Trading incentives to meet the analyst forecast

The inability to measure a manager’s intent or “scienter” directly is well known in the legal literature (Bainbridge, 2000). Economists (and lawyers) therefore use the concept of “revealed preference” to infer intent from various patterns in observed data (Kreps, 1990, Chapter 2). From an institutional perspective, the only ex ante disclosure that managers have to file is Form 144, in which they document their intent to sell. However, managers are not obligated to sell all the shares they say they will, so most managers just keep a large open balance of shares on Form 144. Therefore, we cannot use Form 144 as a measure of the intent to sell.

Durtschi and Easton (2005) has the greatest relevance in our setting, as our focus is on the zero analyst forecast error discontinuity. They argue that the kink around zero analyst forecast error is largely driven by analyst optimism versus pessimism—essentially that when analysts miss, they miss big, creating a dearth of observations in the just missed region that is not necessarily due to earnings management. Their conjecture does not prohibit our examination of this region. The shortage of firms in this region should only serve to weaken the power of our tests, as we examine whether managers who would have just missed exert additional effort to jump the hurdle when they plan to sell their shares. Our claim is not that the discontinuity around the zero analyst forecast is exclusively due to earnings management, but that trading incentives accentuate this kink.

In related work, Ke (2004) examines the prior earnings threshold and finds that managers with more equity incentives are more likely to manage earnings to report longer earnings strings. Ke (2004) also finds that these managers sell significant amounts of stock in the two to six quarters prior to a break in a string of consecutive earnings increases. Cheng and Warfield (2005) also examine the role of managerial motives in the context of the analyst forecast error discontinuity. They find that managers with high levels of equity incentives are more likely to have met the analyst forecast.

These empirical studies are validated by theoretical work that shows that stock prices are extraordinarily sensitive to small events such as firms just missing earnings thresholds. This theoretical result obtains both when a) investors have behavioral biases (e.g., Rabin, 2002), and (b) when investors are rational, but have incomplete information about the firm’s future prospects (Hart & Kreps, 1986; Stein, 1987). The intuition in Hart and Kreps (1986) is that speculators are always in search of the “the next big stock,” resulting in extreme volatility in price relative to small changes in fundamentals.

We realize that in a more complex setting, a rational forward-looking manager will conduct a long-range optimization, weighing the impact of managing earnings today on future period earnings and future period stock incentive dynamics. We abstract from this complex setting and focus on a single period setting (see Bolton, Scheinkman, & Xiong, 2006 for a multi-period model).

To shed light on the possibility of price-drop-based penalties, we examine the subsequent returns following the trades by management, beginning one day following the managerial trade and ending one day before the manager files the trade with the SEC. The subsequent market-adjusted returns following managerial sales, though significantly lower than for those where the manager purchased shares, remain positive. This finding is consistent with Bartov et al. (2002)—that regardless of how the benchmark is met, there is a premium to meeting the benchmark. This premium alleviates concerns about price-drop-based insider trading penalties.

Forecast management is the lowering of the analyst forecast to a beatable level through guidance by managers (e.g., Bartov et al., 2002; Cotter, Tuna, & Wysocki, 2004; Matsumoto, 2002; Richardson et al., 2004).

The quarterly consensus analyst forecast is the median EPS forecast computed over the set of the analysts’ most recent forecasts that are no earlier than two months before the quarterly earnings release date. This procedure avoids the problem of stale analyst forecasts. We use the unadjusted I/B/E/S forecasts to avoid losing the precision in the decimal places of the forecasts due to the I/B/E/S adjustments of prior forecasts for subsequent stock splits (Baber & Kang, 2002; Payne & Thomas, 2003). Actual earnings realizations are also obtained from I/B/E/S. Note that stock splits are not an issue for insider stock sales because we scale this measure by contemporaneous insider stock holdings.

Also note that our tests include all firm-quarters in which the firm just met or just missed the analyst forecast. We do not use a matched sample approach, which Palepu (1986) argues biases the results by distorting the baseline proportions of the treatment firms.

The use of realized outcomes as a proxy for intent is widespread in studies that examine earnings patterns prior to an event such as stock issuance, option grants and cash pay (e.g., Aboody & Kasznik, 2000; Rangan, 1998; Teoh, Welch & Wong, 1998). In all these cases, when managers take the action prior to the event, they are anticipating that the event of interest will subsequently happen. Like our study, these studies use the actual realization of the event in their empirical tests, not the managers’ ex ante anticipation.

