Short-horizon incentives and stock price inflation

Journal of Corporate Finance xxx (xxxx) xxxx

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

Short-horizon incentives and stock price inflation Jianxin Daniel Chia, Manu Guptab, Shane A. Johnsonc,

⁎

a

University of Nevada, Las Vegas, USA Virginia Commonwealth University, USA c Texas A&M University, USA b

ARTICLE INFO

ABSTRACT

Keywords: Executive compensation Incentives Stock returns Insider trading Earnings management

Do managerial incentive horizons have capital market consequences? We find that they do when short-sale constraints are more binding. Firms experience significant stock price inflation when their CEOs have short horizon incentives. The short-horizon CEOs sell more shares at inflated prices and generate greater abnormal trading profits. The stock price inflation is partly explained by greater earnings surprises and more positive investor reaction to the surprises. To inflate stock prices, short-horizon firms are more likely to employ income-increasing discretionary accruals. Consistent with theoretical predictions, all these effects are attenuated or statistically insignificant when short-sale constraints are less binding.

JEL classifications: M52 G14 G34

1. Introduction The literature has long recognized that short-horizon managers are prone to myopic behaviors (e.g., Dechow and Sloan, 1991). It is not clear, however, even theoretically, whether these behaviors have any material capital market consequences. For example, in the signal-jamming model of Stein (1989), short-horizon managers attempt to inflate stock price, but they are unable to because in an efficient market investors already anticipate and discount these behaviors. What if the market is less than fully efficient? Bolton et al. (2006) model a speculative market with optimistic investors and short-sale constraints, and show that short-horizon managers will attempt to inflate stock price and can succeed in doing so.1 Thus, the unsettled question in the literature is not whether short-horizon managers will attempt to inflate stock prices, which they clearly have incentive to do, but rather whether the attempts will succeed. In addition, through what channels can short-horizon CEOs successfully inflate stock price? And finally, do these CEOs personally gain from any stock price inflation? To answer these questions, we first need to overcome the challenging task of estimating incentive horizons for CEOs. Imagine a CEO with overlapping equity grants awarded at different points in time with different vesting schedules, and that over time, the CEO sells vested shares from different grants at different times. Thus, the CEO's incentive horizon is a function of vesting schedules of overlapping grants and the CEO's intertemporal sale decisions about vested grants. We develop an algorithm that estimates a dynamic incentive horizon measure that reflects this complexity. For any given year, our algorithm can estimate the incentive horizon of any executive in the ExecuComp database, and thus enables us to study capital market implications of CEO incentive horizons for a broad

Corresponding author. E-mail addresses: [email protected] (J.D. Chi), [email protected] (M. Gupta), [email protected] (S.A. Johnson). 1 Also see the theoretical analysis in Peng and Röell (2014) and Goldman and Slezak (2006) on how short incentive horizon can lead to managerial manipulation to inflate stock price. ⁎

https://doi.org/10.1016/j.jcorpfin.2019.101501 Received 27 April 2018; Received in revised form 3 August 2018; Accepted 18 August 2019 0929-1199/ © 2019 Elsevier B.V. All rights reserved.

Please cite this article as: Jianxin Daniel Chi, Manu Gupta and Shane A. Johnson, Journal of Corporate Finance, https://doi.org/10.1016/j.jcorpfin.2019.101501

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

sample of firms. Theory suggests that to detect stock price inflation, our empirical tests should be conditional on short-sale constraints. In Stein (1989), the stock market is efficient, and investors can fully discount managerial manipulation. In Bolton et al. (2006), short-sale constraints prevent arbitrageurs from fully discounting managerial manipulation, so that stock price inflation occurs. Thus, we condition our tests on the level of short-sale constraints, and hereafter focus our discussion on firms with high short-sale constraints.2 We find that firms exhibit significant stock price inflation as their CEOs' incentive horizon becomes shorter. During the 12 months prior to CEO incentive horizon dropping below 1 year, stocks of firms with short-horizon CEOs outperform those with long-horizon CEOs by 37.3 basis points monthly, or 4.48% annually. The inflated price reverse slightly in the first half of the short-horizon year. In the second half, however, the inflation reversal is stronger – short incentive horizon firms underperform those with long horizon by 52.7 basis points per month. Over the longer term from months +13 to +36, short incentive horizon firms underperform long incentive horizon firms by 35.5 basis points per month, or 4.26% per year. In Fig. 1, we plot cumulative alphas of the arbitrage portfolio for firms with high short-sale constraints. We form the arbitrage portfolio by buying firms with short-horizon CEOs and shorting firms with long-horizon CEOs. The positive abnormal returns and the later reversal are consistent with the hypothesis that short-horizon CEOs successfully inflate stock price when they have more vested equity incentives to sell. Given the evidence of short-horizon CEO incentives leading to stock price inflation, we then examine whether short-horizon CEOs exploit the stock price inflation. We employ a propensity score approach to match firms with short-horizon CEOs to firms with longhorizon CEOs on multiple characteristics that potentially affect both incentive horizon (the treatment variable) and insider trading activity (the outcome variable). Of particular importance, we match on the level of equity incentives to isolate the effect of the horizon of incentives. One advantage of a matching procedure over a regression specification is to allow the control variables to enter the empirical model through a less restrictive non-parametric functional form, and thus provide more effective control and sharper identification.3 We find that short-horizon CEOs sell significantly more shares in the short-horizon year than long-horizon CEOs do. A key question, however, is whether these CEOs earn abnormal profits through such sales. We find that they do. Short-horizon CEOs earn significantly positive abnormal trading profits. Given our evidence that stock price inflation coincides with the periods that CEOs have short horizon incentives, we next ask whether and what short-horizon CEOs do to inflate stock price. The literature has documented a myriad of myopic actions that a short-horizon CEO can potentially take,4 but the ultimate objective is to convince investors that the firm should be valued higher. No other corporate events that occur on a regular basis are more information intensive, attention attracting, and therefore value relevant than earnings announcements (e.g., see Beaver, 1968; Landsman and Maydew, 2002; and Bartov et al., 2002). Engelberg et al. (2018) study 97 stock-return anomalies and find that abnormal returns on earnings announcement days account for nearly all of the abnormal returns in those anomalies. Earnings announcements are often accompanied by conference calls with investors, which give CEOs an opportunity to influence investors' interpretation of the earnings and the company's future prospects. If a CEO intends to influence investors' valuation of the firm's stock, earnings announcements should be an opportune venue to do so. Thus, we examine whether short-horizon firms provide greater earnings surprises and whether investors react differently to these earnings surprises. We find that during the stock price inflation period, short-horizon firms have significantly larger earnings surprises than matched long-horizon firms. Moreover, we find that investors react more strongly to earnings surprises for short-horizon firms than for longhorizon firms. The larger surprise and larger investor response partly explain the stock price inflation for short-horizon firms. We also find marginally significant evidence that short-horizon firms' stocks have more negative drift after earnings announcements, which is consistent with the interpretation that their positive earnings announcement CARs represent over-reaction. A next natural question is whether the positive earnings surprises of short-horizon firms reflect true fundamentals or are they accomplished through managing discretionary accruals. The relation between CEO incentive level and earnings management has been widely studied, but the literature has yet to reach a consensus on whether there is a reliable relation (e.g., Armstrong et al., 2010). Gopalan et al. (2014) find that pay duration is negatively related to discretionary accruals, but they do not examine the effect on stock valuation. Again here, we employ a propensity score approach that matches firms with short-horizon CEOs with firms with longhorizon CEOs on many different dimensions, including the level of equity incentives. We find strong evidence that short-horizon firms exhibit abnormal discretionary accruals prior to the short-horizon year as well as during the short-horizon year. Armstrong et al. (2010) discuss hidden bias and the resulting endogeneity challenge when testing the effect of CEO incentives on earnings management. They highlight the benefits of non-parametric matching tests over parametric regression tests and employ a diagnostic tool for potential endogeneity – the sensitivity analyses developed by Rosenbaum (2002). Rosenbaum (2011) introduces new sensitivity tests that substantially increase the power over the tests in Rosenbaum (2002). Based on Rosenbaum's (2011) sensitivity tests, we conclude that the relation between CEO incentive horizon and income-increasing discretionary accruals is unlikely to 2 Although we do not discuss the results in this introduction section, we report and discuss later that, as expected, for low short-sale constraints firms, the magnitude and statistical significance of abnormal stock returns are much lower. Results for other tests in the paper are generally insignificant for firms with low short-sale constraints. 3 Armstrong et al. (2010) offer an excellent discussion on why a matching procedure can provide sharper identification than a linear regression test. In essence, including control variables in a regression per se does not address the fact that the treated group may have very different characteristics from the control group when control variables have poor distributional overlap, e.g., see Heckman et al. (1998) and Dehejia and Wahba (2002). In comparison, a matching procedure can match the treated group and the control group on multiple dimensions and to ensure that along each of the dimension, the treated and control observations have identical distributions. As a result, the researcher has higher confidence to infer causal effect of the treatment variable because the treated group and the control group are identical except for the treatment variable. 4 For example, see Bhojraj et al. (2009), Hribar et al. (2006), and Bergstresser et al. (2006).

