Management Intent and CEO and CFO Turnover around Earnings Restatements: Evidence from the Post-Enron Era

Karen Hennes Smeal College of Business

Penn State University khennes@psu.edu

Andrew Leone Smeal College of Business

Penn State University ajl14@psu.edu

Brian Miller Smeal College of Business

Penn State University bpm173@psu.edu

First Version: September 2006 Current Version: January 2007

Abstract: This study examines the extent to which management’s intention to mislead investors affects the probability that CEOs and CFOs are terminated around restatement announcements. Using a sample of 188 restatements from 8-Ks filed in 2002-2005, we find that turnover rates are extremely high for intentional violations compared to unintentional ones: observed turnover rates are 5 to 6 times higher (log-odds are 11 to 25 times higher after controlling for other explanatory variables) for our intentional GAAP violations than for unintentional violations. This evidence suggests that boards take swift action to dismiss managers when restatements are the result of intentional misbehavior and that the relatively low turnover rates documented in earlier restatement research is likely due to the inclusion of unintentional violations. We also compare the turnover rates in this post-Enron era sample to those of sample of restatements from 1997-1998, and find that without conditioning on intent, turnover rates appear to have declined. However, after conditioning on intent, we find that turnover rates are similar across the two time-periods. We conclude that regulation introduced to encourage boards to discipline managers for intentional misreporting had little impact on turnover rates. This is likely due to the fact that turnover rates were already very high in the pre-Enron era (you cannot fix what is not broken). Our findings highlight the importance of distinguishing restatements by a meaningful measure of severity when conducting research in a restatement setting and provide some of the first evidence on CEO and CFO penalties for misreporting in the post-Enron period.

The authors thank the Smeal College of Business for financial support. We also thank Marty Butler, Rich Frankel, Scott Richardson, Nicole Thorne Jenkins, Tzachi Zach and workshop participants at Barclays Global Investors, Boston University, Dartmouth College, University of Kentucky, University of Minnesota, Penn State University and Washington University-St. Louis for helpful comments.

2

Management Intent and CEO and CFO Turnover around Earnings Restatements: Evidence from the Post-Enron Era

1.0 Introduction

This study examines the extent to which management’s intent to mislead investors affects the

probability that CEOs and CFOs are terminated around restatement announcements. Our research is

motivated by the fact that, although recent studies report that executive turnover rates around restatements

are statistically significantly different from turnover rates in a control sample of non-restaters, some

question whether the observed turnover rates are too low. 1 Abelson, for example, is one member of the

press who comments that when managers are “manipulating a company’s financial numbers to mislead

investors, the punishment is often anything but sharp and swift” (1996). Further, in a recent study on

restatements, Collins, Reitenga, and Sanchez-Cuevas (2005) conclude that “half the sample appear to

have taken little or no action to penalize management.”

One explanation for the seemingly low turnover rates is that recent studies, such as Collins et al.

(2005), draw from the U.S. General Accounting Office (GAO) database (2003), which contains

substantial variation in the types of the restatements included. Although this database has generally been

characterized as one containing cases of “aggressive accounting,” there are a large number of

restatements resulting from seemingly honest bookkeeping errors or from misinterpretations of somewhat

ambiguous GAAP. This raises the question of whether the relatively low turnover rates for executives

around restatements found in prior research are due to widespread governance problems (e.g.,

management entrenchment) or to the mixing together of both intentional GAAP violations, which merit

management turnover, and unintentional GAAP violations, which may not.

We predict that restatements due to intentional GAAP violations are much more likely to lead to

management turnover for the following reasons. First, Palmrose, Richardson, and Scholz (2004) report

that the market reaction to restatements caused by fraud is over three times more negative than in non-

1 For examples of turnover rates in studies where there is statistically significant difference in turnover rates for restatement firms see: Desai, Hogan, and Wilkins (2006), Land (2006), Arthaud-Day, Certo, Dalton, and Dalton (2006), and Jayaraman, Mulford, and Wedge (2004).

3

fraud cases (-20% versus -6%). Given that investors view the restatements involving fraud as more

severe, management’s intention to deliberately misreport is likely to play a significant role in boards’

decision to terminate managers when these misstatements are identified.2 Second, termination of a top

executive may be partly to punish the manager for the loss in shareholder value caused by the restatement,

but the termination is also a highly visible means of restoring financial reporting credibility.

Consequently, we predict that the turnover rates for CEOs and CFOs are much higher for restatements

related to intentional GAAP violations than for restatements related to unintentional errors.

In contrast to prior research, our study does not focus on whether turnover is in fact higher for

restating firms than for non-restating firms. Instead, we attempt to explain cross-sectional variation in

turnover rates conditional on firms restating earnings. We argue that the severity of a restatement is better

captured as a function of management intent (or at least the perception of managements’ intent as actual

intent is never fully known to anyone but the managers) than by any of the previously used measures.3

We proxy for management intent with the presence or absence of an independent investigation or with the

presence or absences of the word “irregularity” in discussions of the GAAP violation. We consider the

occurrence of investigations by (or funded by) the board of directors or investigations by external

regulatory bodies (e.g., SEC, Office of the Attorney General, U.S. Department of Labor, etc.) to indicate

that there is at least some suspicion of managerial misbehavior because, as discussed further in Section

2.1, restatements that initiate or evolve from independent investigations typically involve allegations of

intentional misreporting. As an accounting “irregularity” is an intentional misstatement by definition, we

classify these cases as intentional misstatements as well. 4

2 By definition, an intentional misstatement (that is material) is considered to be fraud. We use intent rather than fraud because, unlike most prior research, we are including cases of suspected fraud as well as cases of prosecuted fraud. Hence, our sample is likely different from studies looking at restatements and fraud in other contexts. It should also be noted that settlements with the SEC in cases of fraud can often include stipulations that require certain executives to resign from the firm. However, these settlements almost always occur at a date several years after the initial restatement is announced. 3 As discussed in Section 2, prior research measures severity by the magnitude of the restatement (Collins et al. 2005) or by who initiated the restatement (Arthaud-Day et al. 2006 and Desai et al. 2006). 4 Statement on Auditing Standards (SAS) No. 53.

4

For our analysis, we identify 456 restatements from 2002-2005 that are announced in an 8-K

filing, include reference to an accounting error, irregularity or investigation, and meet our data

requirements. From this, randomly select 83 of the 351 (about 25%) restatements involving an error or

misapplication of GAAP but that are not prompted by and do not prompt an investigation, which we

classify as unintentional misstatements. We add to that all 105 restatements that we classify as

unintentional, giving as a total sample of 188 restatements (83 unintentional and 105 intentional). For all

188 restatements, we review 10-K filings, 8-K filings, and proxy statements to identify turnover of the

CEO or CFO in the six months preceding and the six months following the date of the initial restatement

announcement.

We find that overall CEO (CFO) turnover rates in the 13 months surrounding the restatements

(six months before to six months after) are 18% (25%). In 30% of the cases, either the CEO or CFO

leaves the firm. Turnover appears higher for CFOs, who are directly responsible for financial reporting,

than for CEOs. As expected, when we partition the restatements by managerial intent, turnover rates are

much higher for intentional GAAP violations. For CEOs (CFOs) the turnover rate is 49% (64%) but only

8% (12%) for unintentional violations. We also find that misstatements that occur in a subsidiary lead to

lower turnover rates for CEOs and CFOs than those that occur at the parent-level.5 After excluding

subsidiary-level restatements and expanding the turnover window to four-years (2 years before and 2

years after), the (untabulated) turnover rates for intentional restatements are 67% for CEOs and 85% for

CFOs. In 91% of these intentional misstatement cases, either the CEO or CFO leaves the firm. This

evidence suggests that the relatively low turnover rates documented in past research are due largely to the

inclusion of unintentional GAAP violations that typically do not warrant firing senior managers. When

the restatements are the result of intentional misbehavior, boards take swift action to dismiss managers.

To control for other factors that might give rise to CEO/CFO turnover, we estimate logistic

regressions for both CEO and CFO turnover. We find that additional controls, including leverage, the 5 Although senior management turnover is lower for subsidiary-level restatements, we find that in virtually all cases where an intentional misstatement occurs at the subsidiary-level, the subsidiary-level managers responsible are fired (e.g., subsidiary president and controller).

5

annual or quarterly nature of the restatement, ROA, CEO ownership, and long-window (day -90 to day

-8) cumulative abnormal returns (CARs), do not diminish the significance of our severity measure

(management intent). Our multivariate tests show that, excluding restatements related to subsidiaries, a

CEO (CFO) is almost 11 (25) times more likely to turnover if the restatement involved an intentional

violation rather than an unintentional error.

We also conduct a case-by-case analysis of non-subsidiary GAAP violations that we classify as

intentional and where neither the CEO or CFO left the firm within the 13-month turnover window. Of the

16 such cases, we find only one case where it appears that an intentional GAAP violation occurred but

there is no relevant termination of either the CEO or CFO within the 24 months before or after the

restatement. In this one exceptional case, Asconi Corp., the CEO and CFO owned 90% of the firm.

We next examine whether the high turnover rates we document for intentional misreporting are a

new phenomenon stemming from increased political and regulatory pressure in the wake of Enron and

other accounting scandals. In a 2002 statement to the U. S. Senate Committee on Banking, Housing, and

Urban Affairs, Senator Sarbanes proposed that managers who step outside of GAAP “ought to be

punished, and punished very severely.”6 To the extent that this pressure induced boards to act more

swiftly in response to misreporting, the turnover rates we document in the post-Enron era should be

higher than those in the pre-Enron era.

In order to compare board behavior in the post-Enron era to a pre-Enron period, researchers need

to consider whether the mix of restatements has changed. Since Sarbanes-Oxley restatement frequencies

have increased dramatically (GAO, 2002 and 2006), but the increase in restatements is due partly to more

restating for minor reporting infractions,7 many of which are errors (e.g., spreadsheet errors) discovered

during the course of SOX 404 compliance. This suggests that although SEC enforcement of more

6 Senator Paul S. Sarbanes, Senate Floor Statement on July 8, 2002 on the Public Company Accounting Reform and Investor Protection Act of 2002. 7 In a speech at the Financial Executives International Meeting on November 17, 2006, Scott A. Taub, Acting Chief Accountant of the SEC, reports that about 55% of recent restatements were due to simple data errors or unintentional misapplications of GAAP.

6

egregious misstatements has increased in recent years, 8 the frequency of restatements for minor errors

may have increased even more. Therefore, to assess turnover rates across time it is important to also

consider changes in the mix of restatements across time.

