The Incentive Effects of CEO Stock Option Grants on Firm Value

Transcription

1 The Incentive Effects of CEO Stock Option Grants on Firm Value By Craig A. Olson School of Labor & Employment Relations University of Illinois-Champaign/Urbana Revised June 2010 Paper presented at the 3 rd International Meeting of EALE/SOLE June 2010, London. This paper was previously titled The Long-Run Impact of CEO Stock Option Grants on Firm Value.

2 Abstract From the early 1990s through 2007 the median number of stock options held by CEOs increased by an average of 8.8 percent per year. These option grants led to a dramatic growth in CEO pay and, if boards of directors were acting to promote shareholder interests, these grants should have increased shareholder wealth. However, empirical evidence showing a link between option grants to executives and firm performance is limited. Using newly disclosed information on CEO option holdings at the end of the 2006 fiscal year, this study investigates the impact of CEO option holdings on firm stock returns for a sample of over 1000 firms. The numerous option grants received by CEOs produce a nonlinear relationship between the firm s stock price and the incentive effects of the options. Therefore, the incentive effect of options was constructed to capture the impact of the entire incentive schedule faced by CEOs. Two sets of incentive measures were constructed. One set was based on the total intrinsic value of the options held by CEOs and the other set was based on the total Black-Scholes value of the options. No significant effect on firm value was found using a 9 day event window surrounding the disclosure of new information on CEO option holdings. However, statistically significant effects were found using both sets of incentive measures when the model captured the nonlinear incentive schedule and abnormal returns were estimated over the 78 weeks following the disclosure of new CEO option information. The market reacted at a gradual and constant rate to the option information over the weeks following firm disclosure. At week 45 the average CAR for incentive measures based on their intrinsic value implies a 4.9 percentage point difference in firm returns between a firm at the 20 th percentile of the option incentive measure and a firm at the 80 th percentile of the distribution. I also find the Black-Scholes value of underwater options had a large negative effect on firm value.

3 The Incentive Effects of CEO Stock Option Grants on Firm Value A large literature in finance, economics and management has focused on practices and policies designed by firms to monitor senior management and reward executives so they are motivated to make decisions that are in the interests of shareholders. Much of the literature in this area has focused on the composition of boards of directors (Hermalin & Weisbach 1998, Adams, Hermalin & Weisbach 2008), the impact of firm performance on CEO turnover (Coughlan & Schmidt. 1985, Warner, Watts & Wruck 1988, Dahya, McConnell & Travlos 2002, Kaplan & Minton. 2006) and the strength of the relationship between changes in the value of CEO compensation or wealth and changes in firm value to shareholders (Jensen & Murphy 1990, Hall & Lieberman 1998, Aggarwal & Samwick 1999, Murphy 1999). The groundbreaking empirical study on the latter topic was the Jensen & Murphy (1990) analysis that found a very weak link between pay and firm value over the timeframe; average CEO compensation under the control of the board of directors changed by $.75 for every $1000 change in shareholder wealth. 1 Subsequent research finds that since the late 1980s the sensitivity of CEO compensation to changes in firm value has increased significantly because of the growing importance of stock options in CEO compensation contracts (Murphy 1999). For a sample of large firms in 1994 Hall & Lieberman (1998) find the median change in the Black-Scholes value of options held by CEOs for a $1000 change in firm value was $2.15. Other data suggest the number of options held by CEOs has grown substantially over the last 15 years and this growth is an important factor explaining both the rise in total CEO compensation and the stronger link between the payouts CEOs receive under their compensation contracts and changes in shareholder wealth. In 1993 the median number of options held by CEOs in the ExecuComp database was 175,000 and by 2006 the median had grown to 583,000 options. Results from a 1 When the stock owned by CEOs is included in the calculation Jensen & Murphy find the average sensitivity of CEO wealth to changes in firm value is significantly greater than $.75. 1

4 median regression using only a linear time trend show the median number of options held by CEOs increased by 8.8 percent per year from An important gap in the existing research is an evaluation of the impact of CEO stock option holdings on shareholder wealth. Although changes in the value of CEO compensation and changes in a firm s stock price are mechanically linked through the formula that defines the ultimate payout executives receive from options (Max(0, Stock Price Exercise Price)) when they are exercised, it is not clear that there is a causal connection between the option payperformance link imbedded in executive compensation contracts and shareholder wealth. While agency theory predicts the use of stock options and restricted stock will reduce agency costs and increase firm value to shareholders, direct evidence in the management and economics literatures establishing this empirical relationship is limited (Daily, Dalton & Cannella 2003, Lazear & Gibbs 2009). Does the pay-performance link in the compensation contracts cause executives to make decisions that are better for shareholders? If there is a link between the structure of CEO option holdings and firm performance, what characteristic(s) of the options held by CEOs cause executives to make better decisions for shareholders? This study investigates these questions by conducting two stock price event studies where the disclosure hypothesized to influence investor valuation of firms was the disclosure in firms 2006 proxy statements that revealed new information about the composition of CEO option holdings that determined the financial return CEO could earn from their option holdings when making decisions that increase firm value. The typical CEO holds a portfolio of stock options in their compensation package that have different expiration dates and strike prices. Because of new executive pay disclosure rules, information needed by investors (and researchers) to calculate the financial incentives CEOs face from their stock option holdings over a range of stock prices became available for the first time when firms filed their 2006 proxy statements with the SEC. This means analysts can precisely calculate the change in the value of CEO option holdings from 2 Author s calculations based on ExecuComp data. 2