Ke et al. (2003) use a similar trading window (see pp. 322–323). Results are not sensitive to using the entire quarter, which is reasonable, because insiders typically do not sell prior to earnings releases. In our sample, 92% of all manager insider trades in a quarter (and 88% of non-manager insider trades) occur after the earnings announcement in that quarter.

The officer shareholdings listed on the SEC filing of a trade can sometimes be very small. To mitigate the generation of outliers due to this small denominator effect, we add back the shares traded in a transaction to the corresponding ownership level figure for all observations. We also exclude option exercises, because the conversion of an option to a share is not really a true purchase. Of course, we include all sales of such shares. Finally, note that we ignore option holdings in the denominator, because the insider trading data do not include option holdings. Substantiating the validity of our metric, alternative insider trading measures such as the square root of the dollar value of insider sales (Noe, 1999, p. 311), and the Net Shares Traded metric (Beneish & Vargus, 2002, p. 761) are correlated with our metric at .86 and .94, respectively.

Hribar and Collins (2002) prescribe the cash flow method to calculate accruals. However, using quarterly data, the cash flow method for our sample period results in a reduction of more than half of the sample. We do, however, replicate our results on this subset of the sample and note our results in Sect. 3.2.

Skinner and Sloan (2002) find that high-growth firms have a more negative stock price reaction to missing versus meeting the analyst forecast than non-high-growth firms. Consistent with these firms having a greater incentive to meet the analyst forecast, we find that growth, as a control variable, is positively associated with just meeting versus just missing the analyst forecast. However, it is also possible that there is an interactive effect. To explore this, we interact I/B/E/S expected long-term growth and insider sales. The sign on the interaction term is positive, as expected, but only marginally significant (p-value = .109; not tabulated).

As noted in Sect. 2.2.3, Hribar and Collins (2002) prescribe the cash flow method to calculate accruals. However, using quarterly data, the cash flow method for our sample period results in a reduction of more than half of the sample (from 11,939 in Table 4 to 5,204 firm-quarter observations). We replicate our results on this subset of the sample and results are similar, though weaker (.018; p-value = .196 vs. .015; p-value = .076 in Table 4). This appears to be due to low power, as the main effect on managerial sales is also very weak (.035; p-value = .240 vs. .035; p-value = .017 in Table 4).

As an additional robustness check (not tabulated), we re-estimate Eq. 3 and include the ranked interaction of DWC _q with the quarter q (lagged) insider sales measure. The addition of this new interaction term does not change the results and the new interaction term itself is not significant, suggesting that our results are not simply an artifact of firms that have more insider sales in general. Rather, the timing of insider sales matters.

Note that these results are not consistent with those in Richardson et al. (2004). However, we focus only on those firms that just met the analyst forecast, rather than all firms that met the analyst forecast. Furthermore, Richardson et al. (2004) exclude all observations that met the analyst forecast but began the period with a pessimistic forecast. If we regress meeting the analyst forecast versus missing the analyst forecast for the entire earnings forecast region on the interaction of the unexpected forecast and managerial insider sales, we obtain a positive and significant coefficient on the interaction term, consistent with Richardson et al. (2004).

Clearly, there are many differences between managerial and non-managerial insiders beyond their ability to affect earnings outcomes. This test alone does not provide definitive evidence that managers are actively meeting the analyst forecast before selling shares, but instead complements the findings in Hypothesis 1 and Hypothesis 2, adding to our aggregate evidence on active earnings or forecast management by managerial insiders to meet the analyst forecast.

The existence of a majority of outside directors may mitigate this opportunistic behavior through various mechanisms. For example, these firms may be subject to greater accounting scrutiny (e.g., Klein, 2002) or have additional restrictions on insider trading (Seyhun, 1998). We do not explore the mechanism by which outside directors mitigate this opportunistic behavior.

Because the comparable prior period earnings may have been managed, Dopuch et al. (2003) use a time-series model to arrive at a proxy for prior period earnings. This estimation requires at least 16 quarters of data to estimate. For simplicity and to maximize our sample size, we simply use one year ago quarterly earnings.