To examine changes in turnover rates and demonstrate the importance of controlling for the type

of misstatement, we collect a sample of 139 restatements from the 2003 GAO database for the period

1997-1998. As with our post-Enron sample, we classify these restatements as intentional and

unintentional. Frequency analysis suggests that overall CEO and CFO turnover rates around restatements

have actually decreased in the period after Enron. However, we show that this change is due to major shift

in the mix of restatements over time. We find that the turnover rates for intentional misstatements have

changed very little over time. However, we do not interpret the lack of a change in turnover rates for

intentional misstatements in the post-Enron era as regulation being ineffective at increasing board

effectiveness (i.e., terminating senior managers when intentional misstatements are discovered). Instead,

we conclude that given the extremely high turnover rates for intentional misstatements (in both the pre-

and post-Enron era), boards appear to have been effective in this regard all along. Consequently, any

regulation attempting to increase board diligence in disciplining managers for intentional misreporting,

will yield only negligible changes (i.e., you cannot fix something that is not broken).

In summary, our findings highlight the importance of distinguishing restatements by a meaningful

measure of severity. Although not a perfect measure of managerial intent (because management intent is

impossible to actually observe), our severity proxy is fairly easy to construct and appears to be very

effective at capturing the seriousness of the restatement: observed turnover rates are 5 to 6 times higher

(log-odds are 11 to 25 times higher after controlling for other explanatory variables) for our intentional

GAAP violations than for unintentional violations.

For researchers, our findings suggest that partitioning on our (fairly easy to construct) proxy for

management intent can significantly enhance the power of tests in most studies related to restatements.

8 SEC Commissioner, Harvey Goldschmidt, in a December 2, 2002 speech at Fordham University Law School, reported that SEC enforcement actions for potential accounting fraud increased from 79 in 1999 to 163 in 2002.

7

This includes not only studies on turnover around restatements but also studies on other topics (e.g.,

insider trading, cost of capital, information content, etc.) that utilize a restatement setting. We counted

more than twenty existing studies (either published or in working paper form) that rely on the GAO

database. Because the GAO sample includes both intentional and unintentional GAAP violations, our

study suggests that the power of the test in most of these studies could be greatly enhanced with our proxy

for intent. Our study also shows that most turnover (roughly 80%) linked to a restatement occurs within a

thirteen-month window surrounding the restatement announcement (six months before and six months

after). Consequently, tests that use shorter windows (as opposed to early research that used windows as

long as five years) are likely more powerful.

The remainder of the paper is organized as follows. Section 2 discusses prior research on turnover

around restatements and develops our predictions on the relation between turnover and restatement

severity (management intent). Section 3 describes our sample selection procedures and reports descriptive

statistics. Section 4 discusses our results, and Section 5 concludes.

2.0 Prior Research on Executive Turnover around Restatements

Management of restating firms likely face reduced compensation, decreased credibility, loss of

employment, and even criminal charges depending on the severity of the GAAP violation. Although

research on the effect of restatements on managerial compensation has recently emerged (e.g., Collins et

al. (2005) and Glass, Lewis, and Co. (2005)), we follow earlier research and focus on loss of employment

as the most severe punishment implementable by the board of directors.

Some of these prior turnover studies find no evidence that restatements, even restatements linked

to explicit fraud, significantly affect the odds of CEO turnover. For example, Beneish (1999) finds no

difference in CEO turnover for firms that violate GAAP during 1987-1993 as compared to a control

sample of compliant firms, and Agrawal et al. (1999) similarly find little evidence that firms suspected of

fraud (including accounting fraud) between 1978 and 1992 have any higher executive turnover than non-

fraud firms.

8

Other studies do find that accounting restatements increase the likelihood of managerial turnover,

although the number of misstating firms that do not experience any turnover is still somewhat higher than

researchers can fully explain. The percentage of firms experiencing executive turnover in prior samples

varies depending on the number of executives and the time windows considered: Desai et al. (2006) find

51% of restating firms in 1997-1998 have turnover of their CEO, Chairman, or President within 2 years

after restatement; Land (2006) estimates 45% of firms restating between 1996 and 1999 have CEO

turnover in the year after the restatement; in the 2 years after the restatement, Arthaud-Day et al. (2006)

observe CEO turnover in 43% and CFO turnover in 55% of their 1998-1999 sample of restating firms;

and Jayaraman, et al. (2004) find that 48% of their restating firms experience turnover of their CEO,

Chairman, or President and 45% experience turnover of their CFO, Treasurer, or Controller in the 4 years

after being listed in the 1999-2000 Accounting and Auditing Enforcement Releases (AAERs). Overall,

past research suggests that a considerable portion of restating firms do not replace management for

financial reporting failures.

2.1 Restatement Severity, Management Intent, and Turnover

Much of the recent research on restatements and executive turnover relies on the GAO sample,

which does not distinguish between intentional (fraudulent) and unintentional GAAP violations.9 The

information provided by the GAO is limited, so it is difficult to sort restatements by severity. The only

potential severity measures available in the GAO database are the nine categories of restatements and the

prompter of the restatement, which studies have used with somewhat mixed success (Arthaud-Day et al.

2006 and Desai et al. 2006). However, as the GAO admits, the prompter of the restatement is hard to

9 For example, in 2005 the SEC sent a letter to the AICPA that clarified the SEC’s position on the application of GAAP for various operating lease issues. This interpretation letter prompted widespread restatements (all of which are captured in the 2006 GAO database) across the retail sector, but these lease adjustments are likely not indicative of aggressive accounting.

9

identify, so the coding of this variable may be inaccurate or nonexistent.10 Therefore, we consider an

alternative measure of restatement severity.

A recent study by Palmrose et al. (2004), suggests that restatements caused by fraud are likely to

be considered the most severe. The authors hypothesize that the stock price reaction is likely greater in

cases of fraud for two reasons. First, the revelation of fraud is likely to increase the discount rate because

it reduces the reliability of management disclosures. Second, it increases the cost of litigation and

regulatory actions, additional monitoring costs, and future regulatory scrutiny. The authors classify a

restatement as being due to fraud if the there is an associated AAER, or the firm acknowledges the

restatement is due to fraud/irregularity. They report that the market reaction to these restatements is -20%

compared to only -6% for non-fraud cases.

Whether management deliberately misreports has been shown to affect the costs of the

restatement to the firm as a whole, but management intent has not yet been fully explored as a factor in

executive turnover decisions. We predict that restatements that occur as a result of the company’s efforts

to intentionally mislead investors and other stakeholders are much more likely to be related to turnover as

compared to cases that result from either errors in interpretation of GAAP or clerical-type errors.

The Chicago Bridge and Iron Company (CB&I) provides an example of a restatement where the

misreporting is suspected to have been intentional: On October 31, 2005, CB&I announced that “the

delay in releasing third quarter 2005 financial results was precipitated by a memo from a senior member

of CB&I’s accounting department alleging accounting improprieties.”11 The accusation of misbehavior

prompted an investigation by the board of directors, which revealed that there were deliberate accounting

irregularities that would necessitate a restatement. Given the revelation of conscious misreporting, we

would classify this restatement as an intentional GAAP violation.

10 The GAO indicates the prompter is unknown in 35% of the restatements in their database. It is also likely that many of the restatements attributed to the company (49%) were actually instigated by auditors’ or regulators’ concerns that were not disclosed. 11 http://www.sec.gov/Archives/edgar/data/1027884/000095012905010329/h29789exv99w1.htm

10

In contrast, an Applebee’s International restatement announcement that was included in the 2006

GAO database suggests a misinterpretation of GAAP but no deliberate attempt to mislead investors: “On

February 9, 2005… like many other companies in the restaurant, retail and other industries, it had

determined that it would correct its accounting treatment for leases.”12 As discussed previously, many

retailers restated due to lease accounting after the SEC issued a letter on February 7, 2005 regarding the

treatment of leases. It is likely that investors would believe that this common error was unintentional.

Although this error could represent a large dollar amount, the reduction in management’s reporting

credibility over this issue is likely to be quite small. Consequently, we classify this restatement as an

unintentional restatement.

We argue that managers’ intent is an important determinant of the board’s perception of severity

and should thus predict turnover around restatements. Besides the costs associated with a fraudulent

misstatement, described above, boards must also consider ways to restore credibility of financial

disclosure. Credible financial reports are vital for access to capital markets and firms must regain any

credibility lost over a restatement. For firms that unintentionally violate GAAP, restoring credibility is

likely easier, since the cause of the GAAP violation can be attributed to problems that can be corrected.

For example, clerical errors can be reduced by implementing systems to check for these potential errors.

However, the best way to credibly eliminate the problems associated with intentional misstatements is to

terminate the employees who are ultimately responsible (e.g., CEO and CFO).

2.1.1 Classifying Intentional and Unintentional Misstatements

Ideally, a firm explicitly discloses that the restatement relates to an accounting irregularity. In

those cases, we can clearly classify the GAAP violation as being intentional. Unfortunately, such explicit

disclosure is not always the case.13 Based on our reading of numerous restatement disclosures, when the

words fraud or irregularity are not explicitly used, our best distinction between an intentional and

12 http://www.sec.gov/Archives/edgar/data/853665/000085366505000063/restatement8k.txt 13 In roughly 60% of the observations that we classify as intentional, where we suspect the firm intentionally misstated earnings, the word “irregularity” or “irregularities” is used.

11

unintentional GAAP violation is whether or not an independent investigation into an accounting matter

was initiated.

Typically, when an accounting irregularity is identified either by the firm or the firm’s auditor, an

independent investigation funded by the board will follow. Similarly, if the SEC suspects an accounting

impropriety, it will initiate its own formal or informal investigation. Therefore, we classify a restatement

as intentional if the disclosure discusses an irregularity, a board-initiated independent investigation, or an

external regulatory inquiry (e.g., SEC, the Attorney General’s Office, the Department of Labor, etc.).

Measuring managerial intent in this manner is not perfect, but it appears to be very effective as a proxy.

The most common cause of a misclassification using this methodology is when an SEC investigation

results from a disagreement about a particular accounting treatment rather than from allegations of

misconduct, but these instances are relatively rare and will only bias against our predictions. We test the

validity of our proxy for severity by reviewing disclosures, analyzing the relative market reaction to

intentional versus unintentional GAAP violations, and examining the class action lawsuits alleging fraud

in our sample (see Section 3.3).

2.2 Intentional Misstatements at the Parent-Level Versus Subsidiary-Level

In addition to intent, we also consider the level of the infraction. In many cases the GAAP

violation is isolated to a foreign subsidiary or distinct business unit within the firm. For example, Amcon

Distributing Company filed an 8-K disclosing management's investigation into potential accounting

irregularities that were discovered in the inventory accounting records of Hawaiian Natural Water Co.,

Inc., a wholly owned AMCON subsidiary. In response, the firm fired both the president and the chief

financial officer of the subsidiary. However, given that the irregularity was entirely contained within the

subsidiary, we might not expect the board to terminate the CEO or CFO at the firm level.

