forecast strategies of earnings manipulators

8/9/2019 Forecast Strategies of Earnings Manipulators

1/55

Forecast Strategies of Earnings Manipulators and Associated Consequences*

Mei FengAssociate Professor of Accounting

Katz Graduate School of Business, University of [email protected]

Weili GeAssociate Professor of Accounting

Michael G. Foster School of Business, University of [email protected]

Chan LiAssistant Professor of Accounting

Katz Graduate School of Business, University of Pittsburgh

[email protected]

Nandu J. NagarajanProfessor of Accounting

Katz Graduate School of Business, University of [email protected]

First version: January 2011Current version: April 2014

Abstract:

This paper investigates how earnings manipulation firms adjust their guidance strategies whilemisstating earnings, and the associated consequences. We find that firms manipulating earningsare significantly more likely to issue earnings guidance than matched control firms. Earningsguidance issued by manipulators delays the detection of misstatements. However, misstatingfirms issuing guidance, when detected, experience larger stock price declines than those notissuing earnings guidance. Misstating firms and their managers also incur additional legal costs,as reflected in the higher settlement amounts paid by these firms and a higher likelihood thatCEOs, responsible for the misstatements, are banned from serving as officers, directors, oraccountants of public companies in the future.

* We thank the Berg Center for Ethics and Leadership at the Katz Graduate School of Business, University ofPittsburgh for research support. We thank Harry Evans, Vicky Hoffman, Valerie Li, Dawn Matsumoto, SarahMcVay, Don Moser, Terry Shevlin, D. Shores, Anup Srivastava, Lloyd Tanlu, Tsahi Versano, Wendy Wilson, YouliZou, workshop participants at Duke University, the University of Connecticut, the University of Pittsburgh, theUniversity of Washington, and the University of Waterloo, conference participants at the 2011 AAA meeting, the2011 Accounting Research Conference in honor of Nick Dopuch, the 2012 FARS meeting, the 2012 HKUSTAccounting Research Symposium, and the 2013 Penn State Conference for their helpful comments.


2/55

1

Forecast Strategies of Earnings Manipulators and Associated Consequences

Oh what a tangled web we weave

When first we practice to deceiveSir Walter Scott (Marmion, 1808)

1. Introduction:

Firms that are involved in material earnings manipulations tend to inflate reported

earnings significantly to hide diminishing performance (Dechow et al., 2011).1 Given that

earnings guidance is an important mechanism used by managers to preemptively disclose

earnings information (Hirst et al., 2008; Beyer et al., 2010), firms that manipulate earnings are

also likely to adjust their earnings guidance strategies.

2

To the extent that the adjusted guidance

strategies are part of the overall manipulation scheme, these strategies could affect the detection

of manipulation, shareholder losses due to the manipulation, and any associated legal penalties

borne by the firms and managers. In this paper, we investigate the characteristics and

consequences of earnings guidance issued by earnings manipulators.

To empirically investigate the earnings guidance strategies of manipulation firms, we

identify a sample of firms that, according to SEC enforcement actions, inflated earnings for

multiple years. Unlike the case of restatements, the SEC enforcement actions focus on the most

egregious cases, where managers are more likely to have conspired and intentionally engaged in

earnings manipulation schemes (Dechow et al., 2011). In addition, it is likely that multi-year

1 We define material earnings manipulations to be earnings management activities that violate Generally Accepted

Accounting Principles (GAAP). In terms of the magnitude of the misstatement amount, for example, Palmrose andScholz (2004) report that firms in their restatement sample inflate earnings by 19%, on average. A recent report bythe Committee of Sponsoring Organizations of the Treadway Commission documents that for firms manipulatingearnings, as identified by the SEC between January 1998 and December 2007, the average cumulative earningsmisstatement is $397.7 million while the average sales revenue is $2,557.3 million based on the last financialstatements issued prior to the beginning of the fraud period. Further, firms in our sample inflate their earnings, onaverage, by 21.6% of sales.2 Beyer et al. (2010) show that about 55% of accounting-based information (measured by stock returns on dayswhen accounting disclosures are made) come from management forecasts, suggesting that management forecasts play an important role as a source of information to investors about forthcoming reported earnings.


3/55

2

earnings manipulators coordinate their accounting policies over longer periods of time, and, thus,

are more likely to also strategically adjust their guidance issuance. Hence, compared with firms

that misstate earnings for a single year, multi-year earnings manipulators constitute a more

powerful setting in which to investigate the management guidance strategies used by

manipulation firms. We compile a matched sample of non-manipulation firms, based on

industry and firm size, and hand collect earnings guidance data for one year before and during

the misstatement period for both manipulation and matched control firms.3

We begin by showing that the earnings guidance issued by manipulators is more

consistent with the reported manipulated earnings than with the restated actual earnings. Further,

investors and analysts, who would be unaware of the earnings manipulations when the guidance

is issued, revise their beliefs according to these inflated earnings forecasts. Specifically, we find

that earnings guidance issued by manipulation firms significantly influences stock prices, analyst

forecast revisions, and analyst forecast dispersion, and the extent of the effect is similar to the

effect of guidance issued by non-manipulation firms. Hence, earnings guidance issued by

earnings manipulators appears to mislead investors and analysts into believing the inflated firm

performance.

We next find that after commencing earnings falsification, manipulation firms are

significantly more likely to issue earnings forecasts than non-manipulation firms. The

percentage of firms that issue earnings forecasts increases by 20.8% (from 35.5% to 56.3%) once

firms commence earnings manipulation, which is about three times the increase for the matched

control firms (7.1%) during the same period. Given that the misstatement decision may be

driven by the systematic differences between manipulation and non-manipulation firms, we also

3 We do not use management guidance data from the First Call database because many manipulation firms in oursample (33%) are not covered by First Call and Chuk et al. (2012) document that First Call exhibits systematic biases in the way it collects management guidance from company press releases.


4/55

3

conduct a difference-in-differences analysis on the likelihood of forecast issuance, which allows

us to better control for firm-specific sticky earnings guidance strategies and time-series earnings

guidance trends. We find that manipulation firms increase earnings guidance issuance during the

misstatement period to a significantly greater extent than the matched control firms during the

same period. Further, our AAER content analyses show that among earnings manipulators, firms

issuing guidance are more likely to be subject to capital market pressure than firms not issuing

guidance. Taken together, the above evidence suggests that firms manipulating earnings,

particularly those under significant capital market pressure, appear to increase their earnings

guidance issuance during the misstatement period.

The increased frequency of issuing earnings guidance reports that are in line with the

misreported earnings can benefit earnings manipulators in two ways. First, more earnings

guidance can help reduce the cost of capital and lift stock prices for manipulation firms (e.g.,

Botosan 1997; Verrecchia 2001; Lambert et al. 2007; Baginski and Rakow 2012). Second,

increased earnings guidance can help reduce perceived information asymmetry and information

uncertainty, leading to a higher reputation for transparent reporting (e.g., Baginski et al. 1993;

Coller and Yohn 1997; Graham et al. 2005). This would, in turn, help to allay any doubts about

the underlying economic reality of the firms in the minds of individuals who could potentially

detect the manipulation, such as investors, analysts and the media, thus avoiding or delaying

detection of the earnings misstatement.4 Our empirical analyses on the consequences of the

earnings guidance strategies corroborate these arguments.5

4 We note that managers will only increase management guidance issuance if the expected benefits outweigh anyexpected additional legal costs associated with the increased earnings guidance. We discuss this further on pages 7-9.5 Alternatively, managers may issue earnings forecasts that are ex post overoptimistic, and then have to manipulateearnings to meet their own forecasts (Kasznik 1999). Forecasting is positively related to earnings manipulation because the need to meet the forecasts leads to earnings manipulation. This argument is consistent with Schrand andZechman (2011)’s conclusion that managers of firms misstating earnings tend to be over-confident about future


5/55

4

Specifically, if management earnings forecasts issued during the manipulation period

alleviate investors’ concerns regarding the information asymmetry and uncertainty associated

with the firm, the revelation of the manipulation should reawaken and reinforce such concerns as

investors would now realize that the actual information asymmetry and uncertainty were not

mitigated during the manipulation period. Hence, compared with manipulators who do not issue

earnings guidance, those who do would be likely to suffer a larger stock price decline following

the public revelation of accounting misstatements, after controlling for the severity of

manipulation. Our empirical results confirm this. Next, because the risk of manipulation may

also influence the management forecasting decision, we examine whether forecasting delays the

detection of manipulation using a simultaneous equations approach. We find that managers are

more likely to issue guidance when the detection risk is high (i.e., the lead time to detection is

shorter). More importantly, we also find that it takes longer to detect accounting manipulations

for firms issuing earnings forecasts than for those not issuing earnings forecasts, consistent with

the conjecture that increased earnings guidance benefits manipulation firms by delaying

detection of the misstatements.

