I need help with the article. I need to do 2 pages summary for the article ( CEO Turnover and Audit Pricing) I attatch . I attach the summary sample of article ( On the Timing of CEO Stock Option Awards) to do the same format .

MANAGEMENT SCIENCE informs Vol. 51, No. 5, May 2005, pp. 802-812 issn 0025-1909 eissn 1526-5501 05 5105 0802 doi 10.1287/mnsc.1050.0365 2005 INFORMS On the Timing of CEO Stock Option Awards Erik Lie Henry B. Tippie College of Business, University of Iowa, Iowa City, Iowa 52242-1000, erik-lie@uiowa.edu T his study documents that the abnormal stock returns are negative before unscheduled executive option awards and positive afterward. The return pattern has intensied over time, suggesting that executives have gradually become more effective at timing awards to their advantage, and possibly explaining why the results in this study differ from those in past studies. Moreover, I document that the predicted returns are abnormally low before the awards and abnormally high afterward. Unless executives possess an extraordinary ability to forecast the future marketwide movements that drive these predicted returns, the results suggest that at least some of the awards are timed retroactively. Key words: CEO stock option awards; timing History: Accepted by David Hsieh, nance; received February 24, 2004. This paper was with the author 1 1 months for 1 revision. 2 1. Introduction CEO option grants between 1981 and 1992, and these returns are actually more negative for the sample of scheduled awards than for their overall sample. However, they nd little evidence of positive abnormal returns following the awards.1 The somewhat conicting results in extant literature suggest that further analysis of the stock price returns around option awards would be fruitful. The purpose of the initial part of this study is to provide such an analysis. I gather a sample of 5,977 CEO stock option awards from 1992 through 2002, 1,668 of which have sufcient information to be classied as unscheduled and 1,426 as scheduled. The stock return pattern for unscheduled awards is strong and striking. The average abnormal return during the 30 trading days leading up to the awards is 3%, most of which occurs during the 10 days immediately before the award. After the unscheduled awards, there is a sharp reversal; during the rst 10 days afterward the average abnormal return is 2%, and it is almost another 2% during the next 20 days. I document a similar pattern for scheduled awards. However, it is considerably weaker, presumably because awards that I classify as scheduled allow less leeway in setting the grant date. In particular, the abnormal stock return is roughly 1% during the 30 days before the scheduled awards Stock options are generally granted with a xed exercise price equal to the stock price on the award date. If executives can inuence the timing of a grant, they might therefore time it to occur (i) after an anticipated future stock price decrease, (ii) after a recent price decrease that they perceive to be unwarranted by fundamentals (in which case the price would gradually increase in the future), or (iii) before an anticipated stock price increase. In any of these cases, self-serving behavior by executives should manifest itself in stock price decreases before stock option grants and/or stock price increases afterward. Yermack (1997) examines the stock returns around 620 stock option awards to CEOs between 1992 and 1994. While the stock returns leading up to the award dates are normal, the stock returns during the 50 trading days afterward exceed those of the market by more than 2%. He interprets these results as evidence that executives opportunistically time awards to occur before anticipated stock price increases. Aboody and Kasznik (2000) investigate a sample of 2,039 scheduled option awards to CEOs between 1992 and 1996. They focus on scheduled awards to remove the possibility that the results are attributable to opportunistic timing of the awards. The abnormal returns before scheduled awards are statistically indistinguishable from zero, while the abnormal returns during the subsequent 30 days are almost 2% and statistically different from zero. They interpret these ndings to suggest that executives opportunistically time the release of information around xed option awards. Using a different data source, Chauvin and Shenoy (2001) nd evidence of negative abnormal returns before 783 1 One might argue that the results in Chauvin and Shenoy (2001) are not directly comparable to those of Yermack (1997) and Aboody and Kasznik (2001). Whereas Yermack and Aboody and Kasznik compile their samples of option awards from yearly proxy statements, Chauvin and Shenoy get their sample from backled reports supplied by the companies shortly after reporting changes in May 1991. Thus, it is conceivable that the results in Chauvin and Shenoy suffer from sample-selection bias. 802 Lie: On the Timing of CEO Stock Option Awards 803 Management Science 51(5), pp. 802-812, 2005 INFORMS and roughly 1% during the 30 days afterward. In the remainder of my analysis, I focus on the sample of unscheduled awards, because the main focus of this paper is the timing of awards (rather than the timing of information releases around awards). Next, I examine whether the documented return trends arising from award timing have changed over time. The exposure of opportunistic behavior in Yermack (1997) or recent scandals such as that involving Enron might have made executives more reluctant to engage in such behavior. Alternatively, executives might have become more effective in timing the awards to their advantage. I show that the abnormal return trends around unscheduled awards have intensied over time. This suggests that executives are getting better or more aggressive at opportunistically timing awards during the sample period. Thus, Yermack's exposure of award timing did little to minimize these activities. If anything, it might have had the opposite effect. The results might further explain why Yermack nds no evidence of poor stock price performance leading up to awards using a sample from 1992 through 1994, as the effect seems rather modest for this earlier period, and why Chauvin and Shenoy (2001) nd scant evidence of good stock price performance after awards using a sample from 1981 through 1992. If the distinct stock return pattern around scheduled awards is entirely attributable to executives timing awards relative to expected future price patterns, their collective ability to forecast future price movements based on inside information is striking. I propose a new hypothesis that could also explain the documented return patterns. In particular, the awards might be timed ex post facto, whereby the grant date is set to be a date in the past on which the stock price was particularly low. Such retroactive timing obviously requires little skill, although outsiders might perceive it to be fraudulent. In any event, it is unlikely that outsiders would ever learn of it, because the company does not publicly report the grant date until months thereafter. To test the ex post facto timing hypothesis, I examine the predicted stock returns from the threefactor model of Fama and French (1993) around the unscheduled awards. I nd that these returns are abnormally low before the awards and abnormally high afterward. Unless executives have a superior ability to forecast future short-term marketwide movements that drive the predicted stock returns, the results indicate that at least some of the ofcial grant dates must have been set retroactively. However, some caveats are in order. First, it is not impossible that insiders are able to predict future shortterm marketwide movements, and that this explains some of my results. Indeed, Lakonishok and Lee (2001) provide evidence based on insider trading that insiders can at least forecast long-term marketwide movements, but I am not familiar with evidence that insiders can forecast short-term (such as daily or weekly) marketwide movements. Second, it is not clear that retroactive timing is in violation of the stock option plan. In fact, the standard legal document behind the stock option plan does not specify whether a grant date can be set retroactively. Finally, my analysis is designed to uncover evidence of retroactive timing in the aggregate, and might be useless in identifying exactly which rms engage in such activities. Despite these caveats, my new evidence of selfdealing is likely to generate substantial controversy and perhaps prompt changes in both the stock option plan and how companies report their option awards. The remainder of the paper is organized as follows. The next section discusses potential opportunistic behavior around executive stock option awards and the predicted stock price patterns. Section 3 describes the sample. Section 4 presents empirical results. Finally, 5 summarizes and concludes. 