Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2019

Meeting-or-Beating Earnings Benchmarks: The Effect of Natural DJonigshaan Psatrkers

Follow this and additional works at the DigiNole: FSU's Digital Repository. For more information, please contact [email protected] FLORIDA STATE UNIVERSITY

COLLEGE OF BUSINESS

MEETING-OR-BEATING EARNINGS BENCHMARKS:

THE EFFECT OF NATURAL DISASTERS

By

JONGHAN PARK

A Dissertation submitted to the Department of Accounting in partial fulfillment of the requirements for the degree of Doctor of Philosophy

2019 Jonghan Park defended this dissertation on March 27, 2019. The members of the supervisory committee were:

Tianming Zhang Professor Directing Dissertation

Yingmei Cheng University Representative

Bruce Billings Committee Member

Spencer Pierce Committee Member

The Graduate School has verified and approved the above-named committee members, and certifies that the dissertation has been approved in accordance with university requirements.

ii

I dedicate my dissertation to my loving parents, Wonkyu Park and Changsook Yoon and my sister, Soohyun Park. I am truly thankful for their endless love, encouragement and sacrifices.

iii ACKNOWLEDGMENTS

I would like to express my sincere gratitude to my dissertation chair, Tianming Zhang for his mentorship, guidance, and support. Without his constructive feedback and continuous encouragement, I would not have been able to complete my dissertation. I also would like to thank Bruce Billings, Yingmei Cheng, and Spencer Pierce for their invaluable guidance and service on my dissertation committee. In addition, I would like to thank Kurt Gee for generously allowing me to use the non- GAAP earnings data. I also acknowledge helpful comments from Roby Lehavy, Ted Christensen, Robert hills and workshop participants at Florida State University, 2018 BYU Accounting Research Symposium, the University of New Hampshire, Hong Kong Baptist University, and the Chinese University of Hong Kong (Shenzhen).

iv TABLE OF CONTENTS

LIST OF TABLES ...... vi

ABSTRACT ...... vii

1. INTRODUCTION ...... 1

2. BACKGROUND AND HYPOTHESIS DEVELOPMENT ...... 9

3. DATA DESCRIPTION ...... 16

4. METHODOLOGY ...... 17

5. EMPIRICAL TESTS AND RESULTS ...... 21

6. CONCLUSION ...... 33

APPENDICES ...... 35

A. VARIABLE DEFINITIONS ...... 35

B. EXCERPTS FROM 10-K REPORTS ...... 38

C. TABLES ...... 39

REFERENCES ...... 57

BIOGRAPHICAL SKETCH ...... 62

v LIST OF TABLES

1 SHELDUS Hazard types...... 39

2 Descriptive Statistics...... 40

3 Natural Disaster and Meeting-or-Beating Earnings Benchmarks ...... 43

4 Test of Non-GAAP Exclusions ...... 46

5 Test of Expectation ...... 47

6 Test of Market Response to Beaters ...... 48

7 Test of Earnings Management ...... 49

8 Test of Entire Distribution of Earnings Forecasts...... 52

9 Natural Disaster and Beating Analyst Forecasts ...... 53

10 Test of Using a Sample of Small Firms ...... 54

11 Test of Using Non-GAAP Earnings Disclosed by Managers ...... 56

vi ABSTRACT

I examine whether managers are more likely to meet-or-beat earnings benchmarks when firms experience natural disasters. When a natural disaster strikes and causes significant damages, it could negatively affect a firm’s financial performance, potentially resulting in firms missing earnings targets. In this type of situation, incentives for managers to meet static earnings benchmarks (i.e., zero or last year’s earnings) might weaken because they can shift the blame for poor performance to natural disasters. On the other hand, managers may have stronger incentives to meet dynamic earnings benchmarks (i.e., analyst forecasts) because analysts take into consideration the effect of disasters and missing this benchmark could signal that the effect of disasters was worse than expected. In addition, subjective estimation of losses or charges related to disasters may enable managers to more easily engage in non-GAAP exclusions management to increase the likelihood of meeting analysts’ estimates. Using a comprehensive dataset of natural disasters occurring in the U.S. since 1989, I find that firms affected by natural disasters are more likely to meet-or-beat analyst forecasts through non-GAAP exclusions management. This paper extends the earnings benchmark literature by providing evidence that, when facing unexpected external shocks such as natural disasters, managers could utilize a crisis to make opportunistic accounting choices.

vii CHAPTER 1

INTRODUCTION

The economic consequences of natural disasters in the United States have increased steadily for several decades, resulting in extensive losses in both tangible assets and human capital.

Nevertheless, few studies examine the capital market and financial reporting effects of natural disasters (Cheng et al. 2018). When a natural disaster strikes and inflicts significant damage, it is likely to negatively affect a firm’s operating performance. As a result, affected firms are more likely to miss earnings targets.1 Further, managers may be less incentivized to engage in costly earnings management to achieve earnings benchmarks, as they can easily and credibly shift the blame for the current poor performance to the natural disaster.

On the other hand, the economic magnitude of natural disasters is typically difficult for investors or analysts to quantify, providing managers an opportunity to utilize strategic non-GAAP exclusions, that are harder for external entities to detect, to meet-or-beat earnings benchmarks.

Therefore, it is not clear whether managers are more or less likely to meet-or-beat earnings benchmarks when firms experience natural disasters. Utilizing a novel database covering comprehensive natural disasters occurring in the United States, I examine whether managers are more likely to meet-or-beat earnings benchmarks when firms experience natural disasters.

Natural disasters are well suited for earnings benchmark study for the following reasons.

First, natural disasters are external shocks that are exogenous to firm and managerial characteristics (Dessaint and Matray 2017). Previous evidence of earnings management study

1 Prior research finds that managers have strong incentives to achieve earnings targets. For example, previous evidence suggests a disproportionally high number of firms reporting earnings per share that meet or slightly exceed these earnings benchmarks. (e.g., Hayn 1995; Burgstahler and Dichev 1997; Degeorge et al., 1999).

1 could have potential endogeneity problems such as reverse causality or unobserved heterogeneity because a firm’s likelihood of meeting-or-beating earnings benchmark is highly correlated with many firm and managerial characteristics.2 On the other hand, my findings with respect to a firm’s strategic accounting choice followed by natural disasters cannot easily to be attributed to firm or managerial characteristics (Dessaint and Matray 2017).

Second, natural disasters provide an ideal setting to observe differential managerial incentives to achieve the three earnings benchmarks (i.e., zero earnings, last year’s earnings, and analyst earnings forecasts). While prior research consistently documents that mangers tend to avoid missing the three earnings benchmarks, there is very little evidence on managers’ differential motivations to achieve these three benchmarks.3 For example, in some situations, managers may have strong incentives to achieve one but not the other benchmarks.

Natural disasters could weaken a manager’s motivation to achieve static earnings benchmarks such as zero- or last year’s earnings. When a natural disaster strikes, managers likely need to engage in excessive earnings management to reach the targets (i.e., avoiding losses or avoiding earnings decreases) because they have to cover additional losses and charges caused by natural disasters. Furthermore, managers could conveniently claim that poor performance is attributable to natural disasters rather than management failures. 4 Hence, in the presence of

2 For example, if a study examines the association between firm-level event such as SEO and the likelihood of meeting-or-beating earnings benchmarks, it is hard to distinguish whether the empirical evidence is attributable to the effect of SEO or just better performing firm’s higher likelihood of achieving the benchmarks (Teoh, Welch, and Wong 1998). 3 Prior studies suggest that incentives such as career concerns, litigation risk, and greater market rewards motivate managers to achieve these targets (Bartov, Givoly, and Hayn 2002; Mergenthaler, Rajgopal, and Srinivasan 2012; Matsunaga and Park 2001). 4 Anecdotal evidence suggests that firms often blame poor financial performance on external factors. For example, Ford Motor, Winn-Dixie, WebMD, and Delta Air Lines blamed negative earnings surprises on the September 11, 2001 terrorist attacks (Barton and Mercer 2005). 2 reasonable excuses from external shocks, managers may be less likely to engage in costly earnings management to overcome the circumstances outside of their .

In contrast, natural disasters could amplify a manager’s motivation to achieve dynamic earnings benchmark (i.e., analyst forecasts). Unlike static earnings benchmarks, to some extent, analyst forecasts incorporate the effects of disaster-related economic damages upon a firm's financial performance. Accordingly, missing analysts’ estimates could imply that losses or charges related to natural disasters are worse than expected, exacerbating investors’ negative of the firm’s future performance uncertainty. In this respect, meeting-or-beating analyst forecasts could be a way to demonstrate a firm’s ability to manage the crisis, mitigating shareholders’ concerns. Therefore, managers may be highly motivated to meet-or-beat analyst earnings forecasts after natural disasters strike.

Third, natural disasters could change managers’ preference of management choice.5 Prior research suggests that managers tend to use within-GAAP earnings management first and use non-GAAP reporting management as a last resort (Black, Christensen, Joo, and

Schmardebeck 2017). However, natural disasters enable managers to easily engage in relatively less costly forms of perception management such as non-GAAP exclusions management or analysts’ expectation management to influence a firm’s likelihood of meeting-or-beating analyst forecasts. When firms experience natural disasters, disaster-related expenses are reflected in special items, but analysts generally exclude those unusual expenses in estimating earnings since non-recurring items are often considered to be less predictive of future cash flows or abnormal returns (Doyle et al. 2003; Curtis et al. 2014). In this type of situation, analyst forecasts are likely

5 Prior research suggests that managers employ various perception management tools including real earnings management, accruals management, expectation management and non-GAAP reporting management to manage stakeholder perceptions (Black et al. 2017). 3 to be less accurate because the actual economic disaster-related damages are unknown ex ante.

Managers may utilize this circumstance to employ non-GAAP exclusions management. For example, managers could overestimate disaster-related unusual expenses to increase non-GAAP earnings (i.e., core earnings), which helps firms to meet-or-beat analyst forecasts.6 The subjective estimations by management are difficult for investors or analysts to quantify due to their lack of knowledge about the true economic effect of the natural disaster. Relatedly, the difficulty in estimating earnings may induce analysts to seek management input for additional access to private information regarding the disasters. In other words, natural disasters enable managers to easily guide analyst forecasts to be walked down to a level that they can meet-or-beat. In sum, while natural disasters could weaken a manager’s motivation to employ within-GAAP earnings management to reach static earnings targets, they may provide managers a good opportunity to engage in perception management to meet-or-beat dynamic earnings target.

I first examine whether firms affected by natural disasters are more likely to meet-or-beat the three earnings benchmarks. To directly test managers’ intentional management behavior, I test whether managers are more likely to “just meet-or-beat” or “just miss” earnings benchmark when firms experience natural disasters.7 In order to identify which firms are affected by natural disasters,

I utilize the Spatial Hazard Events and Losses Database for the United States (hereafter,

SHELDUS), which provides information about direct losses caused by natural disasters at the county-level. If a firm is located in a county where a natural disaster occurred, it is considered to

6 Non-GAAP earnings metrics are a customized version of GAAP earnings which are not prepared in accordance with generally accepted accounting procedures. They have been called ‘pro forma’, ‘Street’, or ‘core’ earnings in prior research (Doyle et al. 2013). 7 I use firm-year observations with forecast errors within a two-cent interval for analyst earnings benchmarks (Filzen and Peterson 2015). For zero- and last year’s earnings benchmarks, I use a ten-cent interval to define whether firms just meet or just miss, because using a two-cent interval shrinks my sample size significantly. To stay consistent with the test of analyst forecast errors, in untabulated analysis I use a two-cent interval for zero- and last year’s earnings benchmarks. I obtain qualitatively similar results. 4 have experienced the natural disaster. To ensure that a natural hazard has significantly affected a firm’s financial performance, I exclude natural disasters which inflict little or no damage.8 Next, I test how managers influence a firm’s likelihood of meeting-or-beating earnings benchmarks. I examine whether firms use opportunistic non-GAAP exclusions of non-recurring items; and whether these firms would fail to meet-or-beat analyst forecasts if they did not use these exclusions.

I then examine whether managers are more likely to guide analysts’ expectations to a level that they can meet-or-beat when firms experience natural disasters. Lastly, I examine how investors interpret meeting-or-beating analyst forecasts when firms experience natural disasters. Prior research finds that investors either reward or discount the earnings surprise of firms achieving earnings benchmarks. If achieving the target during the disaster year mitigates investors’ concerns and conveys a positive signal for future performance, investors might reward beaters with the premium (Bartov et al. 2002). On the other hand, if meeting-or-beating analyst forecasts is a consequence of management intentional intervention, investors tend to discount the earnings surprise (Doyle et al. 2013). To test investors’ response to the earnings surprise of firms meeting- or-beating analyst forecasts, I compare the three-day cumulative abnormal returns around earnings announcements of firms affected by disasters to those that are not affected.