2

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Cumula ve Arbitrage Alphas 6.00% 4.00%

Cumula ve Arbitrage Alphas

2.00% 0.00% -12

0

12

24

36

-2.00% -4.00% -6.00% -8.00% -10.00% -12.00% -14.00%

Event Months Fig. 1. Cumulative arbitrage alphas for high short-sale constraint (SSC) portfolios formed on CEO incentive horizon. This figure presents cumulative arbitrage alphas from a calendar-time portfolio of high SSC firms formed on CEO incentive horizon. (The portfolios are formed in calendar time – the plot in event time is for illustrative purpose only.) We use results reported in Table 2 Panel A to create cumulative arbitrage alphas. Each month we assign firms to one of the four portfolios based on long-or-short CEO incentive horizon and high-or-low firm SSC, both measured at the beginning of the year. We require at least ten observations per portfolio. We use a four-factor model, which includes the three factors from Fama and French (1993) and the momentum factor from Carhart (1997), and calculate abnormal returns from equally-weighted portfolios. We create an arbitrage portfolio by taking a long position in firms with short-horizon CEOs and a short position in firms with long-horizon CEOs. The figure shows that compared to firms with long-horizon CEOs, firms with short-horizon CEOs experience significant stock price inflation followed by reversal. All variables are described in Table 1. The sample period is from 1992 to 2014.

be driven by endogeneity. Taken together, the empirical results show that short-horizon incentives lead to stock price inflation, that CEOs take advantage of the inflation to sell more shares and generate greater profits, and that beating earnings expectations and managing discretionary accruals are important tools to achieve and sustain the price inflation. Interestingly, these empirical findings are conditional on a higher level of short-sale constraints. When short-sale constraints are low, stock price inflation is greatly attenuated, and other tests become generally insignificant. The contrast reflects precisely the different theoretical predictions in Stein (1989) and Bolton et al. (2006) that arise from different assumptions of market efficiency, specifically whether short-sale constraints are binding. As an essential tool to mitigate agency problems, equity-based incentives comprise two key dimensions: incentive size and incentive horizon. That is, how much incentives to provide and how soon to allow the agent to cash out the incentives. While the literature on incentive size is extensive,5 significantly less is understood about incentive horizon. One hindrance to a better understanding is that incentive horizon is not readily observable. Earlier studies on managerial horizon and its effects often focus on retiring CEOs because horizons for younger CEOs are difficult to measure (e.g., Dechow and Sloan, 1991; Cassell et al., 2013). Xu (2013) hand-collects CEO contract terms and expands the horizon study to younger CEOs. Contributing to this literature, our algorithm of estimating incentive horizon can be easily implemented for all executives in ExecuComp and provides a dynamic horizon measure that incorporates all forms of equity incentives (stock, option, vested, unvested, newly granted, and previously granted incentives). Our incentive horizon measure enables us to study a large panel of firms and document a significant stock valuation effect of executive incentive horizon. As mentioned earlier, it is not surprising that CEOs desire higher stock prices before selling their stocks. For example, Brockman et al. (2010) suggest that firms release more good news before CEOs sell their exercised stock options. However, manipulating the timing of news releases does not necessarily lead to stock overvaluation. By documenting stock price inflation and the subsequent reversal, we provide sharper evidence of incentive-horizon-induced overvaluation, and thus extend Brockman et al. (2010). Furthermore, we provide the first empirical evidence that short-horizon CEOs successfully exploit the overvaluation through selling stocks at inflated prices. We also identify beating earnings expectations and managing accruals as important tools in achieving and sustaining the overvaluation. The evidence illustrates how the temporal structure of managerial incentives can have consequential capital market effects, and complements the literature that studies various economic outcomes of managerial horizon and incentive structures (e.g., Dechow and Sloan, 1991; Edmans et al., 2017). We describe the incentive horizon measure in Section 2 and formulate our hypotheses in Section 3. We then describe our data and methods in Section 4, present the results in Section 5, and conclude in Section 6.

5

This literature is too large to survey here, see e.g., Cheng and Warfield (2005) and Nagar et al. (2003). 3

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

2. Incentive horizon measure The literature has long recognized that managerial horizon is an important element of agency conflicts. Earlier papers on managerial short-termism or myopia measure horizon with CEO age (Dechow and Sloan, 1991), contract terms (Xu, 2013), or holdings of restricted stocks (Johnson et al., 2009). Recent studies utilize grant-level vesting schedule data to measure horizon (e.g., Cadman and Sunder, 2014; Gopalan et al., 2014; Egger and Radulescu, 2014). However, as mentioned earlier, a manager's incentive horizon is determined by new grants and existing grants, as well as the vesting schedules and exercising or sale decisions on previously awarded and vested grants. Thus, while identifying the vesting schedule of a new grant is fairly straightforward, keeping track of previous grants and sale decisions on previous grants are tricky and can require grant and sale data going back many years. Plus, it is difficult to track when an executive sells stocks and from which grant he sells, thus complicating the tracking of vesting schedules of overlapping grants. Our first objective is to develop a comprehensive horizon measure that encompasses new and existing grants and previous sale decisions. Since our following tests will focus on CEOs, we describe the horizon measure below in the context of a CEO, but note that the measure can be computed for any executive in the ExecuComp database. We need a measure that captures the horizons of CEOs' vested stock and stock options, which technically have a horizon of zero, and their unvested stock and stock options, which may have varying horizons depending on the original vesting schedule and when they were granted. The ideal measure would also capture the relative size of the incentives at each horizon. Unfortunately, there is not a machine readable database containing such a measure for a large sample of firms. Thus, we construct an approximation of the CEO incentive horizon based on a refined version of the algorithm first developed by Chi and Johnson (2009) and Chi et al. (2011).6 We describe details of the algorithm next. We infer the vesting period for restricted stock as follows. For each CEO for each firm-year, we use the data in ExecuComp to calculate the number of restricted shares that vest in each of the subsequent 3 years. The number of restricted shares that vest in a particular year is computed by the accounting identity as the number of restricted shares that the executive had at the prior year-end, plus newly granted restricted shares, and minus the number of restricted shares at current year-end. We then compute a timeweighted average of the numbers of shares that vested across the 3 years. For example, the proportion of shares that vests in year one is multiplied by one; the proportion that vests in year two is multiplied by two, and so on. Remaining shares not vested by year three are assumed to vest in year four. The computation produces a horizon measure in units of years. We adjust for stock dividends and stock splits in this calculation. This algorithm approximates the number of years that it takes for the initial unvested holdings of shares in a given firm-year to vest. If no shares vest during the 3 years, we assume somewhat arbitrarily that the vesting horizon is 4 years. The censoring at 4 years should help to capture cases when incentives are subject to cliff vesting because Cadman et al. (2013) find that only 1% of grants cliff vest beyond 5 years. We perform a parallel computation for unvested stock options to approximate the vesting horizon for stock options.7 A numerical example illustrates the algorithm. Suppose CEO A has 300 shares of restricted stock at year 0, and 100 shares vest at the end of each of the next 3 years. CEO B also has 300 restricted shares, but the 300 shares vest all at once at the end of year 3. Even though both CEOs have all their shares vest in 3 years, CEO A's effective incentive horizon is shorter than CEO B's. Our algorithm captures the difference in the two effective incentive horizons – the estimated incentive horizon for CEO A is 2 years (computed as 1*100/300 + 2*100/300 + 3*100/300), and for CEO B is 3 years (computed as 1*0/300 + 2*0/300 + 3*300/300). In practice, some firms specify performance vesting, which is when the vesting of stock options or shares depends on achieving specified accounting-based or other targets (see e.g., Bettis et al., 2010, 2016). Performance vesting is an important issue because there could be reverse causality between vesting and the outcome variables that we are examining. For example, a CEO could inflate the stock price to a threshold that triggers future vesting. We would observe a relation between short CEO incentive horizon and the abnormal stock return preceding the short-horizon period, but the causality would run from stock return to incentive horizon instead of vice versa. For firms that employ performance vesting, there is no obvious way to compute an ex ante vesting horizon based on the vesting rules they specify. Our algorithm determines the vesting horizon based on how many shares actually vest over time for each executive. Thus, our algorithm should produce reasonable ex ante estimates of performance-based vesting horizons under the assumption that managers and investors have unbiased expectations about the relevant future performance outcomes that determine the vesting. When presenting the empirical results in a later section, we show the robustness of the results to performance vesting. Unrestricted stock holdings and vested stock options technically have vesting horizon lengths of zero. We recognize that implicit or explicit expectations by a board of directors may prevent a CEO from selling her entire unrestricted stock or vested option holdings even though those holdings have no vesting restriction.8 Indeed, across the CEOs in our sample, fewer than 1% of them sell unrestricted stock and vested option holdings down to zero during their employment periods. We use the observed minimum incentives over a CEO's time series as an estimate of the minimum level of incentives the CEO is expected to hold throughout her tenure. Instead of assuming an incentive horizon of 0 year for these minimum levels of vested incentives, we assume that these incentives have a

6 Note that we need the horizon of incentives from new stock and stock option grants, as well as the horizon of incentives from existing stock and stock option holdings held by the CEO in a particular year. While data for the former are in firms' Form 4 filings, data on the latter are not. 7 If a CEO leaves the company within three years, her unvested incentives may cliff vest. This may produce an ex post horizon measure that is shorter than the ex ante horizon. For this reason, we exclude all firm-year observations that see the CEO departing within the following three years. This exclusion also means that our findings are different from and complement the earlier findings on retiring CEOs (e.g., Dechow and Sloan, 1991; Cassell et al., 2013). 8 Cai and Vijh (2007) present several arguments for why managers cannot freely sell even unrestricted shares.