2. Opportunistic Behavior Around Option Awards and Predicted Price Effects Once a company has adopted a stock option plan (which requires a vote of approval by shareholders), the board of directors generally assigns the administration of the plan to the compensation committee. The compensation committee ofcially determines the size and timing of stock option grants, but there are several reasons to suggest that executives affect these decisions. First, Yermack (1997) nds that executives often propose the parameters of the stock option grant, whereas the compensation committee merely raties these proposals. Second, executives might inuence the committees' decisions via their close friendships with individual committee members. Third, executives might inuence the timing of the compensation committee meetings, which regularly coincide with the award date. (Yermack 1997, Chauvin and Shenoy 2001 describe this process in greater detail.) A key feature of executive option awards is that the exercise price typically equals the stock price on the day of the award. Because option values decrease with the exercise price, executives naturally prefer for the stock price to be as low as possible (and ideally lower than the fundamental value) on the award date to increase the value of their compensation. This preference might give rise to opportunistic behavior.2 2 Zhang (2002) discusses an analogous type of opportunistic behavior around option awards, in which executives grant options to Lie: On the Timing of CEO Stock Option Awards 804 If the awards are unscheduled (i.e., the options are not awarded on the same date every year), executives might use their inuence to time the awards on a date when the stock prices are particularly low. First, if executives perceive the current prices to be higher than the fundamental value and/or expect the prices to fall in the near future, they might try to push back the award date. For example, if they expect that the capital market will be disappointed in the current quarter's earnings, they might postpone any award until after the earnings announcement date. Such behavior should manifest itself in poor stock price performance leading up to the award dates. Second, if executives perceive recent price drops to be unwarrantedfor example, because of rumors about the company or its products that executives know to be falsethey might promote immediate awards to take advantage of the articially low prices. As the capital market realizes that the stock is undervalued, the price should increase. Third, if executives expect future price increases, irrespective of past price performance, they might also advocate immediate awards. An example of this would be that managers believe that the current period's earnings will pleasantly surprise the capital market when announced in the future. In this study, I propose an alternative way of opportunistically timing the awards that does not require the ability to forecast future stock price movements. In particular, the grant date could simply be set to be a past date on which the market price was particularly low. A necessary condition for such retroactive timing is that the grant date precedes the decision date. Three compensation experts with whom I have been in contact say that they are aware of cases in which the grant date preceded the decision date. One expert indicated that the options involved were usually \"promised\" to an executive (perhaps through an employment agreement or in connection with an IPO), but not formally granted until later. Another indicated that while serving on the compensation committee of a large-company board, the committee was called upon to ratify a decision made \"internally\" (purportedly by the human resource staff) to award options with a past grant date in one or two instances. There are several reasons to believe that retroactive timing occurs in practice. First, it rank-and-le employees when the shares are overvalued. Such behavior, combined with the inability of rank-and-le employees to affect the timing of grants, might explain why stock prices do not fall before the option regrants in Lavelle's (2004) sample, which mostly excludes grants to top executives. Further, Huddart and Lang (2003) report that option exercises by both senior and junior employees precede negative abnormal returns, suggesting that they time option exercises to occur before poor stock performance, which is also consistent with value-maximizing behavior. Management Science 51(5), pp. 802-812, 2005 INFORMS would be a very effective and simple way of boosting the value of the awards. Second, stock option plans (which are standard legal documents) are vague as to how the grant date should be determined, and do not specically prohibit the grant date from preceding the decision date. Finally, it is difcult for outsiders to uncover such practices, because individual stock option agreements are signed and dated by the employee-recipient, but are not publicly disclosed. Why doesn't the compensation committee instead simply boost the value of the award by either awarding more options or awarding options with an exercise price lower than the market price at the award date? Paul Dorf, managing director at Compensation Resources, Inc., offers several reasons. First, the number of options awarded is often determined by past awards and/or industry norms. Second, the stock option plan limits the number of options that can be awarded. Third, stockholders dislike the potential dilutive effect generated by a large number of outstanding options. Fourth, accounting rules require a charge to earnings for grants that are issued in-themoney. Fifth, stockholders are averse to the notion of issuing options \"at a discount\" to executives. If the awards are scheduled, executives could instead try to control the release of information to the capital market in an effort to depress the price on the award date (see Aboody and Kasznik 2000). However, any stock price effect is likely to be weaker around scheduled awards for two reasons. First, all of the techniques described above could be used to inate the value of unscheduled awards, whereas the only way to inate the value of scheduled awards is to control the information ow. Second, scheduled awards are partially predictable by the capital market, thus creating trading opportunities that, when exploited, will tend to remove any price effect. Because the focus of this study is the timing of awards rather than the timing of information releases of awards (which, incidentally, is the focus of Aboody and Kasznik 2000), I focus on unscheduled awards. Nevertheless, I include scheduled awards in my analysis for comparison purposes. 