Distress risk and earnings management

Transcription

1 Distress risk and earnings management Zhongling Qin a *, Xiao Ren b a University of Georgia, Athens GA USA b University of Georgia, Athens GA USA (First version: January 2017) Abstract We provide evidence that distress risk increases managers use of income-deflating discretionary accruals, and that the effect shows up and is magnified in the population of managers that decide to deflate earnings. Further empirical evidence shows that the increase in negative discretionary accruals when firms are in distress is 1) a reversal of income-inflating total accruals in previous years and 2) a switch to income-increasing real-activities earnings management from the equilibrium in the trade-off between accrual-based and real-activities earnings management in healthy times. In addition, ex ante monitoring reduces the effect of distress risk on discretionary accruals. This effect is distinct from the predictions of debt-covenant hypothesis in accounting literature. JEL Classification: G3; G33; M4; M41 Keywords: Earnings management; Distress risk We are grateful for help from Jack He, Celim Yildizhan, Bryan Oh, Huan Yang, and for helpful comments from participants in seminar at University of Georgia. The authors are responsible for any errors or omissions. *Corresponding author. zq04523@uga.edu 1

2 The characterization of distress risk and earnings management is a natural line of questioning because financial distress changes the expected payoffs of all claimants involved. Distress-related events such as suboptimal fire sales of assets, extra administration and litigation, and protracted and conflicted debt renegotiations may all detract from an efficient emergence by a firm from distress. Government in seeking tax payments, employees in seeking wages, debtholders seeking payment, and equity holders looking to value what s remaining are agents who have a stake in the value of a firm after the resolution of distress. Under the threat of a costly bankruptcy, managers have the incentives to 1) change their financial reporting choices and 2) operate the firm differently, both of which affect a firm s reported earnings. Therefore, this paper seeks to empirically study the behavior of earnings management during ex-ante distressed periods. When studying the relation between financial distress and earnings management, much research attention has focused on earnings behavior around the event of debt covenant violations, otherwise referred to as technical defaults. For example, DeFond and Jiambalvo (1994), Sweeney (1994), and DeAngelo et al. (1994) all examine earnings around debt covenant violations. Generally, as the authors themselves acknowledged, these early findings on this topic suffer from small sample sizes and selection bias for a variety of reasons 1. In addition, these studies present mixed results on the direction of the earnings behavior and posit varying theories explaining the direction of the earnings behavior. More recent and statistically powerful studies such as Dichev and Skinner (2002), Jaggi and Lee (2002), and Jha (2013) remain focused on the event of debt covenant violations but also find mixed results on the direction of earnings behavior. 1 For example, as Defond and Jiambalvo (1994) acknowledge, these measures create sample selection bias, since the firms in the sample are known to have violated debt covenants or filed bankruptcy, and the firms that successfully avoided violation/bankruptcy by manipulation are omitted from the sample. 2

3 We argue that debt covenant violation events are not sufficient for capturing the distress risk of firms for three reasons. First, debt covenant violations often occur outside of financial distress. Direct survey evidence from Gopalakrishnan and Parkash (1995) find that upon violation, less than 1-2% of the time do debt service default (bankruptcy) or acceleration of loan terms happen. In fact, upon these technical defaults, 93% of responding lenders do not regard such events as dangers to a firm s ability to service its debt. Second, debt covenants are written for a myriad of purposes such as influencing firm investment, disciplining management in normal times, or information sharing (Tirole 2010). As such, debt covenants are only among many channels for management motivations to lead to accounting choices, and are not necessarily how a manager might end up manipulating earnings when the firm is under high perceived financial distress. So, debt covenants represent a well-studied channel that affects earnings management, while we study ex-ante financial distress as another distinctive channel that affects earnings management. Third, our study of earnings management behavior can be more generally interpreted as an ad-hoc model for the conditional expectation of earnings management, in which we condition on financial distress as part of a manager s decision making information set. For our question, it is pertinent and required that the information on the state of financial distress be an ex-ante measure. In terms of decision making realism, a manager cannot make an earnings decision based on hindsight (expost) knowledge of a covenant violation, but must rely on an expectation of financial distress conditional on his contemporaneous information. So in the context of studying financial distress, we argue that a more sufficient and robust measure for capturing distress risk is needed. Given the above arguments, previous studies focusing on debt covenant violations may or may not select into a sample that contains simultaneously highly distressed firms, and are possibly not equilibrium evidence for the question of earnings management behavior under financial 3