When intentional misstatements are isolated to a subsidiary, the CEO/CFO can attribute the

problem to subsidiary-level management and restore financial reporting credibility by firing the

subsidiary-level management team. We thus predict that CEO/CFO turnover is less likely to occur for

12

intentional misstatements that occur at the subsidiary-level than for intentional misstatements that cannot

be attributed to an isolated business unit.

2.3 Other Factors Influencing Turnover

In addition to our severity and subsidiary variables discussed in the previous sections, we also

control for other factors that prior research suggests are likely related to turnover. Prior restatement

studies (e.g., Palmrose and Scholz, 2004) argue that restatements of unaudited interim reports are viewed

as less severe than restatements of audited annual reports, so we include a dummy for annual

restatements. To control for potential entrenchment effects such as those documented in Denis, Denis, and

Sarin (1997), we consider board and CEO ownership. Many prior studies, including Gilson (1989), find

that turnover is more likely for financially distressed firms, so we also control for distress with

LEVERAGE (debt to total assets).14 Finally, we control for past stock performance, measured as the CAR

in the 90 days prior to the restatement announcement up to 8 days prior to the announcement. This

controls for potential performance-related reasons for terminating the CEO/CFO, such as those

documented by Warner, Watts, and Wruck (1988) or Mian (2001). We also control for size by including

indicator variables for size quintiles. To test our predictions for our severity measures while controlling

for all the factors described above, we estimate the following model for CEO, CFO, and CEO or CFO

turnover:

TURNOVER=β0+β1INTENTIONAL+β2SUBSIDIARY+β3INTENTIONAL*SUBSIDIARY +β4ANNUAL+β5CEO_EQUITY +β6LEVERAGE +β7ROA+β9CAR(t-90-t-8)

+β8Size Quintile 1+β9Size Quintile 2+ β10Size Quintile 4+ β11Size Quintile 5+ε

(1)

where

TURNOVER=1 if the CEO (or CFO) announces that he/she is leaving the firm within the 6 months before or the 6 months after the restatement announcement and 0 otherwise. INTENTIONAL=1 if the restatement involved either a board investigation or SEC investigation and 0 otherwise.

14 We also used the Altman (1968) Z-Score to measure financial distress but this measure caused the loss of more observations than the loss of observations using leverage. As Z-Score was not significant in our regressions anyway, we use LEVERAGE instead to retain more observations and reduce the likelihood of introducing a selection bias.

13

SUBSIDIARY=1 if the GAAP violation occurred in a subsidiary and 0 otherwise. ANNUAL=1 if the firm restated a 10-K and 0 if the firm restated only 10-Qs; CEO_EQUITY=the percentage of equity ownership of the CEO in the year prior to the restatement; LEVERAGE=Debt (#9+#34) / Assets (#6); ROA= Operating income after depreciation scaled by average assets (Compustat #178/#6); CAR(t-90-t-8)= The firm’s cumulative abnormal returns from 90 trading days prior to the restatement announcement to 8 trading days prior, and expected returns are the CRSP value-weighted returns inclusive of dividends; SIZE QUINTILEi= An indicator variable for size quintiles based on total assets.

3.0 Sample Selection and Descriptive Statistics 3.1 Sample Selection

Table 1 summarizes our sample selection. Restatement firms are identified by searching the 8-K

filings on the SEC Edgar site from January 1, 2002 – June 15, 2006.15 To be included in the sample, a

firm must announce an earnings restatement and reference an accounting error, an irregularity, or an

internal or SEC investigation into accounting matters. Our procedure yields a total of 630 restatements

where there is reference to an accounting error or to an independent investigation.16 Of these, 460 are

classified as unintentional GAAP violations and 170 classified as intentional. A restatement is considered

to be the result of an intentional GAAP violation if the firm announced either an SEC inquiry or a board-

sponsored independent investigation, or if the firm referred to the misstatement as an irregularity.

Consistent with prior research, we further exclude firms in the financial services industry (SIC codes

6000-6999) as well as firms that do not have data available on Compustat. This leaves us with 361

unintentional and 132 intentional cases.

15 We began this data collection before the updated GAO data (2006) were released. 16 Beginning on August 14, 2004, the SEC now requires firms to file an 8-K with Item 4.02 (Non-Reliance of Previously Issued Financial Statements) whenever it is determined that previously issued financial statements should no longer be relied upon. This disclosure change likely enables us to identify more restatements in the time-period after 2004 for two reasons. First, prior to this change, although firms often disclose restatements either at the time a pending restatement is expected or at the time of an earnings announcement, such disclosure was not explicitly required. Second, the new coding scheme that includes a separate category explicitly for restatements greatly increases the accuracy of our search procedures. We do not believe this biases our tests. The specific regular expressions used to identify restatements are available from the authors upon request.

14

To reduce the cost of data collection, we randomly select approximately 25% of the restatements

arising from unintentional violations. We include all intentional violations to maintain power in our tests

of intentional GAAP violations. After reading restatement announcements, we drop 10 additional

observations where the restatement had no income effect. Lastly, we eliminate 27 cases where the

accounting investigation ultimately did not lead to a restatement or is not yet completed. This leaves 83

unintentional and 105 intentional observations.

To adjust for the disproportional sampling (we select all intentional GAAP violations

restatements but only a fraction of the unintentional), we follow Manski and Lerman (1977) and weight

by estimated population proportions. Unintentional (intentional) restatements are assigned a weight of

3.9(1). The weight of 3.9 for the unintentional sample is computed by dividing the total number of

restatements identified in the population (361) by the total number of observations sampled for analysis

(93).

3.2 The Turnover Window

In contrast to many prior restatement studies, we consider turnover both before and after the

announcement date for two reasons. First, if management turnover occurs prior to the announcement, it

may be the case that new management discovered GAAP violations perpetuated by prior management. In

such a case, we would not expect current management to be dismissed for the acts of previous

management. If these pre-announcement management changes are not considered, researchers will

incorrectly conclude that the boards failed to take swift action to terminate at-fault managers. A second

reason to include the pre-restatement announcement period is that an initial investigation into

irregularities can occur many months before a restatement is publicly announced. CEOs and CFOs may

be terminated during an investigation but before the restatement is announced. These measurement

problems suggest that expanding the turnover window to include a timeframe before the restatement

announcement is appropriate.

15

In our primary analysis, we use a 13-month window (six months before and six months after)

around the restatement announcement.17 However, given the lack of consensus in past research on the

appropriate measurement window for executive turnover around restatements, we review the timing of

turnover to assess the most appropriate window. We plot the cumulative turnover rates for restatements

caused by intentional and unintentional GAAP violations beginning six months before to two years after

the restatement announcement. Figure 1 plots the cumulative turnover rates for CEOs beginning six

months prior to the restatement announcement. The plot shows that most of the turnover for both

intentional and unintentional GAAP violations occurs very close to the announcement between six

months before and six months after the restatement announcement. A total of 50 (89% of all turnover)

CEOs announce they are leaving their firm over this time period for intentional GAAP violations. Very

little turnover occurs after six months following the restatement announcement. The turnover rate is much

lower for unintentional restatements (a total of 9), but it is still concentrated around the five months

before and five months after the announcement where 7 (77%) CEO departures are announced.

Figure 2 plots cumulative CFO turnover. As in the case of CEO turnover, most of the CFO

turnover occurs close to the restatement announcement. Roughly 92% of the CFO turnover (63 out of 79)

occurs between six months prior to and six months following the restatement announcement for

intentional GAAP violations. CFO turnover is less concentrated for the unintentional cases where 10 out

of 18 restatements (55%) occur between six months prior to and six months following the restatement.

Overall, this analysis suggests that most turnover occurs close to the restatement announcement

and that using windows that cover six months before and six months after the restatement will safely

capture most of the turnover related to restatements. More importantly, the fact that most of the turnover

occurs close to the restatement suggests that the turnover is likely related to the restatement. Using longer

windows will tend to introduce more noise into the analysis (i.e., more cases of turnover unrelated to the

restatement).

17 Identifying exact announcement dates related to restatements is challenging. We use the date that an intention to restate is first made (not the date that an initial investigation or a potential restatement is announced).

16

3.3 Validity of Severity Measure

We argue that investigations (internal and external) generally signal that the GAAP violation is

likely due to an intentional misstatement and that these misstatements are more severe than unintentional

errors. To validate that these intentional GAAP violations are correctly capturing severity, we analyze the

stock returns of our restatement firms between 90 days prior to 90 days following the restatement

announcement. Figure 3 reports population-proportion adjusted mean and median CARs from t-90 to

t+90. Expected returns used to compute CARs are CRSP value-weighted returns with dividends. To avoid

survivor bias, we do not require firms to trade on all 180 days. The number of observations ranges from

152-169.

Consistent with prior research, there is a noticeable dip in returns in the days surrounding the

restatement announcement with mean (median) CAR of -4.9% (-3.1%) from seven days before to seven

days after the announcement. These returns are not as negative as reported in earlier research. For

example, Palmrose et al. (2004) report mean (median) 2-day CARS of -9.5% (-5.1%), for restatements

announced between 1995 and 1999. However, this is potentially explained by a change in the mix of

restatements over time. We also note that returns decline prior to the announcement and we believe this

decline is largely attributable to the difficulty in measuring the “information event.” We use the date that

the firm announces that it will restate earnings as the announcement date, but firms may disclose that they

are investigating accounting issues in an earlier press release. These earlier announcements likely lead to

trading in anticipation of the restatement.

Figure 4 reports the mean and median CARs grouped by intentional and unintentional violations.

The number of observations ranges from 84 to 101 for intentional and 68 to 69 for unintentional

restatements. The CARs for our unintentional sample do not drift far from zero over the entire 180-day

period and exhibit very little reaction to the restatement announcement. The returns are actually positive

leading up to the announcement but mean (median) CARs are -1.93% (-.90%) in the 15-day window

around the announcement. In contrast, the CARs for the intentional group decline substantially. The mean

17

(median) CARs are -13.64% (-19.14%) for the intentional group around the same event window. In

addition, CARs drifted -13% (-12%) from t-90 to t-8 for the intentional restaters. Again, this downward

drift prior to the announcement dates suggests that a number of firms in our sample disclosed an expected

restatement before they announced that the actually would restate earnings. Overall, this evidence is

consistent with our proxy for intent capturing the market’s perception of the seriousness of the

restatement.

As a second validity check, we compare the frequency of class action lawsuits claiming fraud for

our INTENTIONAL and UNINTENTIONAL samples. Assuming that, all else equal, it is easier to file a

class action lawsuit for fraud (intentional misstatement), we should find that occurrence of a class action

lawsuit is highly correlated with our proxy for intent. Table 2, Panel A, reports a frequency count of our

classification of INTENTIONAL versus the filing of a class action lawsuit.18 The first three columns of

Panel A provide counts based on our classification criteria (Irregularity, SEC inquiry, Board-initiated

internal investigation). Some firms have more than one of the required conditions (e.g., SEC investigation

and internal investigation), hence the three columns sum up to more than the “Total” column. Overall, we

find only one case where a class action lawsuit was initiated but we did not code the firm as an

investigation firm.