5 decisions that change firm value by 1, 5, or 50 percent change. This study approximates the option incentive schedule for each CEO and tests for a market reaction to this schedule immediately before and after each firm s 2006 proxy filing date. Results from this analysis are contrasted with a model that measures option incentives at only the firm s current stock price. Since the information on CEO option holdings revealed in the 2006 proxy statement had never been revealed previously investors, the market reaction to this information may not have been immediate. Therefore, one event study estimates the market reaction from three trading days prior to the proxy filing thru five trading days after the filing date. In the second study the event period extends from four weeks prior to the filing date thru 78 weeks after the filing date. Two measures are constructed to capture the financial incentives that CEOs face from their option holdings. One measure is based on the total intrinsic value of an executive s options and the other measure is based on the total Black-Scholes value of the options. Neither of these measures exactly captures how a rational, utility maximizing executive is likely to value their options (Hall & Murphy 2002) but both are likely to be highly correlated with executive and market valuations of the financial incentives attributable to CEO stock options. The Incentive Effects of CEO Option Holdings Agency theory (Jensen & Meckling 1976) predicts that executives will make decisions that increase shareholder wealth when those decisions also increase executive utility. Thus, the incentive effect stock options have on executive behavior favorable to shareholders depends on how the utility of their option holdings change when the firm s stock price changes. I denote this value as VSO/ SP where VSO is the value of stock options to an executive and SP is the firm s stock price. Determining VSO/ SP is a difficult problem because executives are hard to survey and typically receive a grant of some number of options on an annual (or more frequent) basis where the strike or exercise price (EP) for options in a grant typically equals the stock price on the grant date. Thus, at any point in time the incentive effect from holding options requires an 3

6 executive (and researcher) to value a heterogeneous set of options with different exercise prices that expire at different times in the future (Hall & Lieberman 1998). In 2006 the median number of different option grants held by the CEOs studied here was six. A plausible model for how an executive values his/her options assumes the total value of all options held is a weighted linear function of the value of an option in each grant weighted by the number of options in the grant: (1) VSO t = (Opt t-1 t )(Val t-1 t) + (Opt t-2 t )(Val t-2 t) + (Opt t-3 t )(Val t-3 t) +... (Opt 1 t )(Val 1 t) In this equation Opt t-k t is the number of unexercised options held at time t that were granted at time t-k and Val t-k t is the value to the executive at time t of an option granted at time t-k. Eq. 1 implies the incentive effect these option holdings have on executive behavior beneficial to shareholders depends on (2) VSO t / SP t = (Opt t-1 t ) (Val t-1 t)/ SP + (Opt t-2 t ) (Val t-2 t)/ SP + (Opt t-3 t ) (Val t-3 t)/ SP +... (Opt 1 t ) (Val 1 t)/ SP This equation shows the total incentive effect of an executive s option holdings is a weighted sum of the number of options held across all option grants where the number of options from each grant is weighted by the incentive effect of each option in the grant. While Opt t-k t is now easily measured and observed because of recent changes in CEO pay disclosure requirements (SEC, 2006), the (Val t-k t)/ SP t terms are difficult to measure because they are likely a complex function of a large number of observable and unobservable variables time left until the option expires, the option s exercise price, the executive s forecast about what will happen to the firm s stock price before the option expires, uncertainty around this stock price forecast, the level and composition of the executive s total wealth, executive risk aversion and the discount rate the executive uses to evaluate payoffs that can occur at different times. How do executives account for these different factors in their valuation of the options they receive? One possible model is the Black and Scholes (1973) that describe how competitive markets with utility maximizing participants value stock options where the owners are free to sell 4

7 their options. They show the holder of a tradable option to buy a share of a non-dividend paying stock at a fixed price in the future will never exercise the option prior to its expiration date because the option owner can always make more money from selling the option than the profit (Max(0, SP-EP)) gained by exercising it. The price obtained from selling the option is more than what can be gained from exercising it because buyers are willing to pay a premium over Max(0, SP-EP) because of the rational expectation that expected future stock price changes will increase the value of (SP-EP) on the expiration date. The premium option owners can earn by selling the option rather than exercising it is equal to the difference between the Black-Scholes value and (Max(0, SP-EP)). The finance literature has long recognized that the Black-Scholes model doesn t describe how executives value the options they hold in their compensation contracts (Huddart and Lang 1996, Carpenter 1998, Hall & Murphy 2002). One important reason for this difference is that executives are typically prohibited from selling their options because firms want executives to hold their options so they will make decisions that promote the interests of shareholders. This means an executive s choice each period after options from a grant have vested is not between selling the option to someone else or holding it another period but between exercising the option or deciding to hold it another period. Since executives cannot sell their options and hold an undiversified portfolio of assets that includes their option holdings and their firm specific human capital, risk averse executives may choose to exercise their options before they expire when they observe a high stock price because this decision allows them to lock-in the gains and diversify their wealth (Hall & Murphy, 2002). This willingness to exercise an option before its expiration date influences the current value of an option and, therefore, VSO t / SP t. 3 3 Option vesting requirements and forfeiture requirements if the CEO leaves the firm implies executives could also exercise options before the expiration date. Executives with short tenure expectations are likely to exercise options earlier than executives with longer tenure expectations which will cause the former group of executives to value options less than the latter group because there is a lower probability of a large stock price increase prior to executive turnover. 5