3. Sample Since 1992, the Securities and Exchange Commission (SEC) has required rms to disclose certain information in proxy statements about stock option grants to top executives during the scal year. While rms are not required to disclose the award dates, they can be inferred from the stated maturity dates in combination with information about the beginning and end of the scal years and the assumption that the maturities of the options are in whole years. Note that because the proxy statements are generally led Lie: On the Timing of CEO Stock Option Awards 805 Management Science 51(5), pp. 802-812, 2005 INFORMS several months after the end of the scal year (the median is about three months afterward), it is not possible to exploit systematic stock price patterns around award dates, perhaps unless the awards are predictable.3 My sample of CEO stock option awards is taken from Standard & Poor's ExecuComp database. ExecuComp includes information about stock option grants from proxy statements for more than 2,000 large companies, which are or were members of the S&P 1500 (S&P 500, S&P 400 MidCap, and S&P 600 SmallCap). The initial sample contained 11,949 grants to CEOs during the scal years from 1992 through 2002. After having excluded observations that (a) lacked grant data, (b) were not in CRSP, (c) lacked price data in CRSP around the inferred grant date, or (d) were repricings or reloads, the sample contained 11,249 grants.4 Next, I obtained closing prices in CRSP from two days before through two days after the inferred grant date to identify the date whose closing price matched the share price from ExecuComp. For the purposes of my study, I dene this date to be the exact grant date. This leads me to a sample of 5,977 grants.5 Following Aboody and Kasznik (2000), I dene an award to be scheduled if it occurs within one week of the one-year anniversary of the prior year's award date and unscheduled if it does not occur within one week of this anniversary or if no options were awarded during the prior year. If no award information is available for the prior scal year, such 3 For a random sample of 100 unscheduled option awards, I searched public news announcements from one month before through one month after the award date for evidence that the awards were made public, but I found no such evidence. Note, however, that effective August 29, 2002, the SEC changed the reporting regulations with respect to stock option grants. Specically, rms must now report executive stock option grants within two business days. This is likely to affect the timing of stock option grants documented herein. 4 Repricings occur when the exercise price of outstanding options is lowered (generally to the current market price). As with general option awards, it is in the CEO's interest that the price is temporarily low at the time of the repricings. A separate analysis reveals that the stock return pattern around the sample of excluded repricings is similar to that documented in Callaghan et al. (2004). The returns during the months preceding the repricings are abnormally low, explaining why the options are repriced to restore their incentive effect. The returns during the days immediately after the repricings are abnormally positive, consistent with the notion that executives opportunistically time the repricing date or information releases around the repricing date. 5 It is unclear why I am unable to identify the exact grant date for many of the observations. Perhaps the inferred date is more than a couple of days away from the exact grant date, perhaps the company did not simply use the closing price as the exercise price for the executive options, perhaps the award represents a reload, or perhaps ExecuComp made mistakes when adjusting the postsplit prices reported in the proxy statements back to the actual market prices around the grants (indeed, I uncovered several such mistakes). as for those in 1992, I leave the award unclassied. This yields a nal sample of 1,426 scheduled awards, 1,668 unscheduled awards, and 2,883 unclassied awards, although I exclude the scheduled awards for most of the analysis.6 Table 1 presents the sample of option grants to CEOs during 1992-2002 by scal year, calendar month, and scal quarter. The number of awards is considerably lower for the rst couple of years, but this is at least partially due to more spotty coverage by ExecuComp during those years. Further, the number of unscheduled awards has gradually increased during the sample period, whereas the trend for scheduled awards is more stable. Option awards, especially scheduled awards, occur more frequently during the months of January, February, and December than during other months. Further, half of the scheduled option awards take place during the rst scal quarter, whereas 43% and 48% of the unscheduled and unclassied awards, respectively, take place in this quarter. Table 2 provides descriptive statistics for the scal year preceding the option awards. The sample rms are large, with an overall average book value of assets of $6.9 billion. Interestingly, rms that award options on a scheduled basis appear to be more mature than rms that award options on an unscheduled basis, as evidenced by their greater size and protability and lower market-to-book ratio. Firms with unclassied awards resemble rms with unscheduled awards, consistent with the notion that a majority of the unclassied awards are actually unscheduled. 4. Empirical Results 4.1. Abnormal Returns Around Option Awards Figure 1 displays the average cumulative abnormal returns around unscheduled, scheduled, and unclassied awards.7 I calculate abnormal returns around option awards as the difference between the stock returns of the awarding rm and the returns predicted by Fama and French's (1993) three-factor model, where the estimation period is the year ending 50 days before the award date. For the samples of unscheduled and unclassied awards, the stock prices (when adjusted for market effects) start to decline more than a month before the awardrst gradually, and then more dramatically during the days immediately before the awards. However, there is a sharp reversal of the price trend on the award dates. Immediately after the awards, the prices tend to increase. 6 For 65 of the 5,977 observations, I lack data to estimate the predicted returns based on the Fama and French (1993) three-factor model. 7 Yermack (1997) and Aboody and Kasznik (2000) provide similar types of graphs for samples of all awards and scheduled awards, respectively. Lie: On the Timing of CEO Stock Option Awards 806 Management Science 51(5), pp. 802-812, 2005 INFORMS Table 1 Sample Distribution Across Time Unscheduled awards n = 1 668 Fraction of sample (%) Fraction of universe (%) Scheduled awards n = 1 426 Fraction of sample (%) Fraction of universe (%) Unclassied awards n = 2 883 Fraction of sample (%) Panel A: Option awards by scal year 1992 0 1993 1 1 1994 6 6 1995 9 9 1996 10 10 1997 9 9 1998 10 10 1999 11 10 2000 14 13 2001 15 15 2002 16 17 0 2 8 11 11 9 9 11 12 12 14 Panel B: Option awards by calendar month January 14 February 13 March 9 April 10 May 8 June 4 July 7 August 6 September 5 October 6 November 7 December 10 16 18 6 6 7 4 5 6 5 5 4 17 14 15 8 8 9 6 7 5 5 7 6 11 Panel C: Option awards by scal quarter Quarter 1 43 Quarter 2 23 Quarter 3 15 Quarter 4 19 50 16 10 24 Fraction of universe (%) 48 22 13 18 0 3 7 10 9 8 8 9 10 11 13 3 11 11 8 8 10 10 12 10 10 7 22 27 21 15 15 17 16 18 16 17 13 Notes. Distribution of the sample of option grants awarded to CEOs during the scal years 1992-2002 by scal year, calendar month, and scal quarter. An award is classied as scheduled if it occurs within one week of the one-year anniversary of the prior year's award date, and unscheduled if it does not occur within one week of this anniversary or if no options were awarded during the prior year. If insufcient information is available to classify an award, it is left unclassied. Table 2 Descriptive Statistics Unscheduled awards Scheduled awards Unclassied awards Mean Assets ($MM) Market-to-book ratio Operating income/assets Total debt/assets Cash/assets Median Mean Median Mean Median 5,757 2.208 0.008 825 1 568 0 138 9,732 1.941 0.145 1,281 1.442 0.162 6,163 2 265 0 174 749 1 552 0 145 0.216 0.159 0 200 0 065 0.235 0.095 0.231 0.034 0 220 0 137 0 200 0 058 Notes. Descriptive rm statistics for the scal year prior to the option awards. An award is classied as scheduled if it occurs within one week of the one-year anniversary of the prior year's award date and unscheduled if it does not occur within one week of this anniversary or if no options were awarded during the prior year. If insufcient information is available to classify an award, it is left unclassied. The price increase is more pronounced during the rst few days, but continues for at least a month. Though this pattern is also evident for scheduled awards and awards with uncertain grant dates, it is considerably less pronounced. The similarity of the patterns for unscheduled and unclassied awards suggests that unclassied awards generally are unscheduled. 