In terms of legal consequences, we find that when the misstatement is revealed, the

settlement amounts following class action and SEC lawsuits are significantly higher for

manipulators issuing earnings guidance than for those not issuing guidance, after controlling for

performance. This alternative story is inconsistent with our empirical findings, as described below. First, we findthat manipulation firms are more likely to issue earnings guidance when the manipulation detection risk is high,

suggesting manipulation and the associated risk influence the guidance decision, not just vice versa. Second, ourdifference-in-differences analysis results show that manipulation firms are more likely to issue earnings guidance inline with misreported earnings, after we exclude the first misstatement year. This finding is inconsistent with thealternative explanation because a rational manager who issues honest, yet overoptimistic, forecasts in the firstmanipulation year and has to manipulate earnings to meet these forecasts, would be unlikely to continue issuingoveroptimistic forecasts in subsequent misstatement years, thereby avoiding the need for further earningsmanipulations. Third, in our robustness checks, we continue to find that manipulation firms increase their earningsguidance during the misstatement years after controlling for CEO overconfidence, where we measure CEOoverconfidence using industry-adjusted in-the-money unexercised exercisable options held by the CEO followingSchrand and Zechman (2012).


6/55

5

the amount and duration of the manipulation, as well as the market reaction to the disclosure of

the misstatement.6 Moreover, CEOs of manipulation firms issuing earnings guidance are more

likely to be banned from serving as officers, directors, or accountants of public companies in the

future. These findings suggest that firms and individual managers bear higher legal costs when

they not only misstate earnings but also issue false guidance reports consistent with the misstated

earnings.

Our study contributes to both the earnings management and voluntary disclosure

literatures. While prior research has extensively documented the characteristics and

consequences of earnings manipulations (e.g., Beneish 1999; Karpoff et al., 2008a, 2008b;

Dechow et al., 2011), only a few studies have investigated other decisions made by earnings

manipulators. These studies (Sadka 2006; McNichols and Stubbern 2008; Kedia and Phillipon

2009) have focused on the operational and investment decisions of firms manipulating earnings.

For example, Kedia and Phillipon (2009) find that in order to reduce the risk of detection,

manipulation firms also distort their hiring and investment decisions to match their inflated

earnings performance. To the best of our knowledge, our study is one of the earliest to

empirically investigate how firms strategically adjust other disclosures, specifically earnings

guidance, when manipulating their mandatory financial reports. Our results suggest that when

firms choose to manipulate their reported earnings, they increase the issuance of earnings

forecasts that are in line with manipulated earnings. Moreover, these forecasts have substantial

6 It is possible that managers become more reluctant to increase management earnings guidance when the legal costfollowing the detection of manipulation is high (i.e., a negative association between management guidance issuanceand legal costs). We do not use simultaneous equations to control for this possibility because this analysis isconducted at the firm-level and there are only 72 observations. However, not controlling for this will bias againstthe finding that manipulation firms issuing earnings guidance bear higher litigation costs (i.e., a positive association between management guidance issuance and legal costs).


7/55

6

consequences, such as misleading investors and analysts, delaying detection of the manipulation

and increasing associated legal penalties for manipulation firms and their managers.

Our paper also contributes to the management forecast literature which has generally

assumed that reported earnings are within GAAP and not significantly misstated. These studies

have found that management earnings forecasts provide useful information to investors (e.g.,

Ajinkya and Gift 1984), reduce information asymmetry and cost of capital (e.g., Coller and Yohn

1997; Baginski and Rakow 2012), decrease earnings management (Jo and Kim 2007), and

increase reporting transparency (Graham et al. 2005). Baginski et al. (2011), a concurrent

working paper that also examines earnings guidance issued by manipulation firms, concludes

that fraud firms continue to engage in quality voluntary disclosures while committing fraud. Our

findings suggest that, instead of providing useful information, management earnings forecasts

issued by manipulation firms actually further mislead information users.7 Rather than being

beneficial to investors, these earnings forecasts are harmful to investors, as evidenced by larger

stock price declines following the public revelation of earnings misstatements and delayed

detection of manipulators issuing earnings guidance.

The remainder of the paper is organized as follows. Section 2 reviews previous research

and develops our hypotheses. Section 3 describes our sample and research design. Section 4

presents empirical results, and Section 5 concludes the paper.

7 Another major difference between our study and Baginski et al. is that we examine the consequences of theforecasting strategies while they do not. In addition, our samples and research designs differ considerably. Wefocus on firms that have misstated at least two years’ financial statements in order to alleviate the concern of reversecausality. We also hand collect management forecast data due to the limitations of the First Call database (Chuk etal., 2012). These differences in samples and research designs most likely explain the differences in our results onmanagement forecast characteristics.


8/55

7

2. Prior research and hypothesis development

2.1 Management earnings forecast issuance of manipulation firms

Prior research has demonstrated that capital market concerns, such as the desire to raise

capital and to reduce the cost of equity, serve as a major source of incentives for managers to

manage earnings (Dechow and Skinner 2000; Dechow et al., 2010). For example, a few studies

have shown that firms engaging in earnings manipulations are also more likely to be raising

capital during the misstatement years (e.g., Dechow et al., 1996; Beneish 1999). Earnings

manipulation helps firms hide diminishing performance thereby avoiding any associated stock

price declines, which would otherwise impede the likelihood of successfully raising external

financing as well as reduce the dollar amount of capital raised (e.g., Dechow et al., 2011).

Consistent with managers’ incentives to maintain high market valuations, Dechow et al. (2011)

find that manipulation firms tend to have higher price-to-fundamental ratios as well as strong

stock price performance prior to earnings manipulation. In sum, firms that manipulate earnings

appear to face high market pressure and respond to the pressure by violating GAAP and

artificially inflating their financial performance.

Issuing earnings guidance consistent with the misreported earnings can help manipulators

address higher capital market pressure. Prior research has documented that management

forecasts are informative to investors and analysts (e.g., Ajinkya and Gift 1984; Jaggi, 1980;

Waymire, 1986; Baginski and Hassell 1990), improve the accuracy of analyst forecasts

(Waymire 1986), reduce the dispersion in analyst forecasts (Baginski et al. 1993; Clement et al.

2003), and reduce bid-ask spreads and the cost of capital (Coller and Yohn 1997; Verrecchia

2001; Lambert et al. 2007; Baginski and Rakow 2012). Reductions in the cost of capital likely

lead to higher stock prices. Similarly, we expect that management forecasts issued by


9/55

8

misstatement firms during the manipulation period would also reduce the cost of capital and

increase stock price as investors and analysts would be unaware of the manipulation when the

forecasts are issued.

Issuing management earnings forecasts can also potentially help firms avoid or delay

detection of earnings manipulations. By providing useful information to investors and reducing

perceived information asymmetry, management forecasts clearly help firms improve their

reputation for reporting transparency. Consistent with this, Graham et al. (2005) report in their

survey that managers’ primary incentive for voluntary disclosures is to develop a reputation for

transparent reporting. Following this logic, if increasing voluntary disclosures, such as earnings

guidance, helps firms establish a reputation for reporting transparency, they are likely to do so to

reduce potential scrutiny for earnings manipulation. The reduced scrutiny can, in turn, decrease

or delay the likelihood of the detection of manipulation, thereby helping to avoid the extremely

negative consequences for both firms and managers when material accounting misstatements are

detected and disclosed (Karpoff et al., 2008a; Karpoff et al., 2008b).8 We note that in order to

reap the above two benefits associated with management forecasts, a manipulation firm needs to

issue forecasts consistent with reported manipulated earnings rather than the true earnings.

Earnings guidance benefits earnings manipulators through reducing the cost of capital

and delaying detection. However, there are also costs associated with issuing false earnings

guidance. In particular, earnings guidance in line with the manipulated earnings may

substantially increase legal costs for firms and managers once the manipulation is detected,

8 The psychology literature suggests another possibility by providing considerable evidence that the mere repetitionof the same or similar but non-identical statements enhances their perceived validity (e.g., Hasher et al.1977; Bacon1979; Schwartz 1982; Gigerenzer 1984; Hawkin and Hoch 1992; Joe, 2003). If manipulation firms issue earningsforecasts in line with their anticipated misstated earnings and then release similar but not necessarily identicalmanipulated earnings, it is possible that the earnings number will look familiar to market participants, potentiallyincreasing its credibility. Thus, the psychology literature’s arguments also support a greater association betweenvoluntary forecasts and earnings manipulation. However, we acknowledge that it is difficult to provide directevidence in support of this psychology-based explanation.