8 Hall and Murphy (2002) develop a model that describes how a rational, utility maximizing, risk averse executive values the options from a single grant. 4 Building on methods developed for pricing market traded options that were developed after Black-Scholes (Cox et al., 1979), the log-normal distribution of stock prices unfold over time in a two-dimensional binomial tree. At each node in the binomial tree the probability of a price increase or decrease and the price change associated with these two price movements are chosen to produce the stock price distribution predicted for the firm given an estimate of the normal distribution that generates the firm s stock returns. The current value of an option to an executive in the Hall & Murphy model is the value of the option at the first price node (SP 0 ) of the binomial tree and this value is determined by solving by backward induction the executive s valuation problem beginning with the distribution of terminal prices. The value of an option in the current period and at the current price depends on what the executive will do with the option in the next period when faced with a different stock price. The executive starts at the terminal prices, makes a rational choice at each price node and then works backwards through the tree (and time) to solve the utility produced at each node based on whether or not the option is exercised or held another period. In other words, for the terminal prices in the tree in period T, the option will be exercised if the stock price is greater than the exercise price. Then the executive moves to each node at time T-1 and decides whether to exercise or not exercise the option at that price based on whether the utility from exercising at T-1 is more or less than the expected utility from holding the option until period T. This produces an expected utility at each node in period T-1. This logic continues back through the tree until the expected utility of the option for the current period is calculated. Determining the utility of an option to an executive using the Hall & Murphy model requires knowledge about the functional form of an executive s utility function, their risk aversion, their beliefs about the mean 4 They do not solve the model for the more difficult and more realistic situation where an executive owns options from several grants with overlapping terms. 6

9 and variance of the distribution describing how future firm stock returns are generated, the quantity of stock they own in the firm and the value of their other non-firm assets. At the grant date, they find that a dollar change in a firm s stock price raises the executive s valuation of each option by $.50 to $.60 for plausible values of risk aversion and executive wealth. This value changes based on the option s intrinsic value and the time left before the option expires. 5 Their research represents a substantial step forward; before this study researchers (and executives) didn t have a plausible estimate of the incentive effects of options from a single grant for a risk averse, utility maximizing executive. For this study, a number of important issues are left unresolved. Most importantly, their simulations are based on a CEO holding a set of options that have identical terms. Thus, the model fails to incorporate an important feature of CEO option holdings - grants with different expiration dates and strike prices. Therefore, this study does not attempt to apply the Hall & Murphy model to the valuation problem faced by the executives in this study. However, one implication of the HM study is that executive valuation of their option holdings are likely to be highly correlated with two measures that are easily calculated by researchers, executives and investors. One measure is the total intrinsic value (TIV) of the CEO's option holdings and the second measure is the total Black-Scholes value (TBSV) of the option holdings. Since executives receive option grants at multiple times over their tenure the total intrinsic value of unexercised options at time t is (3) TIV t = (Opt t-1 t )(Max(0,SP t EP t-1 )) + (Opt t-2 t )(Max(0,SP t EP t-2 )) + (Opt t-3 t )(Max(0,SP t EP t-3 )) +. (Opt 1 t )(Max(0,SP t EP 1 )) Since each term in Eq (3) is equal to zero if the option is underwater (SP t EP t-k ), if I t-k t equals an indicator variable equal to 1 if SP t EP t-k and 0 otherwise, Eq (3) can be rewritten as the total value of all in-the-money (SP > EP) options held at time t: 5 See Hall & Murphy (2002), Table 3. 7