4 distress. We attempt to remedy aforementioned selection issues as well as clarify the mixed evidence on the subject by empirically analyzing a large Compustat/CRSP sample to achieve evidence of overall equilibrium outcomes of earnings behavior under distress risk. We use the Merton (1973) option view of the firm to construct a measure to proxy for distress risk. Such a measure is appealing for three reasons. First, the measure has the favorable interpretation as an exante probability of bankruptcy, which will not suffer from selection bias related to debt covenant violations. Second, the wide availability of the option inputs relieves data constraints on sample size, which will better approximate the overall population and equilibrium outcomes. Finally, such a measure is popularly used in the finance literature for its return implications, and has been found to favorably predict bankruptcy out of sample (Bharath and Shumway 2008). To capture managers earnings management behavior, we use measures of discretionary accruals as calculated by the Jones model and the modified Jones model. Following Kothari et al. (2005), we choose to include an intercept in order to ensure that the error terms of the models are on average zero, which better avoids misspecification. The results show that there is a significant association between earnings management and distress risk. An increase in distress risk leads to a decrease in signed discretionary accruals and an increase in the absolute value of discretionary accruals. We then separate the sample into a subsample of firms with positive discretionary accruals and another subsample of firms with negative discretionary accruals and find that distress risk increases the magnitude of both positive and negative discretionary accruals. One of the potential concerns of the Merton (1973) model that generates our probability of default measure is that it views firms with no debt as having no distress risk, which results in 38% of firm-year observations in our sample having zero probability of default. Therefore, we repeat the analysis with probability of default truncated at zero, and find 4

5 that the association between distress risk and earnings management weakens in the subsample with positive discretionary accruals, while the association in the negative subsample remains strong. Thus, when probability of default is non-trivially zero and if the manager decides to deflate earnings, an increase in distress risk is associated with a significant further deflation in earnings on average. Another important consideration is the manager s incentives to deflate earnings when the firm is in financial distress. Previous studies suggest several possible mechanisms. For example, DeAngelo et al. (1994) argue that managers of financially distressed firms deflate their earnings to give a signal to the board of directors acknowledging the current or expected financial difficulties and that they are willing to change strategy by streamlining operations. Jaggi and Lee (2002) suggest that managers may do so as part of creating leverage for creating better refinancing terms in debt renegotiations. Finally, Jha (2013) argues that managers tend to inflate earnings before debt covenant violation, and when violation occurs, they are forced to manage earnings downwards to reverse previously built-up positive discretionary accruals. We argue that managers in distressed firms manage earnings downwards because 1) they need to reverse the positive discretionary accruals accumulated in healthy times and 2) the higher scrutiny the firms face during financial distress makes it costlier to use accrual-based earnings management, forcing the managers to switch to the use of real activities earnings management. To test the first hypothesis, we include an interaction of probability of default with a dummy variable which equals to 1 if a firm has positive accruals in the past year in the cross-sectional analysis. Results show that the managers who have previously built up positive accruals manage their earnings downwards even more during financial distress. To test the second hypothesis, we use 5

6 the methodology of Zang (2012), and find that distress risk increases the use of real activities earnings management, substituting away from the use of accruals earnings management. Our research contributes to the literature in the following ways. First, we use a probability of default as the measure of distress risk, which 1) is an ex ante measure that can proxy for being a part of managers decision-making information set, 2) is robust to the selection bias issue as mentioned in Defond and Jiambalvo (1994) and several other papers, and 3) provides a much larger sample size for making inference on the population cross section of firms earnings management. The advantage of using probability of default enables us to comprehensively analyze in equilibrium how distress risk affects managers earnings management decision. Second, we examine the possible mechanisms of such phenomenon. Our empirical evidence is consistent with two explanations: the reversal story and the trade-off story. Finally, while our study complements and extends the debt covenant hypothesis literature, we show that the effect of probability of default on earnings management is distinct from the effect of debt covenant violation of earnings management. This paper is organized as follows: in section one, we review prior literature; in section two, we present our hypotheses; in section three, we describe our research design and explain our data sources; in section four, we present our main results; in section five, we do cross sectional analysis to provide evidence in support of two specific explanations; in section six, we conduct a robustness test; and in section seven, we make conclusions. 1. Prior Literature A large body in the accounting literature studies the debt covenant hypothesis which states that managers will make accounting choices to reduce the likelihood their firms will violate 6