I find that managers are more likely to just meet-or-beat than to just miss analyst forecasts, and they are more likely to use non-GAAP exclusions management to meet-or-beat analyst forecasts when firms experience natural disasters. In contrast, I do not find evidence of firms having a higher propensity to achieve zero or last year’s earnings benchmarks. This evidence suggests that while managers are less incentivized to meet zero- or last year’s earnings benchmarks, analyst forecasts are still important targets to achieve when firms experience natural disasters. I

8 The process of defining natural disasters is explained in more detail in the Methodology section. 5 also find results suggesting that investors do not appear to discount the earnings surprise of firms meeting-or-beating analyst forecasts during the disaster year. This is consistent with previous evidence that it is challenging for investors to detect manager’s opportunistic behavior related to special items (Doyle et al. 2013).

I conduct five additional analyses to validate my results. First, I examine whether managers are more likely to engage in accruals or real earnings management to meet-or-beat analyst forecasts when firms experience natural disasters. Prior research suggests that managers prefer to use within-

GAAP earnings management as a first choice, so managers could employ earnings management to meet-or-beat earnings benchmarks during the disaster year. However, I do not find evidence that firms manipulate accruals or real activities to meet-or-beat analyst forecasts when firms experience natural disasters, suggesting that managers are more likely to use opportunistic non-

GAAP exclusions rather than perform costly earnings management. Second, I examine the entire distribution of earnings forecast, using an alternative benchmark of meeting analyst forecasts, which differs from comparing just missing versus just meeting. 9 I define achieving analyst forecasts as beating by 0-1 cent (Cheng and Warfield 2005; Frankel et al. 2002). Using the entire distribution of earnings forecasts, I find qualitatively similar results. Third, I examine the association between natural disasters and the likelihood of beating analyst forecasts by more than two cents. Beating analyst forecasts by a large amount is often considered as an economic consequence rather than managers’ intentional management behaviors. I find evidence that firms affected by disasters are less likely to beat analyst forecasts by a large amount, which corroborates my findings that managers are more likely to intentionally adjust non-GAAP earnings to meet

9 Test of entire distribution of meeting analyst forecast using one cent interval has been commonly used in prior research (Cheng and Warfield 2005; Frankel et al. 2002).

6 analyst forecasts when firms experience natural disasters. Fourth, to mitigate the concern that firms’ headquarter locations may not always reflect manufacturing plants or operation sites, I repeat my tests using a sample of small firms whose headquarter locations are more likely to be located in the same area as manufacturing plants or operation sites. While potential measurement error is not removed completely by this analysis, it is likely to be less severe. Using a sample of small firms,

I find evidence that is consistent with my main findings. Lastly, I repeat my tests using an alternative dataset of non-GAAP earnings. According to Bentley et al. (2018), using I/B/E/S actual

EPS to examine managers’ non-GAAP reporting incentives significantly biases against finding evidence of aggressiveness in managers’ non-GAAP metrics. Using the dataset of non-GAAP earnings that managers actually disclose in their earnings announcement, I find evidence that managers are more likely to report non-GAAP earnings on their earnings announcement and that such reported non-GAAP earnings are more likely to meet-or-beat analyst forecasts when firms experience natural disasters.

My study contributes to the current body of accounting literature and presents intriguing implications. Most notably, I investigate the effect of a natural disaster on managers’ strategic behavior, which is largely unexplored in prior literature. My results suggest that managers could utilize natural disaster as a good opportunity to engage in non-GAAP exclusions management to meet-or-beat analyst forecasts. Foremost, my empirical evidence is less plagued by endogeneity issues (e.g., unobserved heterogeneity or reverse causality) because natural disasters are exogenous shocks to firm and managerial characteristics (Dessaint and Matray 2017; Cheng et al.

2018). Second, my results extend the benchmark-beating literature by providing evidence that, when facing unexpected exogenous shocks, managers could have different motivations to achieve the aforementioned three different earnings benchmarks. While prior studies document that

7 managers have strong incentives to achieve three earnings benchmarks, there is very little evidence on managers’ differential motivations to achieve the three benchmarks. Third, my study extends research examining the association between current period earnings management decisions and non-GAAP reporting. Black et al. (2017) provide evidence that managers tend to first use within-

GAAP earnings management before employing non-GAAP disclosures. However, I provide evidence that managers may exclude the negative effects via opportunistic non-GAAP exclusions rather than perform costly earnings management as a first choice when firms are affected by large negative exogenous shocks such as natural disasters. Lastly, since the negative economic effects of natural disasters have substantially increased over time, the effect of a natural disaster on a firm’s accounting choices could be of considerable interest not only to academics but also to practitioners and regulators.

8 CHAPTER 2

BACKGROUND AND HYPOTHESIS DEVELOPMENT

Prior research consistently suggests that managers have strong incentives to achieve earnings-based benchmarks. Specifically, zero, last year’s earnings, and analyst earnings forecasts are well established benchmarks that firms tend to avoid missing. Burgstahler and Dichev (1997) provide evidence of a discontinuity in the earnings distribution, indicating that more firms report small profits than firms reporting small losses. Barth et al. (1999) find that firms that consistently exceed prior years’ earnings have higher price-earnings multiples than firms that do not. Moreover, former SEC Chairman Arthur Levitt has stated that the stock market is unforgiving to firms that miss analysts’ earnings forecasts (Levitt 1998).10

In order to achieve these earnings-benchmarks, managers use various perception management tools (e.g., earnings management or disclosure of non-GAAP earnings) to manage stakeholders’ perceptions. Prior research suggests managers’ choice from among various perception management techniques is depending on firm-specific circumstances and the relative costs of each perception management tool (Black et al. 2017). In this respect, I expect that natural disasters could be an intriguing exogenous shock to affect the relative costs of each perception management tool, thereby influencing managers’ differing incentives to achieve the three earnings benchmarks.

When natural disasters strike and inflict significant damages, affected firms are more likely to have lower than expected earnings. Consequently, in order to achieve static benchmark

10 Prior research also find the cost of missing analyst forecasts, such as reduced management credibility (Bartov,Givoly, and Hayn 2002); a higher likelihood of litigation; lower cash bonuses for the Chief Executive Officer (Matsunaga and Park 2001); and a higher probability of forced management turnover (Mergenthaler, Rajgopal, and Srinivasan 2012). 9 (i.e., zero- or last year’s earnings), managers would likely need to perform more aggressive earnings management to cover additional losses or charges from the disaster-related economic damages. In this type of situation, managers are less motivated to perform costly earnings management such as accruals, which reverse the following period, or real earnings management which can affect earnings over serval periods (Black et al. 2017).11 Rather, managers can easily blame the firm’s current poor performance on natural disasters. For example, when Hurricane

Katrina struck, Entergy Corp. claimed that significant lost revenue was attributable to the hurricane.12 Society in general is likely to be sympathetic to those firms that suffered natural disasters. Therefore, markets could perceive a manager’s assertion that poor financial performance is a consequence of unexpected negative external shocks rather than management failure.

Collectively, managers are less incentivized to engage in costly earnings management to achieve static earnings targets when firms experience natural disasters.

On the other hand, the effect of natural disasters on managers’ incentives to achieve dynamic earnings benchmarks (i.e., analyst forecasts) could be different. First, to some extent, analysts’ estimates are more likely to take into consideration the information regarding disaster- related economic damages affecting a firm’s financial performance. One substantial difference between analyst earnings forecasts and zero- or last year’s earnings benchmark is that analysts often exclude non-recurring items in estimating earnings since those items are transitory, not representing core earnings.13 In other words, analysts are more likely to exclude disaster-related

11 Prior research suggests that both accruals and real earnings management could have potentially significant costs because accruals reverse in the subsequent period and real earnings management such as cutting R&D, , or adjust inventory levels expenses can affect earnings over serval periods (Black, Christensen, Joo and Schmardebeck 2017). Moreover, these earnings management are subject to scrutiny from various monitoring agency including auditors, investors and stakeholders. 12 Appendix B includes more detailed descriptions which is excerpted from Entergy Corp.’s 10-K reports. 13 Prior research provides discussions about the actual process by which GAAP items are excluded from analysts’ actual earnings (Christensen et al. 2011; Doyle et al. 2013). The decision is ultimately made by a consensus of the analysts covering the firm, but these studies suggest that managers can influence that decision (Filzen and Peterson 10 costs in their estimates when firms experience natural disasters, suggesting that market participants might not easily accept a manager’s excuse that missing analysts’ estimates is a consequence of the negative effects of disasters. Missing analysts’ estimates imply that losses or charges related to disasters might be worse than expected, which could be an unfavorable signal increasing the firm’s future performance uncertainty. Therefore, managers are still incentivized to meet-or-beat analyst forecasts to convey that the firm can handle the crisis, thereby lessening shareholders’ concerns regarding the firm’s future performance uncertainty (Bartov, Givoly, and Hayn 2002).14

Moreover, this circumstance that analysts exclude non-recurring expenses in their estimates enables managers to easily engage in perception management. When analysts estimate disaster-related unusual expenses, they are more likely to include estimation errors because true economic effects of the natural disaster are difficult for outside entities to quantify. If managers estimate disaster-related exclusions greater than analysts do, this would help firms to meet-or-beat analyst forecasts. Unlike costly earnings management manipulating accruals or real activities, this non-GAAP exclusions management is relatively costless form of perception management (Black et al. 2017). Consequently, managers could opportunistically estimate losses and charges related to disasters to influence the likelihood of achieving analyst forecasts. Given the arguments presented previously, the effect of natural disasters on incentives to meet-or-beat earnings

2015). If the company reports exclusions on a different basis than the majority of analysts, the forecast data provider will adjust the number reported in the press release to reflect the consensus analyst reporting basis and record that number as its actual earnings number (Doyle et al. 2013). In other words, the actual earnings number reported by I/B/E/S (which I use as my main non-GAAP earnings proxy) already includes any ex-post adjustments made by analysts to undo any detected managerial opportunism. According to prior literature, however, although analyst have some ability to detect and unwind opportunistic managerial exclusions, analysts do not fully reverse all recurring expenses excluded by management and market is still partially fooled by some non-GAAP exclusions (Gu and Chen 2010; Doyle et al. 2013). 14 Bartov et al. (2002) provide evidence that the premium of the earnings surprise is larger for firms in a financial distress, suggesting that markets perceive a distress-firm’s beating earnings expectation as information about its ability to survive. 11 benchmarks could be different depending on the types of earnings benchmarks. This leads to my first hypotheses as follows:

H1a: Firms affected by natural disasters are less likely to meet-or-beat zero- or last

year’s earnings benchmarks than firms not affected by natural disasters.

H1b: Firms affected by natural disasters are more likely to meet-or-beat analysts’

earnings estimations than firms not affected by natural disasters.

Prior research introduces four different ways of how managers can influence a firm’s likelihood of meeting analyst forecasts: (1) analysts’ expectation guidance, (2) exclusion of non-

GAAP specific items from actual earnings, (3) accruals manipulation, and (4) real activities manipulation (Doyle et al. 2013; Burgstahler and Eames, 2006; Matusmoto 2002; Filzen and

Peterson 2015). As mentioned previously, when firms experience natural disasters, managers are more likely to engage in relatively less costly forms of perception management to meet-or-beat analyst forecasts rather than employ costly earnings management. I posit that analysts’ difficulty of identifying the economic magnitude of losses or charges related to disasters enables managers to be better able to engage in a behavior of either non-GAAP exclusion or expectation guidance to meet-or-beat analyst forecasts.

The non-GAAP exclusion hypothesis, which suggests that firms could increase their non-

GAAP EPS by excluding unusual expenses, is based on the assumption that analysts do not fully unwind the opportunistic non-recurring item exclusions. If analysts are able to identify all actual exclusions in their forecasts, they can unwind these expenses and adjust their estimates (Doyle et al. 2013). Prior research provides evidence that analysts demonstrate some ability to differentiate between informative and opportunistic managerial exclusions, but they do not fully reverse all

12 recurring expenses excluded by management and that market is still partially fooled by some non-

GAAP exclusions (Gu and Chen 2010; Doyle et al. 2013).15

Especially, detecting manager’s opportunistic exclusions could be more challenging when firms experience natural disasters. Most charges or losses, such as restructuring charges, impairments, and other asset write-downs due to natural disasters, are reflected in special items, which are classified as non-recurring items. Since the exact amounts of disaster-related economic damage are unknown ex ante, analysts more likely estimate losses caused by disasters with errors.

In other words, analysts are less likely to fully adjust their forecasts for opportunistic unusual expenses in the presence of natural disasters, enabling managers to have more discretion to define non-GAAP earnings. For example, managers may reclassify some actual recurring expenses as non-recurring exclusions, in the process of overestimating the losses caused by natural disasters.