4

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

horizon of 4 years. Vested stock and stock options above this minimum level are assumed to be able to be sold at the CEO's discretion, so we assume a horizon of 0 year for them.9 Our measure captures the vesting horizon length going forward from each firm-year. The measure also explicitly incorporates differences in granting behavior across firms and differences in the prior exercise and stock sales behavior across CEOs. With the horizons from each source of incentives in hand, we combine them into weighted measures that reflect the relative magnitudes of the incentives from each source. The weighted measure should capture the overall time horizon element of the incentives that a CEO faces. For example, a CEO may have incentives that take a long time to vest, but those incentives are small in magnitude compared to her short-term incentives. This CEO faces very different incentives to boost short-term stock prices compared to a CEO with relatively small short-term incentives but relatively large long-term incentives. To capture such differences, we compute weighted incentive horizon measures based on the relative magnitudes of each source of incentives. We use stock and stock option deltas to measure the magnitudes of incentives. Appendix A describes the details of calculating deltas. For each CEO, we compute the overall weighted incentive horizon as the sum of (1) the restricted stock horizon (in years) multiplied by the proportion of total delta that is provided by restricted stock; (2) the unvested stock option horizon (in years) multiplied by the proportion of total delta that is provided by unvested stock option; (3) 4 years times the proportion of total delta that the time-series minimum represents (i.e., 4 years times those incentives that we assume the CEO must hold even though they are vested); and (4) 0 year times the proportion of total delta provided by the vested stock and stock options above the time-series minimum level. Finally, we define incentive horizon less than 1 year as “short” and greater than 2 years as “long.” We therefore exclude firm-years where a CEO's incentive horizon is between 1 and 2 years to achieve greater dispersion in horizon. The incentive horizon measure that we compute omits the effects of bonuses, which are frequently tied to accounting figures. While such bonuses may also induce CEOs to manipulate accounting figures, the bonuses do not depend upon fooling investors into overvaluing a firm. Although we do not see a clear way to incorporate a bonus horizon into our weighted incentive horizon measure, we can report that all of our results are robust to controlling for bonus scaled by total compensation (ExecuComp variables bonus/ TDC2). 3. Hypotheses development Several theoretical models formalize the link between short-horizon incentives and stock price inflation. In Goldman and Slezak (2006) and Peng and Röell (2014), CEOs who face relatively shorter incentive horizons are more likely to attempt strategies designed to boost stock prices artificially in the short run. Bolton et al. (2006) provide similar arguments, but in their model current shareholders deliberately structure the horizon of CEO incentives to induce managers to pursue such strategies. They model a firm's stock price as having a long-term fundamental value component and a short-term speculative component. CEOs can pursue strategies that fool a subset of investors into believing that firm value is higher than its true long-term value. If short-sale constraints bind, the trading actions of the fooled investors lead to an increase in the short-term speculative component of the stock price. Investors eventually realize that the stock is overvalued (or short-sale constraints relax), and correction occurs as the stock price falls. Because current shareholders value the option to sell their stock to the more optimistic investors in the near term at an inflated price, they optimally incentivize CEOs to undertake strategies to inflate prices in the short run by giving them relatively more short-horizon incentives. While these models differ from each other on various dimensions, a common prediction is that CEOs with short incentive horizons are more likely to attempt to inflate stock prices. If CEOs are successful in their inflation attempts, we should observe two outcomes: 1) stock price inflation and reversal; and 2) CEOs exploit the inflation by selling more stock and selling it at overvalued prices. This leads to the first two hypotheses we test, stated in alternative form: H1. Firms whose CEOs have short-horizon incentives exhibit evidence of share price inflation (positive abnormal returns) and eventual correction (negative abnormal returns) around the short-horizon period. These effects should obtain only among stocks with high short-sale constraints because these constraints limit downward pressure on the stock by arbitrageurs. H2. CEOs who have short-horizon incentives sell significantly more stock and make positive abnormal profits than comparable longhorizon CEOs do. These two hypotheses focus on the potential outcomes of share price inflation and measure stock returns over relatively long periods. As such, the hypotheses do not depend on identifying specific actions or information releases that drive share price inflation. Earnings releases are arguably one of the most value-relevant disclosures that firms make on a routine basis, and are therefore plausible drivers of stock price inflation. Engelberg et al. (2018) find that earnings-announcement-day returns account for nearly all of the abnormal returns in 97 known anomalies. Thus, we formulate hypotheses about (1) the magnitude of earnings surprises and (2) the magnitude of stock price responses to earnings surprises, stated in alternative form as: 9 It is possible that CEOs who hold stock above the time-series minimum simply prefer to hold equity long term, which is opposite of our assumption about it having a horizon of zero. To investigate the likelihood of this, we regress future (year-ahead) stock sales on the proportion of incentives from vested stock and options above the time-series minimum, the proportion of incentives from unvested options, and the proportion of incentives from unvested (restricted) stock. We find that future stock sales are positively related (p-value < .05) to the proportion of incentives provided by vested stock and options above the time-series minimum, which validates our assumption that these incentives have a short horizon.

5

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

H3. Firms whose CEOs have short-horizon incentives provide greater earnings surprises than matched long-horizon firms do. As in H1, these effects should obtain only among firms with high short-sale constraints. H4. Firms whose CEOs have short-horizon incentives exhibit greater stock price responses to earnings surprises than matched longhorizon firms. As in H1, these effects should obtain only among firms with high short-sale constraints. Sloan (1996) finds that investors misprice the accrual component of reported earnings, and Beneish and Vargus (2002) find that the mispricing occurs primarily for income-increasing accruals. Investors appear to interpret the income-increasing accruals to imply higher future earnings even though the accruals ultimately reverse and reduce earnings. Various studies show that managers seem to be able to fool investors through accruals management, e.g., Sloan (1996) and Teoh et al. (1998). Thus, accruals management can be important tool for short-horizon managers to inflate stock prices. Stated in alternative form, we hypothesize that: H5. Firms with short-horizon CEOs are more likely to employ income-increasing discretionary accruals, and these accrualsmanagement efforts should concentrate among firms with high short-sale constraints. 4. Data and methods 4.1. Sample To construct our data, we start with all firm-years in Standard and Poor's ExecuComp database over the 1992–2017 period that have the data required to compute the various incentive measures for CEOs. We then merge the ExecuComp data with data for stock returns, insider trading, earnings announcements, and discretionary accruals. We describe the details of various dataset construction in the following subsections. 4.2. Short-sale constraints We measure short-sale constraints (SSC) using three different measures: idiosyncratic risk of a stock, level of short interest, and total market capitalization of the firm's stock. Idiosyncratic risk of a stock presents considerable challenges for arbitrageurs. Pontiff (2006) surveys related literature and concludes that “idiosyncratic risk is the single largest barrier to arbitrage.” Consistent with Pontiff's argument, Mashruwala et al. (2006) report a higher concentration of accruals anomaly in stocks with high idiosyncratic volatility. The association between idiosyncratic volatility and SSC is also documented by Fu (2009). We measure idiosyncratic volatility by regressing stocks' daily excess returns on the three-factor model of Fama and French (1993).10 To reduce the influence of infrequent trading, we require at least 15 trading days in a month for each regression. We compute idiosyncratic risk as the standard deviation of regression residuals multiplied by the square root of the number of trading days. Our second measure of SSC is the level of short-interest scaled by monthly trading volume. We collect data on short interest from Compustat Monthly Short Interest File. A higher level of short interest indicates more binding short-sale constraints because borrowing shares to short becomes more difficult if the stock already has a very high level of short interest (Desai et al., 2002; Asquith et al., 2005; Boehme et al., 2006).11 Our third measure of SSC is market value of equity. Smaller firms likely have greater information asymmetry and less liquidity, and therefore are likely to have higher short-sale constraints. We combine the three measures of SSC into one composite measure defined as the sum of firm's quartile rank (1 through 4) for each measure. The minimum is 3 indicating a firm is in the first quartile on all three measures (the lowest SSC); the maximum is 12 indicating a firm is in the top quartile on all three measures (the highest SSC); the median rank of a firm is 6 on the scale of 1–12. We classify firms with above median composite short-sale constraints as facing high SSC, and those below median as facing low SSC. 4.3. Calendar-time portfolio returns We employ the calendar-time-portfolio method to measure abnormal stock return around CEOs' short-incentive-horizon years. Combining stocks in calendar-time portfolios corrects for lack of independence among firms' returns measured contemporaneously (Fama, 1998). We form two equally-weighted portfolios every month based on incentive horizon, one for firms with CEO incentive horizon less than a year and another for firms with CEO incentive horizon longer than 2 years. We also form an arbitrage portfolio that buys the short-horizon portfolio and sells short the long-horizon portfolio. We then regress portfolio returns on market, SMB, HML, and momentum factors (Fama and French, 1993; Carhart, 1997). The regression intercept is the abnormal stock return. 4.4. Insider trades and profits To examine whether short-horizon CEOs exploit price inflation, we examine these CEOs' trading activities. We identify insider 10

We find nearly identical results if we use the single market factor. A limitation of this measure is that in some situations, a lower level of short interest might indicate higher, rather than lower, short-sale constraints because a lower level of short interest might be the result of short sellers unable to borrow the stocks to sell short. 11

6

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

trading activity from Thomson Reuters' Insider Filing Data Feed. Following Jagolinzer et al. (2011), we accumulate a CEO's transactions each day while netting out purchases from sales. We then regress the firm's excess returns of up to 180 days on market, SMB, HML, and momentum factors. The regression intercept is the abnormal return. CEO's transactions are profitable (unprofitable) if the stock experiences a negative (positive) abnormal return, i.e., the regression intercept is negative (positive). We multiply the intercept by −1 for ease of interpretation and exposition. Finally, we weight the intercept by the size of transaction to estimate the dollar value of the CEO's trading profit. 4.5. Earnings surprises and market reaction to earnings Earnings announcements are an important corporate information event and plausibly an opportune venue for CEOs to influence investors' valuation of their firms. We first examine the size of quarterly earnings surprises around the short-horizon period. We measure unexpected earnings by the difference between actual earnings and the I/B/E/S analyst consensus earnings. We standardize the unexpected earnings by the standard deviation of unexpected earnings during the past 20 quarters, requiring at least eight quarters of available data (e.g., Doyle et al., 2006). This standardization is important for our test because we want to control for the typical inherent profit volatility of each firm and capture the unexpected earnings beyond that. Standardizing by stock price does not control for the inherent profit volatility and would potentially contaminate our test because of the mispricing we document later in Table 2. We measure investor reaction to earnings announcements by day (−1,+1) cumulative abnormal return (CAR). We winsorize SUE and CAR at the 1st and 99th percentiles to reduce the influence of extreme outliers. 4.6. Earnings management To identify firms that use income-increasing discretionary accruals, we begin with a discretionary accrual measure obtained from a modified version of the Jones (1991) model. We run annual cross-sectional regressions of the following model for each of the Fama and French (1997) 48-industry groups:

TAit Assetsit

= 1

1 Assetsit

+ 1

1

Salesit ARit + Assetsit 1

2

PPEit Assetsit

+ 1

it ,

(1)

where ∆Salesit is the change in sales, ΔARit the change in accounts receivables, and PPEit property, plant, and equipment. Following Hribar and Collins (2002), we calculate total accruals, TAit, as Compustat data item ibc (income before extraordinary items and discontinued operations from the Cash Flow Statement) minus OCF (net cash flow from operating activities (oancf) minus extraordinary items and discontinued operation (xidoc)). We winsorize the variables in Eq. (1) at the 1st and 99th percentiles before running the regressions and exclude those industry-years that have fewer than eight observations. We obtain discretionary accruals as the residuals from Eq. (1). Kothari et al. (2005) recommend adjusting discretionary accruals using performance-matched firms. They match each firm with another firm from the same industry based on their return on assets. Performance-adjusted discretionary accruals are then defined as the difference in discretionary accruals of the subject firm and the matched firm. Given our focus on how managerial incentive horizon affects the discretionary accrual measures, we need to ensure that such matching procedures do not wipe out any effect of incentive horizon. If incentive horizon affects discretionary accruals, then netting accruals for firms with similar incentive horizons could net out any effects of the incentive horizon. Thus, we follow Kothari et al. except that we impose the additional requirement on the matched firm that it has a different incentive horizon than the sample firm. Specifically, for every observation with an incentive horizon shorter than the sample median, we choose the match (based on the Fama-French 48 industry, year, and ROA) from among firms with incentive horizons longer than the sample median. We do the converse for every observation with an incentive horizon longer than the sample median. This approach preserves the performance matching and also ensures that we retain any variation that arises from the differences in incentive horizon. Beneish and Vargus (2002) find that the accrual mispricing in Sloan (1996) occurs primarily for income-increasing accruals, and the theoretical models that underpin our study focus on managerial attempts to cause overvaluation. Thus, in addition to examining the continuous discretionary accrual measure, we define a dummy variable equal to one to indicate firm-years in which the performance-adjusted discretionary accrual is income-increasing (i.e., positive), and zero otherwise. 4.7. Matching The propensity score matching method has been shown to be more effective than OLS in matching firms experiencing a particular treatment (for example, CEOs with short incentive horizon) with those without the treatment (for example, CEOs with long incentive horizon). As Armstrong et al. (2010) note, controlling for determinants of a treatment in an OLS regression imposes a functional form on the relation between treatment and firm characteristics.12 Applying the propensity score matching methodology, Armstrong et al. find that the positive relation between the level of managerial incentives and accounting irregularities disappears. Following this line of research, we employ the propensity score matching methodology to study the effects of the horizon of managerial incentives. In particular, we estimate the probability of short CEO incentive horizon using the following probit model: 12