There are 21 cases (out of 105) where we classify the restatement as intentional but there is no

corresponding class action lawsuit. Additional analysis, reported in Panel B, reveals that most these

restatements appear to be cases that might have otherwise warranted a class action law suit but the firms’

stock was not being traded at the time of the restatement (n=3), the market value of the firm was below

$25 million (n=8), the restatement occurred in a subsidiary (n=7), or the stock returns around the

restatement were positive (n=3). These findings suggest that our proxy for intent appears to work quite

well at identifying intentional misstatements. Further, it is likely better than relying on class action

18 We obtained class action lawsuit filings from the Stanford Class Action Clearinghouse (http://securities.stanford.edu/companies.html) and Lexis/Nexis.

18

lawsuits as a proxy because using class action lawsuits will cause researchers to misclassify cases where

fraud occurs but lawyers do not find it beneficial to sue (e.g., smaller firms or limited damages).

3.4 Descriptive Statistics

Descriptive Statistics are reported in Table 3. The table contains statistics for our intentional and

unintentional GAAP violations as well as for a random sample of firm-year observations, stratified on 2-

digit SIC code and year, for comparison purposes. To mitigate the impact of outliers, all variables are

winsorized at the bottom and top 1%. All variables are measured in the year prior to the restatement

announcement.

Somewhat surprisingly, measures of size (sales, total assets, and market value) suggest that the

high-severity restatement firms are the largest. For example, mean (median) sales in millions of dollars of

the intentional, unintentional, and random sample groups are 3,984 (497), 2,725 (533), and 2,252 (163),

respectively. Mean (median) ROA is lower for the INTENTIONAL group (mean=-1.8%, median=3.0%)

compared to the UNINTENTIONAL group (mean=3.3% and median=5.6%), though the mean is higher

than that of the random sample (mean=-12.1%).

Mean (median) net income/assets is -2.3(2.3%), -14.2% (-3.6%), and -22.7% (2.2%) for

unintentional, intentional and random sample groups, respectively. LEVERAGE is higher for the

INTENTIONAL sample (mean=.29, median=.25) compared to the UNINTENTIONAL sample

(mean=.25, median=.17). Stock ownership of the CEO is similar between UNINTENTIONAL (mean =

9%, median = 3%) and intentional (mean = 7%, median = 2%).19 Overall, the performance characteristics

and stock ownership of the unintentional and intentional groups are fairly similar. Both groups, however,

appear to be larger than a random sample of the Compustat population.

19 We do not hand-collect ownership information for our random sample.

19

4. 0 Results 4.1 Univariate Analysis

Table 4 reports frequency information on CEO and CFO turnover rates across various groups.

Beginning with the 324 unintentional violations (actual count is 83 but count is weighted by 3.9 for

comparability to the intentional sample), we find that the turnover rate is substantially higher for CFOs;

more specifically, we observe that 8.4% (12.0%) of the CEOs (CFOs) turned over within between 6

months before and 6 months after the restatement announcement. In 15.7% of the unintentional

violations, either the CEO or CFO resigned.20 The turnover rates for the intentional cases are much

higher than the unintentional cases. In the intentional cases, the turnover rate for CEOs (CFOs) is 48.6%

(63.8%) and the combined turnover rate for either a CEO or CFO is 73.3%. This implies that a CEO

(CFO) leaves more frequently around intentional GAAP violations than around unintentional

misstatements, which is consistent with our expectation that boards are more likely to terminate CEOs

and CFOs when the GAAP violation that necessitated the restatement is perceived as intentional.

We next compare the turnover rates of the subsidiary-level restatements to the parent-level

restatements. As expected, the turnover rates for subsidiary-level restatements are lower that of parent-

level restatements in the case of intentional violations. CEO (CFO) turnover rates for subsidiary-level

restatements are 29.2% (41.7%) compared to 54.3% (70.4%) for parent-level restatements. This is

consistent with the firm attributing the intentional misstatements to managers at the subsidiary. Also

consistent with this explanation is that, in virtually every case of an intentional violation at the subsidiary-

level, we find that the firm announced that the subsidiary-level managers were terminated.21 In

untabulated results, we find that turnover rates for parent-level restatements are extremely high when a

20 It is difficult to compare the turnover rates that we document to prior restatement research because past research

investigates different time periods, different groupings of officers, and different window lengths. For non-restatement samples, Yermack (2004) finds an unconditional annual rate of CEO turnover of 13.8% in Fortune 500 firms from 1994 to 1996. More recent research by Kaplan and Minton (2006) finds that CEO turnover in Fortune 500 firms is 16.5% over the period from 1998 to 2005. We find an annual turnover rate (6 months before to 6 months after) of 48.6% (8.4%) for the intentional (unintentional) restatement firms in our sample. 21 Although we make made no prediction about turnover rates for unintentional subsidiary-level violations, it is somewhat surprising that the observed turnover rates are actually higher than in the case of unintentional parent-level restatements. However, there are only 15 such cases (unweighted), which limits inferences.

20

four-year window (two years before to two years after) is considered. Over this time period, we find that

the CEO (CFO) turnover rate for intentional/parent-level restatements is 66.7% (85.2%) and in 91.4% of

the cases either the CEO or CFO leaves the firm.

Finally, we compare turnover related to annual restatements to those of quarterly restatements.

Given that quarterly financial statements have not undergone the audit process, we expect quarterly

restatements to be viewed as being less severe than annual restatements. We consider a restatement to be

annual when the 10-K is restated and quarterly when only 10-Qs are restated. The frequency of quarterly

restatements in our sample is much lower than annual restatements. There are 372 annual restatements

(281 unintentional and 91 intentional) and only 57 quarterly restatements (43 unintentional and 14

intentional). As expected, the turnover rates are higher for annual restatements in the intentional group.

Somewhat surprisingly, the turnover rates are higher for quarterly restatements in the unintentional

sample.

In summary, we observe that turnover appears to be strongly related to the underlying intent

behind the GAAP violation. Intentional violations appear to lead to greater turnover of both CEOs and

CFOs, but the turnover rates are reduced when the problem can be isolated to a subsidiary. In the next

section, we provide logistic regression analysis to further support these observations.

4.2 Multivariate Analysis

Our observations from Table 4 support our predictions that intentional GAAP violations give rise

to higher CEO and CFO turnover rates and that the turnover rates for intentional GAAP violations will be

lower when the irregularity causing the restatement occurs at the subsidiary-level. However, other factors

that cause turnover are likely to be correlated with intentional GAAP violations. To control for these

factors, we estimate logistic regressions for CEO, CFO and CEO or CFO turnover.22

22 Given the strong relation between bankruptcy and CEO turnover described in past research (e.g., Beneish 1999) we test the sensitivity of our results to bankruptcy filings. We identify all firms that file for bankruptcy at any point after the restatement announcement. There are 21 firms with restatements classified as intentional and only 2 with restatements classified as unintentional. Although the turnover rate for the remainder of the intentional group

21

Table 5 reports results for our turnover regressions. Consistent with our prediction, the coefficient

on INTENTIONAL is positive and significant (p<0.01). Given a parent-level restatement, a CEO is

roughly 11 times more likely to turnover if the restatement was an intentional violation rather than

unintentional. To test whether turnover rates are lower for intentional, subsidiary-level restatements than

for intentional, parent-level restatements, we test β2+β3<0. As reported in table 5, β2+β3=-1.32 and is

significantly less than zero (p<.05 one-tailed test). All p-values in the remainder of this paper are reported

as one-tailed when the sign is predicted. Of the control variables, only CEO_Equity is significant and in

the expected direction (p<.05). The negative coefficient is consistent with the CEOs using their ownership

in the firm to make it more difficult for boards to fire them.

Results for CFO turnover are very similar to our CEO turnover results. The signs and significance

levels on INTENTIONAL and β2+ β3 are consistent with our predictions. The coefficient on

INTENTIONAL is 3.22 (significant at p<.01), suggesting that, conditional on a parent-level restatement,

a CFO is almost 25 times more likely to turn over if the restatement is an intentional violation rather than

an unintentional violation. β2+β3=-1.79 and is significantly less than zero (p<.05), suggesting that

turnover for CFOs is also lower for intentional misstatements if the misstatements occur at the subsidiary

level. The coefficient on CEO_EQUITY is also weakly significantly negative in the CFO regression

(p<.10) suggesting that the CEO exerts influence to retain the CFO. Finally, the size controls suggest that

the probability of CFO turnover declines with firm size.

When CEO and CFO turnover are combined, results are very similar. The signs and significance

levels of the coefficients for our predictions remain. For all models, most of our performance-related

control variables are not significant. This is consistent with our descriptive statistics in Table 3,

suggesting that performance is fairly similar for INTENTIONAL and UNINTENTIONAL restaters.

Given the CAR distributions reported in Figures 3 and 4, one question is whether our proxy for

severity (intent) is any better than simply using restatement announcement returns. In other words, our

declines, it is still significantly higher than the unintentional group. The CEO turnover rate for firms that do not file for bankruptcy is 42.9% for intentional restatements and 8.6% for unintentional restatements.

22

measure INTENTIONAL could simply be capturing the market reaction to restatements, which is an

overall measure of severity. In table 6 we compare the effectiveness of this alternative proxy by re-

estimating our model substituting announcement CARs (CAR(t-7 – t+7)) for INTENTIONAL. For all

turnovers (CEO, CFO, CEO/CFO), the main effect, CAR(t-7 – t+7), is similar in sign and significance to

INTENTIONAL, though β2+β3 is not significantly different from zero. However, the explanatory power

of the model (measured by Psuedo-R2 or Log Likelihood) are much lower. For example, the log

likelihoods using CAR(t-7 – t+7) as a proxy for severity are 21.9, 33.5, and 31.5 compared with log

likelihoods of 37.2, 61.6, and 64.1 when using INTENTIONAL as a proxy for severity, for the CEO,

CFO and CEO/CFO turnover regressions, respectively. These findings suggest that, although short-

window CARs are a good proxy for severity, INTENTIONAL is more effective, at least in explaining

turnover.

4.3 Descriptive Review of Intentional Violations with No Turnover

Although the high turnover rates that we document are suggestive of boards acting swiftly to

remove managers involved in irregularities, it is still somewhat surprising that intentional, Parent-Level

GAAP violations do not always lead to the resignation of the CEO and CFO. To understand what factors

explain why in 20% of the cases (16 of 81) neither the CEO nor the CFO left within the 13-month

window, we perform a detailed examination of each of these cases. Our review of these cases is

summarized in Table 7.