This process does not necessarily change the bottom line EPS, but increase non-GAAP EPS, which helps firms to appear to meet-or-beat analyst forecasts on a non-GAAP basis. Therefore, I expect that managers are more likely to use income-increasing non-GAAP exclusions to meet-or-beat analyst forecasts when firms are affected by significant natural disasters.

However, it is also plausible that managers could use expectation management to meet-or- beat analysts’ estimates when firms experience natural disasters. As mentioned previously, natural disasters could increase forecast errors due to the difficulty of identification of actual economic damages and the increase of uncertainty regarding the affected firms’ future earnings. Since analysts value accurate forecasts (Mikhail et al. 1999; Hong and Kubik 2003; Groysberg et al.

2011), it is possible that the difficulty in induces analysts to seek management input

15 Doyle et al. (2013) suggest that markets tend to discount the earnings surprise when income-increasing exclusions are used, but the discounts of market responses are primarily for firms that use other exclusions rather than special items. In other words, manager’s opportunistic behavior related to special items is challenging for investors to detect. 13 and guidance. Therefore, analysts could be motivated to curry favor with managers to have additional access to private information with regards to the disasters. (Libby et al. 2008; Francis and Philbrick 1993; Feng and McVay 2010; Filzen and Paterson 2015). Since this incentive is aligned with managers who want to convey the signal of the firms’ competence in handling the crisis, it may be easier for managers to guide analysts’ expectations to be walked down to a level that they can meet-or-beat. In sum, given the arguments presented here, how managers can influence a firm’s likelihood of meeting-or-beating analyst forecasts is an open question when firms experience natural disasters. This leads to my second set of hypotheses as stated as follows:

H2a: Firms affected by natural disasters are more likely to meet-or-beat analysts’

expectations through non-GAAP exclusions management than firms not affected by

natural disasters.

H2b: Firms affected by natural disasters are more likely to meet-or-beat analysts’

expectations through walking down the expectations than firms not affected by

natural disasters.

Prior research consistently finds that firms that meet-or-beat analysts’ earnings expectations enjoy a higher return than firms missing expectations. Bartov et al. (2002) provide evidence that this type of a premium to beaters is even higher when meeting-or-beating analyst estimates is less likely to be achieved, suggesting that meeting-or-beating analysts’ estimates imply a positive signal of the firm’s future performance. In the same manner, if meeting-or-beating analyst forecasts during the disaster year can convey the firms’ competence of handling the crisis, this might mitigate shareholders’ concerns with regard to the firms’ future performance uncertainty.

As a result, market could reward those firms with an additional premium.

14 On the other hand, prior research also provides evidence of market discounts on the earnings surprise when meeting-or-beating analyst forecasts is a consequence of management intentional intervention. Doyle et al. (2013) find that investors discount the earnings surprise when accompanied by income-increasing exclusions from GAAP earnings. Hypothesis 2 argues that, when firms experience natural disasters, managers are more easily to engage in non-GAAP reporting to influence a firm’s likelihood of achieving analysts’ benchmarks on a non-GAAP basis, even though they miss GAAP earnings benchmarks. In this case, markets might be suspicious of the positive earnings surprise of firms affected by disasters since the surprise could be a consequence of managers’ opportunistic perception management. Consequently, markets might discount the premium of meeting-or-beating analyst forecasts. In sum, based on the arguments presented here, whether the market rewards or discounts firms meeting-or-beating analyst forecasts during the disaster year, is an empirical question. This leads to my third hypothesis as stated in the null form.

H3: There is no difference in market reactions when firms meet-or-beat analyst forecasts

between firms affected by natural disasters and firms not affected by natural disasters.

15 CHAPTER 3

DATA DESCRIPTION

To construct my sample, I obtain the natural disaster data from the SHELDUS (Spatial

Hazard Events and Loss Database for the United States) database.16 SHELDUS covers natural hazards that occurred in the U.S territory, including thunderstorms, hurricanes, floods, wildfires, earthquakes and tornados, as well as perils such as flash floods and heavy rainfall. 17 The

SHELDUS data provides the information regarding direct losses of property damage, crop damage, and human losses, including injuries or fatalities caused by natural disasters at the county-level.

Table 1 provides the category of natural hazard types and the frequency and total damages caused by each disaster in my sample.

Based on those reported damages, I identify natural disasters that are significant enough to affect a firm’s financial performance. I discuss this process in more detail in the Methodology section. I obtain corporate financial data and stock returns from Compustat and CRSP. I obtain analyst forecast data from the I/B/E/S Summary unadjusted files and actual EPS numbers from the

I/B/E/S Actual unadjusted files. The final sample contains 4,960 firm-year observations which only includes forecast errors within a two-cent interval from 1989 to 2015. I begin my sample in

1989 because it is the earliest year for which SHELDUS includes the information regarding property and crop losses. For sensitivity analyses, I expanded my sample to 19,345 which includes the entire distribution of earnings-based benchmarks.18

16 SHELDUS was developed by the Hazards and Vulnerability Research Institute at the University of South Carolina. Since 2018, the Arizona State University Center for and Homeland Security supports and maintains SHELDUS (Cheng et al. 2018). 17 See https://cemhs.asu.edu/sheldus/. 18 The sample of 4,960 for main test only includes observations just missing and just beating (i.e -0.02≤ forecast error ≤ 0.02). However, in sensitivity analysis, I expand it to observations of entire distributions rather than comparing only just meet-or-beat versus just miss. 16 CHAPTER 4

METHODOLOGY

4.1 Measure of Natural Disasters

In order to determine whether the firm is affected by natural disasters, I use the zip code of a firm’s headquarter location from Compustat.19 Since the SHELDUS database provides the list of counties affected by a natural disaster, I match a firm’s headquarter zip code from Compustat to the county information from SHELDUS. One concern could emerge that the headquarter location may not always reflect the place of manufacturing plants or operation sites which could actually experience natural disasters (Cheng et al. 2018). Since I do not have access to plant-level information, I follow the assumption from prior research which suggests that, on average, plants are located in the same area as a firm’s headquarters (Chaney et al. 2014; Dessaint and Martay

2017).20

To verify that natural disasters significantly affect a firm’s financial performance, I determine a few criteria to define natural disasters. First, in order to focus on short-term external effects on U.S firms or industry, I eliminate droughts from the natural hazard sample. Droughts usually last for multiple years. Thus, droughts are not a surprise to the managers or investors during a multi-year period (Cheng et al. 2018).21 In other words, I focus on short-lived natural disasters that do not last for more than a few months. The occurrence of these types of natural disasters

19 Prior research suggests that they match the SHELDUS data to Compustat using county information (Dessaint and Matray, 2017). However, most of the county information in Compustat is missing. Therefore, I match zip code to county FIPS information from census.gov and then match this information to the SHELDUS data. 20 This potential measurement error could decrease the power of my tests, but there is no reason to expect that it would induce systematic bias into my results (Cheng et al. 2018). Also, to mitigate this concern, I perform sensitivity analyses using a sample of small firms and financial or utility firms whose headquarter locations are more likely to be in the same area as their operation sites. 21 Droughts make up only about 0.02% of the total natural disasters in my sample. 17 provides no information on the probability of the reoccurrence in the near future, so firms are not able to predict future occurrence (Dessaint and Matray 2017).22 Second, I also exclude natural disasters that cause no or minor property or crop damages, which do not directly affect a firm’s financial performance. Since the SHELDUS data only provides the information of property and crop losses at the county-level, I estimate the firm-level damage (adjusted for CPI) caused by natural hazards. Following the way in which SHELDUS distributes losses equally between the affected counties, I assume that firms are equally damaged if they are located in the same county.23

Accordingly, firm-level damage is calculated by the sum of county-level property and crop damage divided by the number of firms located in a county that year. I only keep firm-year observations that losses or charges related to natural hazards are more than 10% of a firm’s net income.24

4.2 Measure of Three Earnings Benchmarks

To define whether firms meet-or-beat analysts’ forecasts, I use the latest annual analysts’

EPS consensus as a benchmark. Analysts’ estimates and actual EPS numbers are obtained from

22 Dessaint and Matray (2017) examine how a sudden external shock such as a hurricane affects managers’ risk assessment. In their study, they find that there is no increasing or decreasing trend in the frequency of hurricanes, indicating that hurricane strikes are not serially correlated. Moreover, annual reports from the SHELDUS database suggest that the type of hazards causing serious losses is different for each year, indicating that it is hard to predict the frequency or magnitude of a future natural disaster. 23 The SHELDUS database explains how they distribute losses caused by natural hazards in a data description. For instance, a thunderstorm event affecting Richland and Lexington County in South Carolina and causing property damages of $50,000 will be entered into the database as an event affecting Richland County with $25,000 and Lexington County with $25,000 worth of damage (See http://hvri.geog.sc.edu/SHELDUS/index.cfm?page=faq). Following their definition, I distribute losses equally between the affected firms to estimate firm-level damage. In SHELDUS, the definition of "affected county" is broad: a county enters the database when monetary or human losses are strictly positive. In other words, some firms in counties of disaster zones could be little harmed by natural hazards (Dessaint and Matray 2017). However, I do not have access to firm-level disaster information and this concern of classification error works against finding a result. For robustness, I distribute county-level losses based on a firm’s size (total assets) and I obtain qualitatively similar results. 24 The cut-off of net income indicates that, if firms report poor performance (i.e., negative income), those can be classified as experiencing natural disasters even though being not significantly affected by natural disasters. I do not exclude those in my sample because these disasters might be the ones that management most likely opportunistically uses as an excuse for poor performance. I use a 10% cut-off for my test, but I change this cut-off to 5%, 15%, 30% and 50% and my results continue to hold. 18 the I/B/E/S summary unadjusted file and Actuals unadjusted file, respectively. To directly test managers’ intentional management behavior, following Filzen and Peterson (2015), I compare the firms that just meet-or-beat analyst forecasts to the firms that just miss analyst forecasts. To define meeting-or-beating the benchmark, I classify JustMOB as one, if a firm’s reported earnings equal or exceed analyst forecast consensus by two cents or less (i.e., 0 ≤ forecast error ≤ 0.02). Otherwise,

JustMOB equals zero if a firm’s reported earnings are less than analyst forecasts by two cents or less (i.e., -0.02 ≤ forecast error < 0). In sensitivity analyses, I use a one cent interval as an alternative benchmark, which has been consistently used in prior research.25

To examine whether firms meet-or-beat zero or last year’s earnings benchmark, I also compare the firms’ likelihood of a small miss versus a small beat. I limited the range of zero and last year’s earnings benchmarks within ten cents interval since my sample size shrinks significantly

26 within the two cents interval. I classify Zero_beati,t as an indicator variable that equals one if earnings per share before extraordinary item (Compustat data item epspx) is greater than zero and less than ten cents (i.e., 0 ≤ epspxt ≤ 0.10). Otherwise, Zero_beati,t equals zero if earnings per share before extraordinary item is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ epspxt

< 0). For the test of the last year’s earnings benchmark, I classify Last_beati,t as an indicator variable that equals one if the change of earnings per share between current and the previous year is greater than zero, but less than ten cents. (i.e., 0 ≤ actualt - actualt-1 ≤ 0.10). Otherwise,

Last_beati,t equals zero if the change of earnings per share between current and the previous year

25 The interval for establishing “just beat” versus “just miss” from Filzen and Peterson (2015), has not been consistently measured. Athanasakou et al. (2009) use the interval _£0.02 ≤ forecast error < £0.02. Cheng and Warfield (2005) and Frankel et al. (2002) define “just beat” as beating by 0–1 cent. Yu (2008) utilizes an interval of -0.08 ≤ forecast error ≤ 0.04. Prawitt et al. (2009) choose to test the sensitivity of their primary scaled measure with an unscaled interval of -0.02 ≤ forecast error ≤ 0.02. Therefore, for sensitivity analysis, I follow most recent studies that use a one cent interval as a benchmark to meet-or-beat analyst forecasts (Doyle et al. 2013; Huang et al. 2017). 26 To stay consistent with the test of analyst forecast errors, in untabulated analysis I use a two-cent interval for zero- and last year’s earnings benchmarks. I obtain qualitatively similar results.

19 is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ actualt – actual t-1 < 0). In sensitivity analyses, I examine the entire distribution of zero- and last year’s earnings benchmark rather than comparing a small miss versus a small beat.

20 CHAPTER 5

EMPIRICAL TESTS AND RESULTS

5.1. Meeting-or-Beating the Three Earnings Benchmarks

To test my first hypothesis, I examine whether firms affected by disasters are more likely to meet-or-beat earnings benchmarks. First, I estimate a model, where JustMoBi,t is the dependent variable, to examine the association between natural disasters and the likelihood of meeting-or- beating analyst forecasts. Disasteri,t, the main variable of interest, is an indicator variable that equals one if a firm is in a county that experiences natural disasters in year t. Otherwise, Disasteri,t equals zero if a firm does not experience natural disasters. As stated in my first hypothesis, I expect the coefficient of Disasteri,t to be positive. Next, using Zero_beati,t and Last_beati,t as the dependent variables, I examine the association between natural disasters and the likelihood of meeting-or-beating zero- and last year’s earnings benchmarks. If firms affected by disasters are less likely to avoid small negative earnings or small earnings decreases, the coefficient of

Disasteri,t is expected to be negative.