Also see Core (2010) for a discussion of Armstrong et al. (2010). 7

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Pr(Short Horizoni ) =

+

1

× IncentiveRatioi +

(CFO )i +

7

2

× Log

× CapIntensityi +

8

M B

+ i

3

× Log (Total Assets )i +

× InstiOwni +

9

× BoardIndPcti +

4 10

× Leveragei + × BoardSizei +

5

× ROAi + 11

× Yeari +

6

× Stdev i

(2)

As shown in Eq. (2), we employ a wide range of firm characteristics that (1) potentially explain managerial incentive horizon and (2) are plausibly correlated with the outcomes that we study, namely insider trading volume and profits, earnings surprises and market reactions, and discretionary accruals. To control for the effect of the level of incentives, we include the incentive ratio from Bergstresser and Philippon (2006) as a control variable. They define the incentive ratio as the delta of the manager's stock and stock options divided by the sum of that delta, salary, and bonus. We use the logarithm of market-to-book ratio to control for firm valuation and growth potential. Watts and Zimmerman (1990) argue that large firms face higher political costs and thus have a stronger incentive to use accounting discretion to reduce these costs. Dechow and Dichev (2002) posit that larger firms can estimate accruals more accurately as they have more stable operations. We include log of book assets (Compustat item at) to control for firm size. We also control for financial leverage (items dltt/at). Higher financial leverage increases the volatility of net income and gives the manager a stronger incentive to manage earnings to avoid covenant violations and preserve their credit ratings. On the other hand, higher leverage is a proxy for closer monitoring from debtholders and could be related to less earnings management. Firm performance potentially affects the incentive level and structure, so we control for firm performance with the return-on-assets ratio (items ni/at). Firms with more volatile business have a greater incentive to manage earnings to reduce their apparent risk level (Dechow and Dichev, 2002). Thus, we include the standard deviation of cash flows from operations (items oancf/at) for the last 5 years (requiring a minimum of 3 years of data). Francis et al. (2004) recognize the effects of differences in measurement and recognition of tangible and intangible assets on accrual quality. We control for differences in asset structure through capital intensity, the ratio of net fixed assets to total assets (items ppent/at). Corporate governance mechanisms plausibly affect the incentive structure and managerial behavior, so we control for three governance variables: total institutional ownership from Thomson Reuters 13F filings, percentage of independent directors on the board, and board size, both from Risk Metrics. To preserve the sample size, we set missing governance values to zero, and create a dummy for each variable indicating whether the variable value is missing. The dummies are then included in the propensity score model. Next, using the propensity scores predicted from Eq. (2), we match firms with short CEO incentive horizons with firms with long CEO incentive horizons. We retain matches with the smallest difference in propensity scores. To establish the validity of the matching process, we test whether characteristics of firms with short incentive horizon are statistically different from those with long incentive horizon. We then compare various outcomes for the test variables defined earlier, i.e., insider trading and profits, earnings surprises, and discretionary accruals, for the matched pairs. 4.8. Summary statistics Using 1992–2017 ExecuComp data to compute 1992–2014 CEO incentive horizons, we are able to estimate the horizon for 19,445 CEOs. After requiring that a CEO remain in office for the future 3 years, we are left with 16,510 observations. We also remove the CEOs with horizon between one and 2 years to create greater contrast in horizon, and 11,615 observations remain. After requiring the availability of the control variables and the outcome variables, we have 9970 observations for the abnormal return tests. Table 1 reports the summary statistics. The average CEO incentive horizon is 1.83 years, and the median is 1.78 years. The interquartile dispersion of 1.64 years should provide enough variation to test the effects of incentive horizon. 35% of the CEO-years in our sample have a “short horizon” (i.e., incentive horizon less than 1 year). When we look closer, we find that a typical CEO experiences sizable time-series variation in her incentive horizon: for any given year that a CEO's horizon is “short,” there is a 36% probability that the horizon is not short in the previous year, and a 26% probability that the horizon is not short in the following year; and this probability of “not short” increases as the CEO moves further away from the “short” year. This within-CEO variation, together with the cross-CEO variation, enable us to identify the horizon effect. The other variables in Table 1 appear to have reasonable distribution characteristics. 5. Results 5.1. Abnormal stock returns As developed in Hypothesis 1, a key prediction of the theoretical models that motivate our study is that firms with short-horizon CEOs will exhibit stock price inflation. To test this hypothesis, we form calendar-time portfolios based on incentive horizon and examine the portfolio abnormal returns in periods surrounding the short-horizon year. Recall that incentive horizon is measured at the end of each fiscal year, which we label as months 1–12. We compute abnormal returns for the year leading up to the short-horizon year (months −12 to −1), during the short-horizon year (months 1–12), and 2 years after (months 13–36). The results are in Table 2. The left side of Panel A shows that for the high-SSC subsample, short-horizon firms have a significantly positive alpha of 0.556% per month in months −12 to −1, while long-horizon firms have an alpha of 0.183%. The arbitrage portfolio has a large positive alpha of 0.373% (p-value = .02). During months 1–12 and 13–36, the arbitrage portfolio alpha turns significantly 8

Vesting years weighted by deltas of all equity incentives (restricted and unrestricted stocks, vested and unvested options). A dummy variable equal to one if the weighted incentive horizon < 1 year Delta (measuring the sensitivity of an executive's equity-based incentive to the company's stock price) scaled by the sum of delta, salary, and bonus, as in Bergstresser and Philippon (2006). Dollar value of CEO's trades CEO's abnormal trading profits following Jagolinzer et al. (2011) Standardized unexpected earnings Earnings announcement cumulative abnormal return for (−1,1) window in percent Performance-adjusted discretionary accruals scaled by total assets, in %. A dummy variable equal to one if DACC is positive. The natural logarithm of market value of assets to book value of assets. The natural logarithm of total book assets. Long-term debt over total book assets. Income before extraordinary items over total book assets. Standard deviation of cash flow from operations over total book assets for the last 5 years. Net fixed assets over total book assets. Total institutional ownership. Fraction of board members who are independent. Number of directors on the board.

Weighted CEO incentive horizon (all incentives) Short-horizon dummy Incentive ratio

9

4,529,141 1,026,450 1.14 6.47 10.80 0.50 0.67 1.53 0.16 0.11 0.11 0.22 0.36 0.34 4.60

0.48 0.20

0.35 0.24 1,843,384 −26,265 0.37 0.58 −0.56 0.48 0.88 7.21 0.18 0.05 0.05 0.29 0.44 0.49 6.56

1.02

Std

1.83

Mean

116,244 −127,256 −0.09 −3.04 −6.59 0.00 0.41 6.11 0.02 0.02 0.02 0.12 0.00 0.00 0.00

0.00 0.10

1.03

p25

506,467 −7655 0.26 0.39 −0.40 0.00 0.81 7.06 0.16 0.06 0.04 0.23 0.53 0.60 8.00

0.00 0.18

1.78

p50

1,920,000 62,196 0.90 4.18 5.60 1.00 1.27 8.21 0.27 0.09 0.06 0.41 0.76 0.78 10.00

1.00 0.33

2.67

p75

All variables are at the firm-year level. Incentive horizon and incentive ratio are estimated using ExecuComp data. Institutional ownership data are from Thomson Financial. Board data are from IRRC. All other variables are calculated using Compustat data. The sample has 9970 observations. The sample period is from 1992 to 2014.

Trading volume (in $) Abnormal trading profits (in $) SUE CAR(−1,1) Discretionary accruals (DACC) Positive DACC dummy Log(M/B) Log(total assets) Long-term debt ROA Std dev of cash flows Fixed assets Institutional ownership Board independence Board size

Definition

Variable name

Table 1 Summary statistics.

J.D. Chi, et al.

Journal of Corporate Finance xxx (xxxx) xxxx

10

0.341 (0.118) −0.300 (0.109) 1414

0.766 (0.000) 0.226 (0.195) 3396

−0.424 (0.085) −0.527 (0.026)

0.341 (0.070) −0.043 (0.772) 1338

Horizon < 1 year

0.620 (0.000) 0.221 (0.103) 0.050 (0.669) 1338

Horizon < 1 year

Arbitrage Portfolio

0.373 (0.020) −0.617 (0.000) −0.355 (0.006)

Low short-sale constraints Horizon > 2 years

0.183 (0.235) 0.582 (0.000) 0.320 (0.011) 3396

High short-sale constraints

0.556 (0.000) −0.035 (0.802) −0.034 (0.815) 1414

Horizon < 1 year

Arbitrage portfolio

Horizon < 1 year

Horizon > 2 years

Low short-sale constraints

High short-sale constraints

0.530 (0.000) −0.071 (0.537) 3822

Horizon > 2 years

0.432 (0.000) 0.259 (0.008) 0.236 (0.031) 3822

Horizon > 2 years

−0.189 (0.288) 0.028 (0.848)

Arbitrage portfolio

0.188 (0.040) −0.038 (0.705) −0.186 (0.050)

Arbitrage portfolio

This table presents average monthly abnormal returns (alphas) from calendar-time portfolios formed on CEO's incentive horizon and short-sale constraints. Each month we assign firms to one of the four portfolios on the basis of CEO's incentive horizon and firm's short-sale constraints, both measured at the beginning of the year. We require at least ten observations per portfolio. We use a four-factor model, which includes the three factors from Fama and French (1993) and the momentum factor from Carhart (1997), and report abnormal returns from equally-weighted portfolios. All variables are described in Table 1. The sample period is from 1992 to 2014. Panel A reports monthly average abnormal returns held for annual intervals, while Panel B reports average abnormal returns held for 6months intervals. p-values are in parentheses. Significant arbitrage-portfolio alphas (with p-values < .05) are in boldface.