7 accounting-based debt covenants (Dichev and Skinner 2002). While the prediction seems intuitive, the literature has not yet reached a consensus about the direction the managers in distressed firms tend to manage their earnings. Several papers document the finding that, consistent with positive accounting theory, firms approaching covenant violation will make income-increasing accounting choices to loosen their debt constraints (Watts and Zimmerman 1986). For example, Sweeney (1994) uses a sample of 130 firms which encountered first-time debt covenant violations to show that these firms adopt changes in accounting choices, such as the depreciation method or choice of inventory valuation, to increase their earnings. Defond and Jiambalvo (1994) show that firms use abnormal discretionary accruals before debt covenant violation to manage their earnings upwards. Dichev and Skinner (2002) use a larger sample to show that firms tend to slightly beat covenant targets, although they do not explicitly suggest that the managers use income-inflating accounting choices. Rosner (2003) acknowledges that bankruptcy and financial distress are distinct and categorized firms in a two by two sort of bankrupt/non-bankrupt and stress/non-stressed states. She shows that firms that are stressed and eventually go bankrupt use more income-inflating earnings management than non-stressed, non-bankrupt firms. Our study is similar to her study in the view that we are studying the effect of firms financial distress on managers earnings management behavior, as opposed to the effect of actual bankruptcy on earnings management. Our contribution above her study is that we formally calculate a probability of distress, which is continuous in nature, as opposed to her use of various operating performance metrics as proxies of distress, which also have potential selection-bias issues. Another segment of the literature argues that firms approaching debt covenant violation tend to manage their earnings downwards. For example, DeAngelo, DeAngelo and Skinner (1994) 7

8 document that 76 NYSE firms with persistent losses and dividend reductions show a decrease in total accruals. They argue that managers do so to acknowledge financial difficulties and release a signal that they are serious about streamlining the operation. Some other papers find mixed results. For instance, Jaggi and Lee (2002) show that managers who are able to obtain waivers for debt covenant violations use income-increasing accruals, while the managers whose waivers are denied use income-decreasing accruals. Jha (2013) finds that managers use income-increasing discretionary accruals in the quarters preceding a debt-covenant violation, but use income-decreasing discretionary accruals in the quarter a violation occurs. 2. Hypotheses We hypothesize that distress risk, the probability that a firm goes into debt-service bankruptcy, should influence a manager s earnings management behavior. Several papers show that, in healthy times, managers tend to manipulate earnings upwards (e.g., Bergstresser and Philippon (2004), Cheng and Warfield (2005)). When firms are in distress, managers may have even stronger incentives to inflate earnings to conceal poor financial performance. Managers may have such an incentive because their stock-based compensation is affected by markets reacting badly to lower-than-expected earnings or they fear of job/reputation loss. For example, Hotchkiss et al (2008) show that distress often can result in managers being fired. Managers may also deflate earnings for several reasons when firms are in distress. One reason is that distress risk increases the level of monitoring by parties involved in the resolution of financial distress. Several papers present evidence that lenders increase their monitoring on financially distressed firms. For instance, Lasfer, Sudarasanam and Taffler (1996) show that banks 8

9 play a significant role in monitoring the fire sales of financially distressed firms. Lee and Mullineaux (2004) show that syndicated loans are structured to enhance monitoring efforts and to facilitate renegotiation in the case of financial distress. Rosner (2003) has complementary evidence suggesting that auditors scrutinize a firm s financial statements more carefully when the firm in question is in distress. The motivation on the part of the firm s auditors is to avoid SEC sanctions if the firm goes bankrupt and reporting fraud or misstatements is discovered. Therefore, the increase in monitoring and attention may make managers reverse previous positive earnings management due to the increased likelihood of discovering earnings manipulation. A second reason is that the incentives for manipulating earnings upward remain but there is a trade-off between accruals manipulation and real activities management. Generally, accruals manipulation is easier to discover because it is based on simple changes in numbers; however, it is also easier to discover as many market agents and regulators can act to find evidence of abnormal accruals changes by reconciling accounting formulae. On the other hand, real activities management is harder to discover because it involves changing the way the firm is operated, and consequently, it is difficult to prove that the manager has the intention to manipulate over being incompetent. Zang (2012) provides some of the first evidence for the relative intensity of the scrutinization of accruals and real-activities earnings management. So in healthy times, there is an equilibrium between the two. Distress then will increase the likelihood of accruals manipulation discovery which shifts managers to increase the use of real-activities management. A final reason is argued by DeAngelo, DeAngelo, and Skinner (1994), stating that managers may deflate earnings to send a signal to outside stakeholders (e.g. labor unions, or debtholders) to convince them to give concessions during contract renegotiation. 3. Methods and Data Sources 9