In order to control for firm characteristics, following prior research, I include firm size

(Log of market value of equity), book-to-market (BTM), sales growth (Growth), return on assets

(ROA), profitability, (Profitable), implicit claims with stakeholders (R&D expense and Labor intensity), the level of operating cash flows (CFO), analyst following (NUMEST), the number of shares (Logshares), net operating assets (NOA), cash flow volatility (CFVOL), and audit quality

(BigN) (Cheng and Warfield 2005; Doyle et al. 2013; Davis et al. 2009; Matsumoto 2002; Frankel et al. 2002; and Prawitt et al. 2009). Appendix 1 provides a detailed explanation of the variables.

I also include year and industry fixed effects and cluster standard errors by firm to control for potential serial correlation in errors terms (Petersen 2009). I estimate this equation using both a

21 logistic model and a linear probability model. I use the linear probability model to mitigate concerns with regards to the incidental parameters problem that could occur with logistic regressions when including fixed effects (Cheng et al. 2018). The regression model, which tests the first set of hypotheses, is stated as follows:

P(Meeting-or-Beatingi,t) = β0+ β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t +

β10 CFOi,t + β11 Logsharesi,t + β12 CFVOLi,t + β13 NUMESTi,t +

β14 BigNi,t + Fixed Effects + εit (1)

Table 3 presents the results for the estimation of H1 testing. Panel A presents the result of the logistic regression and Panel B presents the results of the linear probability model. Consistent with my expectation, I find that Disasteri,t is positively and significantly associated with JustMoBi,t in both regressions.. This evidence suggests that firms affected by natural disasters are more likely to just meet-or-beat analysts’ expectations than to just miss. In contrast, the coefficient of

Disasteri,t is negatively significant for Zero_beat and not significant for Last_beat, indicating that firms affected by natural disaster are less likely to avoid losses and not more likely to beat last year’s earnings. This result implies that, while managers might have less incentive to achieve static benchmarks, analyst forecasts are still important targets to achieve when firms experience natural disasters.

5.2. Non-GAAP Exclusions Management Tests

To test H2a, I examine whether firms affected by natural disasters are more likely to use non-GAAP exclusions to meet-or-beat analyst forecasts than firms that do not experience natural disasters. If firms determine “Actual” EPS to meet-or-beat analyst forecasts, non-GAAP earnings

22 are more likely to be higher than GAAP EPS due to the exclusion of income-decreasing non- recurring items. First, I examine whether firms affected by natural disasters are more likely to have higher non-GAAP earnings than GAAP earnings. Following Doyle et al. (2013), I use a logistic regression where the dependent variable is Excl_use which equals one if actual EPS from I/B/E/S is greater than GAAP EPS before extraordinary items and discontinued operations (i.e. Epspx from

Compustat), and zero otherwise. The regression model is stated as follows:

P(Excl_usei,t) = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t +

β11 Logsharesi,t + β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t +

Fixed Effects + εit (2)

Next, I examine whether non-GAAP exclusions help firms to meet-or-beat analysts’ expectations during the disaster year. I restrict my sample to only firms that just meet-or-beat analyst forecasts and examine whether those firms would not be able to meet-or-beat analyst forecasts if they did not use non-GAAP exclusion. I use a logistic regression where the dependent variable is Excl_mgmti,t which equals one if the firm would have missed expectations without the use of the exclusions, and zero otherwise. The main variable of interest, Disasteri,t is expected to be positive in both regressions 2 and 3. The regression model is stated as follows:

P(Excl_mgmti,t) = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t +

β11 Logsharesi,t + β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t +

Fixed Effects + εit (3)

Table 4 presents the results of the analyses of non-GAAP exclusions management. First, I find that Disasteri,t is significantly and positively associated with Excl_usei,t, indicating that firms

23 affected by natural disasters are more likely to have higher non-GAAP earnings than GAAP earnings. Also, I find a positive association between Disasteri,t and Excl_mgmti,t, indicating that firms affected by natural disasters would be more likely to miss analyst forecasts if they do not use non-GAAP exclusions. Consistent with my expectation, the results suggest that managers are more likely to manage or define non-GAAP earnings to meet-or-beat analyst forecasts when firms are affected by natural disasters.

5.3. Expectation Management Guidance Tests

To test H2b, I perform two tests to examine whether firms are more likely to use expectation management to meet-or-beat analyst forecasts when firms experience natural disasters.

First, I perform a logistic regression where the dependent variable is Exp_mgmt. Exp_mgmt is defined as one if the firm would have missed analyst expectations using the first consensus forecast of the period but meets-or-beats analysts’ expectations using the last consensus forecast of the period. Otherwise, Exp_mgmt equals zero. Exp_mgmt is only defined in cases where firms just meet-or-beat analyst forecasts. The Logistic regression for this test is stated as follows:

P(Exp_mgmt i,t) = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t +

β11 Logsharesi,t +β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t +

Fixed Effects + εit (4)

Second, I examine the association between firms’ forecast walk-downs and natural disasters when firms meet-or-beat analyst forecasts. I estimate forecast walk-downs as the ending forecast minus the beginning forecast scaled by stock price, i.e., (analyst forecastend - analyst forecastbeg) / |Stock priceend | (Bradshaw et al. 2016). In other words, a walk-down is represented

24 by a negative value. If firms are more likely to guide analysts’ expectations to be walked down to meet-or-beat during the disaster year, I expect to find a negative association. The regression model is specified as follows:

Walkdowni,t = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t + β6 ROAi,t +

β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t + β11 Logsharesi,t +

β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t + Fixed Effects + εit (5)

Table 5 presents the results of the analyses of expectation management. The result of logistic regression suggests that Exp_mgmt is not significantly associated with natural disasters.

This evidence suggests that firm affected by natural disasters do not necessarily miss analyst expectations using the first consensus forecast of the period when they meet analyst forecasts using the last consensus of the period. I also do not find a significant association between natural disasters and walk-down of forecasts. In other words, the results suggest that firms affected by natural disasters do not necessarily engage in expectation management to meet-or-beat analyst forecasts, which corroborates the results of the non-GAAP exclusions management hypothesis.

5.4. Market Reaction to Beaters During the Disaster Year

To test my third hypothesis, I first restrict my sample to only firms that meet-or-beat analyst forecasts. Then, I estimate the following regression which captures the market’s response to the earnings announcement for firms that meet-or-beat analyst forecasts:

CAR(-1,1)i,t = β0 + β1 Disasteri,t + β2 Surprise + β3 Disaster*Surprisei,t+ β5 Firmsizei,t +

β6 BTMi,t + β7 Growthi,t + β8 Accrualsi,t + Fixed Effect + εit (6)

The dependent variable, CAR (-1, 1) represents the three-day cumulated abnormal return centered on the earnings announcement date. The coefficient of main variable of interest is

25 interaction between natural disasters and earnings surprise. If investors reward the earnings surprise of firms affected by disasters with an additional premium, the coefficient is expected to be positive. On the other hand, if investors discount the earnings surprise of firms affected by disasters, the coefficient is expected to be negative.

Table 6 presents the results of the analysis of market response to beaters. The coefficient on the interaction between natural disasters and earning surprise is not statistically significant. In other words, investors’ response to just meeting-or-beating expectations is not significantly different for firms affected by disasters versus firms not affected by disasters. This evidence suggests that market participants do not appear to discount the earnings surprise of firms meeting analyst forecasts during the disaster year in which management interventions are more likely to be engaged.

5.5 Sensitivity Analyses

I perform five additional sets of analyses to ensure that my results are robust. First, I examine whether managers are more likely to employ accruals or real earnings management to meet-or-beat analyst forecasts when firms experience natural disasters. Accrual or real earnings management has been consistently documented as a common tool of earnings management. Prior research finds that managers tend to employ within-GAAP earnings management techniques first before employing other perception management tools (Black et al. 2017). For example, subjective estimation of losses or charges from disasters might enable managers to manipulate accruals to increase short-term operating performance to recover unexpected losses caused by natural disasters.

26 Relatedly, managers might cut R&D and other discretionary expenses to meet-or-beat earnings benchmarks during the disaster year. 27

For the test of accruals management, following Filzen and Peterson (2015), I estimate

“unmanaged accruals”, using discretionary accruals from the performance-matched Jones (1991)

28 model developed by Kothari et al. (2005). Then, I classify Acc_mgmti,t as one if firms meet -or- beat expectations, but would have missed expectations if discretionary accruals were removed from earnings. Consistent with the non-GAAP exclusions management test, I restrict my sample to only firms that just meet-or-beat analyst forecasts. The logistic regression, where the dependent variable is Acc_mgmt, is stated as follows:

P(Acc_mgmti,t) = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t +

β11 Logsharesi,t + β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t +

Fixed Effects + εit (7)

I also examine the association between discretionary accruals and natural disasters.

Abnormal accruals are estimated using the performance-adjusted modified Jones model (Jones

1991; and Kothari et al. 2005). If firms affected by disasters are more likely to have greater abnormal accruals, the coefficient of disasters is expected to be positive. The regression model is specified as follows:

27 According to the survey result from Graham et al. (2005), to meet earnings benchmarks, managers prefer to engage in real earnings management such as cutting R&D and other discretionary expenses, before turning to accruals management. 28 Unmanaged accruals are calculated as following: First, the estimate of discretionary accruals is multiplied by lagged total assets, and then scaled by shares of outstanding. Next, this amount is subtracted from the Actual EPS from I/B/E/S to get the unmanaged level of earnings. Acc_mgmti,t equals one if this unmanaged level of earnings is less than the mean of Analysts’ EPS consensus (Filzen and Peterson 2015). 27 Accrualsi,t = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t + β6 ROAi,t +

β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t + β11 Logsharesi,t +

β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t + Fixed Effects + εit (8)

For the test of real earnings management, following Badertscher (2011), I define REM, a comprehensive measure of real earnings management, as a sum of three measures of real earnings management (abnormal cash from operations, abnormal expenditures, and abnormal production of inventory) which have been used in prior research.29 Then, following Black et al. (2017), I classify

REM_mgmti,t as one if firms meet-or-beat expectations, but would have missed expectations if real earnings management was removed from earnings. The logistic regression, where the dependent variable is REM_mgmt, is stated as follows:

P(REM_mgmti,t) = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t +

β6 ROAi,t + β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t +

β11 Logsharesi,t + β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t +

Fixed Effects + εit (9)

I also examine the association between the magnitude of real earnings management and natural disasters. If firms affected by disasters are more likely to engage in real earnings management, the coefficient of disasters is expected to be positive. The regression model is specified as follows:

29 The three real earnings management measures are commonly used in prior research (Cohen et al. 2008; Gunny 2010; and Zang 2012). Abnormal cash from operations captures sales manipulation (e.g., the acceleration of the timing of sales or abnormal discretionary expenditure), abnormal expenditures captures R&D, advertising, and selling, general, and administrative expense, and abnormal inventory production captures managers’ overproduced inventory, which lowers overhead costs assigned to each unit of inventory under absorption costing. (Badertscher 2011; Black et al. 2017) 28 REMi,t = β0 + β1 Disasteri,t + β2 FirmSize i,t + β3 BTMi,t + β5 Growthi,t + β6 ROAi,t +

β7 Profitablei,t + β8 R&Di,t + β9 Labori,t + β10 CFOi,t + β11 Logsharesi,t +

β12 CFVOLi,t + β13 NUMESTi,t + β14 BigNi,t + Fixed Effects + εit (10)

Panel A of Table 7 presents the results of the analyses for the use of accruals management to meet-or-beat analyst forecasts. I do not find significant results that firms would have missed analyst forecasts if managers do not employ accruals management. I also do not find a significant association between Disaster and Accruals. Panel B of Table 7 presents the results of the analyses for the use of real management to meet-or-beat analyst forecasts. The results do not show that firms would have missed analyst forecasts if managers do not employ real earnings management.

I also do not find a significant association between Disaster and REM. Overall, the results suggest that accruals or real management does not help firms to meet-or-beat analyst forecasts when firms experience natural disasters.

Second, I use alternative benchmarks of meeting-or-beating earnings targets and test the entire earnings distribution, rather than only comparing just meet-or-beat versus just miss. For the test of the zero- and last year’s earnings benchmarks, I use Zero_EMi,t and EA_increasei,t as the dependent variables where Zero_EMi,t equals one if earnings per share before extraordinary item

(Compustat data item epspx) is greater than zero and where EA_increasei,t equals one if the change of earnings per share between current and the previous year is greater than zero (actualt – actual t-

1 > 0). Otherwise, Zero_EMi,t or EA_increasei,t equals zero. For the test of analyst forecasts benchmark, I use a one-cent interval which has been commonly used in previous research. I use a logistic regression where the dependent variable, MEETi,t equals one if actual EPS equals or exceeds analyst forecast consensus by one cent (i.e., 0 ≤ forecast error ≤ 0.01). Otherwise, MEETi,t equals zero.