Months 1–6 Months 7–12 N

Portfolio duration

Panel B:

Months −12 to −1 Months 1–12 Months 13–36 N

Portfolio duration

Panel A:

Table 2 Abnormal returns for portfolios formed on CEO incentive horizon and firm short-sale constraints.

J.D. Chi, et al.

Journal of Corporate Finance xxx (xxxx) xxxx

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

negative to −0.617% and −0.355% per month, respectively. The negative alphas more than offset the positive alpha in months −12 to −1. Thus, the evidence is consistent with the prediction that short incentive horizon firms exhibit stock price inflation (positive abnormal returns) followed eventually by correction (negative abnormal returns). Fig. 1 plots the arbitrage-portfolio alphas over time for the high-SSC subsample and provides a clear visual presentation of the stock price inflation and reversal. While we see a similar pattern of alphas for the low-SSC subsample, the magnitude and the statistical significance of the alphas are much lower. The contrast between the high versus low-SSC subsamples provides further confidence in our interpretation that the observed arbitrage portfolio alphas are the result of mispricing. In Panel B of Table 2 we zero in on the 1–12 months alphas. For the high-SSC subsample, we see that the reversals in the shorthorizon portfolio alpha and the arbitrage portfolio alpha occur mainly during months 7–12. That is, it seems that the short-horizon CEOs are able to maintain the stock price inflation for a while, i.e., the arbitrage alpha during months 1–6 though negative, is only marginally significant. For the low-SSC subsample, we again see relatively muted reversals in the short-horizon portfolio alpha and the arbitrage portfolio alpha. To assess whether performance vesting effects are likely to bias our findings, we repeat our main stock return tests in Table 2 while excluding sample years beyond 2002. Performance-vested option and stock grants began to increase in frequency and value in 2003; in 2002, only 9% of the value of grants was performance based according to Bettis et al. (2016). When omitting years 2003 and beyond, untabulated results for the stock return analysis are qualitatively similar to those reported in Table 2. Our results are also robust to removing data beyond 2006. In addition, the contrast between high vs. low-SSC groups also suggests that our findings cannot be explained by performance vesting. Performance vesting cannot explain why the stock price inflation is much more significant in the high SSC group, but our mispricing explanation can. In sum, Table 2 presents evidence that in the presence of higher short-sale constraints, firms with short CEO incentive horizon experience stock price inflation leading up to the short-horizon year; the price inflation persists for several months, and then reverses. Do short-horizon CEOs exploit the temporary stock price inflation? We examine this question next. 5.2. Insider trading volume and profits Table 3 tests Hypothesis 2 by comparing insider trading dollar volume and dollar trading profits between short-horizon CEOs and long-horizon CEOs. We find that irrespective of short-sale constraints, short-horizon CEOs sell significantly more stock than longhorizon CEOs. This is likely driven by the fact that short-horizon CEOs simply have more stocks available to sell during the shorthorizon year; and that they also sell more vested stocks prior to the short-horizon year because they anticipate that more stocks will vest soon. Note that CEOs selling their stock holdings is not surprising – boards design the compensation package this way that CEOs have to regularly sell vested stocks to satisfy their liquidity needs, and the more vested stocks they have, the more they likely will sell. The key question is that when short-horizon CEOs sell their holdings, whether they make abnormal profit. We answer this question next. We show in Panel B of Table 3 that in the year leading up to as well as during the short-horizon year, short-horizon CEOs generate larger abnormal trading profit. When short-sale constraints are high, short-horizon CEOs generate a positive abnormal profit of $28,641 in the year leading up to the short-horizon year, and $16,359 during the short-horizon year. The difference between short versus long-horizon CEOs is $92,681 (t-stat = 3.50) and $52,043 (t-stat = 1.95) in each of the 2 years. In contrast, when short-sale constraints are low, short-horizon CEOs lose money on insider sales, and even underperform long-horizon CEOs in the year leading up to the short-horizon year. The abnormal trading profits are consistent with the earlier findings for abnormal stock returns.13 Next in Panel C of Table 3, we compare how well our sample of short-horizon firms matches with long-horizon firms. The first column reports slope coefficients for a probit regression for determinants of short incentive horizon. The last column reports standardized bias after matching (Rosenbaum and Rubin, 1985) and p-value for significance of standardized bias. The only matching variables for which standardized bias are significant at p-values of 5% or less are standard deviation of cash flows and board independence. The magnitude of the difference is still small. In case of standard deviation of cash flows, the short-horizon sample has a mean of 5.72%, while that of the long-horizon sample is 6.60%. Similarly for board independence, the short-horizon sample has a mean of 60.59%, while that of the long-horizon sample is 58.20%. We also compute the Rubin's B and Rubin's R statistics, both of which suggest the bias is not significant.14 These findings suggest validity of matches used in this comparison. To summarize the findings so far, firms with short-horizon CEOs experience temporary stock price inflation, and short-horizon CEOs sell more stocks and make greater abnormal profits. The abnormal stock returns and abnormal CEO trading profits are driven by firms with high short-sale constraints, which is consistent with the mispricing interpretation. 13 We compare the magnitude of insider trading profits with other studies in the literature. Unfortunately, most of these studies only report alphas (e.g., Jagolinzer et al., 2011; Gao et al., 2014). Dai et al. (2015) report aphas as well as average dollar trades for executives. Dai et al. report that the average daily alpha for a 6 month period following the insider trade is 0.006. Mean size of a trade is exp(10.213) = $27,255. Therefore, their average dollar trading profit is = $27,255 * 0.006 = $163 per day. Assuming 180 calendar days over a six month period, the total abnormal profit = 163 * 180 = $29,340. 14 Rubin's B measures the absolute standardized difference of the means of the linear index of propensity score in the treated sample and matched non-treated sample. Matching is considered biased if B > 25%. Rubin's R is the ratio of treated to matched non-treated variances of the propensity score index. Matching is considered biased if R is outside [0.5; 2].

11

4,975,953 2584

Low short-sale constraints ATT 6,873,405 N 2564

−45,954 2584

Low short-sale constraints ATT −216,711 N 2564

12

Board independence

Institutional ownership

Fixed assets

Std dev of cash flows

ROA

Long-term debt

Log(total assets)

Log(M/B)

Incentive ratio

Year

−170,756

92,681

1,897,452

713,131

0.6059

0.4400

0.582

0.4212

0.2316

0.2328

0.0577

0.1639

6.7098

0.932

0.3301

2006

Mean for horizon > 2

−98,144 2568

−35,683 4959

5,965,967 2568

1,532,088 4959

0.0660

−79,137 2445

16,359 2855

7,141,283 2445

2,475,012 2855

Horizon > 2 years

0.0572

0.0627

0.1621

6.7422

0.9151

0.3351

2006.1

0.061 (0.000) 0.958⁎⁎⁎ (0.000) 0.052⁎ (0.066) 0.0002 (0.990) 0.921⁎⁎⁎ (0.000) 0.034 (0.819) −0.268⁎⁎⁎ (0.002) −0.249⁎⁎⁎ (0.002) 0.486⁎⁎⁎ (0.000) 0.117 (0.316) ⁎⁎⁎

Mean for horizon < 1

−3.39⁎⁎⁎

3.50⁎⁎⁎

6.33⁎⁎⁎

6.54⁎⁎⁎

Coefficient (p-value)

High short-sale constraints in year t

Panel C: Goodness of matching between short vs. long-horizon CEOs

−64,039 4713

High short-sale constraints ATT 28,641 N 2875

Panel B: Abnormal trading profits (in $)

1,401,680 4713

High short-sale constraints ATT 2,114,811 N 2875

Panel A: Trading volume (in $)

t-stats

Horizon < 1 year

Difference

Horizon < 1 year

Horizon > 2 years

In year t

In year t − 1

Table 3 Insider trading volume and profits – sorted on CEO incentive horizon and firm short-sale constraints.

1.3 (0.608) 2.0 (0.405) −2.7 (0.279) 2.9 (0.226) −1.1 (0.666) 4.4⁎ (0.071) −13.6⁎⁎⁎ (0.000) 0.6 (0.825) 4.6⁎ (0.087) 7.2⁎⁎⁎ (0.004)

Bias % (p-value)

0.32

1.95⁎⁎

3.54⁎⁎⁎

7.48⁎⁎⁎

t-stats

(continued on next page)

19,006

52,043

1,175,316

942,923

Difference

J.D. Chi, et al.

Journal of Corporate Finance xxx (xxxx) xxxx

6.861

0.043 (0.000) −123.776⁎⁎⁎ (0.000) 8314 0.0938 2.9% 21.9% 0.77

6.713

Mean for horizon > 2

3.9 (0.108)

Bias % (p-value)

Panel A presents CEO's trades in the year before (year t − 1) and the year during (year t) the short-horizon year. Panel B reports the abnormal profits earned on those trades in the 180-day period following each trade. The average treatment effect for the treated (ATT) is reported for propensity-score matched samples. We identify CEO trading activity from Thomson Reuters' Insider Filing Data Feed. We accumulate CEO's transactions each day while netting out purchases from sales. We then estimate abnormal returns for up to 180 days as the intercept of a regression of firm's returns in excess of risk free rate on market, SMB, HML, and momentum factors (Fama and French, 1993; Carhart, 1997). CEO's transactions are profitable (unprofitable) if the stock experiences a negative (positive) abnormal return, i.e., the regression intercept is negative (positive). We multiply the intercept by −1 and weight the intercept by the size of transaction to estimate dollar value of CEO's trading profit. Panel C reports slope coefficients for a probit regression for determinants of short incentive horizon (horizon < 1 year) and compares firm characteristics of short horizon with long horizon (> 2 years). All variables are described in Table 1. The sample period is from 1992 to 2014. ⁎⁎⁎, ⁎⁎, and ⁎ indicate p < .01, < .05, and < .1, respectively. Rubin's B: the absolute standardized difference of the means of the linear index of propensity score in the treated sample and matched non-treated sample. Matching is biased if B > 25%. Rubin's R: the ratio of treated to matched non-treated variances of the propensity score index. Matching is biased if R is outside [0.5; 2].