4.2. Return Patterns over Time An interesting question is whether the documented trends have changed in intensity over time. Executives might have become more effective in timing the awards to their advantage, especially as executive options have become increasingly more common. If so, it could explain why Yermack (1997) and Chauvin and Shenoy (2001) nd weaker stock return patterns using earlier samples than this study. Lie: On the Timing of CEO Stock Option Awards 807 Management Science 51(5), pp. 802-812, 2005 INFORMS Figure 1 Cumulative Abnormal Stock Returns Around Stock Option Grants 0.01 Unscheduled Scheduled Unclassified 0 - 30 - 20 -10 0 10 20 30 - 0.01 - 0.02 - 0.03 - 0.04 Day relative to option grant Notes. This gure displays the cumulative abnormal stock returns from 30 days before through 30 days after stock option grants to CEOs. Abnormal stock returns are estimated using the three-factor model described in Fama and French (1993), where the estimation period is the year ending 50 days before the award date. An award is classied as scheduled if it occurred within one week of the one-year anniversary of the prior year's award date, and unscheduled if it did not occur within one week of this anniversary or if no options were awarded during the prior year. If insufcient information is available to classify an award, it is left unclassied. Alternatively, the exposure of opportunistic behavior in Yermack might have made executives more reluctant to engage in such behavior for fear of criticism by outsiders. To answer this question, I examine the return patterns for three groups based on the year of the awards. The rst group consists of unscheduled awards during 1992 through 1994 (to correspond with Yermack's sample period), while the last two groups split the remaining eight years into two fouryear periods. If the awards are scheduled, the executives cannot time them to their advantage, and they are therefore excluded from this analysis. However, I include unclassied awards, because the prior evidence suggests that they are primarily unscheduled. Thus, the results for unclassied awards might validate the results for unscheduled awards. Figure 2a shows the cumulative abnormal returns from Day 30 through Day +30 for the three groups of unscheduled awards, while Figure 2b shows the same returns for the three groups of unclassied awards. The trends for unscheduled awards have become more distinct over time. The pattern for the rst two years (which admittedly only consists of 113 unscheduled awards) is rather vague, whereas the pattern for the last four years is very strong. The pattern for the middle period falls roughly in-between the patterns for the other two periods. The results for unclassied awards are very similar, thus corroborating the results for unscheduled awards. That is, the pattern for the rst period is weakest and the pattern for the last period is strongest. Overall, the return trends around awards have become more pronounced during the sample period. This is consistent with the notion that executives have become more effective over time in timing the awards to their advantage. Further, it might explain, at least partially, why Yermack (1997) nds no evidence of stock price declines before awards using a sample from 1992 through 1994, and why Chauvin and Shenoy (2001) nd scant evidence of good stock price performance after awards using a sample from 1981 through 1992. 4.3. Predicted Returns Around Option Awards The sharp decline in prices immediately before unscheduled and unclassied awards followed by a sharp reversal immediately afterward suggests that executives collectively have a remarkable ability to time the awards to their advantage. One might even say that the executives' collective ability is uncanny, especially considering that the compensation committee formally makes the decisions regarding the option awards. This prompts the question as to whether some of the awards are timed ex post facto. That is, when the decision regarding the ofcial award date is made, the ofcial award date (and, hence, the exercise price of the options) might be determined to be an earlier date that had a particularly low price.8 Because the terms associated with the awards are revealed much later, outsiders would not learn of this, thus preventing them from crying foul. In any event, the stock option plans that I have looked at do not explicitly prohibit such activities. The plans generally state that the exercise price should be the market price at the grant date, but do not state that the grant date cannot precede the decision date. The hypothesis that the awards are timed ex post facto is novel. Unfortunately, it is not possible to ascertain from an examination of the abnormal returns whether the awards are timed proactively or retroactively. However, an examination of the predicted returns from the three-factor model might provide valuable insight. Suppose that executives have superior forecasting ability for future rm-specic price changes, but not for future marketwide movements. The intuition for this is that while executives clearly possess unique information about their rms' future cash ows and imminent public announcements that is not generally available to other market participants, it is less likely that they possess unique information that pertains to the overall market.9 If so, 8 Similarly, it has recently been revealed that some mutual funds have allowed hedge funds to trade at closing prices long after the market has closed. This has allowed hedge funds to take advantage of information that has surfaced after the market closing, because this information has not yet been incorporated into the prices at which they have been allowed to trade. 9 Note, however, that Lakonishok and Lee (2001) report evidence that long-term market returns are higher after insiders buy stock. Lie: On the Timing of CEO Stock Option Awards 808 Figure 2 Management Science 51(5), pp. 802-812, 2005 INFORMS Cumulative Abnormal Stock Returns Around Stock Option Grants by Year (a) 0.03 1993-1994 1995-1998 1999-2002 0.02 0.01 0 -0.01 - 30 - 20 0 -10 10 20 30 -0.02 -0.03 -0.04 -0.05 Day relative to option grant Unscheduled awards (b) 0.03 0.02 1992 -1994 1995-1998 1999- 2002 0.01 0 -0.01 - 30 - 20 -10 0 10 20 30 -0.02 -0.03 -0.04 -0.05 -0.06 -0.07 Day relative to option grant Unclassified awards Notes. This gure displays the cumulative abnormal stock returns from 30 days before through 30 days after stock option grants to CEOs. Abnormal stock returns are estimated using the three-factor model described in Fama and French (1993), where the estimation period is the year ending 50 days before the award date. An award is classied as scheduled if it occurred within one week of the one-year anniversary of the prior year's award date and unscheduled if it did not occur within one week of this anniversary or if no options were awarded during the prior year. If insufcient information is available to classify an award, it is left unclassied. executives might be able to time future award dates to coincide with low prices that are attributable to the arrival of rm-specic information to the market, but not to overall market movements. This would manifest itself in negative abnormal returns before the awards and/or positive abnormal returns afterward. In contrast, the predicted returns from the marketmodel should be normal both before and after the awards. Any evidence of predicted returns that are abnormally low before awards and/or abnormally high after awards is consistent with the notion that some awards are timed ex post facto.10 Thus, we cannot preclude the possibility that insiders are able to predict shorter-term marketwide movements also. 10 One might argue