10 (4) TIV t = (I t-1 t) (Opt t-1 t )(SP t EP t-1 ) + (I t-2 t) (Opt t-2 t )(SP t EP t-2 ) + (I t-3 t) (Opt t-3 t )(SP t EP t-3 ) +. (I 1 t) (Opt 1 t )(SP t EP 1 ). TIV t / SP t equals the total number of options at time t that are not underwater : (5a) TIV t / SP t = (I t-1 t) (Opt t-1 t ) + (I t-2 t) (Opt t-2 t ) + (I t-3 t) (Opt t-3 t ) +. (I 1 t) (Opt 1 t ) This derivative is not a continuous function but has kink points at the exercise price for each option grant as shown by the solid line in Figure 1. The kink points (EP A, EP B, EP C ) correspond to the exercise prices for three different option grants and the slope of each line segment equals the number of above water options at P t ; if all of the options are underwater (P t < EP A ) TIV t / SP t equals zero. There is some indirect evidence to suggest executives do evaluate the incentive effects of options using IV t / SP t. 6 The decisions of some firms to reprice underwater options by lowering the exercise price or issuing new options at a lower price suggests these decisions are made because of the perception that underwater options do not provide executives with an incentive to improve firm performance (Chance et al. 2000, Pollack, Fischer & Wade 2002). This suggests executives may not be motivated by underwater options because decisions that raise firm value have little perceived effect on the option s value to the executive. The second incentive measure used in this analysis is the change in the total Black- Scholes value (BSV) of an executive s option holdings for a dollar change in the firm s stock price or TBSV t / SP t. This equals: (5b) TBSV t / SP t = ( BSV t / SP t ) t-1 (Opt t-1 t ) + ( BSV t / SP t ) t-2 (Opt t-2 t ) + ( BSV t / SP t ) t-3 (Opt t-3 t ) +. ( BSV t / SP t ) 1 (Opt 1 t ) There are important differences between TIV t / SP t and TBSV t / SP t. As Figure 1 shows, TIV t / SP t changes by a dollar for each option that is above-water at SP t and is unchanged by 6 I don t mean to suggest an option s total value to an executive equals the option s intrinsic value. An option s intrinsic value seriously understates the total value of options to executives because it ignores the increase in an option s intrinsic value that is likely to occur because of stock price increases that are likely to occur before the option expires. If the intrinsic value of an option represented the value of the option to an executive we d expect executives to exercise all of their above-water options as soon as they vest. This is clearly not the case; in 2006 the median intrinsic value of vested options held by CEOs in the sample studied here was $15.9 million. 8

11 options that are underwater. Thus, the impact of an option on TIV t / SP t depends only on whether the stock price is above or below the option s exercise price. In contrast, since TBSV t / SP t depends on the BSV of each option, the incentive effect depends on all the variables that determine an option s BSV. Thus, unlike TIV t / SP t, underwater options are given positive weight in an executive s evaluation of the financial incentives he or she faces. The dashed line in Figure 1 shows the total BSV for the same set of options used to construct the TIV schedule. Notice that because the Black-Scholes function is a continuous function in the firm's stock price, the TBSV schedule is a smooth function that does not have the kink points found in the TIV schedule. Also note that while TBSV > TIV, for stock prices above EP A TIV t / SP t and TBSV t / SP t are similar. For firms studied here, the mean value of BSV/ SP calculated across CEOs in the sample studied here was $.81 and $.88 calculated over the options that are in-the-money (SP > EP). This latter value is close to the value of 1.0 for TIV/ SP when SP> EP. Hall and Murphy (2002) provide some evidence to suggest how TIV t / SP t and BSV t / SP t compare to (Val t )/ SP. For options that are above-water (Val t )/ SP < BSV t / SP t < TIV t / SP t = 1 and for underwater options 0 = TIV t / SP t < (Val t )/ SP < BSV t / SP t. 7 These relationships imply the magnitude of the incentive effects constructed from the IV and BSV of an executive s option holdings relative to (Val t )/ SP t depends on the mix of under and above-water options. When all of an executive s options are above-water TIV t / SP t and TBSV t / SP t overstate the incentives confronting the executive. On the other hand, VSO t / SP t will fall between TIV t / SP t and TBSV t / SP t if the executive holds a sufficient number of underwater options. In the two event studies presented below the incentive effect of options are measured using both TIV t / SP t and TBSV t / SP t. 7 See Figure 4 in Hall and Murphy (2002). 9

12 Measures of the Incentive Effects of CEO Option Holdings The new executive pay disclosure requirements in 2006 provide for the first time the information researchers and investors needed to calculate equations (5a) and (5b) for each executive. 8 Table 1 summarizes the incentive measures constructed from the TIV and TBSV of the options held at the end of fiscal year 2006 for all the CEOs included in ExecuComp for the 2006 fiscal year. The firms (n=1286) these executives lead are the firms used in the event studies discussed in the next section. 9 All of the values in Table 1 are calculated using the closing stock price for fiscal year Row (a) reports the distribution of the total number (1000s) of unexercised options held by CEOs in the sample, row (b) reports the number of in-the-money (above-water) options and row (c) reports the distribution of the fraction of options that are inthe-money. Rows (a) and (b) show a skewed distribution in both the number of options and the number of in-the-money options. The median number of options held by CEOs was 565,000 and the median number of in-the-money options was 414,000. The gap between the median and the 80 th percentile for the number of in-the-money options was over 800,000 options while the difference between the 20 th percentile and median was about 350,000 options. Row e shows the distribution for TBSV/ SP. For an executive at the median the BSV of his/her options changes by $438,000 for a dollar change in the firm s stock price. This value is very close to the median 8 Hall & Lieberman (1998) estimate the BSV of executive option holdings by making assumptions about the order and price at which options held by CEOs are exercised. 9 The calendar dates for a firm s fiscal year can vary from firm to firm depending on when they choose to begin their fiscal year. All of the values in Table 1 are based on the ending fiscal year 2006 stock price. In the event studies the options are evaluated at a stock price 5 weeks before the firm filed its 2006 proxy statement. 10 The Black-Scholes value for each option held by a CEO was calculated using data on the standard deviation of firm returns and dividend yields reported by ExecuComp. Firms with very high volatility (>1.149) or very low volatility (<.167) were recoded to these threshold values. Firms with dividend yields greater than 5.8 percent were recoded to 5.8 percent and firms paying dividends with a yield of less than.2 percent were recoded to.2 percent. For options that expire in 7 or fewer years, the time remaining until the option expires was left unchanged. For options that had more than 7 years remaining until they expire, the term of the option was set equal to 70 percent of the time left until expiration. This was done to adjust for the fact that executives often exercise options before they expire. The risk free interest rate was set equal to yields on U.S. treasury bonds where the term of the bond was matched to the time remaining until the option expires. 10