There are eight cases where either the CEO or CFO left the firm in the 18 months preceding the

13-month window (i.e., from two years to six months prior to the restatement). In these cases, it is likely

that the incoming CEO (CFO) was not blamed for (and may in fact have discovered) the misstatement as

he/she was able to attribute the problem to his/her predecessor. There are three cases where the CEO or

CFO left the firm after the six-month window but the departure occurs around the time the investigation

was concluded. These “no turnover” cases are attributable to our use of a shorter turnover window.

Finally, there are four cases where it was subsequently concluded either from the independent

23

investigation or the SEC investigation that no intentional GAAP violations occurred. This leaves only one

case, Asconi Corp., where it appears that an intentional misstatement occurred but neither the CFO nor

the CEO is terminated.23 However, Asconi’s CEO and CFO together hold over 90% of the company’s

stock. We conclude from this analysis that boards rarely fail to discipline senior management when

intentional misstatements are discovered.

4.4 Changes in Turnover Rates Pre- versus Post-Enron

In this section we compare turnover rates in the pre-Enron era to the post-Enron era. Our pre-

Enron sample firms are identified by the GAO database (2003) as having a restatement that occurred

between January 1, 1997 and December 31, 1998. We select this time period because it provides a clean

window before the scandals and subsequent regulation is comparable to other recent research on the pre-

Enron period (i.e., Desai et al. (2006)). The pre-Enron sample consists of 194 unique firms restating in

1997 or 1998 as identified by the GAO database. Of these, we exclude 16 firms in the financial services

industry (SIC codes 6000-6999) to be consistent with prior research, and 29 firms that have missing data.

We drop an additional 13 observations where further investigation reveals that the restatement had no

income impact, pertained only to the representation of now-discontinued operations, or was announced

contemporaneously with a merger. This results in a final sample of 136 restatements during the pre-Enron

period.

Table 8 reports logistic regression results estimating model (1) with the addition of an indicator

variable, POST, which is equal to 1 if the restatement occurred in the post-Enron era (2002-2005) and 0 if

it occurred in the pre-Enron era (1998-1999). For ease of exposition, we exclude the interaction term

between INTENTIONAL and SUBSIDIARY. Size quintile controls are included in the regression but not

reported in the table for brevity. Column (1) reports results with CEO Turnover as the dependent variable

23 Asconi Company issued ten million shares during 2003 to the company’s CEO and CFO. The issuances were originally treated as equity transactions. SEC conducted a formal investigation and determined that the market value of the ten million shares of Common Stock issued during 2003 should have been charged against the Company’s income statement. The restatement resulted in the recording of $40.4 million of stock issuance expense.

24

and INTENTIONAL left out of the regression. Without INTENTIONAL, the coefficient on POST is

negative and significant (p<.01), suggesting the overall turnover rates actually declined in the post-Enron

Era. Similar results are obtained in columns (3) and (5) for CFO and CEO/CFO turnover.

When the INTENTIONAL indicator variable is added to the regressions, the significance level on

POST disappears in both the CEO turnover and CFO turnover regressions. However, the coefficient on

POST remains negative and significant in the CEO/CFO turnover regression. It is possible that the

coefficient on POST for the CEO/CFO regression is being driven by differences in turnover rates from

the unintentional restatements. To alleviate this concern and focus on our primary interest of how Boards

respond to severe misstatements (INTENTIONAL), we repeat our analysis including only

INTENTIONAL restatements.

As shown in table 9, the coefficient on POST is insignificant across all three regressions. This

indicates that turnover rates for INTENTIONAL restatements have not changed significantly in the post-

Enron era. Given the extremely high turnover rates in the post-Enron era, we do not interpret this as

ineffective governance continuing even after additional regulation was implemented. Instead, we interpret

this as indicating that boards continue to be effective at taking action in response to intentional

misreporting by managers.

5.0 Concluding Comments

Past research on the relation between senior executive turnover and restatements has been

surprisingly mixed. Until recently, researchers found no evidence that restatements increased the

probability of senior management turnover. Although recent research finds a statistically significant

difference between the restatements and the likelihood of turnover, the turnover rates still appear to be

relatively low. In this study, we attempt to explain the cross-sectional variation in turnover for firms that

restated earnings between 2002 and 2005. We predict that intentional GAAP violations are much more

likely to lead to CEO or CFO turnover.

25

As predicted, we find that turnover rates are much higher for intentional violations than for

unintentional errors: the turnover rate is 48.6% (63.8%) for CEOs (CFOs) for intentional cases but is only

8.4% (12.0%) for unintentional cases. This evidence suggests that boards do take swift action to dismiss

managers when the restatements are the result of intentional misbehavior and that the relatively low

turnover rates documented in earlier restatement research may be due to the inclusion of unintentional

violations in prior restatement samples. We also find that turnover caused by restatements generally

occurs within a one-year window surrounding the restatement announcement (six months before and six

months after), suggesting that shorter windows spanning both the pre-and post-announcement periods will

allow for more powerful tests.

Finally, we find that conditioning on intent is critical to interpreting changes in turnover rates

over time. Without controlling for intent, we obtain the counter-intuitive result that turnover rates have

declined in the post-Enron era. However, examining only cases where turnover would be expected

(intentional restatements), we find that turnover rates have remained constant when comparing the pre-

and post-Enron eras.

For regulators, our findings suggest boards take swift action to remove managers when

intentional misreporting occurs and contradict concerns by Abelson (1996), who writes that when are

“manipulating a company’s financial numbers to mislead investors, the punishment is often anything but

sharp or swift.” We conclude that board governance appears to be quite effective at disciplining firm

management for financial reporting failures. In particular, CEOs and CFOs appear to face heavy penalties

for any willful misreporting.

For researchers, our finding suggests that the power of tests on the consequences of restatements

(e.g., turnover, market reaction, etc.) can be greatly enhanced by classifying restatements by a proxy for

management intent. Data sources, such as the GAO sample, mix together both intentional and

unintentional violations of GAAP, which are substantially different in perceived severity. Our proxy for

intent, presence of an independent investigation is very effective at identifying intentional misstatements

and is relatively easy to construct.

26

References

Abelson, R. 1996. Truth or Consequences? Hardly. New York Times (June 23): F1. Agrawal, A., J. F. Jaffe, and J. M. Karpoff. 1999. Management Turnover and Governance

Changes following the Revelation of Fraud. Journal of Law and Economics, 42(1): 309-342.

Altman, E. 1968. Financial Ratios, Discriminant Analysis and the Prediction of Corporate

Bankruptcy. Journal of Finance, 23: 589-609. Arthaud-Day, M. L., S. T. Certo, C. M. Dalton, and D. R. Dalton. 2006. A Changing of the

Guard: Executive and Director Turnover Following Corporate Financial Restatements. Academy of Management Journal, forthcoming.

Beneish, M. D. 1999. Incentives and Penalties Related to Earnings Overstatements that Violate

GAAP. Accounting Review, 74(4): 425-457. Collins, D., A. L. Reitenga, and J. M. Sanchez-Cuevas. 2005. The Managerial Consequences of

Earnings Restatements. Working Paper, University of Memphis and University of Texas at San Antonio.

Denis, D. J., Denis, D. K., and Sarin, A. 1997. Ownership Structure and Top Executive

Turnover. Journal of Financial Economics, 45: 193-221. Desai, H., C. E. Hogan, and M. S. Wilkins. 2006. The Reputational Penalty for Aggressive

Accounting: Earnings Restatements and Management Turnover. Accounting Review, 81(1): 83-112.

Gilson, S. C. 1989. Management Turnover and Financial Distress. Journal of Financial

Economics, 25(2):241-262. Glass, Lewis, & Co., LLC. 2005. Restatements—Traversing Shaky Ground: An Analysis for

Investors. Trend Alert (May 31) available September 15, 2005 from Meeting of SEC Advisory Committee on Smaller Public Companies: http://www.sec.gov/rules/other/265-23/glasslewis091405.pdf.

Jayaraman, N., C. Mulford, and L. Wedge. 2004. Accounting Fraud and Management Turnover.

Working Paper, Georgia Institute of Technology. Kaplan, S. N. and B. A. Minton. 2006. How was CEO Turnover Changed? Increasingly

Performance Sensitive Boards and Increasingly Uneasy CEOs. Working Paper, NBER. Land, J. K. 2006. CEO Turnover Following Earnings Restatements. Working Paper, North

Carolina Central University.

27

Manski, C. F. and S. R. Lerman. 1977. The Estimation of Choice Probabilities from Choice Based Samples. Econometrica, 45(8): 1977-1988.

Mian, S. 2001. On the Choice and Replacement of Chief Financial Officers. Journal of Financial

Economics, 60(1): 143-175. Palmrose, Z., V. J. Richardson, and S. Scholz. 2004. Determinants of Market Reactions to

Restatement Announcements. Journal of Accounting and Economics, 37(1): 59-89. Palmrose, Z. and S. Scholz. 2004. The Circumstances and Legal Consequences of Non-GAAP

Reporting: Evidence from Restatements. Contemporary Accounting Research, 21(1): 139-180.

U. S. General Accounting Office (GAO). 2002. Financial Statement Restatements: Trends,

Market Impacts, Regulatory Responses, and Remaining Challenges. GAO-03-138. Washington, D. C.: General Accounting Office.

U. S. General Accounting Office (GAO). 2003. Financial Statement Restatements Database.

GAO-03-395R. Washington, D. C.: General Accounting Office. U. S. General Accounting Office (GAO). 2006. Financial Statement Restatements: Updates of

Public Company Trends, Market Impacts, and Regulatory Enforcement Actions. GAO-06-678. Washington, D. C.: General Accounting Office.

Warner, J., R. Watts, and K. Wruck. 1988. Stock Price Drops and Management Changes,

Journal of Financial Economics 20: 431-60.

Yermack, D. 2004. Renumeration, Retention, and Reputation Incentives for Outside Directors. Journal of Finance, 59(5): 2281-2308.

28

Table 1 – Sample Selection Unintentional Intentional Total Number of restatements identified from 8-K filings 460 170 630 Financial Services 43 19 62 Firms not on Compustat 56 19 75 Total Available 361 132 493 Less firms not randomly selected 268 268

Observations 93 132 225 Less cases where restatement had no income impact 10 0 9 Less cases where investigation did not lead to restatement or is not complete 0 27 27

Final Sample 83 105 188 Weighting by population proportion 3.9 1.0

Weighted Sample 324 105 439 Notes: Our sample was obtained as follows. We identified all 8-K disclosures where firms disclosed a restatement or intended restatement due to an error in previously reported financial statements between January 1, 2002 and December 31, 2005. This includes 8-Ks filed specifically to announce a restatement, to announce a change in auditor, or to announce quarterly or annual financial statements. If a firm announced more than one restatement during this time-period, we selected the first restatement disclosed. This process yielded 630 restatements. Of these 460 are restatements arising from unintentional GAAP violations and 170 are from intentional violations. We classify violations as intentional if announced either an SEC or internal investigation into the accounting misstatement or referred to the misstatement as an irregularity. Of these we exclude firms in the financial services industry (SIC codes 6000-6999) as well is firms that do not have data available on Compustat. This leaves us with 361 unintentional and 132 intentional cases. To reduce the cost of data collection, we randomly select approximately 1/4 of the restatements caused by unintentional violations. We include all intentional GAAP violations to maintain power in our tests for this group. From this we drop 10 observations from the unintentional group where the restatement had no income impact. These generally related to reclassifications or corrections to shares outstanding used in the computation of EPS. Finally, we eliminate 27 intentional GAAP violation cases where the investigation ultimately did not lead to a restatement or where the investigation is underway but a restatement has not yet been announced.