10/55

9

because the guidance serves to compound the attempt to fool the investors. Prior literature (e.g.,

Francis et al. 1994; Skinner 1994; Rogers and Stocken 2005) has shown that managers face high

litigation risk when they fail to release accurate and timely earnings news, and such risks

substantially affect management forecast strategies. Given these costs and benefits, ex ante it is

unclear how earnings manipulation decisions influence the likelihood of issuing earnings

guidance. The foregoing arguments lead to our first hypothesis, stated in alternative form.

H1: Manipulation firms are more likely to issue earnings forecasts during misstatementyears than non-manipulation firms.

2.2. Consequences associated with earnings guidance of manipulation firms

As discussed above, earnings guidance issued by manipulation firms will be misleading

because it would be consistent with misreported earnings. Given that investors are unaware of

this deception, earnings guidance may still reduce the perceived informational asymmetry and

cost of capital when issued. However, once the manipulation is detected and publicly disclosed,

investors would realize that the earnings forecasts actually misled them and that the actual

informational asymmetry was not mitigated. As a result, the investors are likely to reverse the

effect of earnings forecasts issued by manipulators during the manipulation period, and

correspondingly adjust these firms’ cost of capital upward, leading to even lower stock prices

following the revelation of the manipulation. Hence, we expect that the stock market reaction is

more negative for manipulation firms issuing false earnings forecasts while they are misstating

earnings. This leads to our next hypothesis.

H2: Stock market reactions to the revelation of accounting manipulations are morenegative for firms issuing earnings forecasts during the misstatement period, compared tomanipulation firms not issuing forecasts.

Dyck et al., (2010) document that fraud can be detected by a wide range of agents, such

as equity investors, analysts, the media, auditors, and regulators. Many of these agents are


11/55

10

likely to pay attention to or use management forecasts (Hirst et al. 2008). As a result,

management earnings forecasts should be helpful in enhancing the perception of reporting

transparency, thereby serving to allay any third party suspicions about the financial performance

of the firm. This, in turn, will reduce outsiders’ incentives to collect information that could cast

doubts on the underlying performance of the firm and, ultimately, delay detection of the

misstatements. We formally state our hypothesis below:

H3: It takes longer to detect the accounting misstatements of manipulation firms issuingearnings forecasts than for manipulation firms not issuing earnings forecasts.

Prior research has also shown that misstatement firms and managers responsible for the

misstatements bear substantial legal penalties when manipulations are discovered (Karpoff et al.,

2008a; Karpoff et al., 2008b). The legal costs not only include substantial monetary penalties,

but also non-monetary penalties such as employment restrictions or criminal charges. It violates

Section 17(a) of the Securities Act and Section 10(b) of the Exchange Act to compound the

misstatement by also knowingly issuing false guidance that is consistent with the manipulation.

It seems reasonable to expect that the punishment would fit the crime, i.e., violating more

antifraud provisions could lead to more legal penalties being imposed on both firms and

managers. Therefore, we expect manipulation firms and their managers to bear higher legal

costs when misstatements are detected if firms have also issued false earnings guidance.9

H4: Firms and managers issuing earnings forecasts during the misstatement period facehigher legal penalties when the misstatements are detected compared to manipulationfirms not issuing forecasts.

9 We do not examine additional labor market costs due to false earnings guidance because Karpoff et al. (2008b)find that 93 percent of the managers who are involved in their firms misstating earnings lose their jobs followingdiscovery of fraud. Therefore, there is unlikely to be enough cross-sectional variation in turnover rates to test theimpact of falsifying earnings guidance on the turnover rate.


12/55

11

3. Sample selection

Our sample begins with 2,261 Accounting and Auditing Enforcement Releases (AAERs)

issued by the SEC from May 17th

, 1982 through June 10th

, 2005.10

The SEC has issued AAERs

against companies, auditors, and officers for alleged accounting and/or auditing misconduct

since 1982. The misstatements identified in the AAERs typically are egregious cases of material

misstatements, involving GAAP violations that have been detected by the SEC. One limitation

of the AAER misstatement sample is that the SEC is unlikely to have identified all the

manipulations; thus some of our control firms might have "undetected" manipulations. However,

this sample limitation would only affect any inference related to the association between

earnings forecast characteristics and manipulations if the SEC were to systematically pursue

companies with certain forecast characteristics. We are not aware of any evidence supporting

this possibility.11 The literature generally suggests that the SEC relies on a wide range of

information sources (e.g., internal reviews, restatements, anonymous tips, news reports) to

identify firms for review (Dechow et al., 2011; Dyck et al., 2010). Nevertheless, we may be

unable to generalize our conclusions on the forecasting strategies of manipulation firms to

“undetected” manipulators.

Table 1 Panel A reports our sample selection procedure. A total of 415 accounting

manipulation firms referenced in the AAERs have CNUM data. We first exclude 227 firms, for

which accounting manipulations occurred prior to 1994 because management forecast data are

difficult to manually collect before 1994, resulting in 188 manipulation firms. Next, we keep

10 We would like to thank Dechow et al. (2011) for sharing their AAER database. We note that, out of 2,261AAERs, there are many cases unrelated to earnings misstatements as well as cases related to firms without CUSIP(see the detailed description in Dechow et al., 2011). We do not include these cases in our analyses.11 In addition, if our result of increasing forecast issuance is driven by the SEC’s selection bias, we would expect toobserve more timely detection for manipulation firms with earnings guidance. In contrast, we find that it takeslonger to detect the manipulation by firms issuing earnings guidance, suggesting that the SEC’s selection bias isunlikely to explain our results.


13/55

12

firms that misstated at least two years’ financial statements by removing 94 firms that only

manipulate earnings for one year. This helps us identify a powerful sample to investigate the

management guidance strategy of manipulation firms. We then exclude three firms that managed

earnings downward. Finally, we require manipulation firms to have data on fiscal year, total

assets, and SIC code in order to find matched non-manipulation firms. The final accounting

manipulation sample contains 76 firms with 197 corresponding firm-years.12

We next compile a control sample. For each manipulation firm, we identify firms that: (1)

are in the same industry, (2) have total assets in the range of 80 percent to 120 percent of the

manipulation firm’s total assets in the year prior to the first manipulation year, and (3) still exist

during the last misstatement year of the corresponding manipulation firm. We follow Frankel et

al.’s (2002) SIC-based industry classification scheme. If there is more than one matched firm

available, we keep the control firm whose total asset value is closest to that of the sample firm in

the year immediately before the manipulation years.

We hand collect management forecast data to avoid the potential systematic bias

introduced by the coverage of Thomson First Call’s Company Issued Guidance (CIG). Chuk et

al. (2012) find that the CIG database has limited coverage and tends to cover firms that are

performing well. Therefore, CIG likely under-reports management forecast data for

manipulation firms in particular because these firms tend to be poor performers (Dechow et al.,

12 We note that this small sample size enables us to hand collect numerous data such as management forecasts,restatement amount, AAER content, manipulation revelation dates, legal penalty, and CEO compensation. Thehand-collected data helps us measure the variables more accurately and garner some important insights about

management forecasts issued by earnings manipulation firms. Moreover, regarding whether we can generalize ourresults to other manipulation firms, we believe the issue of coordination in management forecast strategies is morecommon than our sample size indicates. Firms are unlikely to always decide to manipulate earnings at the lastminute. If not, firms need to consider and adjust other strategies from the moment they decide to misstate earnings.We also find that our sample firms are similar to other manipulation firms in terms of total assets, leverage, andlikelihood of having losses one year prior to manipulations although our sample has higher return on assets thanother manipulation firms. Finally, our sample size, though small, is comparable to other studies using accountingmanipulation samples. For example, Erickson et al. (2006) analyze executive compensation for 50 accounting fraudfirms and Farber (2005) examines corporate governance for 87 accounting fraud firms. Nevertheless, weacknowledge that our conclusions might not be generalizable to all manipulation firms.


14/55

13

2011). We gather EPS forecasts, including point, range, open-ended and qualitative forecasts,

from the Dow Jones News Wire for both the manipulation and matched samples. We collect

management forecasts for the year prior to and during the misstatement period. The year prior to

the misstatement period serves as an additional benchmark aside from the control firms. Further

sample attribution processes for testing other hypotheses are discussed in related analyses.

