stock returns? Quarterly Journal of Economics 107, 13031331. Shiller, R.J., 1984. Stock prices and social dynamics. Brookings Papers on Economic Activity, 457498. Stambaugh, R.F., 1999. Predictive regressions. Journal of Financial Economics 54, 375421. Torous, W., Volkanov, R., Yan, S., 2004. On predicting returns with nearly integrated explanatory variables. Journal of Business 77, 937966. Wang, F., 2010. Informed arbitrage with speculative noise trading. Journal of Banking and Finance 34, 304313.

Campbell, J., Yogo, M., 2006. Efcient tests of stock return predictability. Journal of Financial Economics 81, 2760. Chan, K., Ikenberry, D., Lee, I., 2007. Do managers time the market? Evidence from open-market share repurchases. Journal of Banking and Finance 31, 26732694. Chen, L., Zhao, X., 2009. Return decomposition. Review of Financial Studies 22, 52135249. Chowdhury, M., Howe, J.S., Lin, J.C., 1993. The relation between aggregate insider transactions and stock market return. Journal of Financial and Quantitative Analysis 28, 431437. Cohen, R., Gamers, P., Voulteenaho, T., 2002. Who under reacts to cash-ow news? Evidence from trading between individuals and institutions. Journal of Financial Economics 66, 409506. DeLong, B., Shleifer, A., Summers, L., Waldmann, R., 1990. Positive feedback investment strategies and destabilizing rational speculation. Journal of Finance 45, 374397. Eckbo, B.E., Smith, D., 1998. The conditional performance of insider trades. Journal of Finance 53, 467498. Eleswarapu, V.R., Reinganum, M.R., 2004. The predictability of aggregate stock market returns: Evidence based on glamour stocks. Journal of Business 77, 275294. Fama, E., French, K., 1988. Dividend yields and expected stock returns. Journal of Financial Economics 22, 325. Hecht, P., Vuolteenaho, T., 2006. Explaining returns with cash-ows proxies. Review of Financial Studies 19, 159194. Ginglinger, E., Hamon, J., 2007. Actual share repurchases, timing and liquidity. Journal of Banking and Finance 31, 915938. Givoly, D., Palmon, D., 1985. Insider trading and the exploitation of inside information: Some empirical evidence. Journal of Business 58, 6987. Ikenberry, D., Lakonishok, J., Vermaelen, T., 1995. Market under reaction to open market shares repurchase. Journal of Financial Economics 39, 181208. Jaffe, J.F., 1974. Special information and insider trading. Journal of Business 47, 410 428. Jeng, L., Metrick, A., Zeckhauser, R., 2003. Estimating the returns to insider trading: A performance-evaluation perspective. Review of Economics and Statistics 85, 453471. Jenter, D., 2005. Market timing and managerial portfolio decisions. Journal of Finance 60, 19031949.