29 Table 8 presents results of the logistic regression using alternative benchmarks of earnings targets. I find that natural disaster is positively associated with the likelihood of meeting analyst forecasts and negatively associated with the likelihood of avoiding losses and earnings decreases, consistent with the results reported in Table 3. Using alternative benchmarks of earnings targets, I find robust evidence that managers are more likely to meet-or-beat analyst forecast while less likely to meet-or-beat zero- and last year’s earnings benchmarks when firms experience natural disasters.

Third, to ensure that managers are more likely to engage in intentional earnings management, I examine the association between natural disasters and the likelihood of beating analyst forecasts by large amounts. Unlike beating by small amounts, beating analyst forecasts by more than two cents could be considered as an economic consequence rather than intentional earnings management. I use both a logistic model and a linear probability model where the dependent variable BIG_BEAT equals one if actual EPS exceeds analyst forecast consensus by two cents (i.e., forecast error > 0.02). Otherwise, BIG_BEAT equals zero.

Table 9 presents the result of estimation examining the firm’s likelihood of beating analyst forecasts when firms experience natural disasters. Column 1 presents the results of the logistic regression and column 2 presents the results of the linear probability model. In both regressions, I find that natural disaster is negatively and significantly associated with the likelihood of beating the target, suggesting that firms affected by natural disasters are less likely to beat analyst forecast by large amounts. This evidence corroborates my findings that managers are more likely to intentionally define or manage earnings to meet rather than beat analyst forecasts during the disaster year.

30 Fourth, I repeat my test using a sample of small firms to mitigate the concern that firms’ headquarter locations are not always reflect the place of manufacturing plants or operation sites which could actually experience natural disasters (Cheng et al. 2018). Unlike big firms who have expanded their foreign operations, small firms’ headquarters and operation sites are both likely to be located in the same are. Following the definition of “Small” used for Standard & Poor’s Small

Cap Index, I classify small firms with market capitalization ranging from $300 million up to $1.5 billion.30

Table 10 presents the results of estimations using a sample of small firms. The coefficient of natural disasters is positively significant for analyst forecasts, but not significant for zero- and last year’s earnings benchmark, consisted with the results reported in Table 3. In other words, using the sample mitigating measurement errors, I find robust evidence that managers are motivated to meet-or-beat analyst forecasts while less motivated to achieve static earnings benchmarks when firms experience natural disasters.

Finally, I use the alternative dataset including non-GAAP earnings from Bentley et al.

(2018). According to Bentley et al. (2018), non-GAAP earnings from analyst forecast data providers (FDPs) such as IBES could be different with those from managers although there is a substantial overlap between two data sets. Since I examine potential managerial opportunism, I want to ensure that my tests capture what managers are actually disclosing in their earnings announcements (Doyle et al. 2013). Using the dataset from Bentley et al. (2018), I test whether managers are more likely to disclose non-GAAP earnings in their earnings announcement and

30 I do not include overly small firms because these firms might systematically differ from other firms. For example, some small firms were exempted from the disclosure requirements of SFAS No. 89, and the SEC has separate 10K and 10Q filing requirements for small firm (Wall street Journal 1991 and Lang et al. 1993). Also, Dodd-Frank Act of 2010 exempts firms with a public float of less than $75 million. In other words, these kinds of small firms might have different incentives to meet-or-beat analyst forecasts (Holder et al. 2012). Moreover, restricting my sample to small firms with market capitalization less than $300 million significantly reduce my sample size which could weaken the power of the test. 31 whether the use of non-GAAP exclusions increase the propensity for the firm to meet-or-beat analyst expectations.

Table 11 presents the result of analysis using non-GAAP earnings disclosed in earnings announcement. Using non-GAAP earnings disclosed by managers, I find qualitatively and quantitatively similar results. I find evidence that Disaster is positively associated with the likelihood of reporting non-GAAP earnings in earnings announcement. Also, I find that Disaster is positively associated with the non-GAAP exclusions management. Consistent with the argument of managers’ opportunism, managers are more likely to disclose non-GAAP earnings in their earnings announcement and those earnings are more likely to meet-or-beat analyst forecasts.

32 CHAPTER 6

CONCLUSION

The economic consequences of natural disasters have continued to increase in recent years.

They present serious risks for corporate profits and capital markets. Nevertheless, natural disasters have not been explored in detail in the financial reporting literature. Natural disasters could be an intriguing exogenous shock influencing managers’ differing incentives to achieve the three earnings benchmarks. In this study, I empirically examine how natural disasters affect managers’ strategic behavior to achieve earnings benchmarks. Specifically, I examine whether managers engage in aggressive non-GAAP reporting to meet-or-beat analysts’ earnings targets, when facing unexpected natural disasters.

I hypothesize that firms affected by natural disasters are more likely to meet-or-beat analyst forecasts through non-GAAP exclusions management. The subjective estimation of losses or charges related to disasters may enable managers to more easily engage in non-GAAP exclusions management that are harder for outside entities to quantify because the actual economic disaster- related damages are unknown ex ante. Using a comprehensive dataset of natural disasters that occurred in U.S, I examine the association between natural disasters and a firm’s likelihood of meeting-or-beating analysts’ estimates. Consistent with my expectations, I find that firms affected by natural disasters are more likely to meet-or-beat analyst forecasts and that these firms are more likely to use non-GAAP exclusions to meet-or-beat analyst forecasts. My results are robust to tests of other perception that could influence the likelihood of meeting-or-beating analyst forecasts.

My results offer new insights to the benchmark-beating literature by providing evidence that natural disasters could influence managers’ differing incentives to achieve earnings

33 benchmarks. Moreover, my study adds to the line of the earnings management and non-GAAP reporting literature. Unlike previous finding from Black et al. (2017), I provide evidence that managers employ non-GAAP reporting to meet-or-beat analyst forecasts rather than perform costly earnings management as a first choice when firms are affected by large negative exogenous shocks such as natural disasters. With the growing number of occurrences of natural disasters, this evidence of the impact of natural disasters on a firm’s accounting choices would be of considerable interest not only to academics, but also to practitioners and regulators.

34 APPENDIX A

VARIABLE DEFINITIONS

Variable Definition (Alphabetical)

Variable Definitions Accrual management indicator equals one if the firm meets-or-beats expectations but would have missed if accruals are removed from Acc_mgmt = earnings, zero otherwise. In other words, (Actual- accruals=unmanaged levels of earnings) do not meet-or-beat expectation. Only defined in cases where Justmob = 1. Abnormal accruals calculated using the performance-adjusted Accruals = modified Jones model (Jones 1991; and Kothari et al. 2005). An indicator variable equal to one if actual I/B/E/S EPS exceeds BIG_BEAT = analyst forecast consensus by two cents (i.e. forecast error > 0.02), and zero otherwise. An indicator variable equal to one if a firm is audited by a “Big 4” BigN = audit firm, and zero otherwise. BTM = Book value of assets over market value of assets. CFO = Cash flow from operation, scaled by total assets. The standard deviation of the companies’ cash flow from operations CFVOL = for the previous five years scaled by quarterly total assets. Discretionary accruals estimated from modified Jones model DA = developed by Kothari et al. (2005). An indicator variable equal to one if a firm’s headquarter is located Disaster = in a county affected by a natural disaster during the year, and zero otherwise. An indicator variable equal to one if current year’s earnings per share Earnings_increase = is greater than the last year’s earnings per share (i.e., I/B/E/S actual t- 1 < I/B/E/S actual t), zero otherwise. An indicator variable equal to one if a firm meet-or-beat expectation but would have missed if exclusions are removed from earnings, and Excl_mgmt = zero otherwise. In other words, (Actual-exclusion) do not meet-or- beat expectation. Only defined in cases where Justmob = 1 An indicator variable equal one if I/B/E/S Actual EPS exceeds Excl_use = COMPUSTAT EPS (either epspxq or epsfxq depending on the I/B/E/S primary or diluted indicator), and zero otherwise. An indicator variable equal to one if the firm would have missed analyst expectations using the first consensus forecast of the period, Exp_mgmt = but meets-or-beats analyst expectations using the last consensus forecast of the period, and zero otherwise. Only defined in cases where Justmob = 1.

35 Variable Definition (Continued) Forecast errors (Actual I/B/E/S EPS less mean EPS forecast as of the Femean = end of year). Firm size = The natural log of market value of equity.

Growth = Sales growth (Salet-Salet-1/ Salet-1). An indicator variable equal to one if 0 ≤ forecast error ≤ 0.02, and JustMoB = zero if -0.02 ≤ forecast error <0. Labor = Labor intensity (1 - ppegt) / (at + dpact) as in Matsumoto (2002). An indicator variable equal to one if the change of earnings per share between current and the previous year is greater than zero, but less than ten cents (i.e., 0 ≤ actual - actual ≤ 0.10), and zero if the change Last_beat = t t-1 of earnings per share between current and the previous year is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ actualt – actual t-1 < 0). Logshares = The natural log of shares outstanding. ln(cshoq). An indicator variable equal to one if non-GAAP EPS (actual I/B/E/S MEET = EPS) equals or exceeds analyst forecast consensus by one cent (i.e. 0 ≤ forecast error ≤ 0.01), and zero otherwise. An indicator variable equal to one if the earnings announcement MGR_Excl = contains a non-GAAP EPS disclosure, and zero otherwise. An indicator variable equal to one if reported non-GAAP EPS MOB_MGR = (disclosed by managers) equals or exceeds analyst forecast consensus, and zero otherwise. NUMEST = The number of analysts following. An indicator variable equal to one if operating cash flows, operating Profitable = income, or earnings are negative (following Christensen et al. 2008). RD = Research and development expense scaled by total assets (XRD/AT). An indicator variable equal to one if a firm is audited by a “Big 4” BigN = audit firm, and zero otherwise. The sum of the three measures of real earnings management, REM = abnormal cash from operations, abnormal expenditures, and abnormal production of inventory (Badertscher 2011). An indicator variable equal to one if a firm meets-or-beats REM_mgmt expectation but would have missed if the amounts of real earnings management are removed from earnings, and zero otherwise. ROA = Operating income before depreciation, scaled by total assets. I/B/E/S actual EPS minus the median consensus of analyst forecasts Surprise = from I/B/E/S. (Analyst forecast - Analyst forecast ) / Abs(prcc_f). Negative Walkdown = end beg value indicates a walk-down.

36 Variable Definition (Continued) An indicator variable equal to one if earnings per share before extraordinary item (Compustat data item epspx) is greater than zero Zero_beat = and less than ten cents (i.e., 0 ≤ epspxt ≤ 0.10), and zero if earnings per share before extraordinary item is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ epspxt < 0). An indicator variable equal to one if earnings per share before Zero_EM = extraordinary item (Compustat data item epspx) is greater than zero, and zero otherwise.

37 APPENDIX B

EXCERPTS FROM 10-K REPORTS

In August and September 2005, Hurricanes Katrina and Rita caused catastrophic damage to large portions of the U.S. Utility's service territory in Louisiana, Mississippi, and Texas, including the effect of extensive flooding that resulted from levee breaks in and around the greater New Orleans area. The storms and flooding resulted in widespread power outages, significant damage to electric distribution, transmission, and generation and gas infrastructure, and the loss of sales and customers due to mandatory evacuations and the destruction of homes and businesses. Total restoration costs for the repair and/or replacement of the U.S. Utility's electric and gas facilities damaged by Hurricanes Katrina and Rita and business continuity costs are estimated to be $1.5 billion, including $835.2 million in construction expenditures and $664.8 million recorded as regulatory assets. The cost estimates do not include other potential incremental losses, such as the inability to recover fixed costs scheduled for recovery through base rates, which base rate revenue was not recovered due to a loss of anticipated sales. For instance, at Entergy New Orleans, the domestic utility company that continues to have significant lost revenue caused by Hurricane Katrina, Entergy estimates that lost net revenue due to Hurricane Katrina will total approximately $320 million through 2007. In addition, Entergy estimates that the hurricanes caused $32 million of uncollectible U.S. Utility customer receivables.