[Table 1] 4. Empirical results

4.1 Descriptive statistics of management earnings forecasts issued by manipulation firms

We begin our empirical analysis by comparing the likelihood of issuing annual and

quarterly earnings forecasts for manipulation firm-years versus matched non-manipulation firm-

years. We report annual and quarterly earnings forecasts separately, as well as the combined

forecasts in Table 2 Panel A. Column (1) of Panel A shows that while 56.3 percent of

manipulation firms issue EPS forecasts (FORECAST), only 42.6 percent of non-manipulation

firms do so. The difference is statistically significant at the one-percent level. Conditional on

issuing EPS forecasts, manipulation firms issue forecasts more frequently (ForecastNumber) and

are more likely to accompany the forecasts with breakdown forecasts (BREAKDOWN), such as

cash flow forecasts, revenue forecasts, EBITDA, and operating margins etc. There are no

differences between the two groups for other forecast properties, such as forecast specificity and

news content.13

Columns (2) and (3) report annual and quarterly forecast properties separately. Among all

the forecast issuances, 47% are annual forecasts, and 53% are quarterly forecasts (not tabulated).

Consistent with the results for the overall forecasts, manipulation firms are more likely to issue

13 When we examine the percentage of different types of forecasts based on specificity, we do not find significantdifferences between manipulation firms and control firms either. In particular, the forecasts by manipulation(control) firms are 18.9% (14.3%) qualitative, 10.8% (14.0%) open-ended, 37.6% (39.4%) based on range, and 32.7%(32.3%) based on point.


15/55

14

both annual and quarterly forecasts than non-manipulation firms. The annual EPS forecasts

issued by manipulation firms are also more likely to be accompanies by breakdown forecasts

than the EPS forecasts issued by control firms. Compared with quarterly EPS forecasts issued by

control firms, the corresponding forecasts issued by manipulation firms are more specific, less

timely (i.e., shorter HORIZON for manipulation firms), and closer to reported earnings (i.e.,

smaller ABSERROR for manipulation firms). Across the three columns, there are no significant

differences in NEWS (i.e., the difference between management EPS forecast and prior analysts’

EPS forecast, scaled by stock price at the end of prior year) between the two groups, suggesting

that for each management forecast, manipulating firms are not managing expectations downward

to a greater extent than control firms.

Next, we compare the annual management forecast error based on the misreported

earnings |(misreported EPS – management earnings forecast)/stock price at the end of the

previous year| with the forecast error based on the restated earnings |(restated EPS – management

earnings forecast)/stock price at the end of the previous year| for manipulation firms.14

We

restrict our analysis to manipulation firm-year observations during which at least one annual

point or range EPS forecast was issued (i.e., we exclude open-ended and qualitative annual

forecasts from the sample), which reduces the sample size to 69 firm-year observations. Reading

through the subsequent 10-K filings or amended 10-K filings for the 69 firm-year observations,

we are able to identify restated earnings for 45 firm-year observations.15

14 We focus on annual EPS forecast error because quarterly restated earnings are often not available.15 Because management earnings forecasts and the reported EPS in the CIG data are likely to be pro forma earnings- based and the restated EPS is GAAP-based, we use the following method to back out the restated pro forma EPS.First, we calculate the difference between restated earnings and the original misreported earnings in the 10-K filings(restated earnings – misreported earnings, as disclosed in Compustat). Second, we scale the difference by thenumber of shares used to calculate the EPS to obtain the restatement amount per share. Finally, we subtract therestatement amount per share from the pro forma based reported EPS in the CIG data to obtain the restated proforma EPS. One implicit assumption is that the restatement adjustment only applies to earnings before exclusionsfor pro forma earnings. This assumption is consistent with McVay (2006), which provides evidence consistent with


16/55

15

We expect that forecasted EPS would be more consistent with the original misreported

earnings rather than the restated earnings. As shown in the last three rows in Panel A of Table 2,

the mean absolute value of the forecast error based on misreported EPS is 0.007.16

In contrast,

the mean forecast error based on restated EPS is 0.052, which is more than seven times larger

than the mean values based on misreported EPS. The median test yields a similar result

(untabulated). Overall, as expected, forecasted EPS is much more consistent with misreported

EPS than actual EPS.

To verify whether investors and analysts revise their beliefs significantly following

management forecast issuance by manipulators, we next examine the stock market reaction and

analyst response to their earnings guidance based on the following regression:

DEPENDENT VARIABLE = b0 + b1 NEWS + b2MANIPULATE + b3DOWNWARD+ b4HORIZON+ b5MANIPULATE*NEWS+ b6DOWNWARD*NEWS+ b7HORIZON*NEWS (1)

The results are reported in Panel B of Table 2. The dependent variable is the three-day

abnormal return surrounding forecast announcement in Column (1) and analyst forecast revision

in Column (2) (i.e., (the last analyst consensus forecast following the management forecast

within 5 days – the last analyst consensus forecast before the issuance of management forecast) /

stock price at the end of previous year). The definitions of the independent variables are the

same as in Panel A and are provided in Table 1 Panel B. NEWS is significantly positive in both

regressions, suggesting that management forecasts are informative to investors and analysts. The

interaction term, MANIPULATE*NEWS, is not significant in either regression. Thus, investors

and analysts do not appear to differentiate the management forecasts of misstatement firms from

those of control firms. In untabulated analysis, we also find that both manipulation and control

managers having strong incentives to inflate the core earnings (similar to pro forma earnings) than the bottom-linenet earnings.16 If a firm issues more than one forecast in a year, we take the average of all forecasts issued during that year.


17/55

16

firms experience a significant decrease in analyst forecast dispersion following earnings

guidance issuance, and there is no difference in the decline of analyst forecast dispersion

between manipulation firms and control firms.17

This suggests that similar to forecasts issued by

non-manipulation firms, forecasts issued by manipulation firms significantly reduce perceived

information asymmetry and information uncertainty. Overall, the above evidence suggests that

investors and analysts, unaware of the manipulation and potential bias in the earnings guidance,

are misled by such guidance into believing the inflated firm performance.

Panel C compares firm characteristics (measured one year prior to the manipulations) for

manipulation and non-manipulation firms. Manipulation firms have a higher mean return on

assets (ROA) than control firms, but the median ROAs are similar across the two groups. The

other noticeable statistically significant difference is that manipulation firms seem to be less

likely to belong to a litigious industry than non-manipulation firms. We control for these firm

characteristics in our subsequent multivariate analysis. Panel D reports the correlations between

control variables. FORECAST is significantly positively related to MANIPULATE,

performance (ROA), SIZE and analyst following (ANALYST). The correlations among control

variables are generally consistent with the prior literature.

[Table 2]

It is possible that manipulation firms are always more likely to issue forecasts than

matched control firms, even before they start misstating earnings. To investigate this possibility,

we compare the percentage of firm-years during which forecasts are issued (Table 3, Panel A)

and the number of forecasts issued per firm-year (Table 3, Panel B) across manipulation and

control firms for two periods, one year before the misstatement period and during the

17 Specifically, we measure the change in analyst forecast dispersion as the dispersion of the consensus analystforecast following the management forecast within 5 days minus the dispersion of the last consensus analyst forecast before the issuance of management forecast, divided by stock price at the end of previous year.


18/55

17

misstatement period.18 If firms provide more earnings guidance in order to better conceal

misstatements, potentially through enhanced reputational effects, we expect manipulation firms

to issue more forecasts than the matched control firms during the manipulation period, but not

during the year prior to the manipulation period. Consistent with this prediction, as Table 3

Panel A shows, although the percentages of manipulation and control firms that issue earnings

guidance are exactly the same one year prior to the misstatement period (i.e., 35.5%),

manipulation firms are significantly more likely to issue earnings guidance during the

misstatement period than control firms (i.e., 56.3% versus 42.6%, t=3.15 with p-value


19/55

18

in forecast frequency is likely driven by the manipulation decision as managers are more likely

to have anticipated overstating earnings when issuing earnings forecasts in the subsequent

misstatement years than in the first misstatement year.19

Figures 1 and 2 depict the forecast

issuance and frequency for manipulation and control firms. In summary, the results reported in

Table 3 suggest that, compared to control firms, manipulation firms are more likely to issue

earnings guidance when they commence misstating earnings.

[Table 3 and Figures 1 and 2]

4.2 Multivariate analyses of management forecast issuance by manipulation firms

Prior research has shown that forecasting firms are systematically different from non-

forecasting firms (e.g., Hirst et al., 2008). Therefore, we next estimate a multivariate logistical

regression of forecast occurrence to further examine the association between accounting

manipulations and earnings forecast issuance after controlling for those differences.