13 value of TIV/ SP ($414,000). Comparing rows (b) and (e) shows a very similar distribution for TBSV/ SP and TIV/ SP. The correlation between TBSV/ SP and TIV/ SP is.94. TIV/ SP and TBSV/ SP may not describe how CEOs evaluate the incentive effect of holding options because the effort required to change the firm s stock price by a dollar is likely to vary from firm to firm and be related to the firm s stock price. In particular a dollar change in a firm s stock price represents a very large percentage change in the size of the firm if the stock price is low and a very small percentage change in the size of the firm if the stock price is high. Since the CEO effort required to change firm value by a large percentage is likely to be very different from the effort required to change firm value by a small percentage amount, the incentives effects of TIV/ SP and TBSV/ SP are likely to be different at different stock price levels. An alternative incentive metric that avoids this problem is the Jensen and Murphy (1990) which measures the dollar change in either TIV or TBSV for a $1000 change in firm value. 11 The remaining rows (f-j) in Table 1 summarize the distribution of the Jensen-Murphy incentive measures constructed from TIV/ SP and TBSV/ SP. I will refer to these measures as TIV/ ($1000 FV) and TBSV/ ($1000 FV). The top measure in each cell is TIV/ ($1000 FV) and the second measure in parentheses is TBSV/ ($1000 FV). Row f summarizes the distributions of these measures evaluated at the year end 2006 fiscal year stock price. The median value of TIV/ ($1000 FV) was $5.15 and the median for TBSV/ ($1000 FV) was virtually the same, $5.14. While the sample of firms included in this study is different from the Hall & Lieberman (1998) sample, it is worth comparing the estimates of TBSV/ ($1000 FV) in the two samples. The median value in their sample was $2.15 and the the median in this sample $5.14. The difference in these values suggest the financial incentives provided to CEOs by stock option holdings is much stronger now than in The Jensen-Murphy measure equals ( IV t / SP t ) ( SP t / $Firm Value) t (1000) = (No of in-the-money Options)/(No of shares)*1000. See Baker and Hall (2004) for a discussion of alternative incentive measures. 11

14 Inspection of the values of TIV/ ($1000 FV) and TBSV/ ($1000 FV) across the different deciles show these measures are also similarly distributed; they are highly correlated with an r =.89. These two measures are also highly skewed. For CEOs at the 80 th percentile of these distributions a decision that increases firm value by $10 million would increase the total intrinsic value of their options by $139,600 and the BSV of their option holdings by $135,900 or an amount equal to percent of the gain to shareholders. In contrast, for executives at the 20th percentile the same decision increases TIV by $7100 or the BSV by $10,000. The results in row (f) refer to the distribution of the option pay-performance relationships across firms evaluated at each firm s 2006 fiscal year end stock price. Because each CEO typically holds options from multiple option grants at different exercise prices, TIV/ ($1000 FV) and TBSV/ ($1000 FV) change with changes in the firm s stock price. For TIV/ ($1000 FV) the kink points in the option pay-performance schedule shown in Figure 1 causes TIV/ ($1000 FV) to differ across the different line segments based on the total number of in-the-money options. A decision that causes a stock price change large enough to move an executive to another line segment produces an option pay-performance relationship different from the value produced by a small change in the stock price around P t. The relationship between a firm s stock price and TBSV/ ($1000 FV) is not the kinked linear schedule because each option s delta ( BSV/ SP) is a continuous positive function of the firm s stock price. Thus, holding the other variables constant that determine ( BSV/ SP), TBSV/ ($1000 FV) is also positively related to the firm s stock price. Rows g-j of Table 1 show TIV/ (1000 FV) and TBSV/ ($1000 FV) at 50, 70, 130 and 150 percent of the firm s 2006 fiscal year-end stock price. These data show an interesting pattern; the decline in TIV/ (1000 FV) as the stock price declines from 150 to 50 percent of the firm s 2006 year end stock price is much steeper than the decline in TBSV/ (1000 FV). For example, at the median TIV/ (1000 FV) declines from $6.40 at 150 percent of SP to $.16 at 50 percent of SP while TBSV/ ($1000 FV) declines from $5.14 to $3.16. This difference is 12