that executives could time awards to occur shortly after they have observed marketwide declines, in which case the predicted returns would be negative before the awards. However, they would not benet from this. Executives only gain from marketwide declines before awards if (a) they had the ability to predict the decline before it occurred and, therefore, postponed Examining predicted returns from the three-factor model around awards presents unique challenges, because they (i) tend to be positive, such that we need a benchmark other than zero, and (ii) contain both yearly and seasonal variations. To mitigate these challenges, I run a logistic regression of the occurrence of awards against prior and subsequent abnormal stock and predicted returns. This requires generation of a control sample with no awards. For each observation in the original sample (i.e., rm and award date), I generate ve control observations with no awards by using the same rm combined with a random date drawn from the period from six months to one month before the award date or the period from one month to six months after the award date. Thus, the returns for the control observations effectively serve as benchmarks. the award, or (b) they had a superior ability to deem the decline to be unwarranted, in which case we should also observe subsequent marketwide price increases. Lie: On the Timing of CEO Stock Option Awards 809 Management Science 51(5), pp. 802-812, 2005 INFORMS By including dummy variables for the month of the observation, the regression analysis controls for seasonality in returns documented by, e.g., Keim (1983) and Reinganum (1983). Because previous results show that awards vary across the calendar months, the absence of such control variables could give rise to spurious relations between returns and the occurrence of awards. A further advantage of this analysis is that it effectively controls for rm-specic risk factors that affect stock returns, irrespective of whether these factors can be identied. As long as the risk factors are reasonably constant over time, they should be present for the original observation as well as for the associated control observations, thus largely washing out in the overall analysis.11 Table 3 presents the results from the regression analysis. Consistent with earlier evidence, unscheduled awards are more likely to occur after negative abnormal stock returns and before positive abnormal stock returns. The abnormal returns immediately surrounding the awards have the greatest effect on the occurrence of awards, but even abnormal returns at least a couple of weeks before or after the awards have a statistically signicant effect. The most interesting result in Table 3 is that unscheduled awards are more likely to occur after dismal predicted returns and before high predicted returns. The effects of the predicted returns during the two days before and the two days after are particularly strong, with p-values less than 0.01. Unless executives could have anticipated the marketwide returns and, hence, predicted returns from the threefactor model, the results suggest that executives time at least some of the awards ex post facto. To further validate these results, I run the same regression for unclassied awards, a majority of which are likely to be unscheduled. The results are similar to those for unscheduled awards. In particular, the unclassied awards are also more likely to occur after low abnormal and predicted returns and before high abnormal and predicted returns, although the effect from the predicted returns immediately before is weaker. This lends further credence to the results 11 Another potential problem is that the option grant dates might be correlated across rms, which would cause conventional standard errors to be underestimated. In my sample of unscheduled awards, I identied 576 cases where the grant date was the same as the grant date for another observation. As a benchmark, I generated 100 distributions of random grant dates with the same number of observations in each calendar year as the original sample and based on 252 trading days in a year. The average number of cases where a date was similar to another date was 502, and the maximum was 539. Thus, although 576 is statistically different from 502, it does not seem to be so high as to cause a major problem with the conventional standard error. for unscheduled awards and the notion that awards are timed retroactively.12 Finally, I run the regression for scheduled awards for comparison purposes. As expected, the results are generally much weaker than for the other award categories. However, the predicted returns during the two days afterward positively affect the probability even for scheduled awards. I conjecture that this result arises because even with grants classied as scheduled, there might some leeway with the precise date. For example, if executives have a two-week window (i.e., from one week before through one week after the anniversary) in which to make awards, one might argue that they have ample exibility to opportunistically time those awards. An alternative, but related, conjecture is that many unscheduled awards just happened to occur within a week of the one-year anniversary of prior grants, in which case they are incorrectly classied here as scheduled grants. To investigate this further, I tighten the denition of scheduled to include only those that occurred within a day of the one-year anniversary of the prior grant date. If either of my conjectures is correct, any effect stemming from predicted returns should be even weaker for this sample. Table 4 shows the results for the scheduled awards with tight schedules, as well as for other scheduled awards. As expected, the effect from predicted returns is weaker for scheduled awards with tight schedules than for the others. In fact, none of the coefcients on predicted returns differ statistically from zero in the sample of scheduled awards with tight schedules. These results further corroborate my earlier results and interpretations. Overall, the logistic regressions show that awards are timed to occur after price decreases and before price increases. Unlike prior studies, I show that overall market factors cause a portion of the price patterns. Thus, unless executives have an informational advantage in forecasting future market movements, the results suggest that the benecial timing of the awards occurs, at least partially, because executives determine the ofcial grant date to be a date in the 12 I also ran the regression for each of the years from 1994 through 2002 for unscheduled awards (I excluded 1993 due to the small number of observations), and from 1992 through 2002 for unclassied awards. The coefcient on the predicted return during the two days immediately after the option grant is of most interest given its statistical signicance in Table 3 and its implication about grant behavior. This coefcient is statistically signicant at the 1% level for only one year (2001, which has the most observations) for unscheduled awards and for no year for unclassied awards. More importantly, it is positive for all but one year (1994, which has the fewest observations) for unscheduled awards and for all years for unclassied awards. I interpret these results as evidence that the results do not appear to be driven by just a few years, and that a large sample is needed to uncover the underlying relationships. Lie: On the Timing of CEO Stock Option Awards 810 Management Science 51(5), pp. 802-812, 2005 INFORMS Table 3 Logistic Regressions of Awards Unscheduled Coefcient Intercept Abnormal return during days 30 to 10 Abnormal return during days 10 to 5 Abnormal return during days 5 to 2 Abnormal return during days 2 to 0 Abnormal return during days 0 to +2 Abnormal return during days +2 to +5 Abnormal return during days +5 to +10 Abnormal return during days +10 to +30 Predicted return during days 30 to 10 Predicted return during days 10 to 5 Predicted return during days 5 to 2 Predicted return during days 2 to 0 Predicted return during days 0 to +2 Predicted return during days +2 to +5 