38 APPENDIX C

TABLES

Table 1 SHELDUS Hazard Types

Total Damage SHELDUS Category Frequency Percentile (2015 CPI adjusted) Avalanche 4 0.04% $0.3 Coastal 125 1.39% $67,429.5 Earthquake 5 0.06% $34,812.9 Flooding 2,068 22.98% $5,015.5 Fog 36 0.40% $11.5 Hail 665 7.39% $9,878.7 Heat 25 0.28% $1,324.9 Hurricane/Tropical Storm 93 1.03% $123,312.4 Landslide 26 0.29% $171.8 Lightning 1022 11.36% $609.0 Severe Thunderstorm 170 1.89% $8,691.3 Tornado 636 7.07% $19,755.5 Tsunami 1 0.01% $0.3 Wildfire 147 1.63% $11,341.7 Wind 2951 32.79% $25,194.2 Winter Weather 1025 11.39% $17,159.8 *Damages are in millions. * Most frequent natural hazards are ordered as follows: Wind, Flooding, Winter Weather, Lightning, and Hail. ** Natural hazards causing biggest damage are ordered as follows: Hurricane, Coastal, Earthquake, Wind and Tornado.

39 Table 2 Descriptive Statistics

Panel A: Sample of Analyst Forecast Benchmarks N Mean Median SD Q1 Q3 JustMoB 4,960 0.735 1 0.442 0 1 Disaster 4,960 0.353 0 0.478 0 1 Firmsize 4,960 6.302 6.056 1.931 4.910 7.508 BTM 4,960 0.523 0.412 0.419 0.242 0.684 Growth 4,960 0.199 0.100 0.499 -0.005 0.259 ROA 4,960 -0.015 0.025 0.174 -0.034 0.076 Profitable 4,960 0.762 1 0.426 1 1 Logshares 4,960 3.747 3.604 1.350 2.787 4.519 CFO 4,960 0.048 0.071 0.143 0.011 0.128 NOA 4,960 1.107 0.593 1.881 0.334 1.076 NUMEST 4,960 7.922 5 7.605 2 11 CFVOL 4,960 0.055 0.035 0.062 0.01 0.065 RD 4,960 0.0723 0.023 0.120 0 0.103 Labor 4,960 -0.260 -0.248 0.357 -0.447 -0.120 BigN 4,960 0.777 1 0.416 1 1 MEET 4,960 0.107 0 0.309 0 0 BIG_BEAT 4,960 0.402 0 0.490 0 1 Panel A of Table 2 reports the summary statistics for 4,960 firm-year observations which include two-cent intervals of analyst forecast errors from 1989 to 2015. JustMOB is an indicator variable equal to 1 if 0 ≤ forecast error ≤ 0.02, and zero if -0.02 ≤ forecast error <0, and 0 otherwise. Disaster is an indicator variable equal to 1 if a firm’s headquarter is located in a county affected by a natural disaster during the year, and 0 otherwise. Firmsize is the natural log of market value of equity. BTM is the book value of assets over market value of assets. Growth is the Sales growth (Salet-Salet-1/ Salet-1). ROA is the operating income before depreciation, scaled by total assets. Profitable is an indicator variable equal to 1 if operating cash flows, operating income, or earnings are negative, and 0 otherwise. Logshares is the natural log of shares outstanding. CFO is the cash flow from operation, scaled by total assets. NOA is the scaled beginning of the period net operating assets. NUMEST is the number of analysts following. CFVOL is the standard deviation of the companies’ cash flow from operations for the previous five years scaled by quarterly total assets. RD is the research and development expense scaled by total assets (XRD/AT). Labor is the Labor intensity (1 - ppegt) / (at + dpact) as in Matsumoto (2002). BigN is an indicator variable equal to 1 if a firm is audited by a “Big 4” audit firm, and 0 otherwise. MEET is an indicator variable equal to 1 if non-GAAP EPS (actual I/B/E/S EPS) equals or exceeds analyst forecast consensus by one cent (i.e. 0 ≤ forecast error ≤ 0.01), and 0 otherwise. BIG_BEAT is an indicator variable equal to 1 if actual I/B/E/S EPS exceeds analyst forecast consensus by two cents (i.e. forecast error > 0.02), and 0 otherwise. All continuous variables are winsorized at 1% and 99 % of the distribution. (Continued on next page)

40 Table 2 (continued)

Panel B: Sample of Zero Earnings Benchmarks N Mean Median SD Q1 Q3 Zero_beat 1,391 0.511 1 0.500 0 1 Disaster 1,391 0.522 1 0.499 0 1 Firmsize 1,391 5.213 5.073 1.652 4.041 6.085 BTM 1,391 0.661 0.556 0.524 0.294 0.872 Growth 1,391 0.182 0.078 0.454 -0.046 0.263 ROA 1,391 -0.005 0.000 0.048 -0.008 0.006 Profitable 1,391 0.679 1 0.467 0 1 Logshares 1,391 3.360 3.321 1.179 2.547 4.038 CFO 1,391 0.044 0.049 0.092 0.008 0.091 NOA 1,391 0.972 0.577 1.565 0.305 0.986 NUMEST 1,391 4.839 3 5.466 1 6 CFVOL 1,391 0.069 0.047 0.073 0.025 0.085 RD 1,391 0.072 0.033 0.103 0 0.121 Labor 1,391 -0.279 -0.245 0.316 -0.448 -0.123 BigN 1,391 0.732 1 0.443 0 1 Panel B of Table 2 reports the summary statistics for 1,391 firm-year observations which include ten-cent intervals of zero-earnings from 1989 to 2015. Zero_beat is an indicator variable equal to 1 if earnings per share before extraordinary item (Compustat data item epspx) is greater than 0, and 0 otherwise. All continuous variables are winsorized at 1% and 99 % of the distribution. All other variables are defined in Panel A of Table 2. (Continued on next page)

41 Table 2 (continued)

Panel C: Sample of Last Year’s Earnings Benchmarks N Mean Median SD Q1 Q3 Last_beat 3,619 0.580 1 0.494 0 1 Disaster 3,619 0.355 0 0.479 0 1 Firmsize 3,619 6.020 5.710 2.017 4.522 7.354 BTM 3,619 0.562 0.487 0.426 0.268 0.758 Growth 3,619 0.156 0.072 0.392 -0.008 0.210 ROA 3,619 -0.030 0.020 0.190 -0.040 0.063 Profitable 3,619 0.748 1 0.434 0 1 Logshares 3,619 3.636 3.537 1.312 2.740 4.357 CFO 3,619 0.032 0.064 0.158 0.005 0.117 NOA 3,619 1.182 0.623 1.974 0.330 1.189 NUMEST 3,619 6.669 4 6.705 2.000 9.000 CFVOL 3,619 0.055 0.035 0.064 0.018 0.064 RD 3,619 0.076 0.015 0.149 0 0.101 Labor 3,619 -0.258 -0.256 0.390 -0.472 -0.113 BigN 3,619 0.776 1 0.417 1 1 Panel C of Table 2 reports the summary statistics for 3,619 firm-year observations which include ten-cent intervals of last year’s earnings from 1989 to 2015. All continuous variables are winsorized at 1% and 99 % of the distribution. Last_beat is an indicator variable equal to 1 if the change of earnings per share between current and the previous year is greater than zero, but less than ten cents. (i.e., 0 ≤ actualt - actualt-1 ≤ 0.10 and 0 otherwise. All continuous variables are winsorized at 1% and 99 % of the distribution. All other variables are defined in Panel A of Table 2.

42 Table 3 Natural Disaster and Meeting-or-Beating Earnings Benchmarks

Panel A: Logistic Regression

JustMOBi,t Zero_beati,t Last_beati,t 0.179** -0.829*** -0.106 Disaster (2.14) (-5.25) (-1.21) 0.136*** 0.011 0.141*** Firmsize (2.63) (0.12) (2.72) -0.040 -0.099 -0.122 BTM (-0.40) (-0.55) (-1.19) -0.093 0.307* 0.454*** Growth (-0.94) (1.88) (4.02) 0.194 0.775** ROA (0.58) (2.03) 0.352*** 2.706*** -0.027 Profitable (3.21) (14.72) (-0.24) 0.246 -2.360** -0.022 RD (0.67) (-2.30) (-0.06) 0.428** 0.125 0.006 Labor (2.25) (0.36) (0.03) 0.236 1.299 0.272 CFO (0.58) (1.46) (0.61) -0.070 -0.419*** -0.187*** Logshares (-1.17) (-3.54) (-3.12) -0.006 0.098* -0.029 NOA (-0.33) (1.93) (-1.15) 1.466** 3.846*** 1.042 CFVOL (2.15) (3.59) (1.47) 0.002 -0.0266* 0.010 Numest (0.26) (-1.72) (1.32) 0.049 0.255 -0.062 BigN (0.57) (1.57) (-0.67) Year-fixed effect Yes Yes Yes Industry-fixed effect Yes Yes Yes Observation 4,960 1,391 3,619 Adj. R2 3.13% 27.88% 4.04% (Continued on next page)

43 Table 3 (continued)

Panel B: Linear Probability Model Regression

JustMOBi,t Zero_beati,t Last_beati,t 0.034** -0.138*** -0.026 Disaster (2.14) (-5.14) (-1.23) 0.027*** 0.001 0.032*** Firmsize (2.69) (0.01) (2.71) -0.008 -0.022 -0.029 BTM (-0.38) (-0.73) (-1.21) -0.020 0.051* 0.104*** Growth (-0.96) (1.85) (4.23) 0.034 0.184** ROA (0.51) (2.07) 0.070*** 0.491*** -0.005 Profitable (3.17) (18.60) (-0.18) 0.046 -0.384** -0.003 RD (0.71) (-2.17) (-0.04) 0.085** 0.020 0.002 Labor (2.26) (0.35) (0.03) 0.045 0.192 0.063 CFO (0.55) (1.37) (0.60) -0.015 -0.068*** -0.423*** Logshares (-1.27) (-3.54) (-3.14) -0.002 0.013* -0.007 NOA (-0.41) (1.92) (-1.15) 0.277** 0.582*** 0.231 CFVOL (2.13) (3.44) (1.42) 0.002 -0.005* 0.002 Numest (0.20) (-1.93) (1.32) 0.010 0.040 -0.014 BigN (0.61) (1.48) (-0.67) Year-fixed effect Yes Yes Yes Industry-fixed effect Yes Yes Yes Observation 4,960 1,391 3,619 Adj. R2 3.71% 32.82% 5.45% (Continued on next page)

44 Table 3 (continued) Table 3 presents the results of logit and linear probability model regressions testing whether firms that experience natural disasters are more likely to meet-or-beat three earnings benchmarks. Panel A reports the results of the logic regression and Panel B report the results of linear probability model. The dependent variable, JustMoB equals one if firm’s reported earnings equal or exceed analyst forecast consensus by two cents (i.e., 0 ≤ forecast error ≤ 0.02). Otherwise, JustMoB equals zero if -0.02 ≤ forecast error <0. Zero_beat equals one if earnings per share before extraordinary item (Compustat data item epspx) is greater then zero, but less than ten cents (i.e., 0 ≤ epspxt < 0.10). Otherwise, Zero_beati,t equals zero if earnings per share before extraordinary item is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ epspxt < 0). Last_beat equals one if the change of earnings per share between current and the previous year is greater than zero, but less than ten cents. (i.e., 0 ≤ actualt - actualt-1 ≤ 0.10). Otherwise, Last_beati,t equals zero if the change of earnings per share between current and the previous year is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ actualt – actual t-1 < 0). Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. The standard errors are clustered at firm level and are also robust to heteroscedasticity. All of other variables are defined in the appendix. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

45 Table 4 Test of Non-GAAP Exclusions

Variable Excl_usei,t Excl_mgmti,t 0.377*** 0.334*** Disaster (4.25) (3.17) 0.051 0.021 Firmsize (0.83) (0.29) 0.816*** 0.794*** BTM (6.35) (5.16) -0.070 -0.087 Growth (-0.66) (-0.69) -8.324*** -10.071*** ROA (-12.38) (-11.74) 0.879*** 1.058*** Profitable (7.13) (6.86) -1.820*** -2.081*** RD (-2.98) (-2.77) 0.258 0.492* Labor (1.05) (1.68) 4.787*** 5.484*** CFO (7.86) (7.20) 0.256*** 0.314*** Logshares (3.46) (3.65) -0.022 -0.020 NOA (-0.87) (-0.66) -3.810*** -4.372*** CFVOL (-4.40) (-4.20) 0.020** 0.023** Numest (2.32) (2.31) 0.095 0.090 BigN (0.97) (0.78) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 4,960 3,675 Adj. R2 18.27% 21.03% Table 4 presents the results of a logit regression that tests whether firms affected by natural disasters are more likely to meet analyst forecasts using non-GAAP exclusion. The dependent variable, Excl_use is the use of GAAP exclusions indicator that equals one if I/B/E/S Actual EPS exceeds COMPUSTAT EPS, and zero otherwise. Excl_mgmt equals one if a firm meets-or-beats expectation but would have missed if exclusions are removed from earnings, and zero otherwise. Excl_mgmt is only defined when firms just meet analyst forecasts (i.e., JustMoB=1). Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. The standard errors are clustered at firm level and are also robust to heteroscedasticity. All of other variables are defined in the appendix. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