FORECAST = b0 + b1MANIPULATE + b2ROA+ b3CHG_ROA+ b4CHG_EMP+ b5RETURN+ b6 NEWEQUITY+ b7 NEWDEBT+ b8ISSUANCE+ b9SIZE + b10ANALYST+ b11BKMK + b12LITIGATE+ Firm Fixed Effects + Year Fixed Effects (2)

where FORECAST is an indicator variable that is equal to one if firm i issues an annual or

quarterly EPS forecast in fiscal year t, and zero otherwise; and MANIPULATE is equal to one if

firm i manipulated earnings in fiscal year t, and zero otherwise. Both forecasting and

manipulation decisions could be driven by firm performance and capital market related

incentives (Miller, 2002, Lang and Lundholm, 2000, Dechow et al. 2011). Therefore, in

equation (1) we include four performance related control variables and three capital market

19 As a robustness test, we also restrict our sample to manipulation firms with three or more misstatement years andcompare the forecast frequency for the first, the second, and the subsequent misstatement years between themanipulation firms and control firms. The sample includes 28 manipulation firms and 28 control firms. We findthat the forecast frequency difference between the two groups of firms is concentrated in the third and subsequentmisstatement years rather than in the first and second years. The longer the manipulation lasts, the harder it is toargue that managers do not anticipate that they need to manipulate earnings when issuing forecasts; therefore, thisfinding provides additional support to our conjecture that the forecast frequency increase is associated with themanipulation decision.


20/55

19

incentive related variables. The first two performance variables capture accounting performance,

return on assets (ROA) and change in return on assets (CHG_ROA) in year t-1, while the next

two performance variables are market adjusted stock return (RETURN) and abnormal change in

employees (CHG_EMP) in year t-1. Dechow et al. (2011) suggest that abnormal change in

employees is a useful measure of the underlying economic reality of the firm.20

We use the

amount of newly issued equity in year t+1 (EQUITY) and the amount of newly issued debt in

year t+1 (DEBT) to capture the incentives attributable to the capital market. In addition,

following Dechow et al. (2011), we include an indicator variable (ISSUANCE) that measures

whether the firm has issued any securities in year t+1.

We also control for a number of additional variables that prior research suggests are

associated with forecasting. SIZE is measured as the natural logarithm of total assets for firm i at

the end of year t -1. Based on prior research (e.g., Kasznik and Lev 1995), firm size is expected

to be positively related to disclosure frequency. We define ANALYST as the number of analysts

following firm i as of the end of year t-1. A number of prior papers (e.g., Lang and Lundholm

1993) document a positive association between analyst following and disclosure quality. BKMK

is the ratio of book value to market value of equity for firm i at the end of year t-1. Market to

book (the inverse of our measure) is used by Bamber and Cheon (1998) to proxy for proprietary

costs. To control for the likelihood of litigation, we follow Francis et al. (1994) by using a

qualitative variable LITIGATE to indicate membership for firm i in litigious industries.21

All

variables are defined in Table 1 Panel B.

20 Following Dechow et al. (2011), we measure abnormal change in employees as the percentage change in thenumber of employees less the percentage change in total assets. Dechow et al. (2011) argue that this measure mightcapture the underlying firm performance if assets are overstated and the change in number of employees is not likelyoverstated. In addition, managers might reduce employee headcount in an attempt to increase earnings.21 Our results of increasing forecasts for manipulation firms continue to hold when we measure the litigation riskusing the prediction model in Rogers and Stocken (2005); however, using their model results in further reduction of


21/55

20

We estimate this model using firm-year observations before and during the misstatement

period. By including firm fixed effects and year fixed effects, Model (1) is essentially a

difference-in-differences specification that is similar to the methodology in Armstrong et al.

(2011) and Bertrand and Mullainathan (2003). The coefficient on MANIPULATE, b1,

represents the difference in the change of forecast issuance from one year before manipulation to

the manipulation period between manipulation firms and matched control firms. In other words,

when we estimate whether the misstatement firms significantly increase the likelihood of

forecast issuance during the manipulation period, our benchmark is the change in forecast

issuance of matched non-manipulation firms during the same period. This specification allows

us to better control for firm-specific sticky earnings guidance strategies and time-series earnings

guidance trends.

The regression results are reported in Table 4. The coefficient on MANIPULATE is

significantly positive (p-value =0.013), suggesting that manipulation firms are more likely to

issue EPS forecasts. In terms of the economic magnitude of the results, manipulation firms are

36.5% more likely to issue EPS forecasts than non-manipulation firms. As to control variables,

we find that firms with higher return on assets (ROA) have a greater propensity to issue EPS

forecasts, consistent with Miller (2002). However, interestingly, the issuance of EPS forecasts

appears to be negatively associated with annual market-adjusted buy and hold stock returns

(RETURN). One possible explanation for this result is that the decline in stock return (after

controlling for ROA) is correlated with increased demand for information from investors, which

results in more forecast issuance. In addition, the issuance of EPS forecasts is positively

associated with the amount of equity issuance (NEWEQUITY) and the number of analyst

our sample size. Therefore, for the analysis reported in Table 4, we define the litigation risk following Francis et al.(1994) to minimize data loss.


22/55

21

following (ANALYST), consistent with prior studies (e.g., Lang and Lundholm 1993; Lang and

Lundholm 2000).22

In sum, we investigate how manipulation firms change their forecast issuance when they

start manipulation by conducting (1) a univariate cross-sectional analysis (Table 2, Panel A), (2)

a univariate difference-in-differences analysis (Table 3), and (3) a multivariate difference-in-

differences analysis (Table 4).23 All three analyses unanimously show that manipulation firms

are more likely to issue earnings forecasts during the misstatement period than matched non-

manipulation firms.24

[Table 4]

Analysis based on AAER content

In the previous analysis, we rely on archival financial data to investigate the difference in

management forecast issuance between manipulation and control firms. In this section, we adopt

a different approach by using the SEC’s discussion in the AAERs to examine the role of

22 The insignificant coefficient estimate on SIZE may be due to the high correlation between SIZE and ANALYST(Pearson 0.63 while Spearman 0.62 in Table 2, Panel D). We find that SIZE becomes positively associated with

FORECAST when we exclude ANALYST from the regression. We note that we do not control for corporategovernance and executive compensation in Table 4. Including corporate governance or executive compensationvariables would significantly reduce our sample size. Further, prior research documents that poor corporategovernance is positively associated with accounting manipulations (Dechow et al., 1996), but negatively associatedwith the likelihood of management forecast issuance (Ajinkya et al., 2005). Therefore, excluding corporategovernance in our regression model is likely to bias against our finding a higher likelihood of issuing forecasts formanipulation firms. Regarding equity compensation incentives, we did a robustness check by hand-collecting theequity based compensation data for our sample managers when the data is not available in Execucomp. We measureCEO pay-for-performance sensitivity following the method described by Bergstresser and Philippon (2006). Oursample size dropped from 360 to 239. After including CEO pay-for-performance sensitivity as a control variable,MANIPUATE continues to be significantly positive in explaining FORECAST, thus equity-based incentivecompensation is unlikely to be an alternative explanation, either.23 Even though we use a difference-in-differences research design and include twelve time-varying control variables

(including CEO equity-based compensation), we acknowledge that our analyses are still subject to the omittedvariables problem if (1) this variable is correlated with both MANIPULATE and FORECAST, (2) this variable isnot captured by our time-varying control variables, and (3) this variable has changed from the pre-manipulation period to the manipulation period.24 We did two additional robustness tests. First, we investigate the robustness of our results by focusing on thefrequency of management forecasts. We measure the frequency using the log number of management forecastsissued per firm year. The untabulated results for Equation (2) show that manipulation firms issue more EPSforecasts in a difference-in-differences analysis (the coefficient on MANIPULATE = 0.144, p-value = 0.079).Second, we estimate the forecast issuance regression for annual forecasts only. The inferences are quite similar tothe regression results reported in Table 4 (the coefficient estimate on MANIPULATE is 2.683, p-value = 0.006).


23/55

22

incentive factors underlying managers’ decisions to issue earnings guidance for the group of

manipulation firms. An AAER typically describes the SEC’s conclusions following its

investigation of each accounting misstatement case. Even though the amount of information

contained in the AAERs is at the SEC’s discretion and varies with factors such as the depth of

the SEC’s investigation or the availability of evidence (or other ad hoc factors), this approach

allows us to capture the incentives highlighted by the SEC. Given that managers manipulating

earnings tend to be under high pressure from the stock market (Dechow et al., 2011), we expect

that the guidance issuance decision within manipulation firms is also associated with the market

pressure and related benefits.