N Pseudo R2 Median Bias Rubin's B Rubin's R

Constant

Board size

Mean for horizon < 1

⁎⁎⁎

Coefficient (p-value)

High short-sale constraints in year t

Panel C: Goodness of matching between short vs. long-horizon CEOs

Table 3 (continued)

J.D. Chi, et al.

Journal of Corporate Finance xxx (xxxx) xxxx

13

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Table 4 Earnings surprise and market reaction to earnings surprise – sorted on CEO incentive horizon and firm short-sale constraints. In year t − 1 Horizon < 1 year

In year t Horizon > 2 years

Difference

t-stats

Horizon < 1 year

Horizon > 2 years

Difference

t-stats

High short-sale constraints ATT 0.325 N 1046

0.243 2169

0.082

1.81⁎

0.306 1068

0.256 2224

0.050

1.14

Low short-sale constraints ATT 0.456 N 1390

0.477 2854

−0.020

−0.50

0.512 1559

0.473 2882

0.039

1.00

High short-sale constraints ATT 1.132 N 1046

0.542 2169

0.591

2.06⁎⁎

0.783 1068

0.711 2224

0.072

0.25

Low short-sale constraints ATT 0.192 N 1390

0.370 2854

−0.178

−0.86

0.131 1559

0.136 2882

−0.005

−0.03

Panel A: SUE

Panel B: CAR

This table compares quarterly earnings surprises (SUE) and earnings announcement cumulative abnormal returns (CAR) for firms sorted on CEO incentive horizon and firm short-sale constraints. The average treatment effect for the treated (ATT) is reported for propensity-score matched samples. We perform the comparison in the year before (year t − 1) and the year during (year t) the short-horizon year. We measure unexpected earnings by the difference between actual earnings and the I/B/E/S analyst consensus earnings. We standardize the unexpected earnings by the standard deviation of unexpected earnings during the past 20 quarters, requiring at least 8 quarters of available data. We calculate earnings announcement cumulative abnormal return (CAR) over (−1,+1) window and report CAR as a percentage. All variables are described in Table 1. The sample period is from 1992 to 2014. ⁎⁎⁎, ⁎⁎, and ⁎ indicate p < .01, p < .05, and p < .1, respectively.

5.3. SUE and CAR Given the stock price inflation of firms with short-horizon CEOs and that these CEOs exploit the mispricing, we next attempt to identify more precisely what drives the mispricing. In particular, we test Hypotheses 3 and 4 and focus on differences in firms' reported earnings and abnormal returns surrounding the earnings announcements. Table 4 reports the SUE and CAR tests. We first examine the results for year t − 1, the year leading up to the short-horizon year, which is also the year that we find stock price inflation for firms with short-horizon CEOs (as reported in Table 2). To ensure that the surprises and investor responses are truly from the horizon effect and not from the inherent business volatility or information asymmetry, we impose two more matching variables in addition to those listed in Eq. (2): analyst forecast dispersion scaled by the median forecast, and number of analysts. The matching procedure provides satisfying matches as Rubin's B & R statistics do not indicate significant bias. We do not report covariate balance for brevity. In Panel A of Table 4, we observe that in year t − 1 and for firms with high SSC, those with short-horizon CEOs report SUEs of 0.325, which is significantly greater than the figure of 0.243 for firms with long-horizon CEOs (t-stat = 1.81). In contrast, there is no significant difference in SUEs when short-sale constraints are low. The contrast mitigates the concern that firms with short-horizon CEOs are merely better performing firms, and that short horizon is merely the result of better performance. The correlation coefficient between SUE and short-sale constraints is −0.111, which demonstrates that sorting on short-sale constraints is not effectively sorting on performance. In Panel B of Table 4, we use the CAR over (−1,1) window to examine investors' reaction to earnings announcements. In year t − 1 and when short-sale constraints are high, the CAR ratio is significantly greater for firms with short-horizon CEOs (1.132%) than for matched long-horizon firms (0.542%) (t-stat = 2.06). Thus, investors react more strongly to earnings surprises by short-horizon firms. In contrast, the difference is insignificant for firms with low short-sale constraints. The findings in Panels A and B are consistent with our interpretation that high short-sale constraints contribute to the stock price inflation in the year leading up to the shorthorizon year. The results indicate that higher SUEs and higher CAR for short-horizon firms with high short-sale constraints are one contributor to the stock price inflation identified for these firms in Table 2. Also shown in Table 4, we repeat the SUE and CAR tests for year t, i.e., the short-horizon year. For that period, we find no significant differences in SUE or CAR across short and long-horizon firms. We note that larger SUE and CAR, although consistent with an overvaluation interpretation, is not necessarily a sign of overvaluation. Do the positive CARs of short-horizon firms revert over time? If so, it will lend greater credence to the overvaluation interpretation. In an untabulated test, we compute longer-run CARs for days 2–50 post earnings announcements. This is analogous to a post-earnings-announcement-drift (PEAD) test. We find that when short-sale constraints are high, short-horizon firms underperform long-horizon firms by 1.032% (t-stat = 1.83) in year t − 1, and by −0.902% in year t (t-stat = 1.61). When short-sale constraints are low, the differences in CAR (2, 50) as well as the corresponding t-stats are markedly smaller. In sum, this test, although only marginally significant, provides another piece of evidence consistent with the overvaluation interpretation. 14

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Table 5 Discretionary accruals for firms sorted on CEO incentive horizon and firm short-sale constraints. In year t − 1 Horizon < 1 year

In year t Horizon > 2 years

Difference

t-stats

Horizon < 1 year

Horizon > 2 years

Difference

t-stats

High short-sale constraints ATT 0.530 N 1310

0.479 2191

0.051

2.60⁎⁎⁎

0.524 1457

0.437 2993

0.088

4.95⁎⁎⁎

Low short-sale constraints ATT 0.478 N 1177

0.480 1683

−0.002

−0.09

0.479 1264

0.497 2400

−0.018

−0.84

High short-sale constraints ATT 0.705 N 1310

−0.859 2191

1.565

3.29⁎⁎⁎

0.328 1457

−1.715 2993

2.042

4.91⁎⁎⁎

Low short-sale constraints ATT −0.119 N 1177

−0.927 1683

0.808

1.70

−0.778 1264

−0.044 2400

−0.734

−1.72⁎

Panel A: DACC dummy

Panel B: DACC

This table compares discretionary accruals by firms sorted on CEO incentive horizon and firm short-sale constraints measured in the year before (year t − 1) and the year during (year t) the short-horizon year. The average treatment effect for the treated (ATT) is reported for propensity-score matched samples. Discretionary accruals is measured as regression residual using the following industry-year regression: TAit Salesit ARit PPE 1 = Assets + 1 Assets + 2 Assets it + it , where ∆Salesit is the change in sales, ∆ARit the change in accounts receivables, and PPEit Assets it 1

it 1

it 1

it 1

property, plant, and equipment. Following Hribar and Collins (2002), we calculate total accruals, TAit, as ibc (income before extraordinary items and discontinued operations from the Cash Flow Statement) minus OCF (net cash flow from operating activities (oancf) minus extraordinary items and discontinued operation (xidoc)). We follow Kothari et al. (2005) and performance-adjust the discretionary accruals using ROA-matched firms. We set DACC dummy (Panel A) equal to one for firm-years with positive performance-adjusted discretionary accrual, and zero otherwise. DACC (Panel B) is the difference in regression residuals of a sample firm with its matched firm. The sample period is from 1992 to 2014. ⁎⁎⁎, ⁎⁎, and ⁎ indicate p < .01, p < .05, and p < .1.

5.4. CEO incentive horizon and discretionary accruals Earnings management through discretionary accruals is well documented in the accounting and finance literature, but the literature has yet to reach a consensus on whether there is a reliable relation between managerial incentives and earnings management (see a thorough review and analysis in Armstrong et al., 2010). We next test Hypothesis 5 and examine whether shorter CEO incentive horizon is associated with higher discretionary accruals, which may help explain the stock price inflation and greater earnings surprise for firms with short-horizon CEOs. We again employ a propensity score approach that matches firms with short-horizon CEOs with firms with long-horizon CEOs on many different characteristics. Given our focus on the horizon of CEO incentives, we also match on the level of the equity-based incentives. We essentially ask whether similar firms whose CEOs face the same level of equity-based incentives, but differ on when their CEOs can profit from the incentives, differ in their likelihood of employing income-increasing discretionary accruals. We examine both the likelihood of employing income-increasing discretionary accruals and the magnitudes of the discretionary accruals. Again, the matching procedure provides satisfying matches as Rubin's B & R statistics do not indicate significant bias. We do not report covariate balance for brevity. Table 5, Panel A shows that firms with short-horizon CEOs engage in income-increasing accruals management when firms face high short-sale constraints. This is evident from difference in DACC dummy as well as levels of DACC. In the year prior to the shorthorizon year, 53% of short-horizon firms have positive DACC whereas 47.9% of matched long-horizon firms have positive DACC, and this difference is significant with a t-stat of 2.60. Similarly, during short-horizon year, 52.4% of short-horizon firms have positive DACC whereas 43.7% of matched long-horizon firms have positive DACC. Again, the difference is significant with a t-stat of 4.95. In the sample with low short-sale constraints, there is no difference in the proportion of firms with positive DACC for short- versus long-horizon firms. About 48% firms have positive DACC for short- as well as matched long-horizon firm in the year prior to short horizon. Similarly, the difference is insignificant in the year of short-horizon. Moving to the comparison of levels of DACC in Panel B, we find that short-horizon firms have significantly higher DACC than matched long-horizon firms in the year prior to and during the short-horizon year. This difference is significant only for the sample with high short-sale constraints and not for the sample with low short-sale constraints. Combining the earning management result with the return result in Table 2 and the SUE and CAR result in Table 4, we can infer that while short-horizon firms employ earnings management strategies more aggressively than their matched long-horizon counterparts, the effects on stock price, earnings surprise, and earnings announcement return are significant only in the year leading up to the short-horizon year. Thus, during the short-horizon year, the more aggressive earnings management actions of short-horizon firms have the effect of sustaining already-inflated share prices for the first 6 months of the year (i.e., zero abnormal returns over those 15