46 Table 5 Test of Expectation Management

Variable Exp_mgmti,t Walkdowni,t 0.120 0.001 Disaster (1.28) (0.22) -0.204*** 0.015*** Firmsize (-3.25) (2.96) 0.722*** -0.030** BTM (5.19) (-2.40) -0.761*** 0.034*** Growth (-4.92) (4.23) -1.077*** 0.156*** ROA (-2.48) (3.03) -0.523*** 0.042*** Profitable (-4.07) (4.86) -0.205 0.023 RD (-0.40) (0.71) -0.174 0.013 Labor (-0.72) (0.79) 0.864* -0.098* CFO (1.75) (-1.79) 0.111 -0.019 Logshares (1.48) (-3.14) -0.037 0.004** NOA (-1.13) (2.12) -3.418*** 0.070 CFVOL (-4.02) (0.82) 0.005 -0.000 Numest (0.56) (-1.09) 0.082 0.007 BigN (0.80) (1.25) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 3,675 3,675 Adj. R2 9.48% 15.91% Table 5 presents the results that examine whether firms affected by natural disasters are more likely to meet analyst forecasts using expectations manipulation. The dependent variable, Exp_mgmt equals one if the firm would have missed analyst expectations using the first consensus forecast of the period but meets-or-beats analyst expectations using the last consensus forecast of the period, and zero otherwise. Walkdown indicates the difference between beginning consensus and last consensus of forecast. Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. The standard errors are clustered at firm level and are also robust to heteroscedasticity. All of other variables are defined in the appendix. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

47 Table 6 Test of Market Response to Beaters

Variable CAR_VW (-1,1) CAR_EW (-1,1) 0.000 0.001 Disaster (0.03) (0.36) 2.278** 2.357** Surprise (2.14) (2.25) 0.287 -0.016 Disaster*Surprise (0.22) (-0.01) 0.002*** 0.003*** Firmsize (2.81) (2.96) 0.011** 0.010** BTM (2.43) (2.39) -0.007 -0.008 Growth (-1.15) (-1.31) -0.000 -0.000 Accruals (-0.26) (-0.04) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 3,675 3,675 Adj. R2 8.63% 8.61% Table 6 presents the results of a pooled cross-sectional OLS regression that tests whether markets discount firms that use income-increasing exclusions to meet analyst forecasts during the disaster years. The dependent variable, CAR_VW (-1,1) is equal to three-day cumulated abnormal returns of value-weighted portfolios centered on the earnings announcement date. CAR_EW (-1,1) is equal to three-day cumulated abnormal returns of equally-weighted portfolios centered on the earnings announcement date. Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. The standard errors are clustered at firm level and are also robust to heteroscedasticity. All of other variables are defined in the appendix. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

48 Table 7 Test of Earnings Management

Panel A: Natural disaster and Accruals Management

Variable Acc_mgmti,t Accrualsi,t 0.101 -0.077 Disaster (1.00) (-0.31) 0.013 -0.619*** Firmsize (0.19) (-3.69) -0.024 -1.024*** BTM (-0.17) (-3.07) -0.278** 2.381*** Growth (-2.01) (5.45) 12.092*** -1.752 ROA (7.83) (-1.00) 0.557*** 0.374 Profitable (4.17) (1.03) -0.267 4.082 RD (-0.40) (1.63) -0.680** 1.581** Labor (-2.38) (2.21) -15.902*** 1.562 CFO (-10.76) (0.66) -0.107 0.237 Logshares (-1.39) (1.24) 0.043 -0.175** NOA (1.28) (-2.38) -0.910 8.304*** CFVOL (-1.06) (2.97) -0.000 -0.004 Numest (-0.01) (-0.20) -0.071 -0.614** BigN (-0.68) (-1.98) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 3,675 3,675 Adj. R2 21.03% 16.81% (Continued on next page)

49 Table 7 (continued)

Panel B: Natural Disaster and Real Earnings Management

Variable REM_mgmti,t REMi,t 0.226 0.000 Disaster (1.06) (0.02) -0.582*** -0.009 Firmsize (4.89) (-0.65) 0.134 -0.015 BTM (0.62) (0.658) 0.027 -0.045 Growth (0.11) (-1.06) 1.110*** 0.149 ROA (1.03) (0.75) -5.068*** 0.008 Profitable (-19.21) (0.26) -5.029* -0.917*** RD (-1.79) (3.99) -0.495 -0.073* Labor (-1.25) (1.87) -8.290*** -0.800*** CFO (-6.92) (-4.65) 0.528*** -0.009 Logshares (3.78) (-0.62) -0.159* -0.014* NOA (-1.93) (-1.78) -1.899 8.304*** CFVOL (-0.73) (2.97) 0.210 0.000 Numest (1.19) (-0.30) 0.027 -0.004** BigN (0.12) (-0.15) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 3,675 3,675 Adj. R2 10.03% 5.58% (Continued on next page)

50 Table 7 (continued) Table 7 presents the results of analyses that tests whether firms affected by natural disasters are more likely to meet analyst forecasts using earnings management. Panel A present the results for the association between natural disasters and accruals management. The dependent variable, Acc_mgmt equals one if a firm meets or beats expectation but would have missed if accruals are removed from earnings, and zero otherwise. Acc_mgmt is only defined when firms just meet analyst forecasts (i.e., JustMoB=1). Accruals is abnormal accruals calculated using the performance-adjusted modified Jones model (Jones 1991; and Kothari et al. 2005). Panel B present the results for the association between natural disasters and real earnings management. The dependent variable, REM_mgmt equals one if a firm meets-or-beats expectation but would have missed if the amounts of real earnings management are removed from earnings, and zero otherwise. REM_mgmt is only defined when firms just meet analyst forecasts (i.e., JustMoB=1). REM is the sum of the three measures of real earnings management, abnormal cash from operations, abnormal expenditures, and abnormal production of inventory (Badertscher 2011). Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. All of other variables are defined in the appendix. The standard errors are clustered at firm level and are also robust to heteroscedasticity. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

51 Table 8 Test of Entire Distribution of Earnings Forecasts

Variable MEETi,t Zero_EMi,t EA_increasei,t 0.107** -1.968*** -0.192*** Disaster (1.81) (-24.93) (-5.05) -0.361*** 0.572*** 0.024 Firmsize (-11.13) (11.48) (0.99) -0.602*** -0.309*** -0.549*** BTM (-8.98) (-3.58) (-12.06) -0.015 0.311*** 0.978*** Growth (-0.25) (3.26) (15.06) 1.035*** 1.763*** ROA (3.88) (0.000) 0.665*** 4.377*** 0.809*** Profitable (8.18) (39.30) (17.06) -0.015 -6.609*** 0.571** RD (-0.06) (-9.81) (2.55) 0.284** 0.575*** -0.184** Labor (2.32) (2.93) (-2.16) -0.134 7.096*** -0.147 CFO (-0.46) (15.07) (-0.69) 0.514*** -0.864*** -0.012 Logshares (12.97) (-13.95) (-0.40) 0.028* -0.060** -0.041*** NOA (1.80) (-2.14) (-3.24) -0.581 1.959** 3.298 CFVOL (-1.18) (0.034) (8.65) 0.013*** -0.029*** 0.003 Numest (3.08) (-4.37) (0.89) -0.104* -0.184** 0.035 BigN (-1.64) (-2.10) (0.84) Year-fixed effect Yes Yes Yes Industry-fixed effect Yes Yes Yes Observation 19,345 19,345 19,345 Adj. R2 5.65% 63.24% 11.79% Table 8 presents the results of a logistic regression testing whether firms affected by natural disasters are more likely to meet-or-beat analyst forecasts, using the entire distribution of earnings forecasts. The dependent variable MEET equals one if actual I/B/E/S EPS exceed analyst forecast consensus by one cent (i.e. 0 ≤ forecast error ≤ 0.01), and zero otherwise. Zero_EM equals one if earnings per share before extraordinary item is greater than zero, and zero otherwise. EA_increase equals one if current year’s earnings per share is greater than the last year’s earnings per share (i.e., I/B/E/S actual t-1 < I/B/E/S actual t), and zero otherwise. Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. All of other variables are defined in the appendix. The standard errors are clustered at firm level and are also robust to heteroscedasticity. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

52 Table 9 Natural Disaster and Beating Analyst Forecasts

Logit LPM

Variable BIG_BEATit BIG_BEAT,t -0.078** -0.018** Disaster (1.97) (-1.98) 0.282*** 0.063*** Firmsize (11.09) (11.36) 0.156*** 0.033*** BTM (3.58) (3.58) 0.152*** 0.034*** Growth (3.62) (3.62) -0.337** -0.074** ROA (-2.12) (-2.12) 0.478*** 0.105*** Profitable (9.50) (9.59) 0.325* 0.072* RD (1.74) (1.77) -0.063 -0.014 Labor (-0.64) (-0.63) 0.076 0.020 CFO (0.41) (0.48) -0.334*** -0.075*** Logshares (-10.82) (-11.04) -0.049*** -0.011*** NOA (-4.03) (-4.35) 1.051*** 0.231*** CFVOL (3.11) (3.06) -0.001 -0.000 Numest (-0.15) (-0.13) 0.189*** 0.043*** BigN (4.34) (4.38) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 19,345 19,345 Adj. R2 3.84% 5.01% This table presents the results of a logit and linear probability model regression testing whether firms that experience natural disasters are more likely to beat analyst forecasts by a large amount. Column 1 reports the results of the logic regression and column 2 reports the results of the linear probability model. The dependent variable BIG_BEAT equals one if actual I/B/E/S EPS exceeds analyst forecast consensus by two cents (i.e. forecast error > 0.02), and zero otherwise. Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. All of other variables are defined in the appendix. The standard errors are clustered at firm level and are also robust to heteroscedasticity. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

53 Table 10 Test of Using a Sample of Small Firms

Variable JustMOBi,t Zero_beati,t Last_beati,t 0.202** -0.825*** 0.016 Disaster (2.03) (-2.69) (0.90) -0.050 0.410 -0.012 Firmsize (-0.43) (1.22) (-0.54) 0.063 -8.295** -0.048 BTM (0.38) (-1.95) (-1.53) 0.210 0.394*** 0.089*** Growth (1.36) (-3.67) (3.49) 0.311 -0.279** ROA (0.47) (-2.50) 0.601*** 0.394 0.205*** Profitable (3.91) (1.31) (6.68) 0.601 -5.392* 0.033 RD (0.63) (-1.67) (0.28) -0.317 0.610 0.043 Labor (-1.28) (0.84) (0.89) -0.399 1.377 0.072 CFO (0.58) (0.62) (0.57) 0.095 -0.276 -0.009 Logshares (1.12) (-1.27) (-0.53) 0.030 0.190 0.005 NOA (0.57) (1.34) (0.66) 1.777 -2.267 0.364 CFVOL (1.20) (-0.59) (1.51) -0.072*** -0.056 -0.009*** Numest (-5.43) (-1.47) (-3.16) -0.328*** -0.522 -0.017 BigN (-2.58) (-1.37) (-0.69) Year-fixed effect Yes Yes Yes Industry-fixed effect Yes Yes Yes Observation 2,620 380 1,423 Adj. R2 5.33% 18.13% 7.53% (Continued on next page)

54 Table 10 (continued) Table 10 presents the results of logistic regressions using a sample of small firms to test whether firms that experience natural disasters are more likely to meet-or-beat three earnings benchmarks. The dependent variable, JustMoB equals one if firm’s reported earnings equal or exceed analyst forecast consensus by two cents (i.e., 0 ≤ forecast error ≤ 0.02). Otherwise, JustMoB equals zero if -0.02 ≤ forecast error <0. Zero_beat equals one if earnings per share before extraordinary item (Compustat data item epspx) is greater then zero, but less than ten cents (i.e., 0 ≤ epspxt < 0.10). Otherwise, Zero_beati,t equals zero if earnings per share before extraordinary item is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ epspxt < 0). Last_beat equals one if the change of earnings per share between current and the previous year is greater than zero, but less than ten cents. (i.e., 0 ≤ actualt - actualt- 1 ≤ 0.10). Otherwise, Last_beati,t equals zero if the change of earnings per share between current and the previous year is less than zero, but greater than negative ten cents (i.e., -0.10 ≤ actualt – actual t-1 < 0). Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. The standard errors are clustered at firm level and are also robust to heteroscedasticity. All of other variables are defined in the appendix. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests.