Predicted return during days +5 to +10 Predicted return during days +10 to +30 January dummy February dummy March dummy April dummy May dummy June dummy July dummy August dummy September dummy October dummy November dummy Number of observations 1 424 0 689 1 144 2 061 2 417 5 322 1 871 1 513 0 827 1 087 1 011 0 865 3 625 3 545 1 361 1 473 0 826 0 414 0 489 0 175 0 078 0 250 0 934 0 520 0 737 0 891 0 751 0 372 10,003 Scheduled p-value 0.000 0.001 0.002 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.146 0.371 0.002 0.003 0.152 0.040 0.008 0.001 0.000 0.165 0.532 0.052 0.000 0.000 0.000 0.000 0.000 0.006 Coefcient 0 792 0 331 0 496 0 869 0 904 3 339 2 330 1 676 0 228 0 667 0 255 1 928 2 944 4 538 2 855 0 782 0 463 0 032 0 185 1 252 1 204 1 013 1 546 1 375 1 190 1 501 1 394 1 543 8,552 p-value 0.000 0.221 0.344 0.210 0.274 0.000 0.001 0.001 0.317 0.139 0.799 0.162 0.079 0.005 0.034 0.412 0.267 0.775 0.095 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 Unclassied Coefcient 1 457 0 697 0 745 1 677 3 271 5 606 2 282 1 647 0 765 0 973 1 179 1 779 1 087 4 423 0 498 0 928 0 795 0 426 0 662 0 199 0 254 0 134 0 621 0 425 0 819 0 871 0 604 0 580 16,897 p-value 0.000 0.000 0.016 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.050 0.028 0.268 0.000 0.534 0.129 0.003 0.000 0.000 0.046 0.010 0.164 0.000 0.000 0.000 0.000 0.000 0.000 Notes. Logistic regressions of the choice to award options. For each observation in the original sample (i.e., rm and award date), ve control observations are generated by using the same rm combined with a random date drawn from the period from six months to one month before the award date or the period from one month to six months after the award date. The dependent variable equals one for the original observations and zero for the control observations. Independent variables include abnormal stock returns and predicted returns for various periods before and after the grant dates, as well as dummy variables for the calendar month of the observation. Abnormal stock returns are estimated using the three-factor model described in Fama and French (1993), where the estimation period is the year ending 50 days before the award date. Predicted stock returns are the actual returns less the abnormal returns. An award is classied as scheduled if it occurs within one week of the one-year anniversary of the prior year's award date, and unscheduled if it does not occur within one week of this anniversary or if no options were awarded during the prior year. If insufcient information is available to classify an award, it is left unclassied. past. In fact, my study does not preclude the possibility that the entire stock price pattern is due to retroactive timing, rather than proactive timing as suggested in past studies. 5. Summary and Conclusion Using a large sample of stock option awards to CEOs from 1992 through 2002, I nd that the abnormal stock returns are negative before the award dates and positive afterward. While these trends are evident around both scheduled and unscheduled awards, they are much more pronounced around unscheduled awards. The return patterns around unscheduled awards appear to have intensied over time, suggesting that executives have gradually learned how to better time awards to their advantage or become more aggressive in their timing efforts. This could explain the absence of both negative abnormal returns leading up to the awards in Yermack's (1997) sample from 1992 through 1994 and positive returns following awards in Chauvin and Shenoy's (2001) sample from 1981 through 1992. Prior studies have attributed the stock returns around unscheduled awards to executives timing awards relative to expected future price patterns. If so, the distinct stock returns documented here suggest that executives' ability to forecast future price patterns is uncanny, especially for later years. This prompts me to propose a novel alternative hypothesis that the awards are timed ex post facto. That is, the grant date might be set to be an earlier date with a particularly low price. I nd evidence consistent with this ex post facto timing hypothesis. In particular, I report that predicted returns from the three-factor Lie: On the Timing of CEO Stock Option Awards 811 Management Science 51(5), pp. 802-812, 2005 INFORMS Table 4 Logistic Regressions of Scheduled Awards with Fixed vs. Relaxed Schedule Tight schedule Relaxed schedule Coefcient Intercept Abnormal return during days 30 to 10 Abnormal return during days 10 to 5 Abnormal return during days 5 to 2 Abnormal return during days 2 to 0 Abnormal return during days 0 to +2 Abnormal return during days +2 to +5 Abnormal return during days +5 to +10 Abnormal return during days +10 to +30 Predicted return during days 30 to 10 Predicted return during days 10 to 5 Predicted return during days 5 to 2 Predicted return during days 2 to 0 Predicted return during days 0 to +2 Predicted return during days +2 to +5 Predicted return during days +5 to +10 Predicted return during days +10 to +30 January dummy February dummy March dummy April dummy May dummy June dummy July dummy August dummy September dummy October dummy November dummy Number of observations p-value Coefcient p-value 0 758 0 545 0 344 1 591 0 158 0 552 1 521 1 177 0 160 0 873 0 152 1 124 0 106 4 468 0 833 1 757 1 039 0 171 0 035 1 372 1 154 1 145 1 468 1 528 1 154 1 548 1 493 1 625 0.000 0.199 0.666 0.132 0.900 0.646 0.132 0.141 0.649 0.216 0.920 0.586 0.967 0.072 0.688 0.240 0.110 0.287 0.827 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0 854 0 196 0 615 0 051 1 751 6 886 2 834 1 891 0 846 0 509 0 549 2 905 5 193 5 727 4 994 0 240 0 043 0 085 0 338 1 122 1 248 0 884 1 625 1 259 1 223 1 439 1 306 1 450 0.000 0.585 0.379 0.957 0.119 0.000 0.002 0.006 0.019 0.397 0.691 0.129 0.025 0.010 0.006 0.850 0.941 0.601 0.029 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 4,095 4,457 Notes. Logistic regressions of the choice to award options. For each observation in the original sample (i.e., rm and award date), ve control observations are generated by using the same rm combined with a random date drawn from the period from six months to one month before the award date or the period from one month to six months after the award date. The dependent variable equals one for the original observations and zero for the control observations. Independent variables include abnormal stock returns and predicted returns for various periods before and after the grant dates, as well as dummy variables for the calendar month of the observation. Abnormal stock returns are estimated using the three-factor model described in Fama and French (1993), where the estimation period is the year ending 50 days before the award date. Predicted stock returns are the actual returns less the abnormal returns. An award is classied as scheduled if it occurs within one week of the one-year anniversary of the prior year's award date. A scheduled award is further classied as tight if it occurred within one day of the one-year anniversary of the prior year's award date and relaxed otherwise. p-values are given in parentheses. model are abnormally low leading up to the awards and abnormally high afterward. Unless executives have an informational advantage that allows them to develop superior forecasts regarding the future market movements that drive these predicted returns, the results suggest that the ofcial grant date must have been set retroactively. The results are provocative and might cause some investors to cry foul. However, even though retroactive timing of executive stock option awards seems fraudulent, it is not clear that it is in violation of the stipulations in the stock option plans. Further, although I show aggregate evidence that retroactive timing occurs, it is difcult, if not impossible, to prove that such timing takes place in individual cases. Acknowledgment The author thanks, an anonymous referee, the associate editor, the department editor (David Hsieh), David Aboody, Diane Denis, Paul Dorf, Randy Heron, Steve Huddart, Dave Ikenberry, Bill Lewellen, Heidi Lie, Bill Maxwell, John McConnell, Kevin Murphy, Tom Rietz, Anand Vijh, Wanda Wallace, and seminar participants at the University of Iowa, Penn State University, Vanderbilt University, and the College of William & Mary for helpful comments. References Aboody, D., R. Kasznik. 