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

6 months), after which correction sets in and the firms exhibit negative abnormal returns.15 Armstrong et al. (2010) apply the propensity score matching method and find no significant relation between managerial equity incentive level and accounting irregularities. Their finding challenges earlier studies that do show a positive relation between managerial incentive level and earnings manipulation. Applying the same propensity score matching technique as in Armstrong et al. except also matching on the level of equity incentives, we find that the horizon of incentives is related to earnings management behavior. Our results suggest that future studies may gain additional insights by treating the horizon of equity incentives as an important dimension of managerial incentives. 5.5. Endogeneity and other concerns As already discussed, our finding on abnormal stock returns (in Table 2) is consistent with the mispricing explanation, not the performance-vesting explanation, and we showed that the abnormal-return results are robust to studying the sub-period in which performance vesting was rare. To examine the robustness of our other tests to performance vesting, we include in our propensityscore-matching approach a large set of future performance variables that are potentially performance-vesting metrics and thus trigger vesting in the years that enter into our horizon calculation. The performance measures include: year +1 and +2 stock returns, the percentage change in earnings per share, and return on assets. We use unadjusted and industry-adjusted versions of these measures, and also include the squares of the measures to attempt to capture nonlinearities in performance-vesting rules. When controlling for these future performance measures, we obtain results similar to those tabulated in our tests for insider trading and abnormal profits, standard earnings surprises and associated abnormal returns, and earnings management. Across our tests, the consistent contrast between high versus low short-sale constraints groups is very telling. We show that when short-sale constraints are more binding, firms with short-horizon CEOs are more likely to experience stock price inflation, greater CEOs stock sales and sale profits, greater earnings surprises and investor reaction to these surprises, and greater income-increasing accruals management. Yet when short-sale constraints are low, most of these differences disappear. If an omitted variable, not related to mispricing attempts, is driving these results, we cannot find a clear theoretical explanation as to what this omitted variable might be and why it asymmetrically affects short- versus long-incentive-horizon firms and at the same time asymmetrically affects lowversus high-SSC firms. We also evaluate robustness of our results for each component of SSC. For the stock price inflation test, employing each of the three SSC components separately produces very similar results as the composite SSC measure. For other tests, the results sometimes become less significant when employing only one SSC component. But as a whole, we do not see one dominant component driving the composite SSC measure. 5.6. Another endogeneity test – matched pair sensitivity test When conducting an analysis of matched treatment-control pairs based on observational data, one wants the pairs to be identical except for the presence or absence of the treatment. In such a perfect situation, endogeneity concerns are avoided. In reality, as Armstrong et al. (2010) note, there may be “endogenous matching of executives and contracts on unobservable firm and CEO characteristics.” This “hidden bias” may affect inferences because it is possible for an unobservable characteristic to cause nonrandom assignment of an observation to the treatment or the control group and also affect the outcome. Rosenbaum (2002) develops a sensitivity analysis to assess how large a deviation from random assignment could be before one's conclusion about a statistically significant treatment-control difference would be altered. For example, a sensitivity analysis might conclude that a non-zero treatment effect exists even if an unobservable characteristic makes it 50% more likely that an observation is in the treatment group instead of the control group, which would imply a 60%/40% odds ratio of being a treatment vs. control. In this case, the parameter Γ that indicates the departure from random assignment would be 1.5 (=60%/40%) vs. a Γ of 1.0 (=50%/50%) for random assignment. However, if an unobservable characteristic makes it 60% more likely for an observation to be in the treatment group instead of the control group (Γ = 1.6), the sensitivity analysis that produces Γ = 1.6 might indicate that one could no longer be confident in rejecting the null hypothesis of no treatment effect. Similar to other statistical analyses, one can consider the power of a sensitivity analysis. The power of a sensitivity analysis is the probability that it rejects a false null hypothesis of no treatment effect while allowing for a specified level of non-random assignment to the treatment and control groups. A low-power sensitivity analysis has a low probability of rejecting a false null while a higherpower sensitivity analysis has a greater likelihood of rejecting a false null for the same level of hidden bias. In an extension of his 2002 work, Rosenbaum (2011) explores the power of various sensitivity analyses, and concludes that in many situations the Wilcoxon-based sensitivity analysis from his 2002 work has relatively low power compared to an alternative class of sensitivity analyses based on u-statistics. Briefly, the new tests are based on triples m, m, m , where m defines the number of _

treatment minus control differences in outcomes that are sorted in increasing order based on absolute magnitude. One then counts the 15 Of course, we cannot observe the counterfactual if short-horizon CEOs do not engage in income-increasing accruals management – perhaps the alphas will be even more negative during the second half of year t. Note that our results are not necessarily inconsistent with the accrual anomaly (Sloan, 1996) because no study on the accrual anomaly has considered the effect of managerial incentive horizon. Note also that our results are consistent with Johnson et al.'s (2009) finding of statistically zero raw returns over the period in which their sample firms engaged in misreporting.

16

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Table 6 Rosenbaum bounds. Outcome

t-stats for difference in short vs. long sample

(8,7,8)

(8,6,8)

DACC dummy, t (High SSC) DACC dummy, t – 1 (High SSC) DACC, t (High SSC) DACC, t − 1 (High SSC) Insider trading volume, t (High SSC) Insider trading volume, t (Low SSC) Insider trading volume, t − 1 (High SSC) Insider trading volume, t − 1 (Low SSC) Insider trading profits, t (High SSC) Insider trading profits, t − 1 (High SSC) Insider trading profits, t − 1 (Low SSC) SUE, t − 1 (High SSC) CAR, t − 1(High SSC)

4.95⁎⁎⁎ 2.60⁎⁎⁎ 4.91⁎⁎⁎ 3.29⁎⁎⁎ 7.84⁎⁎⁎ 3.54⁎⁎⁎ 6.54⁎⁎⁎ 6.33⁎⁎⁎ 1.95⁎⁎ 3.50⁎⁎⁎ −3.39⁎⁎⁎ 1.79⁎ 2.08⁎⁎

41 3.59 1.37 1.26 1.68 1.56 1.47 1.56 1 1.36 1 1.15 1.20

15.45 3.26 1.36 1.29 1 1.03 1 1.03 1.30 1.47 1 1.12 1.11

This table reports the gamma (Γ) statistics from the sensitivity tests developed in Rosenbaum (2002 and 2011). We show the Γs for the matched-pairdifference tests that are significant in earlier tables. The m, m, m triples are the parameter inputs for the sensitivity test as described in the text. _

number of positive differences among the pairs numbered from m up to m in the sorted set. By choosing the appropriate triple, one _

can place less weight on small-magnitude paired differences to increase power, while also controlling the influence of very largemagnitude paired differences. We refer the reader to Rosenbaum (2011) for the derivation of the test statistics and other details because there is not a compact way to describe them here. For our purposes, we note that Rosenbaum (2011) conducts simulation analyses of many different m, m, m triples to assess their _

power under various assumptions. Based on those analyses, Rosenbaum concludes that m, m, m = (8,6,8) is a “safe choice.” He _

notes that for short-tailed distributions of matched pair differences like the normal distribution or the logistic distribution, (8,7,8) is a better choice. In Table 6, we summarize the resulting Γs for the matched pair difference tests that are significant in our earlier analyses. To give readers examples of how the Γs can change for different triples, we tabulate the Γs for two different triples, (8,7,8) and (8,6,8). The most striking case is the test of whether short-horizon firms are more likely than matched long-horizon firms to employ incomingincreasing discretionary accruals. The Γ of 41 for the (8,7,8) triple indicates that an unobservable characteristic would have to make it 41 times more likely that a firm is a treatment (short-horizon firm) than a control (long-horizon firm) before we would fail to reject the null hypothesis of no treatment effect based on concerns about hidden bias. For the (8,6,8) triple, the Γ is 15.45. The results of the sensitivity analyses are somewhat unsurprising given the economically large magnitude of the difference (recall from Table 5 that short-horizon firms have a 52.4% likelihood of employing income-increasing discretionary accruals vs. a 43.7% likelihood for longhorizon firms). Thus, the conclusion that short-horizon firms are more likely to employ income-increasing discretionary accruals is quite robust. As shown in Table 6, the Γs for the other difference tests are much lower than the one for the likelihood of employing incomeincreasing discretionary accruals. Whether one is willing to alter the conclusions about the statistical significance for the tests depends upon his or her subjective assessments of how large the effects of unobservable characteristics might be on both the likelihood of assignment to the treatment group and the magnitude of the outcome for treatment and control observations. For example, the Γ for the continuous measure of discretionary accruals is 1.37 for the (8,7,8) triple and 1.36 for the (8,6,8) triple. Importantly, the relatively lower values of Γ do not necessarily indicate that there is no true positive effect of short horizon on the continuous accrual measure. Ideally, one would know both how an unobservable characteristic alters the likelihood that an observation is assigned to the treatment group and how that characteristic affects the outcome being tested. If an unobservable characteristic affects the likelihood of assignment to the treatment group but does not affect the outcome, then the original inference remains. Unfortunately, there is no way to measure how much an unobservable characteristic might affect the outcome variable. 6. Conclusion We identify real capital-market consequences of managerial incentive horizon when short-sale constraints are more binding. Firms exhibit significant stock price inflation when their CEOs have short horizon incentives, and the CEOs exploit the price inflation by selling more stock and making larger abnormal profits. Evidence suggests that greater abnormal returns at earnings announcements contribute to the stock price inflation. To create and sustain the price inflation, short-horizon CEOs are more likely to engage in income-increasing accruals management. The findings are consistent with recent theoretical models in which short-horizon incentives induce CEOs to inflate stock prices, and importantly, suggest that they have some success in doing so, and that they profit from it. The findings also shed new light on the role that earnings management plays in this process. Broadly, our findings highlight the importance of incentive horizon in studying the effects of executive incentives and compensation. 17