We read each AAER for our sample firms and gathered information to answer the

following questions that are related to earnings guidance, the various incentives managers face,

and benefits managers reap. The corresponding variables get a value of one when the answer is

yes, and zero otherwise. We then examine and report the association between management

forecast issuance and these variables.

1. Does the AAER mention management earnings forecasts (AAERGUIDE)?2. Does the AAER mention that managers want to meet or beat analyst forecasts (MB_ANALYST)?3. Does the AAER mention that managers want to meet or beat management earnings guidance

(MB_GUIDE)?4.

Does the AAER mention that the firm is under market pressure (MARKET_PRESSURE)?5. Does the AAER mention that managers benefit from insider trading (BENEFIT_TRADE)?6. Does the AAER mention that managers benefit from bonus (BENEFIT_BONUS)?7. Does the AAER mention that managers benefit from equity compensation and/or other benefits

related to stock market performance (BENEFIT_EQUITY)?8.

Are CEOs or CFOs cited in the AAER?

The results are reported in Table 5. Panel A of Table 5 presents the mean of these

AAER-based variables, and Panel B reports the t-test results that compare forecast firms and

non-forecast firms. In addition to presenting the results for the full sample of manipulation firms,

we also show the results separately for firms for which CEOs and/or CFOs are cited by the SEC.


24/55

23

For these firms, there is greater certainty that managers (CEOs/CFOs) whose incentives are

affected also made the management forecast decisions.25 As shown in Table 5 Panel B, not

surprisingly, 12.1 percent of forecast firms were described to have incentives to meet

management forecasts while none of the non-forecast firms were cited for such incentives. We

find that forecast firms are significantly more likely to face capital market pressure than non-

forecast firms (MARKET_PRESSURE equal to 24.1 percent versus 5.0 percent with a p-value of

0.013). In addition, forecast firms’ managers are significantly more likely to have benefited

from manipulations through equity based compensation than non-forecast firms’ managers

(BENEFIT_EQUITY equal to 43.1 percent versus 20.0 percent with a p-value of 0.067). The

results are similar when we limit our analysis to the subsample for which CEOs and/or CFOs are

cited in the AAERs. In fact, the result for market pressure is stronger when CEOs and/or CFOs

are cited (i.e., the difference between forecast and non-forecast firms is 29.4% for such cases

compared to 19.1% for the full manipulation sample). Overall, these results are consistent with

manipulation firms’ managers making forecast issuance decisions strategically. The

management forecast issuance decision appears to be associated both with capital market

pressure, as well as capital market related financial benefits.

[Table 5]

4.3. Consequences associated with earnings guidance of manipulation firms

Our second hypothesis predicts that the stock market reactions to the revelation of

accounting manipulations are more negative for firms issuing earnings forecasts during the

misstatement period. To test this hypothesis, we first obtain the dates when the accounting

manipulations were discovered and publicly disclosed through a Dow Jones News Wire search.

25 We note that not being cited by the SEC does not necessarily mean that CFOs or CEOs are not involved inaccounting manipulations.


25/55

24

Following Karpoff et al. (2008), for the misstatement firms in our sample, we gather the dates of

trigger events that attracted the SEC’s scrutiny (e.g., self-disclosures of accounting irregularity),

announcements of the SEC’s informal inquiry and/or formal investigation, and restatement dates,

starting with the beginning of the misstatement period to the date when the AAER is finally

publicly released. We calculate the abnormal returns by cumulating the three-day abnormal

stock returns around all the above disclosure dates (if they are available) associated with each

misstatement firm. We then estimate the following regression model:

ABRETURN_DISCLOSE =b0+ b1FIRM_FORECAST + b2FIRM_SIZE + b3FPERIOD+ b4MISSTATE_AMOUNT+ Industry Indicators (3)

ABRETURN_DISCLOSE is the cumulative three-day market-adjusted stock returns

around all the above-described disclosure events associated with each misstatement. Note that

we conduct this analysis at the firm level and all independent variables are defined at the firm

level (see detailed variable definitions in Table 1 Panel B). We include the number of

misstatement years (FPERIOD) and the misstatement amount (MISSTATE_AMOUNT) to

control for the severity of the misstatement. The regression results are reported in Table 6.

Consistent with H2, FIRM_FORECAST is significantly and negatively related to the market

reaction (p-value = 0.087).26 Both the coefficients on FPERIOD and MISSTATE_AMOUNT

are significantly negative, suggesting that the stock market reaction is more negative when the

misstatement period is longer and the misstatement amount is larger.

[Table 6]

26 The coefficient estimate on FIRM_FORECAST appears quite large, indicating 17.5 percent larger stock pricedeclines for manipulation firms that issue management forecasts. This is consistent with the finding in Karpoff et al.(2008a) that the majority of the loss of market values upon the revelation of misstatements comes from lostreputation (i.e., 66.6% out of 38% loss of market values per Karpoff et al. (2008a)). We also acknowledge that thiseffect could be a combination of management forecasts and other concurrent expanded voluntary disclosures at thesame time.


26/55

25

H3 predicts that it takes longer to detect misstatements for manipulation firms issuing

earnings forecasts. While H3 focuses on how forecast issuance affects the timeliness of

manipulation detection, it is possible that the lead time to detection of manipulation could also

influence forecast issuance decisions to the extent that the lead time is correlated with managers’

perceived detection risk. We conjecture that managers are likely to sense a higher detection risk

as they get closer to the ultimate date of detection because (1) information disclosures (e.g.,

queries or skepticism about firms’ underlying performance) in the market are likely to precede

discovery of misstatements27

and (2) the magnitude of the effort required to conceal the true

earnings is expected to grow with time. For example, managers might find it more difficult to

conceal fraud when the accumulated misstatement increases significantly over time, the firm gets

into liquidity problems, or when outsiders raise more questions about the firm’s economic

performance. Following this argument, the decision to issue earnings guidance is likely to be

affected by the lead time to the detection of the manipulation. Therefore, we use the lead time to

detection as a proxy for the unobservable detection risk. Following Field et al. (2005), we

address this endogeneity issue by using the following simultaneous equations model to test H3:

Equation (4a) (modeling firms’ earnings forecast decisions):

FORECAST = γ0 + γ1MONTH_DETECT + γ2ROA + γ3RETURN + γ4 NEWEQUITY + γ5SIZE+γ6ANALYST + γ7BKMK + γ8LITIGATE + γ9PRE_FORECAST (4a)

Equation (4b) (modeling the timeliness of manipulation detection):

27 For example, prior to the detection of the earnings manipulation by Tyco International, a Dallas fund manager andshort-seller named David Tice openly criticized Tyco’s accounting practices during its $30 billion acquisition deals.In response to such criticism, Tyco chairman and CEO Dennis Kozlowski called the short-seller’s statement“unfounded and malicious” while providing earnings guidance for the next quarter that misrepresented the true performance of the firm (New York Times, October 15, 1999). As another example, about one year prior to theSEC’s informal inquiry into Rent-Way’s accounting practices, Rent-Way announced that it has received manyinquiries from investors concerning its performance; in response to that, the company issued earnings guidancesuggesting expected performance would meet or beat analyst forecasts (PR Newswire, October 20, 1999).


27/55

26

MONTH_DETECT = b0 + b1FORECAST + b2SEC_HEADCOUNT + b3BIGN_AUDIT +b4SOX + b5ROA + b6RETURN + b7 NEWEQUITY + b8SIZE + b9 ANALYST + b10BKMK + b11LITIGATE (4b)

where MONTH_DETECT is the log number of months from the end of the misstated fiscal year

(i.e., firm i in year t) to the month of the detection date. It captures the lead time remaining for

the manipulation to be detected per firm-year, with smaller values being “higher detection risk.”

The mean of the number of months from the end of the misstated fiscal year (i.e., firm i in year t)

to the month of the detection date is 19.6 months and the median is 18 months.28 All other

variables are defined in Table 1, Panel B.

Equation (4a) investigates whether the risk of detection affects a firm’s decision to issue

earnings forecasts. We expect that managers of firms facing a high detection risk would be more

likely to issue earnings forecasts (i.e., γ1 in equation (4a) to be negative).29

Equation (4b)

examines whether issuing earnings forecasts affects the risk of detection. We expect b1 in

equation (4b) to be positive if management earnings forecasts are effective in reducing the

likelihood of detection and delaying accounting manipulations from being discovered.