29

Tabl

e 2

– C

lass

Act

ion

Law

suits

and

Inte

nt C

lass

ifica

tions

Pa

nel A

Rea

son

for

clas

sific

atio

n as

Inte

ntio

nal

Ir

regu

lari

ty

SEC

/DO

J In

vest

igat

ion

Inte

rnal

In

vest

igat

ion

Tota

l In

tent

iona

l To

tal

Uni

nten

tiona

l C

lass

Act

ion

Law

suit

36

64

60

84

1 N

o C

lass

Act

ion

Law

suit

7

9 16

21

82

To

tal

43

73

76

105

83

Firm

s ar

e cl

assi

fied

into

the

int

entio

nal s

ampl

e if

the

repo

rt an

irre

gula

rity,

an

SEC

or

Dep

artm

ent o

f Ju

stic

e in

vest

igat

ion,

or

an

inte

rnal

inve

stig

atio

n. T

he in

tent

iona

l typ

e co

lum

ns d

o no

t sum

up

“Tot

al I

nten

tiona

l” b

ecau

se s

ome

rest

atem

ents

hav

e m

ore

than

on

e ty

pe o

f rel

ated

inve

stig

atio

n (e

.g.,

an ir

regu

larit

y an

d an

SEC

inve

stig

atio

n an

d an

inte

rnal

inve

stig

atio

n).

Pane

l B- C

lass

ified

as I

nten

tiona

l but

No

Cla

ss A

ctio

n Su

it

Cou

nt

Firm

s not

bei

ng tr

aded

at t

ime

of a

nnou

ncem

ent

3 Fi

rms w

ith m

arke

t val

ue b

elow

$25

mill

ion

8 R

esta

tem

ent w

as a

t the

subs

idia

ry-le

vel

7 Fi

rms w

ith p

ositi

ve sh

ort-w

indo

w re

turn

s 3

Tota

l 21

Le

ss m

ultip

le o

ccur

renc

es

(4)

Firm

s with

at l

east

one

of t

he a

bove

17

M

isst

atem

ents

cod

ed in

tent

iona

l for

pre

senc

e of

SEC

inve

stig

atio

n, b

ut n

o cl

ass a

ctio

n fo

und

4 To

tal

21

30

Tabl

e 3

- Des

crip

tive

Stat

istic

s

U

nint

entio

nal

Inte

ntio

nal

R

ando

m S

ampl

e

N

Mea

n M

edia

n St

d N

M

ean

Med

ian

Std

N

Mea

n M

edia

n St

d Sa

les

80

2,72

5 53

3 6,

356

94

3,98

4 49

7 8,

228

179

2,25

2 16

3 6,

575

Sale

s Gro

wth

80

14

.7%

8.

3%

42.4

%

94

24.0

%

4.7%

10

2.1%

17

2 21

.6%

7.

2%

97.8

%

Tota

l Ass

ets

82

2,78

8 40

5 6,

857

104

4,77

5 56

9 9,

683

180

2,36

7 15

2 7,

252

Mar

ket V

alue

73

2,

247

387

5,54

4 93

3,

212

436

7,25

0 16

1 2,

631

151

7,56

3 R

OA

82

3.

3%

5.6%

13

.6%

10

4 -1

.8%

3.

0%

17.6

%

176

-12.

1%

5.0%

63

.9%

In

com

e/A

sset

s 80

-2

.3%

2.

3%

16.5

%

94

-14.

2%

-3.6

%

34.7

%

173

-22.

7%

2.2%

87

.3%

Le

vera

ge

82

0.25

0.

17

0.27

10

3 0.

29

0.25

0.

24

178

0.34

0.

19

0.53

C

EO S

tock

Ow

ners

hip

83

9 3

15

104

7 2

15

N/A

N

/A

N/A

N

/A

Not

es: D

etai

ls o

f the

sam

ple

sele

ctio

n pr

oced

ure

for r

esta

tem

ent f

irms

are

prov

ided

in T

able

1. I

n ad

ditio

n to

the

rest

atem

ent s

ampl

e, w

e al

so

incl

ude

a ra

ndom

sam

ple

for

com

paris

on p

urpo

ses.

Thi

s ra

ndom

sam

ple

is s

tratif

ied

by in

dust

ry a

nd y

ear

to m

irror

the

repr

esen

tatio

n of

the

rest

atem

ent f

irms.

Var

iabl

es li

sted

abo

ve a

re th

ose

repo

rted

in th

e ye

ar p

rior

to th

e re

stat

emen

t (no

t res

tate

d). S

ales

is C

ompu

stat

#12

. Sal

es

grow

th is

the

chan

ge in

sal

es fr

om t-

2 to

t-1

scal

ed b

y sa

les

in t-

2. T

otal

ass

ets

is C

ompu

stat

#6.

Mar

ket V

alue

is C

ompu

stat

#25

*#19

9. R

OA

is

ope

ratin

g in

com

e af

ter

depr

ecia

tion

scal

ed b

y A

sset

s (C

ompu

stat

#17

8/#6

). In

com

e/A

sset

s is

Com

pust

at #

172/

#6.

Leve

rage

is

Deb

t (#

9+#3

4)/A

sset

s (#

6). C

EO_E

quity

is th

e fr

actio

n of

CEO

ow

ners

hip

in th

e ye

ar o

f th

e re

stat

emen

t. In

cas

es w

here

we

are

unab

le to

obt

ain

owne

rshi

p in

form

atio

n in

the

year

of t

he re

stat

emen

t, w

e us

e th

e pr

ior y

ear.

We

do n

ot c

olle

ct o

wne

rshi

p in

form

atio

n fo

r the

rand

om s

ampl

e.

The

amou

nts

repo

rted

abov

e do

not

alw

ays

capt

ure

the

rest

ated

am

ount

s. W

heth

er o

r no

t th

e C

ompu

stat

val

ues

are

the

initi

ally

rep

orte

d am

ount

s or

the

rest

ated

am

ount

s de

pend

s on

the

timin

g of

the

rest

atem

ents

. If,

for e

xam

ple,

a fi

rm w

ith fi

scal

-yea

r end

of D

ecem

ber 3

1, 2

005

files

a 1

0-K

in M

arch

200

6 bu

t lat

er a

men

ds th

at fi

ling

prio

r to

Com

pust

at’s

nex

t “cu

t” o

f the

dat

abas

e, s

ay N

ovem

ber 2

006,

then

Com

pust

at

uses

the

Nov

embe

r 20

06 d

ata

and

igno

res

the

orig

inal

fili

ng (

in M

arch

200

6). I

n th

ese

case

s, th

e da

ta r

epor

ted

in th

is ta

ble

are

the

rest

ated

fig

ures

. If,

on th

e ot

her h

and,

a c

ompa

ny a

men

ds a

prio

r yea

r afte

r Com

pust

at’s

nex

t “cu

t” o

f the

dat

a, th

e re

stat

ed in

form

atio

n w

ill a

ppea

r in

Com

pust

at’s

spec

ial r

esta

tem

ent v

aria

bles

. In

thes

e ca

ses,

our

des

crip

tive

stat

istic

s will

not

incl

ude

the

rest

ated

am

ount

s.

31

Ta

ble

4 –

Turn

over

Fre

quen

cy

Uni

nten

tiona

l

I

nten

tiona

l To

tal

Turn

over

%

Tu

rnov

er %

Turn

over

%

N

C

EO

CFO

CEO

or

C

FO

N

CEO

C

FO

CEO

or

C

FO

N

CEO

C

FO

CEO

or

C

FO

Tota

l 32

4 8.

4 12

.0

15.7

10

5 48

.6

63.8

73

.3

429

18.3

24

.7

29.8

Su

bsid

iary

vs.

Pare

nt L

evel

S

ubsi

diar

y-Le

vel

43

9.1

9.1

9.1

24

29.2

41

.7

50.0

67

16

.3

20.8

23

.8

Par

ent-L

evel

28

1 8.

3 12

.5

16.7

81

54

.3

70.4

80

.2

362

18.6

25

.5

30.9

A

nnua

l vs.

Qua

rter

ly

Ann

ual

281

8.3

11.1

15

.3

91

49.5

67

.0

74.7

37

2 18

.4

24.8

29

.8

Qua

rterly

43

9.

1 18

.2

18.2

14

42

.9

42.9

64

.3

57

17.4

24

.3

29.5

Not

es:

This

tab

le s

umm

ariz

es t

he t

urno

ver

rate

s fo

r in

tent

iona

l an

d un

inte

ntio

nal

GA

AP

viol

atio

n fir

ms

parti

tione

d on

sub

sidi

ary-

leve

l/par

ent-l

evel

and

Ann

ual/Q

uarte

rly.

Sam

ple

sele

ctio

n an

d de

term

inat

ion

of i

nten

tiona

l ve

rsus

uni

nten

tiona

l cl

assi

ficat

ions

are

de

scrib

ed in

Tab

le 1

. A s

ubsi

diar

y-le

vel r

esta

tem

ent i

s de

fined

as

a re

stat

emen

t tha

t occ

urre

d in

a s

ubsi

diar

y. A

ll ot

her r

esta

tem

ents

are

co

nsid

ered

to b

e pa

rent

-leve

l res

tate

men

ts. A

nnua

l res

tate

men

ts a

re th

ose

that

requ

ired

rest

atem

ent o

f a 1

0-K

filin

g. Q

uarte

rly fi

lings

are

th

ose

that

onl

y af

fect

ed 1

0-Q

filin

gs. T

urno

ver i

s co

nsid

ered

to h

ave

occu

rred

if a

CEO

or C

FO le

ft th

e fir

m in

the

six

mon

ths

befo

re to

si

x m

onth

s af

ter t

he re

stat

emen

t or i

nves

tigat

ion

anno

unce

men

t, w

hich

ever

is fi

rst.

The

freq

uenc

y co

unts

for t

he u

nint

entio

nal s

tate

men

ts

are

wei

ghte

d by

pop

ulat

ion

prop

ortio

n fo

r co

mpa

rison

to

the

inte

ntio

nal

rest

atem

ent

sam

ple.

As

disc

usse

d in

Tab

le 1

, we

sele

cted

a

rand

om s

ampl

e co

nsis

ting

of a

ppro

xim

atel

y 1/

4 of

the

uni

nten

tiona

l ca

ses

that

we

iden

tifie

d bu

t an

alyz

ed a

ll in

tent

iona

l ca

ses.