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Acknowledgements Chi gratefully acknowledges the support from the Key Program of National Natural Science Foundation of China (71731003) and the Young Scholars Program of National Natural Science Foundation of China (71802043). Appendix A. Calculation of deltas To calculate the deltas for stock options, we use the Black and Scholes (1973) model modified by Merton (1973) to incorporate dividends. Executives typically exercise their options before maturity (Hemmer et al., 1996; Huddart and Lang, 1996; Heath et al., 1999), so we reduce the contractual option maturity from ExecuComp by 30%. As a proxy for the risk-free rate, we use the average yield on U.S. Treasury securities that most closely matches the option's (reduced) maturity. We use the standard deviation of stock returns over the prior 60 months to estimate the stock return volatility. We use the average dividend yield over the prior 3 years as a proxy for the future dividend yield. For newly granted options, strike price and maturity are taken directly from ExecuComp. ExecuComp does not report terms and numbers of individual grants for previously granted options, so we use Core and Guay's (2002) 1year approximation method to estimate the strike price of previously granted options. Given the option pricing parameter values and the numbers of vested and unvested options each CEO holds, we use the option pricing model to compute delta, defined as the change in value for a one-percentage-point increase in the market value of the firm's equity, and then multiply by the number of options held. We compute separate measures for vested and unvested options. We compute deltas for restricted and unrestricted stockholdings as 1% multiplied by the stock price and then multiplied by the number of shares held. References Armstrong, C., Jagolinzer, A., Larcker, D., 2010. Chief executive officer equity incentives and accounting irregularities. J. Account. Res. 48, 225–271. Asquith, P., Pathak, P., Ritter, J., 2005. Short interest, institutional ownership, and stock returns. J. Financ. Econ. 78, 243–276. Bartov, E., Givoly, D., Hayn, C., 2002. The rewards to meeting or beating earnings expectations. J. Account. Econ. 33, 173–204. Beaver, W.H., 1968. The information content of annual earnings announcements. J. Account. Res. 67–92. Beneish, M.D., Vargus, M.E., 2002. Insider trading, earnings quality, and accrual mispricing. Account. Rev. 77, 755–791. Bergstresser, D., Philippon, T., 2006. CEO incentives and earnings management. J. Financ. Econ. 80, 511–529. Bergstresser, D., Desai, M., Rauh, J., 2006. Earnings manipulation, pension assumptions, and managerial investment decisions. Q. J. Econ. 121, 157–195. Bettis, C., Bizjak, J., Coles, J., Kalpathy, S., 2010. Stock and option grants with performance-based vesting provisions. Rev. Financ. Stud. 23, 3849–3888. Bettis, C., Bizjak, J., Coles, J., Kalpathy, S., 2016. Performance–Vesting Provisions in Executive Compensation. (Working paper). Bhojraj, S., Hribar, P., Picconi, M., McInnis, J., 2009. Making sense of cents: an examination of firms that marginally miss or beat analyst forecasts. J. Financ. 64, 2361–2388. Black, F., Scholes, M., 1973. The pricing of options and corporate liabilities. J. Polit. Econ. 81, 637–654. Boehme, R., Danielsen, B., Sorescu, S., 2006. Short sale constraints, differences of opinion, and overvaluation. J. Financ. Quant. Anal. 41, 455–487. Bolton, P., Scheinkman, J., Xiong, W., 2006. Executive compensation and short–termist behaviour in speculative markets. Rev. Econ. Stud. 73, 577–610. Brockman, P., Martin, X., Puckett, A., 2010. Voluntary disclosures and the exercise of CEO stock options. J. Corp. Finan. 16, 120–136. Cadman, B., Sunder, J., 2014. Investor horizon and CEO horizon incentives. Account. Rev. 89, 1299–1328. Cadman, B.D., Rusticus, T.O., Sunder, J., 2013. Stock option grant vesting terms: economic and financial reporting determinants. Rev. Acc. Stud. 18, 1159–1190. Cai, J., Vijh, A., 2007. Incentive effects of stock and option holdings of target and acquirer CEOs. J. Financ. 62, 1891–1933. Carhart, M., 1997. On persistence of mutual fund performance. J. Financ. 52, 57–82. Cassell, C., Huang, S.X., Sanchez, J.M., 2013. Forecasting without consequence? Evidence on the properties of retiring CEOs' forecasts of future earnings. Account. Rev. 88, 1909–1937. Cheng, Q., Warfield, T., 2005. Equity incentives and earnings management. Account. Rev. 80, 441–476. Chi, J.D., Johnson, S., 2009. The Value of Vesting Restrictions on Managerial Stock and Option Holdings. Working paper. Available at SSRN. https://ssrn.com/ abstract=1136298. Chi, J.D., Gupta, M., Johnson, S., 2011. The Time Horizon of Managerial Incentives and firms' Information Environment Quality. Working paper. Mays Business School Research Paper No. 2011–5. Available at SSRN. http://ssrn.com/abstract=1571704. Core, J., 2010. Discussion of chief executive officer incentives and accounting irregularities. J. Account. Res. 48, 273–287. Core, J., Guay, W., 2002. Estimating the value of employee stock option portfolios and their sensitivities to price and volatility. J. Account. Res. 40, 613–630. Dai, L., Parwada, J.T., Zhang, B., 2015. The governance effect of the media's news dissemination role: evidence from insider trading. J. Account. Res. 53, 331–366. Dechow, P., Dichev, I., 2002. The quality of accruals and earnings: the role of accrual estimation errors. Account. Rev. 77, 35–59. Dechow, P., Sloan, R., 1991. Executive incentives and the horizon problem: an empirical investigation. J. Account. Econ. 14, 51–89. Dehejia, R.H., Wahba, S., 2002. Propensity score-matching methods for nonexperimental causal studies. Rev. Econ. Stat. 84 (1), 151–161. Desai, H., Ramesh, K., Thiagarajan, S., Balachandran, B., 2002. An investigation of the informational role of short interest in the Nasdaq market. J. Financ. 57, 2263–2287. Doyle, J., Lundholm, R., Soliman, M., 2006. The extreme future stock returns following I/B/E/S earnings surprises. J. Account. Res. 44, 849–887. Edmans, A., Fang, V.W., Lewellen, K., 2017. Equity vesting and investment. Rev. Financ. Stud. 30, 2229–2271. Egger, P., Radulescu, D., 2014. A test of the Bolton–Scheinkman–Xiong hypothesis of how speculation affects the vesting time of options granted to directors. J. Corp. Finan. 29, 511–519. Engelberg, J., McLean, R.D., Pontiff, J., 2018. Anomalies and news. J. Financ. 73, 1971–2001. Fama, E., 1998. Market efficiency, long–term returns, and behavioral finance. J. Financ. Econ. 49, 283–306. Fama, E.F., French, K.R., 1993. Common risk factors in the returns on stocks and bonds. J. Financ. Econ. 33, 3–56. Fama, E.F., French, K.R., 1997. Industry costs of equity. J. Financ. Econ. 43, 153–193. Francis, J., LaFond, R., Olsson, P., Schipper, K., 2004. Costs of equity and earnings attributes. Account. Rev. 79, 967–1010. Fu, F., 2009. Idiosyncratic risk and the cross–section of expected stock returns. J. Financ. Econ. 91, 24–37. Gao, F., Lisic, L.L., Zhang, I.X., 2014. Commitment to social good and insider trading. J. Account. Econ. 57, 49–175. Goldman, E., Slezak, S.L., 2006. An equilibrium model of incentive contracts in the presence of information manipulation. J. Financ. Econ. 80, 603–626. Gopalan, R., Milbourn, T., Song, F., Thakor, A., 2014. The duration of executive compensation. J. Financ. 69, 2777–2817. Heath, C., Huddart, S., Lang, M., 1999. Psychological factors and stock option exercise. Q. J. Econ. 114, 601–628. Heckman, J., Ichimura, H., Smith, J., Todd, P., 1998. Characterizing selection bias using experimental data. Econometrica 66 (5), 1017–1098.

18

Journal of Corporate Finance xxx (xxxx) xxxx

J.D. Chi, et al.

Hemmer, T., Matsunaga, S., Shevlin, T., 1996. The influence of risk diversification on the early exercise of employee stock options by executive officers. J. Account. Econ. 21, 45–69. Hribar, P., Collins, D., 2002. Errors in estimating accruals: implications for empirical research. J. Account. Res. 40, 105–134. Hribar, P., Jenkins, N.T., Johnson, W.B., 2006. Stock repurchases as an earnings management device. J. Account. Econ. 41, 3–27. Huddart, S., Lang, M., 1996. Employee stock option exercises. J. Account. Econ. 21, 5–44. Jagolinzer, A.D., Larcker, D.F., Taylor, D.J., 2011. Corporate governance and the information content of insider trades. J. Account. Res. 49, 1249–1274. Johnson, S.A., Ryan, H.E., Tian, Y.S., 2009. Managerial incentives and corporate fraud: the sources of incentives matter. Review of Finance 13, 115–145. Jones, J.J., 1991. Earnings management during import relief investigation. J. Account. Res. 29, 193–228. Kothari, S.P., Leone, A., Wasley, C., 2005. Performance matched discretionary accrual measures. J. Account. Econ. 39, 163–197. Landsman, W.R., Maydew, E.L., 2002. Has the information content of quarterly earnings announcements declined in the past three decades? J. Account. Res. 40, 797–808. Mashruwala, C., Rajgopal, S., Shevlin, T., 2006. Why is the accrual anomaly not arbitraged away? J. Account. Econ. 42, 3–33. Merton, R.C., 1973. An intertemporal capital asset pricing model. Econometrica 41, 867–887. Nagar, V., Nanda, D., Wyoscki, P., 2003. Discretionary disclosure and stock-based incentives. J. Account. Econ. 34, 283–309. Peng, L., Röell, A., 2014. Managerial incentives and stock price manipulation. J. Financ. 69, 487–526. Pontiff, J., 2006. Costly arbitrage and the myth of idiosyncratic risk. J. Account. Econ. 42, 35–52. Rosenbaum, P.R., 2002. Observational Studies, 2nd ed. Springer Series in Statistics, Berlin. Rosenbaum, P.R., 2011. A new U–statistic with superior design sensitivity in observational studies. Biometrics 67, 1017–1027. Rosenbaum, P.R., Rubin, D.B., 1985. Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am. Stat. 39, 33–38. Sloan, R.G., 1996. Do stock prices fully reflect information in accruals and cash flows about future earnings? Account. Rev. 71, 289–316. Stein, J.C., 1989. Efficient capital markets, inefficient firms: a model of myopic corporate behavior. Q. J. Econ. 104, 655–669. Teoh, S.H., Welch, I., Wong, T.J., 1998. Earnings management and the long-run market performance of initial public offerings. J. Financ. 53, 1935–1974. Watts, R., Zimmerman, J., 1990. Positive accounting theory: a ten year perspective. Account. Rev. 65, 131–156. Xu, M., 2013. The Costs and Benefits of Long–Term CEO Contracts. (Working paper).

19