To estimate this system of simultaneous equations, we need to identify instrumental

variables that are not directly related to MONTH_DETECT but are related to FORECAST in

equation (4a), and vice versa for equation (4b). The instrumental variable in equation (4a) is

PRE_FORECAST. The variable PRE_FORECAST is equal to one if a firm issues at least one

28 To understand the MONTH_DETECT variable further and investigate the percentage of manipulations that areended by detection, we create a firm-level variable, FRAUD_DETECT, which equals the number of months fromthe fiscal year end of the last manipulation year to the detection month. The mean is 7.88 months and the first,

second and third quartiles are 3, 8 and 14 months, respectively. Detection is likely to end fraud if the detection datefalls before the date of release of the next annual report (roughly 15 months after the end of the last fiscal year end).We find that 59 out of the 73 firms (81%) of the firms have FRAUD_DETECT less than or equal to 15 months. Wethen conducted a robustness check after excluding firms with FRAUD_DETECT greater than 15 months. Ourtimeliness results remain similar. The coefficient on FORECAST continues to be significantly positive(coefficient=1.514 with p-value=0.024) while the coefficient on detect is significantly negative (coefficient=-1.033with p-value=0.035).29 The simultaneous equations model is estimated within manipulation firm-years (i.e., excluding matched controlfirm-years). Hence, we do not use a difference-in-differences specification in Equation (2a) by including firm andyear fixed effects. Instead, we include PRE_FORECAST as instrumental variables.


28/55

27

earnings forecast in the year immediately preceding the first manipulation year, and zero

otherwise. Since management forecast issuance is relatively sticky, PRE_FORECAST is likely

to be positively associated with FORECAST. Because PRE_FORECAST is based on the

management forecast decision before manipulation starts, PRE_FORECAST is unlikely to be

related to manipulation and therefore, also unlikely to affect the risk of manipulation detection.

The instrumental variables in equation (4b) are SEC_HEADCOUNT, BIGN_AUDIT, and SOX.

SEC_HEADCOUNT is the log number of the SEC staff; BIGN_AUDIT is equal to one if a firm

is audited by a big N auditor, and zero otherwise; SOX is equal to one if the fiscal year end of a

firm-year is after November 2002, and zero otherwise. We expect the detection risk to be higher

with greater SEC resources, higher audit quality, and after the effective date of the Sarbanes

Oxley Act. However, those variables are unlikely to directly affect management forecast

issuance decisions.30

The estimation results of the simultaneous equations are reported in Table 7. Column (1)

reports the logistic regression results for Equation (4a). The coefficient on MONTH_DETECT is

significantly negative at the five percent level, suggesting that firms with higher detection risk

30 We follow Larcker and Rusticus (2010) and examine whether our instruments are appropriate. First, regardingPRE_FORECAST, when we estimate equation 4(a) while excluding MONTH_DETECT (the first stage model), wefind that PRE_FORECAST is highly significant in explaining FORECAST (coefficient=1.3068 and p-value=0.01).In addition, PRE_FORECAST is not significantly related with the residual in equation (4b) (Pearson correlation=0.03 and p-value=0.70). Therefore, PRE_FORECAST appears to be significantly correlated with FORECAST, butonly related to MONTH_DETECT mainly through FORECAST. Next, regarding the instrumental variables inequation (4b), when we estimate the equation while excluding FORECAST, we find that MONTH_DETECT issignificantly associated with SEC_HEADCOUNT (coefficient=-5.16 and p-value=0.002) and SOX (coefficient=-1.22 and p-value=0.001), but not with BIGN_AUDIT (coefficient=0.12 and p-value=0.65). The partial R-square for

these three variables is 0.268, and the partial F-statistic is 28.73 with p-value less than 0.001. So these threevariables explain a significant part of the variation in MONTH_DETECT. Further, we calculate the correlations ofthe estimated error from equation (4a) with SEC_HEADCOUNT, BIGN_AUDIT, and SOX. The estimated error isnot correlated with any of these three variables (p-values are 0.80, 0.61 and 0.86, respectively). Hence,SEC_HEADCOUNT and SOX appear to be correlated with MONTH_DETECT, but are related to FORECASTmainly through MONTH_DETECT. Finally, we conduct the Hausman test following Larcker and Rusticus (2010).When we include both FORECAST and predicted FORECAST from the first stage model in the regression ofMONTH_DETECT. Predicted FORECAST is significant (coefficient=-1.808; p-value=0.006) while FORECAST isnot significant. Therefore, the Hausman test rejects the null of no simultaneity problems. Taken together, the aboveevidence lends support to the instrument variables used in Equations 4(a) and 4(b).


29/55

28

(i.e., lower MONTH_DETECT, implying that the firm is getting closer to the manipulation

detection date) are more likely to issue earnings forecasts, which is consistent with our prediction.

The coefficient on the instrumental variable in Equation (4a), PRE_FORECAST, is significantly

positive (at the one percent level). Among other control variables, it appears that larger firms

(SIZE) and firms with better operating performance (ROA) are more likely to issue earnings

forecasts, consistent with prior research. Turning to Equation (4b), the second part of the

simultaneous equation where MONTH_DETECT is the dependent variable, the significantly

positive coefficient on FORECAST suggests that it takes longer to detect an accounting

manipulation when the firm issues earnings forecasts, consistent with H3. The coefficients on

two instrumental variables, SEC_HEADCOUNT and SOX, are significantly negative. This

result is consistent with our expectation that greater SEC resources and the effectiveness of the

Sarbanes Oxley Act likely lead to more timely detection of accounting manipulations. Among

other control variables, we find that detection of manipulation is more timely for firms issuing

new equity (NEWEQUITY), firms with greater analyst following (ANALYST), and firms with

lower stock returns (RETURN).

[Table 7]

Our last hypothesis relates to the legal costs associated with issuing forecasts during the

manipulation period. We first analyze legal penalties for manipulation firms after detection,

specifically, shareholder class action lawsuits and SEC lawsuit settlement amounts. We obtain

lawsuit settlement amounts for 72 manipulation firms from Governance Analytics and Audit

Analytics litigation data.31

We compare the settlement amounts for manipulation firms issuing

forecasts with those not issuing forecasts. As presented in Table 8 Panel A, the univariate results

31 We are able to obtain settlement amounts for SEC lawsuits for 29 firms in our sample, 2 of which only have SEClawsuit amounts and 27 have both SEC lawsuit and class action lawsuit amounts. When we exclude the SEClawsuit settlement amounts from the total settlement amounts, our results remain the same.


30/55

29

show that manipulation firms issuing forecasts during the manipulation period pay $247.8

million (with a median value of $21.4 million) as settlement on average, while non-forecast

manipulation firms only pay $64.1 million (median value is 0.8 million), and the difference is

statistically significant. However, the large univariate differences could be driven by other

determinants of settlement amounts that vary between forecasters and non-forecasters (e.g., firm

size, misstatement severity etc.). Therefore, we employ the following regression model to test

H4 after controlling for other factors that could potentially affect the settlement amount:

LN_AMOUNT =b0+ b1FIRM_FORECAST + b2FIRM_SIZE+ b3FIRM_ROA b4FIRM_LEVERAGE+ b5FPERIOD + b6MISSTATE_AMOUNT+

b7ABRETURN_DURING + b8ABRETURN_DISCLOSE+Industry Indicators(5)

where LN_AMOUNT is the natural log of the class action and SEC lawsuit settlement amounts

for firm i. This analysis is also done at the firm level. Because the settlement amount is likely to

be greater for a more severe manipulation, we include four variables to control for the

manipulation severity and the associated damage to investors: the number of misstatement years

(FPERIOD), the misstatement amount (MISSTATE_AMOUNT), market-adjusted stock returns

during misstatement years (ABRETURN_DURING), and the three-day market-adjusted stock

return around the announcement of the misstatement (ABRETURN_DISCLOSE).32

The regression results are reported in Table 8 Panel B. The coefficient on

FIRM_FORECAST is significantly positive (p-value = 0.020), suggesting that manipulation

firms issuing forecasts pay higher settlement amounts, consistent with H4. The results also show

that larger firms and firms with greater severity of manipulation (i.e., misstating earnings for

longer periods and by a larger amount) pay higher settlement amounts.