To

appr

oxim

ate

the

tota

l tur

nove

r w

e w

ould

exp

ect t

o ob

serv

e in

the

popu

latio

n of

all

rest

atem

ents

, we

assi

gn 3

.9 ti

mes

the

wei

ght t

o th

e un

inte

ntio

nal c

ases

.

32

Table 5 – Logistic Regression - Turnover TURNOVER=β0+β1INTENTIONAL+β2SUBSIDIARY+β3INTENTIONAL*SUBSIDIARY +β4ANNUAL+β5CEO_EQUITY +β6LEVERAGE +β7ROA+β8CAR(t-90-t-8)+β9Size Quintile 1 +β10Size Quintile 2+ β11Size Quintile 4+ β12Size Quintile 5+ε

Pred. Sign CEO Turnover CFO Turnover

CEO or CFO Turnover

Intercept -2.22 *** -1.68 * -1.05 (7.22 ) (4.26 ) (1.96 ) INTENTIONAL + 2.35 *** 3.22 *** 3.14 *** (19.09 ) (26.97 ) (29.03 ) SUBSIDIARY ? -0.16 -0.09 -0.55 (0.03 ) (0.01 ) (0.36 ) INTENTIONAL*SUBSIDIARY - -1.16 -1.69 * -1.32 (0.89 ) (1.79 ) (1.17 ) ANNUAL + 0.42 0.41 0.16 (0.39 ) (0.32 ) (0.05 ) CEO_EQUITY - -0.05 ** -0.03 * -0.04 ** (3.43 ) (2.24 ) (4.22 ) LEVERAGE + -0.47 0.75 0.45 (0.25 ) (0.86 ) (0.34 ) ROA - -0.05 -2.05 -0.24 (0.00 ) (1.28 ) (0.02 ) CAR(t-90-t-8) - -0.38 -0.39 -0.78 (0.20 ) (0.21 ) (0.82 ) Size Quintile 1 ? 0.65 -0.15 -0.03 (0.30 ) (0.02 ) (0.00 ) Size Quintile 2 ? 0.69 -0.05 0.07 (0.79 ) (0.01 ) (0.01 ) Size Quintile 4 ? 0.15 -1.11 ** -0.91 * (0.05 ) (2.72 ) (2.13 ) Size Quintile 5 ? -0.27 -2.04 *** -1.42 ** (0.17 ) (7.76 ) (4.85 ) Test of b2+b3<0 - -1.32 ** -1.79 ** -1.87 *** (2.72 ) (4.80 ) (5.57 ) Psuedo-R2 31.13% 45.34% 44.95% Log Likelihood 37.17 61.59 64.13 N 164 164 164

Notes: Logistic regressions of variations of model (1) are reported above. Chi-square statistics are in parentheses. Psuedo-R2 is the Nagelerke Psuedo-R2. Sample selection information is detailed in Table 1. We lose 24 observations due to certain missing data on Compustat, leaving 164 observations. To approximate the total turnover we would expect to observe in the population of all restatements, we estimate the Logistic regression assigning approximately 3.9 times the weight to the low severity cases and then normalizing the sample size. The dependent variable, TURNOVER, is 1 if the executive left the firm in the 6 months before or 6 months after the restatement or investigation announcement, whichever is first. INTENTIONAL is 1 if the restatement announcement discloses either an internal board or SEC investigation and 0 otherwise. SUBSIDIARY is 1 if the restatement occurred in a subsidiary and 0 otherwise. ANNUAL is 1 if the firm restated a 10-K and 0 if the firm restated only 10-Qs. CEO_EQUITY is the fraction of CEO ownership in the year of the restatement. Leverage is Debt (#9+#34)/Assets (#6). ROA is operating income before interest and taxes scaled by Assets (Compustat #178/#6). CAR(t-90-t-8) is the firm’s cumulative abnormal returns from 90 trading days prior to the restatement announcement to 8 trading days prior, and expected returns are the CRSP value-weighted returns inclusive of dividends. ***, **, * represent p-values at the 1%, 5%, and 10%, respectively (p-values are reported as one-tailed when the sign is predicted).

33

Table 6 – Logistic Regression –Turnover – Short-Window CARs as proxy for Severity

TURNOVER=β0+β1CAR(t-7-t+7)+β2SUBSIDIARY+β3CAR(t-7-t+7)*SUBSIDIARY+β4ANNUAL+β5CEO_EQUITY +β6LEVERAGE +β7ROA+β8CAR(t-90-t-8)+β9Size Quintile 1+β10Size Quintile 2+ β11Size Quintile 4+ β12Size Quintile 5+ε

Pred. Sign CEO Turnover CFO Turnover

CEO or CFO Turnover

Intercept -1.50 * -0.90 -0.51 (3.84 ) (1.60 ) (0.58 ) CAR(t-7-t+7) + -3.69 ** -4.50 *** -3.46 **

(4.55 ) (6.95 ) (4.77 ) SUBSIDIARY ? -0.86 -1.19 * -1.49 **

(1.07 ) (1.79 ) (3.05 ) CAR(t-7-t+7)*SUBSIDIARY - -0.13 -1.61 -3.28 (0.00 ) (0.17 ) (0.71 ) ANNUAL + 0.29 0.07 -0.02 (0.20 ) (0.01 ) (0.00 ) CEO_EQUITY - -0.08 ** -0.02 -0.03 *

(3.56 ) (1.25 ) (2.31 ) LEVERAGE + -0.27 0.86 0.70 (0.08 ) (1.22 ) (0.92 ) ROA - -0.62 -2.66 * -1.10 (0.13 ) (2.39 ) (0.46 ) CAR(t-90-t-8) - -1.72 ** -1.72 ** -2.21 ***

(3.81 ) (4.40 ) (7.44 ) Size Quintile 1 ? 0.93 -0.03 0.15 (0.67 ) (0.00 ) (0.02 ) Size Quintile 2 ? 0.00 -0.70 -0.62 (0.00 ) (1.04 ) (0.89 ) Size Quintile 4 ? 0.35 -0.37 -0.26 (0.34 ) (0.43 ) (0.25 ) Size Quintile 5 ? 0.12 -0.86 * -0.45 (0.04 ) (2.25 ) (0.76 )

Test of β2+β3<0 - -0.99 -2.80 -4.77 (0.07 ) (0.37 ) (1.10 ) Psuedo-R2 19.77% 27.41% 24.79% Log Likelihood 21.93 33.54 31.53 N 161 161 161

Notes: Logistic regressions of variations of model (1) are reported above with short-term CARs to proxy for severity. Chi-square statistics are in parentheses. Psuedo-R2 is the Nagelerke Psuedo-R2. Sample selection information is detailed in Table 1. We lose 27 observations due to certain missing data on Compustat or CRSP, leaving 161 observations. To approximate the total turnover we would expect to observe in the population of all restatements, we estimate the Logistic regression assigning approximately 3.9 times the weight to the low severity cases and then normalizing the sample size. The dependent variable, TURNOVER, is 1 if the executive left the firm in the 6 months before or 6 months after the restatement or investigation announcement, whichever is first. CAR(t-7-t+7) is the firm’s cumulative abnormal returns from 7 trading days prior to the restatement announcement through 7 trading days after the announcement. SUBSIDIARY is 1 if the restatement occurred in a subsidiary and 0 otherwise. ANNUAL is 1 if the firm restated a 10-K and 0 if the firm restated only 10-Qs. CEO_EQUITY is the fraction of CEO ownership in the year of the restatement. LEVERAGE is Debt (#9+#34)/Assets (#6). ROA is operating income before interest and taxes scaled by Assets (Compustat #178/#6). CAR(t-90-t-8) is the firm’s cumulative abnormal returns from 90 trading days prior to the restatement announcement to 8 trading days prior. Expected returns are the CRSP value-weighted returns inclusive of dividends. ***, **, * represent p-values at the 1%, 5%, and 10%, respectively (p-values are reported as one-tailed when the sign is predicted).

34

Table 7 – Analysis of Cases where neither the CEO nor the CFO Exits within the 13-month Turnover Window around the Restatement.

Count Total number of observations classified as intentional and parent-level where neither the CEO nor the CFO left within the 13-month window (-6 to +6 months).

16

Either the CEO or CFO left the firm between 24 and 7 months before the restatement announcement. These are cases where the incoming officers likely attributed the misstatement to their predecessors.

8

Either the CEO or CFO left the firm at a date later than 6 months after the restatement but around the time the investigation is concluded.

3

The investigation (independent internal or external) specifically determined that the misstatement was not intentional.

4

Total 15 Case of apparent fraud where there was no turnover (Asconi) 1

Summary of Asconi Case: The Company (Asconi) issued ten million shares during 2003 to the company’s CEO and CFO, who together hold over 90% of the company’s stock. The issuances were originally treated as if they were equity transactions. SEC conducted formal investigation and determined that the market value of the ten million shares of Common Stock issued during 2003 should have been charged against the Company’s income statement as compensation. The restatement resulted in the recording of $40.4 million of expense. On 03/19/05 the SEC issued Wells notice recommending that a civil or administrative enforcement action be brought against the company and Alex Brinister, Asconi’s Vice President for U.S. Operations Interim CAO. Brinister resigned 01/12/06. (Investigation unresolved)

35

Tabl

e 8

– Lo

gist

ic R

egre

ssio

n- P

RE

Ver

sus P

OST

Ana

lysis

TU

RN

OV

ER=β

0+β 1

INTE

NTI

ON

AL+β 2

POST

+β3S

UB

SID

IAR

Y +β 4

AN

NU

AL+β 5

CEO

_EQ

UIT

Y+β

6LEV

ERA

GE

+β7R

OA

+β8C

AR

(t-90

-t-8)

+β9S

ize

Qui

ntile

1

+β 1

0Siz

e Q

uint

ile 2

+ β 1

1Siz

e Q

uint

ile 4

+ β 1

2Siz

e Q

uint

ile 5

+ε

CE

O T

urno

ver

CFO

Tur

nove

r C

EO

or C

FO T

urno

ver

Pr

ed. S

ign

(1)

(2)

(3)

(4)

(5)

(6)

Inte

rcep

t

-1.1

8 * -1

.89 **

* -0

.25

-0.9

9 * 0.

29

-0.3

4

(5

.33 )

(1

1.89

) (0

.30 )

(3

.76 )

(0

.39 )

(0

.43 )

IN

TEN

TIO

NA

L +

1.

75 **

*

2.15

***

2.

39 **

*

(28.

87 )

(4

2.78

)

(47.

71 )

POST

?