2000. CEO stock option awards and the timing of corporate voluntary disclosures. J. Accounting Econom. 29 73-100. Callaghan, S. R., P. J. Saly, C. Subramaniam. 2004. The timing of option repricing. J. Finance 59 1651-1676. 812 Chauvin, K. W., C. Shenoy. 2001. Stock price decreases prior to executive stock option grants. J. Corporate Finance 7 53-76. Fama, E. F., K. R. French. 1993. Common risk factors in the returns on stocks and bonds. J. Financial Econom. 33 3-56. Huddart, S., M. Lang. 2003. Information distribution within rms: Evidence from stock option exercises. J. Accounting Econom. 34 3-31. Keim, D. B. 1983. Size related anomalies and stock return seasonality: Further empirical evidence. J. Financial Econom. 12 13-32. Lie: On the Timing of CEO Stock Option Awards Management Science 51(5), pp. 802-812, 2005 INFORMS Lakonishok, J., I. Lee. 2001. Are insider trades informative? Rev. Financial Stud. 14 79-111. Lavelle, Louis. 2004. The stock-option myth. Business Week (May 6). Reinganum, M. R. 1983. The anomalous stock market behavior of small rms in January: Empirical tests for tax-loss selling effects. J. Financial Econom. 12 89-104. Yermack, D. 1997. Good timing: CEO stock option awards and company news announcements. J. Finance 52 449-476. Zhang, G. 2002. Market valuation and employee stock options. Working paper, Duke University, Durham, NC. Accounting Horizons Vol. 28, No. 2 2014 pp. 297-312 American Accounting Association DOI: 10.2308/acch-50706 CEO Turnover and Audit Pricing Hua-Wei Huang, Robert J. Parker, Yun-Chia Anderson Yan, and Yi-Hung Lin SYNOPSIS: This study examines the relationship between CEO turnover in client companies and the fees charged by their audit firms. We propose that forced CEO turnover (such as dismissals) pose higher business and audit risks for the audit firm than voluntary turnover (such as retirements); further, greater risk leads to higher audit prices. We develop a regression model of audit fees that includes, as predictor variables, type of CEO turnover and control variables identified in prior studies (e.g., ROA, total assets, and corporate governance). Results reveal that companies with forced CEO turnover have significantly higher audit fees than companies with either voluntary turnover or no turnover. Further, we find no difference in audit fees between firms with voluntary turnover and firms without turnover. Keywords: audit fees; audit risk; CEO turnover. Data Availability: The data used in this study are publicly available. INTRODUCTION A long stream of accounting research has investigated audit pricing, i.e., the fees charged by accounting rms for auditing the nancial statements of a client company (e.g., see meta-analysis by Hay, Knechel, and Wong [2006]). Several studies in this literature theorize that the business and audit risks of an audit engagement inuences the audit price and that higher risk is associated with higher fees (e.g., Houston, Peters, and Pratt 2005; Lyon and Maher 2005; Hay et al. 2006; Venkataraman, Weber, and Willenborg 2008). As argued in several studies, auditing rms have three interrelated risks in accepting an audit engagement: (1) business risk of the client that involves the client's protability and survival; (2) audit risk, the risk that the clients' nancial statements are misstated and that the auditor fails to detect the misstatement; and (3) business risk of the auditor that involves potential litigation costs and adverse reputational effects resulting from the engagement (Huss and Jacobs 1991; Johnstone 2000; Stanley 2011). Hua-Wei Huang is an Associate Professor at National Cheng Kung University; Robert J. Parker is a Professor at the University of New Orleans; Yun-Chia Anderson Yan is an Assistant Professor at The University of Texas at Brownsville; and Yi-Hung Lin is a Ph.D. Student at National Cheng Kung University. Hua-Wei Huang gratefully acknowledges the nancial support of the National Science Council, Taiwan, ROC (Project No. NSC 100-2410-H-006-102). Submitted: June 2012 Accepted: January 2014 Published Online: January 2014 Corresponding author: Hua-Wei Huang Email: hwawei7@yahoo.com.tw 297 Huang, Parker, Yan, and Lin 298 We extend the literature that investigates the relation between risk and audit fees by examining a relatively neglected risk factor, CEO turnover in the client company. We distinguish between forced CEO turnover, when the CEO is forced out by the board of directors (such as dismissals), from voluntary turnover (such as retirements). According to the proposed theoretical framework, forced turnover is associated with higher risk for the auditing rm and, consequently, higher audit prices. Based upon prior studies (e.g., Clayton, Hartzell, and Rosenberg 2005), we argue that forced turnover usually results from unacceptable company performance and represents a signal that company strategy and leadership need sweeping changes. The uncertainties surrounding these changes result in higher business risk for the client company and therefore higher risk for the auditor. Further, forced turnover may increase audit risk due to both the pressure on client management for better performance and the turnover of senior management that usually accompanies the CEO departure. In contrast, we argue that voluntary CEO turnovers, such as retirements, reect an orderly management transition that poses relatively few additional risks to the auditor. To test the proposed relations, we collect information from 2004 to 2011 using the Audit Analytics database. We develop a regression model with annual audit fees as the dependent variable (n 13,692). Independent variables include type of CEO turnover (forced or voluntary) and control variables identied in prior studies of audit fees (e.g., ROA, total assets, corporate governance). We also develop a second regression model, in which change in annual audit fees is the dependent variable (n 10,188). Independent variables include type of CEO turnover and changes in the control variables. Results for both models indicate that rms with forced CEO turnover have signicantly higher audit fees than rms with either voluntary turnover or no turnover. Further, we nd no difference in audit fees between companies with voluntary turnover and companies without turnover. The paper proceeds as follows. The second section contains a review of relevant research and development of the hypothesis. The third presents a description of the research model, the sample, the empirical results, and the ndings of additional analyses. The fourth and nal section provides a summary of the major ndings and a discussion of their implications for practice and future research. THEORETICAL DEVELOPMENT According to our theoretical arguments, certain types of CEO turnover increase the risk of an audit engagement and therefore audit fees. As in several prior studies (e.g., Farrell and Whidbee 2003; Clayton et al. 2005), we distinguish between forced and voluntary CEO turnover. Dismissal, when the board of directors ''res'' the CEO, represents one kind of forced turnover. Many CEO resignations, e.g., due to suspected or determined wrongdoing, also are forced, although the ''forced'' nature of the departure may not be as apparent as a dismissal. In contrast, the CEO may depart voluntarily, e.g., an age-related retirement, which represents an orderly transition in the CEO position. In the case of forced CEO turnover, most studies in the business literature assume a causal relationship between poor company performance and the CEO departure (e.g., see DeFond and Park 1999; Huson, Parrino, and Starks 2001; Wowak, Hambrick, and Henderson 2011; Crossland and Chen 2013). As noted by Helwege, Intinoli, and Zhang (2012, 26), ''Previous research shows conclusively that CEOs are more often forced out when their rms perform poorly.'' Within this literature stream, several researchers emphasize the role of the board of