15 explained by the different weight the two measures give to incentives that are underwater underwater options have no incentive effect as measured by TIV/ (1000 FV) and a positive effect as measured by TBSV/ ($1000 FV). This variation in the incentive effects over a range of stock prices is a central part of the empirical analysis reported below. I hypothesize that the incentive effects of options depend on the entire schedule of option values over a broad price range and not simply the incentive effect measured at one stock price. Until 2006 readily accessible detailed information about the number and value of stock options held by CEOs from different grants was not available. From 1992 through 2005 executive compensation disclosure rules required that firms disclose in their annual proxy statement the total number of vested and unvested options held by the CEO, the total intrinsic value of all options held by the CEO, the terms (number, exercise price and expiration date) of new options granted during the fiscal year and an estimate of the cost of the newly granted options to the firm. What was not available prior to 2006 was the data needed to determine TIV/ SP and TBSV/ SP. 12 The new information made available to investors by the new 2006 disclosure rules is hypothesized to have had an impact on firm value. We test for this impact using two stock price event designs that make different assumptions about when and what investors concluded from the option information revealed in each firm s 2006 proxy statement. Each design uses daily return data, including dividends, from CRSP (University of Chicago) and the option information for CEO reported in S & P s ExecuComp database. All of the CEOs from firms included in ExecuComp that reported the information needed to construct Table 1 were included in the analysis (n=1286). The first design assumes investors were able to quickly evaluate the impact on firm value of the incentives provided by the stock options held by CEOs in the days immediately surrounding the filing of the firm s proxy statement for For this design a 12 One could approximate the schedule using the method used by Hall & Lieberman (1998) by tracking option grant dates and estimating which options are exercised each year of an executive s tenure. 13

16 standard event study was conducted using a 9 day event window. The second design allows for the possibility that the market responded only very slowly to the information revealed in the proxy statement about the two incentive measures. Because the information on CEO options needed to construct TIV/ (1000 FV) and TBSV/ ($1000 FV) for each firm had never been previously available to investors for a large sample of firms, the market may have required more time to interpret the new option information and evaluate its impact on firm profitability. For this design we estimate abnormal returns over a period up to 1.5 years (78 weeks) following the proxy filing date. For each event window the analysis was conducted using four different incentive measures. In the first pair of analysis TIV/ (1000 FV) and TBSV/ ($1000 FV) were evaluated at 100 percent of the firm s average closing daily stock price calculated over the trading days in the fifth week preceding the week the firm filed its 2006 proxy statement (event week= -5). These measures are denoted as JM-IV100 and JM-BSV100. In the second pair of analysis the incentive measures are evaluated at five different stock prices: 50 percent, 70 percent, 100 percent, 130 percent and 150 percent of the average price in event week -5. These variables are, respectively, JM-IV50, JM-IV70, JM-IV100, JM-IV130, JM-IV150 for the measures of TIV/ (1000 FV) and JM-BSV50, JM-BSV70, JM-BSV100, JM-BSV130 and JM-BSV150 for the measures of TBSV/ (1000 FV). Comparing the results across the alternative event windows and measures permit tests of whether the estimated impact of option incentives depends on the options intrinsic or BSV, whether incentive effects are captured at a single stock price and how quickly (and if) the market responded to the new information about executive option holdings. A Stock Price Event Study Using the Days Surrounding the Proxy Filing Dates If investors have well defined expectations about how the structure of options in an executive s compensation contract influence firm profitability, new information revealed to 14

17 investors about the structure of option contracts that cause investors to revise their profit expectations for the firm will produce a quick adjustment in the firm s stock price. Under these conditions, evaluating the impact of the option contracts on firm profitability focuses on firm stock prices in the days surrounding the date when investors acquired the new information about CEO held options. Therefore, this section of the paper tests for abnormal stock returns around the proxy filing dates by comparing returns three days before the filing date through five days after the filing date with an estimate of what returns would have been if information on TIV/ (1000 FV) and TBSV/ ($1000 FV) had not been revealed to investors. The difference between the actual return in the event window and the predicted return is the firm s abnormal return (AR i,t ) for period t: AR i,t = (R i,t I i,t = 1) - E(R i,t I i,t = 0) Where (R i,t I i,t = 1) is the observed return for firm i for period t in the nine day event and E(R i,t I i,t = 0) is the predicted return given no new information on CEO options. I i,t is an indicator variable for the information revealed on event day 0 and equals 1 if information on options is revealed to investors and 0 if no information is revealed. The average abnormal return for period t calculated over the N firms in the sample is AAR = ( 1/ N) N t AR i, t i= 1 If the market takes T days to adjust to the new information on options, the mean total effect is the average cumulate average abnormal return or the sum of the market s average reaction over the T days: CAR = 5 t= 3 AAR t In this analysis the AAR t s will be estimated from the coefficients on g(i i,t ) in the following regression model where g(i i,t ) is a function of the information revealed about firm i in period t: 15

18 (6) R i,t -RF t = β 0 + β 1 (MKT t -RF t ) + β 2 (SMB t ) + β 3 (HML t ) + β 4 (MOM t ) + α j *(IND i,j ) + AAR t *g(i i,t ) + ε i,t This model is estimated using daily data on the N firms from the nine day event window and all of the trading days for a firms 2006 fiscal year. Since the proxy filing date for the 2006 fiscal year occurs after the end of the 2006 fiscal year, the 2006 fiscal return data are the control data used to estimate what would have happened in the absence of new information on CEO options. The abnormal average returns depend on a function of the information revealed in the event window and is discussed further below. The standard errors in the model are estimated by clustering on the calendar date to account for the overlap in event times across firms. The first four variables in Eq (6) include the three factors identified by Fama & French (1993) and a fourth factor identified by Carhart (1997). 13 The three Fama-French factors are: the difference between the return on a market portfolio of stocks and the risk free return (MKT t RF t ) for period t, the difference in returns from a portfolio of large firms and a portfolio of small firms (SMB t ) and the difference in the returns of a portfolio of firms with a high book-to-market ratio and a portfolio of firms with a low book-to-market ratio (HML t ). The fourth factor captures market momentum (MOM t ) which has been found to be a significant predictor of returns after controlling for the Fama-French factors (Carhart 1997). These variables are now often used to model returns in event studies (Greenstone, Oyer & Vissing-Jorgensen 2006). The return model is also estimated with and without a set of 2-digit industry dummy variables to capture industry specific risk factors not captured by the four other variables in the model. The results are very similar in the models with and without industry controls so I only present and discuss the results without the industry controls Data for these four factors was downloaded from Kenneth French s web page at mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html. 14 Results for the models with industry controls are available from the author. 16