-0.6

5 **

-0.3

6 -0

.62 **

-0

.34

-0

.92 **

* -0

.71 **

(3.5

9 )

(0.9

7 )

(3.8

0 )

(0.9

5 )

(8.0

6 )

(3.7

0 )

SUB

SID

IAR

Y

- -0

.23

-0.6

7 * -0

.40

-1

.01 **

-0

.56 *

-1.3

2 ***

(0.3

0 )

(2.3

0 )

(1.0

9 )

(5.1

5 )

(2.1

5 )

(8.3

2 )

AN

NU

AL

+ 0.

68 **

0.

47

0.23

0.

05

0.38

0.

22

(3.4

9 )

(1.4

7 )

(0.4

6 )

(0.0

2 )

(1.2

6 )

(0.3

3 )

CEO

_EQ

UIT

Y

- -0

.06 **

* -0

.05 **

* -0

.01 *

-0.0

2 * -0

.02 **

-0

.03 **

*

(8

.23 )

(8

.00 )

(2

.01 )

(2

.65 )

(4

.75 )

(6

.53 )

LE

VER

AG

E +

0.38

0.

33

0.98

**

1.09

**

0.88

**

0.94

*

(0

.44 )

(0

.28 )

(3

.52 )

(3

.68 )

(2

.73 )

(2

.62 )

R

OA

-

-1.5

3 **

-0.8

8 -1

.07 *

-0.2

8 -1

.96 **

-1

.02

(3.9

1 )

(1.2

2 )

(2.1

4 )

(0.1

3 )

(5.3

2 )

(1.4

8 )

CA

R(t-

90-t-

8)

- -2

.01 **

* -1

.18 **

-1

.48 **

* -0

.69 *

-2.0

1 ***

-1.2

2 **

(12.

61 )

(4.1

1 )

(9.2

9 )

(1.7

6 )

(15.

42 )

(4.8

2 )

Psue

do-R

2

22.2

8%

34.2

1%

19.6

1%

37.6

5%

27.2

3%

46.1

4%

Log

Like

lihoo

d

50.2

1 81

.00

46.0

9 95

.85

68.0

2 12

6.35

N

300

30

0

300

30

0

300

30

0

Not

es: L

ogis

tic r

egre

ssio

ns o

f va

riatio

ns o

f m

odel

(1)

are

rep

orte

d ab

ove.

Chi

-squ

are

stat

istic

s ar

e in

par

enth

eses

. Psu

edo-

R2 i

s th

e N

agel

erke

Psu

edo-

R2 . S

ampl

e se

lect

ion

info

rmat

ion

is de

taile

d in

Tab

le 1

. We

incl

ude

the

136

firm

s fro

m t

he p

re-p

erio

d an

d th

e 16

4 fir

ms

from

the

pos

t-per

iod

that

are

not

mis

sing

dat

a. A

s pr

evio

usly

disc

usse

d, in

the

post-

perio

d w

e as

sign

appr

oxim

atel

y 3.

9 tim

es th

e w

eigh

t to

the

low

sev

erity

cas

es a

nd th

en n

orm

aliz

ing

the

sam

ple

size

. The

dep

ende

nt

varia

ble,

TU

RN

OV

ER, i

s 1

if th

e ex

ecut

ive

left

the

firm

in th

e 6

mon

ths

befo

re o

r 6

mon

ths

afte

r th

e re

stat

emen

t or

inve

stiga

tion

anno

unce

men

t, w

hich

ever

is f

irst.

INTE

NTI

ON

AL

is 1

if th

e re

stat

emen

t ann

ounc

emen

t dis

clos

es e

ither

an

inte

rnal

boa

rd o

r SEC

inve

stig

atio

n an

d 0

othe

rwis

e. P

OST

is 1

if th

e re

stat

emen

t occ

urre

d in

19

97 o

r 199

8 an

d 0

if th

e re

stat

emen

t occ

urre

d be

twee

n 20

02 a

nd 2

005.

SU

BSI

DIA

RY

is 1

if th

e re

stat

emen

t occ

urre

d in

a su

bsid

iary

and

0 o

ther

wis

e. A

NN

UA

L is

1

if th

e fir

m re

stat

ed a

10-

K a

nd 0

if th

e fir

m re

stat

ed o

nly

10-Q

s. C

EO_E

QU

ITY

is th

e fr

actio

n of

CEO

ow

ners

hip

in th

e ye

ar o

f the

rest

atem

ent.

In c

ases

whe

re w

e ar

e un

able

to o

btai

n ow

ners

hip

info

rmat

ion

in th

e ye

ar o

f th

e re

stat

emen

t, w

e us

e th

e pr

ior

year

. LEV

ERA

GE

is D

ebt (

#9+#

34)/A

sset

s (#

6). R

OA

is o

pera

ting

inco

me

befo

re in

tere

st a

nd ta

xes

scal

ed b

y A

sset

s (C

ompu

stat

#17

8/#6

). C

AR

(t-90

-t-8)

is th

e fir

m’s

cum

ulat

ive

abno

rmal

ret

urns

fro

m 9

0 tra

ding

day

s pr

ior

to th

e re

stat

emen

t an

noun

cem

ent t

o 8

tradi

ng d

ays

prio

r, an

d ex

pect

ed r

etur

ns a

re t

he C

RSP

val

ue-w

eigh

ted

retu

rns

incl

usiv

e of

div

iden

ds. S

ize

quin

tile

cont

rols

are

incl

uded

in th

e re

gres

sion

but

not

repo

rted

in th

e ta

bles

(for

bre

vity

). *

**, *

*, *

repr

esen

t p-v

alue

s at

the

1%, 5

%, a

nd 1

0%, r

espe

ctiv

ely

(p-v

alue

s ar

e re

porte

d as

one

-taile

d w

hen

the

sign

is p

redi

cted

).

36

Table 9 – Logistic Regression PRE Versus POST, INTENTIONAL Only TURNOVER=β0+β1POST+β2SUBSIDIARY+β3ANNUAL+β4CEO_EQUITY +β5LEVERAGE +β6ROA+β7CAR(t-90-t-8) +β8Size Quintile 1+β9Size Quintile 2+ β10Size Quintile 4+ β11Size Quintile 5+ε

Pred. Sign

CEO Turnover CFO Turnover

CEO or CFO Turnover

Intercept -0.46 0.41 1.09 (0.63 ) (0.51 ) (2.38 ) POST ? -0.17 0.15 -0.18 (0.18 ) (0.14 ) (0.13 ) SUBSIDIARY - -0.97 ** -1.36 *** -1.80 ***

(4.52 ) (8.98 ) (12.59 ) ANNUAL + 0.65 * 0.64 * 0.79 * (2.17 ) (2.20 ) (2.16 ) CEO_EQUITY - -0.05 *** -0.01 -0.03 *** (7.25 ) (0.46 ) (6.61 ) LEVERAGE + 0.94 1.45 * 3.15 *** (1.11 ) (2.63 ) (6.28 ) ROA - -1.46 * -0.84 -3.27 ** (2.19 ) (0.84 ) (4.78 ) CAR(t-90-t-8) - -1.37 ** -0.36 -1.73 ** (4.61 ) (0.37 ) (5.18 ) Size Quintile 1 ? -0.95 -1.03 * -1.82 ** (1.48 ) (1.98 ) (3.81 ) Size Quintile 2 ? 1.12 ** 0.47 0.84 (3.26 ) (0.55 ) (0.96 ) Size Quintile 4 ? 0.48 -0.88 * -0.88 * (0.82 ) (2.64 ) (1.73 ) Size Quintile 5 ? -0.50 -1.30 *** -1.76 *** (0.91 ) (5.49 ) (7.01 ) Psuedo-R2 27.84% 18.86% 37.00% Log Likelihood 39.10 24.95 48.77 N 167 167 167

Notes: Logistic regressions of variations of model (1) are reported above. We exclude all unintentional misstatement and include the 62 firms from the pre-period and the 105 firms from the post-period that are not missing data. Chi-square statistics are in parentheses. Psuedo-R2 is the Nagelerke Psuedo-R2. Sample selection information is detailed in Table 1. The dependent variable, TURNOVER, is 1 if the executive left the firm in the 6 months before or 6 months after the restatement or investigation announcement, whichever is first. CAR(t-7-t+7) is the firm’s cumulative abnormal returns from 7 trading days prior to the restatement announcement through 7 trading days after the announcement, and expected returns are the CRSP value-weighted returns inclusive of dividends. SUBSIDIARY is 1 if the restatement occurred in a subsidiary and 0 otherwise. POST is 1 if the restatement occurred in 1997 or 1998 and 0 if the restatement occurred between 2002 and 2005. ANNUAL is 1 if the firm restated a 10-K and 0 if the firm restated only 10-Qs. CEO_EQUITY is the fraction of CEO ownership in the year of the restatement. LEVERAGE is Debt (#9+#34)/Assets (#6). ROA is operating income before interest and taxes scaled by Assets (Compustat #178/#6). CAR(t-90-t-8) is the firm’s cumulative abnormal returns from 90 trading days prior to the restatement announcement to 8 trading days prior, and expected returns are the CRSP value-weighted returns inclusive of dividends. ***, **, * represent p-values at the 1%, 5%, and 10%, respectively (p-values are reported as one-tailed when the sign is predicted).

37

Figure 1: Percentage of Firms with CEO Turnover over Time Relative to Restatement Announcement Date

Figure 2: Percentage of Firms with CFO Turnover over Time Relative to Restatement Announcement Date

Notes: Figure 1 reports the cumulative CEO turnover percentage from month -6 to month +24, relative to the date of the restatement announcement. The announcement date is the first date the firm announces that it will restate earnings, though it is possible that the firm previously announced an investigation that could potentially lead to an investigation. A firm is grouped as an intentional GAAP violator if the restatement announcement discloses either an internal board or SEC investigation and 0 otherwise. Sample details are reported in Table 1. Figure 2 replicates Figure 1 with CFO turnover. If a firm has turnover more than once during the 30 month window, we select the turnover that occurred closest to the restatement announcement and do not count other turnovers that occur, which understates the overall turnover rates.

38

Figure 3: Cumulative Abnormal Returns 180 Days Surrounding Restatement Announcement

Figure 4: Cumulative Abnormal Returns 180 Days Surrounding Restatement Announcement partitioned by Severity

-35.0%

-30.0%

-25.0%

-20.0%

-15.0%

-10.0%

-5.0%

0.0%

5.0%

10.0%

Mean -Intentional Median -Intentional Mean-Unintentional Median -Unintentional

Notes: Figure 3 reports the mean and median cumulative abnormal returns for all restating firms beginning 90 trading days prior to the restatement announcement and ending 90 days after the restatements. Expected returns are CRSP value-weighted returns with dividends. To avoid survivor bias, we do not require observations to trade over the entire 180-day window. The number of observations ranges from 152 to 169. Figure 4 reports the mean and median cumulative abnormal returns for intentional and unintentional GAAP violations. The number of observations ranges from 84 to 101 for intentional and 68 to 69 for unintentional restatements.