[Table 8]

32 To avoid sample reduction, we assume the misstatement amount to be the sample median value when it is missing.


31/55

30

We next examine whether legal penalties are higher for the managers of firms issuing

earnings forecasts during the manipulation period compared to managers of non-forecasting

manipulation firms. We gather legal penalty data for CEOs from AAERs by determining whether

CEOs were charged by the SEC, and if so, the various types of legal penalties that were sought

by the SEC, ruled by the court, or settled outside of court. We examine three types of legal

penalties: (1) being prohibited from serving as an officer, director, or accountant of a public

company in the future; (2) fines; (3) criminal charges that usually come with the penalty of

prison time or, at least, home detention.

Panel A of Table 9 reports univariate results. The manipulation firms issuing at least one

earnings forecast during the misstatement period (FIRM_FORECAST=1) have significantly

more managers (29% out of 56 firms) being prohibited from being an officer, director, or

accountant compared to the manipulation firms not issuing any earnings forecasts

(FIRM_FORECAST=0, 10% out of 20 firms); the difference is statistically significant at the 5

percent level. This penalty is substantial for managers because it puts significant constraints on

their future labor market opportunities and reduces the value of their human capital. In addition,

36% of the CEOs of manipulation firms issuing forecasts are required by the SEC to pay fines,

while this percentage is 20% for CEOs not issuing any earnings forecasts. Similarly, 4% of the

CEOs of manipulation firms issuing forecasts undergo criminal charges; in contrast, CEOs of

manipulation firms not issuing forecasts are not subject to criminal charges. The above two

differences are statistically significant at the 10 percent level. Overall, the results are consistent

with H4 in that providing earnings guidance is costly for managers of manipulation firms.

Many of the manipulation cases settle without admission of guilt or wrongdoing. In such

cases, directors’ and officers’ insurance often covers managers’ monetary penalties (Dyck et al.,


32/55

31

2010). Moreover, very few managers face criminal charges. Therefore, being prohibited from

serving as an officer, director, or accountant is likely to be the penalty that more directly impacts

managers. Following this argument, we estimate Equation (5) but using “barred from serving as

an officer, director, or accountant” as the dependent variable (Table 9 Panel B). The coefficient

on FIRM_FORECAST is 2.748 with a p-value equal to 0.017. Therefore, the multivariate

analysis continues to suggest that managers bear substantial legal costs when they also issue

earnings forecasts during the misstatement period. The coefficient on FIRM_FORECAST is not

significant when we use “having fines” and “having criminal charges” as dependent variables in

regression analyses (not tabulated).

[Table 9]

4.4 Additional analyses

Delete the first year of manipulation

An alternative explanation for the increased likelihood of management forecast issuance

when firms start to manipulate earnings is reverse causality; i.e., firms issue overoptimistic

management forecasts, and then manipulate their earnings to be consistent with their forecasts

(Kasznik 1999).33

Schrand and Zechman (2011) also provide evidence that managers of firms

misstating earnings tend to be over-confident about future performance. This alternative

explanation, however, cannot explain our finding that firms are more likely to issue earnings

guidance when the risk of manipulation detection is high. To alleviate this concern further, we

repeat our analyses on management forecast characteristics after deleting the first year of

manipulation. If managers decide to manipulate earnings after issuing forecasts, it would be hard

to imagine that the managers continue to issue forecasts in subsequent years and then find that

they have to also manipulate the subsequent years’ earnings. Our results after removing the first

33 This concern of reverse causality applies only to our analysis of the likelihood of issuing earnings guidance (i.e.,H1), not the consequences of earnings guidance strategies (i.e., H2-H4).


33/55

32

year of manipulation are qualitatively similar to our overall sample results. Specifically,

manipulation firms continue to be more likely to issue EPS forecasts in a difference-in-

differences analysis (the coefficient on MANIPULATE = 1.652, p-value = 0.047). Hence, the

reverse causality cannot completely explain the increased management forecast issuance for

manipulation firms.

Walking down analyst forecasts as an alternative explanation

Prior studies have found that firms issue earnings guidance to walk analyst forecasts

downward and to meet or beat analyst forecasts (e.g., Matsumoto 2002; Cotter et al. 2006). In

this section, we investigate whether manipulation firms increase their earnings guidance

frequency during the manipulation period merely to walk down analyst forecasts. We measure

the total walk-down amount by the difference between the first consensus analyst EPS forecast

and the last management EPS forecast for the same reporting period, scaled by stock price at the

end of prior year. We do not find significant differences in the total walk-down amount between

manipulation firms and non-manipulation firms, suggesting that the incentive to walk down

analyst forecasts does not lead to an increased likelihood of forecast issuance for manipulation

firms.

Investment strategy as an alternative explanation

It is possible that managers also use other strategies to avoid or delay detection in

addition to the forecasting strategy. In particular, Kedia and Philippon (2007) argue that

fraudulent firms hire and invest excessively in order to conceal the fraudulent activities.

Therefore, we examine whether the overinvestment strategy explains the results of our

consequence analyses. We follow McNichols and Stubben (2008) to measure the

overinvestment amount and control for it in all of our consequence analyses. In the untabulated


34/55

33

results, the coefficient estimates on FIRM_FORECAST continue to be significantly positive in

all three of the consequence regressions (stock market reaction, settlement amount, and the

probability of being barred from being an officer, director, or accountant) after controlling for

OVERINVEST. In addition, when we include OVERINVEST in the analysis for the timeliness

of detection, we continue to find that management forecasts delay the detection of accounting

manipulations. Overall, the overinvestment strategy does not appear to explain our consequence

results.34

Forecast issuance and the type of whistleblowing

Corporate fraud can be brought to light by various agents, including equity investors,

media, analysts, auditors, employees, and regulators (Dyck et al., 2010). While in Section 4 we

investigate whether earnings guidance issued by earnings manipulators influences when the

misstatements are detected, in this section we investigate whether management forecasts also

affect who detects the earnings manipulations. We expect that, by issuing earnings guidance,

firms would avoid or delay manipulation detection by potential whistleblowers who tend to rely

more on publicly available information such as management earnings guidance.

Dyck et al. (2010) identify the type of fraud whistleblower for 216 cases of alleged

corporate frauds.35

Merging their sample with ours results in a rather small sample of 56 firm-

year observations (i.e., 18 unique firms). The whistleblowers in the merged sample are equity

investors, media, auditors, employees, and the firms themselves. We expect equity investors to

rely more on publicly released management forecasts than other types of whistleblowers, as

34 In another robustness check, we examine whether the importance of earnings to the market is an omitted variablein our penalty analyses. We follow Lenox and Park (2006) and measure the importance of earnings using historicalERC. The coefficient estimates on our forecast issuance variable continue to be significantly positive (p-value=0.043 for lawsuit settlement amount and p-value=0.030 for legal penalties for managers).35 This data are available at http://faculty.chicagobooth.edu/adair.morse/research/data.html. We note that theirsample of fraud cases covers a wide range of frauds, not limited to accounting misstatements.


35/55

34

some of the latter (e.g., employees) are likely to have access to private information. We find that

investors are not whistleblowers for any manipulation firms that issue at least one earnings

forecast during the manipulation period (i.e., 0 out of 10 firms), while investors are

whistleblowers for one firm (i.e., 12.5 percent, 1 out of 8 firms) that never issued earnings

forecasts during the manipulation period. Given that the manipulation can potentially be

revealed in any of the misstatement years, we also conduct a firm-year analysis. While investors

are not whistleblowers for any firm-years with earnings guidance (i.e., 0 out of 39 firm-year

observations), investors are whistleblowers for 23.5 percent of firm-year observations without

earnings forecasts (i.e., 4 out of 17 firm-years). The difference between the two groups is

significance at the one percent level. These results provide some evidence consistent with the

conjecture that issuing earnings guidance reduces the likelihood of manipulations being detected

by investors. However, we do acknowledge that the merged sample is rather small and these

results should be interpreted cautiously.

5. Conclusions

This paper investigates the earnings guidance strategies of manipulation firms and their

associated consequences. We document both statistically and economically significant increases

in the guidance issued by earnings manipulators during the misstatement period, compared with

matched control firms. In terms of consequences, we find that earnings guidance issued by

manipulation firms serves to delay the detection of manipulation. However, when the

misstatement is detected and disclosed, manipulation firms issuing earnings guidance suffer a

significantly larger stock price decline compared to firms not issuing earnings guidance. Finally,

additional legal penalties are imposed on firms and managers that have issued earnings guidance.


36/55

35

Our paper constitutes some of the first empirical evidence on the interrelationship

between manipulated mandatory disclosures and voluntary earnings guidance. Overall, our

findings indicate that when firms choose to manipulate their mandated earnings, they also tend to

falsify their earnings guidance,

forecast strategies of earnings manipulators

Documents