19 In the first specification of Eq. (6) the information revealed in the event window is measured using a simple set of dummy variables (EW i,t ) denoting each of the days in the event window. This specification tests for average abnormal returns on the event days compared to the non-event days in the sample. Because a proxy statement contains a wealth of information about the firm besides information about CEO option holdings, estimates from this specification does not test for an average impact of the CEO option holdings on average firm stock prices; it only tests whether the net effect of all the new information revealed across all of the firms had a significant net positive or negative effect on stock prices. To test for the impact of CEO option holdings on the days surrounding the proxy filing date, g(i i,t ) was defined as a set of nine event day dummies (EW i,t ) were interacted with measures of TIV/ (1000 FV) and TBSV/ ($1000 FV). In the first specification the event window dummies were interacted with JM-IV100 and JM-BSV100. As noted earlier, this specification assumes the incentive effects of options are adequately captured by evaluating each incentive measure at a single stock price. In this specification the coefficient on each interaction term captures the estimated mean impact on daily returns of $1 change in either JM-IV100 and JM-BSV100 on returns. This specification will be inadequate if the incentive effect of options depends on the entire incentive schedule. The second specification controls for the schedule of options by measuring the incentive effects at the five different stock prices. For example, the model for TIV/ (1000 FV) is: (7) R i,t -RF t = β 0 + β 1 (MKT t -RF t ) + β 2 (SMB t ) + β 3 (HML t ) + β 4 (MOM t ) + α j *(IND i,j ) + C -50 t*(ew i,t )(JM-IV50 i,t ) + C -30 t*(ew i,t )(JM-IV70 i,t ) + C 06 t*(ew i,t ) (JM-IV100 i,t ) + C +30 t*(ew i,t )(JM-IV130 i,t ) + C +50 t * (EW i,t )(JM-IV150 i,t ) + ε i,t 17

20 There are several points to note about this specification. First, the interactions between the five points of the incentive schedule and the event dummies produce 45 variables (9 event days x 5 incentive measures) and the coefficients on these variables measure how expected returns vary across firms over the event days based on an approximation of the shape of each firm s TIV/ (1000 FV) schedule. Note that if an executive faces a linear intrinsic value schedule with no kink points then the values of TIV/ (1000 FV) are identical at all five points along the schedule and AAR for period t is (C 50 + C 70 + C 100 C C 150 ) and the estimated CAR equals the running sum of the 45 coefficients: m m CAR[n, m] = ( L + CL + CL + CL + m L= n L= n L= n L= n L= n m m 150 L C C ) A test of the hypothesis that the sum of the 45 coefficients equals zero or C -50 t + C -30 t+ C 06 t + C +30 t + C +50 t ) = 0 tests whether this estimated linear incentive schedule has an impact on firm value over the period from day n to day m. The results show in Figures 2a and 2b plot the CARs for the specifications that interacts JM-IV100 and JM-BSV100 with each of the event day indicator variables. In this figure the CAR is the running sum of the coefficients JM-IV100 and JM-BSV100 on these variables over [- 3 days, + 5 days]. CAR[-3 days, +5 days] will be positive if incentives "matter" and the incentive effects of options on firm value is adequately captured by one of these measures evaluated at a single stock price. The data fail to support this hypothesis for both incentive measures; for each measure CAR[-3 days, +5 days] is close to zero and is not statistically different from zero. Figures 3a and 3b shows the CARs and the 95 percent confidence interval for the 9 days using Eq (7) where the incentives are evaluated at five stock prices. For each event day the AAR is equal to the sum of the coefficients over the five variables. This specification 18