55 Table 11 Test of Using Non-GAAP Earnings Disclosed by Managers

Variable MOB_MGRi,t MGR_Excli,t 0.222** 0.172** Disaster (2.12) (1.98) 0.019 -0.025 Firmsize (0.33) (-0.45) 0.034 -0.05 BTM (0.94) (-1.22) -0.002*** 0.001*** Growth (-4.49) (2.65) 0.213 -3.91*** ROA (0.82) (-7.13) 1.076*** 0.736*** Profitable (7.38) (6.00) 0.821 -1.252 RD (0.89) (-1.47) 0.053 0.824*** Labor (0.28) (2.86) 0.545 1.411** CFO (0.94) (2.47) 0.014 0.242*** Logshares (0.21) (3.52) 0.001 0.418 NOA (0.09) (1.61) 1.812* -4.659*** CFVOL (1.86) (-4.12) 0.001 0.015* Numest (0.19) (1.92) -0.141 0.369** BigN (-0.79) (2.52) Year-fixed effect Yes Yes Industry-fixed effect Yes Yes Observation 3,916 7,342 Adj. R2 6.53% 14.67% Table 11 reports the test of using non-GAAP EPS disclosed in firm’s earnings announcement (i.e., pro-forma earnings). MOB_MGR equals one if firm’s reported earnings equal or exceed analyst forecast consensus, and zero otherwise. MGR_Excl equals one if the earnings announcement contains a non-GAAP EPS disclosure, and zero otherwise. Main independent variable, Disaster is an indicator variable that equals one if a firm experiences a natural disaster in year t, and zero otherwise. All of other variables are defined in the appendix. The standard errors are clustered at firm level and are also robust to heteroscedasticity. *, **, and *** indicate statistical significance at the 0.10, 0.05, and 0.01 levels, respectively, based on two-tailed tests

56 REFERENCES

Abarbanell, J., and R. Lehavy. 2003. Biased forecasts or biased earnings? The role of reported earnings in explaining apparent bias and over/underreaction in analysts’ earnings forecasts. Journal of Accounting and Economics 36 (1–3): 105–46.

Badertscher, B. A. 2011. Overvaluation and the choice of alternative earnings management mechanisms. The Accounting Review 86 (5): 1491–518.

Barth, M., J. Elliot, and M. Finn. 1999. Market rewards associated with patterns of increasing earnings. Journal of Accounting Research 37 (2): 387–413.

Bartov, E., Givoly, D., and Hayn, C. 2002. The rewards to meeting or beating earnings expectations. Journal of Accounting and Economics, 33(2), 173-204.

Barua, A., Lin, S., and Sbaraglia, A. M. 2010. Earnings management using discontinued operations. The Accounting Review, 85(5), 1485-1509.

Bentley, J. W., Christensen, T. E., Gee, K. H., & Whipple, B. C. 2018. Disentangling Managers’ and Analysts’ Non‐GAAP Reporting. Journal of Accounting Research, 56(4), 1039-1081.

Black, E. L., Christensen, T. E., Taylor Joo, T., and Schmardebeck, R. 2017. The relation between earnings management and non‐GAAP reporting. Contemporary Accounting Research, 34(2), 750-782.

Bradshaw, M. T., Lee, L. F., and Peterson, K. 2016. The interactive role of difficulty and incentives in explaining the annual earnings forecast walkdown. The Accounting Review, 91(4), 995- 1021.

Bradshaw, M. T., Christensen, T. E., Gee, K. H., and Whipple, B. C. 2018. Analysts’ GAAP earnings forecasts and their implications for accounting research. Journal of Accounting and Economics, 66(1), 46-66.

Burgstahler, D., and Dichev, I. 1997. Earnings management to avoid earnings decreases and losses. Journal of accounting and economics, 24(1), 99-126.

Burgstahler, D., and M. Eames. 2006. Management of earnings and analysts’ forecasts to achieve zero and small positive earnings surprises. Journal of Business Finance & Accounting 33 (5):633–52.

Cheng, Y., Park, J., Pierce, S., and Zhang, T. 2018. Disaster Cleanup: Big Bath Accounting following Natural Disasters. Working paper.

Cheng, Q., and T. Warfield. 2005. Equity incentives and earnings management. The Accounting Review 80 (2): 441–76.

57 Chaney, T. , Sraer, D. , and Thesmar, D. 2012. The collateral channel: how real estate shocks affect corporate investment. American Economic Review 102, 2381–2409.

Christensen, T. E., Merkley, K. J., Tucker, J. W., and Venkataraman, S. 2011. Do managers use earnings guidance to influence street earnings exclusions? Review of Accounting Studies, 16(3), 501-527.

Christensen, T. E., Paik, G. H., and Stice, E. K. 2008. Creating a bigger bath using the deferred tax valuation allowance. Journal of Business Finance & Accounting, 35(5‐6), 601-625.

Cohen, D. A., A. Dey, and T. Z. Lys. 2008. Real and accrual-based earnings management in the pre- and post-Sarbanes-Oxley periods. The Accounting Review 83 (3): 757–87

Curtis, A. B., McVay, S. E., and Whipple, B. C. 2013. The disclosure of non-GAAP earnings information in the presence of transitory gains. The Accounting Review, 89(3), 933-958.

Davis, L., B. Soo, and G. Trompeter. 2009. Auditor tenure and the ability to meet or beat earnings forecasts. Contemporary Accounting Research 26 (2): 517–48.

Dechow, P. M., Richardson, S. A., and Tuna, I. 2003. Why are earnings kinky? An examination of the earnings management explanation. Review of Accounting Studies, 8(2-3), 355-384.

Degeorge, F., J. Patel, and R. Zeckhauser. 1999. Earnings management to exceed thresholds. Journal of Business 72 (1): 1–33.

Desai, M. A., Foley, C. F., and Hines, J. R. 2009. Domestic effects of the foreign activities of US multinationals. American Economic Journal: Economic Policy, 1(1), 181-203.

Dessaint, O., and Matray, A. 2017. Do managers overreact to salient risks? Evidence from hurricane strikes. Journal of Financial Economics, 126(1), 97-121.

Doyle, J., Lundholm, R., and Soliman, M.,2003. The predictive value of expenses excluded from ‘pro forma’ earnings. Review of Accounting Studies 2–3, 145–174.

Doyle, J., J. Jennings, and M. Soliman. 2013. Do managers define non-GAAP earnings to meet or beat analyst forecasts? Journal of Accounting and Economics 56 (1): 40–56.

Fama, E. F., and K. R. French 1997. Industry Costs of Equity. Journal of Financial Economics 4 3: 153–93.

Farrell, G., 2001. Companies cite attacks in earnings warnings. USA Today (September 27), B01.

Feng, M., and S. McVay. 2010. Analysts’ incentives to overweight management guidance when revising their short-term earnings forecasts. The Accounting Review 85 (5): 1617–46.

58 Francis, J., and D. Philbrick. 1993. Analysts’ decisions as products of a multi-task environment. Journal of Accounting Research 31 (2): 216–30.

Frankel, R., M. Johnson, and K. Nelson. 2002. The relation between auditors’ fees for nonaudit services and earnings management. The Accounting Review 77 (s-1): 71–105.

Freeman, R. N. 1987. The association between accounting earnings and security returns for large and small firms. Journal of Accounting and Economics, 9(2), 195-228.

Filzen, J. J., and Peterson, K. 2015. Financial statement complexity and meeting analysts’ expectations. Contemporary Accounting Research, 32(4), 1560-1594.

Graham, J., C. Harvey, and S. Rajgopal. 2005. The economic implications of corporate financial reporting. Journal of Accounting and Economics 40 (1–3): 3–73.

Groysberg, B., P. Healy, and D. Maber. 2011. What drives sell-side analyst compensation at high status investment banks? Journal of Accounting Research 49 (4): 969–1000.

Gu, Z., and Chen,T. 2004. Analysts' treatment of nonrecurring items in street earnings. Journal of Accounting and Economics 38,129–170.

Gu, F., and W. Wang. 2005. Intangible assets, information complexity, and analysts’ earnings forecasts. Journal of Business Finance & Accounting 32 (9): 1673–702.

Gunny, K. A. 2010. The relation between earnings management using real activities manipulation and future performance: Evidence from meeting earnings benchmarks. Contemporary Accounting Research, 27(3), 855-888.

Haggard, K. S., Howe, J. S., & Lynch, A. A., 2015. Do Baths Muddy the Waters or Clear the Air? Journal of Accounting and Economics 59, 105-117.

Hayn, C. 1995. The information content of losses. Journal of Accounting and Economics 20 _2_: 125–153.

Herrmann, D., O. Hope, J. L. Payne, and W. B. Thomas. 2011. The market’s reaction to unexpected earnings thresholds. Journal of Business Finance & Accounting 38 (1): 34– 57.

Holder, A. D., Karim, K. E., and Robin, A. 2012. Was Dodd-Frank justified in exempting small firms from Section 404b compliance? Accounting Horizons, 27(1), 1-22.

Hong, H., and J. Kubik. 2003. Analyzing the analysts: Career concerns and biased earnings forecasts. The Journal of Finance 58 (1): 313–51.

Huang, S. X., Pereira, R., & Wang, C. 2017. Analyst coverage and the likelihood of meeting or beating analyst earnings forecasts. Contemporary Accounting Research, 34(2), 871-899.

59 Jones, J. 1991. Earnings management during import relief investigations. Journal of Accounting Research 29 (2): 193–228.

Joo, J. H., and Chamberlain, S. L. 2017. The Effects of Governance on Classification Shifting and Compensation Shielding. Contemporary Accounting Research.

Kent, P., Monem, R., and Cuffe, G., 2008. Droughts and big baths of Australian agricultural firms. Pacific Accounting Review 20, 215-233.

Kothari, S. P., A. Leone, and C. Wasley. 2005. Performance matched discretionary accrual measures. Journal of Accounting and Economics 39 (1): 163–97.

Krishnan, G. V., and Yu, W. 2012. Do small firms benefit from auditor attestation of internal control effectiveness? Auditing: A Journal of Practice & Theory, 31(4), 115-137.

Lehavy, R., F. Li, and K. Merkley. 2011. The effect of annual report readability on analyst following and the properties of their earnings forecasts. The Accounting Review 86 (3): 1087–115.

Lev, B., and Penman, S. H. 1990. Voluntary forecast disclosure, nondisclosure, and stock prices. Journal of Accounting Research, 28(1), 49-76.

Levitt, A. 1998. The numbers game. Remarks delivered at New York University Center for Law and Business, New York, NY. September 28. Available at: https://www.sec.gov/news/speech/speecharchive/1998/spch220.txt

Libby, R., J. Hunton, H. Tan, and N. Seybert. 2008. Relationship incentives and the optimistic/ pessimistic pattern in analysts’ forecasts. Journal of Accounting Research 46 (1): 173–98.

Lopez, T., and L. Rees. 2002. The effect of beating and missing analysts’ forecasts on the information content of unexpected earnings. Journal of Accounting, Auditing, and Finance 17 (2): 155–84.

Matsumoto, D. 2002. Management’s incentives to avoid earnings surprises. The Accounting Review 77 (3): 483–514.

Matsunaga, S., and C. Park. 2001. The effect of missing a quarterly earnings benchmark on the CEO’s annual bonus. The Accounting Review 76 (3): 313–32.

Mergenthaler, R., S. Rajgopal, and S. Srinivasan. 2012. CEO and CFO career penalties to missing quarterly analysts forecasts. Working paper, Social Services Research Network. Available at http://ssrn.com/abstract=1152421

Mikhail, M., B. Walther, and R. Willis. 1999. Does forecast accuracy matter to security analysts? The Accounting Review 74 (2): 185–200.

60 Petersen, M. 2009. Estimating standard errors in finance panel data sets: Comparing approaches. The Review of Financial Studies, 22(1), 435-480.

Prawitt, D., J. Smith, and D. Wood. 2009. Internal audit quality and earnings management. The Accounting Review 84 (4): 1255–80.

Teoh, S. H., Welch, I., and Wong, T. J. 1998. Earnings management and the underperformance of seasoned equity offerings1. Journal of Financial economics, 50(1), 63-99.

Waymire, G. 1985. Earnings volatility and voluntary management forecast disclosure. Journal of Accounting Research, 268-295.

You, H., and X. Zhang. 2009. Financial reporting complexity and investor underreaction to 10-K Information. Review of Accounting Studies 14 (4): 559–86.

Yu, F. 2008. Analyst coverage and earnings management. Journal of Financial Economics 88 (2): 245–71.

Zang, A. 2012. Evidence on the trade-off between real activities manipulation and accrual-based earnings management. The Accounting Review 87 (2): 675–703.

61 BIOGRAPHICAL SKETCH

Jonghan Park, was born in Seoul, South Korea in 1985. The majority of his youth was spent in Seoul. After he graduated high school in Korea, he decided to study abroad in the United States.

He attended Brigham Young University and earned a bachelor’s degree in Economics in 2013. In

2014, Jonghan decided to pursue a doctoral degree at Florida State University. During his time at

Florida State University, he has had the opportunity to work with leading scholars in the fields of accounting and management and has taught several courses of Cost Accounting. Jonghan will begin his academic career as an Assistant Professor in Accounting at the Chinese University of

Hong Kong, Shenzhen in July of 2019.

62