directors in CEO departures; accordingly, the board assesses the CEO by evaluating company performance relative to the board's expectations (Puffer and Weintrop 1991; Farrell and Whidbee 2003). When company performance falls short of the board's expectations, the CEO is dismissed or pressured into resigning. We propose that in forced CEO departures, the board not only is dissatised with recent company Accounting Horizons June 2014 CEO Turnover and Audit Pricing 299 performance but also is concerned that unacceptable performance will continue into the future unless signicant change occurs. As argued in Clayton et al. (2005, 1783, ''a forced turnover usually signals that existing rm policies are inadequate and substantial changes are required.'' A forced turnover creates uncertainty regarding both the company's future strategic direction and the ability of the new CEO to effectively manage the required changes (Clayton et al. 2005). Regarding the risk of a new CEO, turnovers that follow weak company performance often backre because many board members are ill-equipped to select a new CEO that can address the fundamental problems of the business (Wiersema 2002). The uncertainties surrounding a forced turnover are reected in a long-term increase in the volatility of the company's stock price as the market attempts to assess both changes in rm strategy and the competency of the new CEO (Clayton et al. 2005). From the perspective of the auditor, the uncertainties surrounding forced CEO turnovers may raise concerns about the future protability of the rm. In other words, the uncertainties increase the business risk of the client company and ultimately the business risk of the auditor. Forced departures also may have consequences for audit risk. As noted in SAS No. 99 (AICPA 2002, 45), pressures for fraud include ''excessive pressure on management or operating personnel to meet nancial targets set up by the board of directors.'' As we noted previously, company performance that is unacceptable to the board often results in forced CEO turnover. Before termination, CEOs with unacceptable performance may be well aware of their precarious situation and the need to report higher performance to maintain their positions. Further, the incoming CEO may feel unduly pressured to meet board expectations given the experience of the previous CEO. Another audit risk involves opportunities for fraud. As noted in SAS No. 99 (AICPA 2002, 46-47), ''unstable organizational structure'' as evidenced by ''high turnover of senior management'' may provide opportunities for fraud. Several accounting studies argue that turnover of senior management is a risk factor for auditors (e.g., Mock and Wright 1993; Helliar, Lyon, Monroe, Ng, and Woodliff 1996; Mock and Wright 1999). Forced CEO turnover may be especially risky; as Fee and Hadlock (2004) report, it results in the departure of not only the CEO but also many senior managers within the company. While most studies in the business literature assume that forced CEO turnover results from poor company performance, some researchers note: (1) determining the cause of CEO turnover is difcult as companies are reluctant to disclose information about it (e.g., Shen and Cannella 2002a, 2002b; Farrell and Whidbee 2003); and (2) CEO turnover may arise from causes other than poor performance, such as CEO misbehavior (Ertugrul and Krishnan 2011). For example, Ertugrul and Krishnan (2011) use newspaper articles (such as the Wall Street Journal) (Ertugrul and Krishnan 2011, 142, Table 4) to try to identify specic reasons for CEO dismissals; they were able to nd explicit reasons for only 26 percent of the dismissals in their sample. They also note that in their sample, when CEO dismissal did not seem to be the result of poor rm performance, the dismissal often followed the discovery of unethical or illegal CEO actions such as accounting manipulations. Unethical or illegal CEO actions pose obvious risks to auditing rms, especially if the actions involve nancial reporting issues. While forced CEO turnover may be associated with increased risk for the auditor, voluntary departures such as retirements are likely not. Retirements reect an orderly CEO change. Presumably, the need for change in strategic direction that characterizes many forced CEO departures is not present with retirement; consequently, with retirement, the business risk of the client rm is lower. With retirement, the audit risk may be lower as there is less pressure on the CEO to meet the performance expectations of the board and, further, turnover of senior managers may be minimal. Formally stated, our hypothesis is (in the alternative form): H1: There is a positive association between audit fees and forced CEO turnover. Accounting Horizons June 2014 Huang, Parker, Yan, and Lin 300 METHOD Models We examine audit fees as a function of the type of CEO turnover and a number of control variables identied in prior studies. We develop two related regression models to examine audit fees.1 The rst model appears below (subscript i denotes rm i in year t): LNAFt;i a0 a1 TURNOVER FORCEDt;i a2 TURNOVER VOLUNTARYt;i a3 LNASSETSt;i a4 REINTAt;i a5 SQSEGSt;i a6 FORGNt;i a7 LIQt;i a8 DAt;i a9 ROAt;i a10 ROA CHANGEt;i a11 LOSSt;i a12 STOCK RETt;i a13 GCt;i a14 MWt;i a15 RESTATEt;i a16 BIG4t;i a17 AUD CHANGEt;i a18 AUD TENUREt;i a19 NAFRt;i a20 BD INDt;i a21 BD SIZEt;i a22 BD MEETt;i a23 DUALITYt;i a24 AC INDt;i a25 AC SIZEt;i a26 AC EXPt;i a27 YEAR a28 IND a29 STOCK EXC et;i : The variables are dened as follows: LNAF natural log of audit fees; TURNOVER_FORCED 1 if the CEO turnover is forced, else 0; TURNOVER_VOLUNTARY 1 if the CEO turnover is voluntary, else 0; LNASSETS NATURAL log of total assets; REINTA proportion of total assets in receivables and inventory; SQSEGS square root of the number of segments; FORGN 1 if the rm has foreign operations, else 0; LIQ current assets=current liabilities; DA total liabilities=total assets; ROA net income=total assets; ROA_CHANGE change in ROA from year t1 to year t; LOSS 1 if rm had loss in current year, else 0; STOCK_RET one year stock return; GC 1 if auditor issues going concern opinion, else 0; MW 1 if reported material weakness in internal controls, else 0; RESTATE 1 if rm restated its current nancial statement, else 0 BIG4 1 if Big 4 auditor, else 0; AUD_CHANGE 1 if auditor change during current year, else 0; AUD_TENURE natural logarithm of the length of the auditor-client relationship; NAFR ratio of non-audit fees to audit fees; BD_IND number of independent directors divided by total number of directors; BD_SIZE number of board members; BD_MEET number of board meetings in current year; DUALITY 1 if the CEO and chair are the same person, else 0; AC_IND number of independent members divided by total number of members on audit committee; AC_SIZE number of members on audit committee; 1 The regression parameter estimates and standard errors in both models are robust because potential clustering along two dimensions, rm and time, has been controlled by SAS cluster identiers (Gow, Ormazabal, and Taylor 2010; Petersen 2009). Accounting Horizons June 2014 CEO Turnover and Audit Pricing 301 AC_EXP number of members with nancial expertise divided by total number of members on audit committee; YEAR year dummy variables; IND industry dummy variables; and STOCK_EXC stock exchange dummy variables. The dependent variable is the natural log of the company's audit fee (LNAF). Information about audit fees is obtained from Audit Analytics database. The key independent variables involve CEO turnover. Two types of turnover are investigated: forced and voluntary. Each type is represented by a separate variable, which is coded 1 if that type of turnover is applicable, and 0 otherwise. We propose that forced CEO turnovers (TURNOVER_FORCED) are positively associated with audit fees while voluntary CEO turnovers (TURNOVER_VOLUNTARY) have no association. For determination of type of CEO turnover, we use the Audit Analytics database.2 We classify turnover as forced if Audit Analytics reports any of the following reasons for turnover: dismissal, investiga