21 also fails to find a significant impact of either TIV/ (1000 FV) or TBSV/ ($1000 FV) on returns over this nine day event window. There are at a variety of possible explanations for these statistically insignificant results. First, the incentive effects of options on CEOs may not be adequately captured by a measure based on either the options intrinsic or Black-Scholes values because stock options do not have a significant impact on executive decision-making. Alternatively, TIV/ (1000 FV) and TBSV/ ($1000 FV) may not capture the financial incentives executives face when making decisions that affect firm value. A third possibility is that the information on executive option holdings disclosed in the 2006 proxy statement was already known to investors and was already reflected in the firms stock prices by the time firms filed their 2006 proxy statements. Fourth, the information about option holdings was new to investors and the nine day event window was not long enough for investors to recognize, process and respond to the entirely new information revealed to investors about CEO option holdings. A number of finance studies that have focused on other events and have found it may take several months for the market to fully adjust to the profit implications of new information (Dharan & Ikenberry 1995, Michaely et al. 1995, Mitchell & Stafford 2000). 15 For example, Lee & Mas (2009) find it takes about a year and a half for stock prices to adjust to new information about a successful unionization drive. In the next section I attempt to narrow down this list of competing explanations by using a 83 week event window. An Analysis of the Long-term Stock Price Reaction to TIV/ (1000 FV) and TBSV/ ($1000 FV) In this section I report results that replicate the method used above but instead of a nine day event window, abnormal returns are estimated for the period four weeks before the firm s 2006 proxy filing date thru the 78th week after the week the proxy statement is filed with the SEC. 15 See Fama (1998) for a discussion of the issues raised by these findings. 19

22 The model is identical to the models reported in the preceding section except the event period indicator variables (EWEEK i ) refer to a calendar week relative to the week a firm filed its 2006 proxy statement. 16 Because daily stock returns data are analyzed, the coefficients on the interaction terms between the event period indicators and the incentive schedule measure(s) describe the impact of a dollar change of either TIV/ (1000 FV) or TBSV/ ($1000 FV) on mean daily returns for the event week. Therefore, CAR[0 wk, +m wk] equals m L= 0 C L *(EWDAYS L ) where C L is the coefficient on the interaction term between the event week indicator and the incentive measure and EWDAYS L is the number of trading days in week L. If each trading week in the event window has the same number of trading days then CAR[0 wk, +m wk] equals m ( EWDAYS). While most trading weeks have exactly five L= 0 C L trading days, a few weeks have either three or four days. This number varies from firm to firm depending on when a firm filed its 2006 proxy statement. Over event weeks 0 through event week 78 the mean number of trading days per event week calculated over the sample of firms was 4.85 days with a standard deviation of.05 days. I assume each event week has 4.85 trading 16 There are other methods that have been proposed to estimate the long-term stock price reaction to the release of new information about the firm. One method is to match each firm with a portfolio of comparable untreated firms and compare the mean difference in returns before and after the release of the information. See Lee & Mas (2009) for an example using this method. This strategy won t produce unbiased results in this study because virtually all firms revealed new information on CEO option holdings for FY 2006 during FY Therefore, the difference between a firm in the sample and the return for the comparable portfolio reflects the impact of options on both the individual firm and the average effect for the portfolio. A second method involves calculating abnormal returns for each firm by estimating a market model for each firm and then calculate AAR over the sample of firms. This method has two problems in the present study. First, it is very hard to get the correct standard errors because of correlated errors created when event periods for different firms in the sample fall on the same calendar period. More importantly, the AAR calculated across firms combines returns for firms that have different intrinsic value schedules and will not identify the impact of TIV/ (1000 FV) or TBSV/ (1000 FV). A third strategy involves constructing calendar time portfolios (Fama 1998) of firms with similar treatment values. This is not easily done in this study because treatment intensity, TIV/ (1000 FV) or TBSV/ (1000 FV), is a function of five different variables. The regression framework used here identifies the effects of TIV/ (1000 FV) and TBSV/ (1000 FV) and the standard errors are corrected for clustering on the calendar day. 20

23 days. 17 For the models that measure incentives at the five different stock prices CAR[0 wk, +m wk] is: m (8) CAR[0 wk, +m wk] = 4.85)( C L + C L + C L + C L + m m L= 0 L= 0 L= 0 L= 0 L= 0 m m 150 L ( C ). Figures 4a and 4b show the plots of CAR[0 wk, +m wk] using the specifications that include the event week indicator variables interacted with either JM-IV100 or JM-BSV100. These plots show the results for long-run abnormal returns are comparable to the results using the nine day event window; the CARs are insignificant over all the weekly event windows from week 0 through week 78 using either incentive measure. Figure 5a plot the CARs for the models that includes the 415 interactions between the 83 event week indicators [-5 wk, +78 wk] and the TIV incentive measures evaluated at the five prices. The CARs shown in these figures are the running sum of the five coefficients multiplied by 4.85 for [0 wk, +78 wk] because the CARs for [-5 wk, -1 wk] were statistically insignificant. In this specification the event window for each firm began 4 weeks prior to the calendar week the firm filed its proxy statement and extended for 78 weeks (1.5 years) beyond the filing week. For the intrinsic value measure, this value equals the CAR for a firm with a linear intrinsic value schedule and estimates the effect on the CAR of a dollar change in the intrinsic value of CEO options for a $1000 change in shareholder wealth. Figure 5a, which plots the CARs for the IV incentive measures, shows a positive linear relationship between CARs and event weeks beginning with event week 0 through about event week 45 to 50 and it then flattens out for the remaining event weeks. Also, over most of the period each CAR is significant or close to significant at the.10 level (2-tail test). This pattern suggests the average daily abnormal return for an event week was roughly constant each week through about event week 45 and implies a relatively slow and constant market adjustment to the information about TIV/ (1000 FV) contained in each firm s proxy statement. The fact that the CARs are relatively constant after 17 All the results remain virtually unchanged when EWDAYS is allowed to vary across event weeks and firms. 21