10 Our sample starts with a panel of firms from 1975 to 2015 from the Compustat/CRSP merged database from which we get all the needed market and accounting information. Following, for example Zhang (2007), we exclude financial firms (6000<=SIC<=6999) and utility firms (4900<=SIC<=4999) because these firms face a different set of regulatory guidelines and therefore, conclusions about these firms accounting choices are not comparable. We require our main sample to have data available for calculating our earnings management measures and the probability of default. In general, we run regressions of the form: EMit = β0 + probdit*β1 + Controlsit*γ + εit, where EMit is an earnings management measure of a firm i at year t, probdit is the probability of default of a firm i at year t, Controlsit are a set of firm-year level controls known to affect EMit. We include year fixed-effects, industry fixed-effects based on SIC2, and all standard errors are robust through clustering at firm level. 3.1 Earnings management measures In constructing earnings management measures, we follow the descriptions of Dechow, Sloan, and Sweeney (1995). We construct two sets of measures of earnings management: discretionary accruals, and real activities. First, we define total accruals as any revenues and incurred expenses that impact the income statement. For discretionary accruals, we choose to calculate the Jones and modified Jones measures, where difference in the two measures is that the modification of total accruals includes accounts receivables. The portion of total accruals that managers do not have discretion over is the 10

11 portion that fluctuates with economic fundamentals of the firm. As such, the Jones procedure runs for each industry-year a procedure of the following form: TA it 1 = k Assets 1 + k i,t 1 Assets 2 i,t 1 ΔSALES it Assets i,t 1 + k 3 PPE it Assets i,t 1 + ε it (1) NA it = k 1 1 Assets i,t 1 + k 2 ΔSALES it Assets i,t 1 + k 3 PPE it Assets i,t 1 (2) DA it = TA it Assets i,t 1 - NA it (3) Discretionary accruals formed by Equation (3) are therefore the residuals of the regressions motivated by Equation (1), where the economic fundamentals are the changes in sales (ΔSalesit), and the plant, property, and equipment (PPEit) of firm i in year t. Discretionary accruals from modified Jones model is calculated by subtracting change in accounts receivable of firm i in year t from change in sales in Equation (2). Second, we define real activities as measures of how a manager may change the operation of the firm at his discretion. Roychowdhury (2006) presents three separate measures of such phenomenon: abnormal cash flows, abnormal production, and discretionary expenses. For example, a manager can temporarily increase the firm s cash flows by offering price discounts or more lenient credit terms, overproduce to reduce his average product costs, or cut discretionary expenses such as research and development. A similar procedure to the Jones procedure is run to calculate discretionary measures, replacing Equation (1) with each of the following models: CFO it 1 = k Assets 1 + k i,t 1 Assets 2 i,t 1 SALES it Assets i,t 1 + k 3 ΔSALES it Assets i,t 1 + ε it (4) PROD it 1 = k Assets 1 + k i,t 1 Assets 2 i,t 1 SALES it Assets i,t 1 + k 3 ΔSALES it Assets i,t 1 + k 4 ΔSALES i,t 1 Assets i,t 1 + ε it (5) 11

12 DISX it 1 SALES = k Assets 1 + k i,t 1 i,t 1 Assets 2 + ε i,t 1 Assets it (6) i,t 1 The procedure produces the residuals of equations (4)-(6), which are our variables of interest. For equations (1), and (4)-(6), we choose to include an intercept in order to ensure that the error terms of the models are on average zero, which better avoids misspecification. 3.2 Probability of default In estimating an ex-ante probability of default, we follow the iterative procedure outlined in Bharath and Shumway (2008). The option view of the firm makes two assumptions: the firm value follows a random walk with a trend, and that the firm possesses a zero-coupon bond that matures in one year. As a result, one can use observable values of total firm assets (V), time to maturity of one year (T), a risk free rate (r), historical volatility of log returns on total firm assets (σv), and debt (F) to infer a value of volatility of log returns using the Black Scholes formula. Because the Black Scholes formula assumes a constant volatility, and observed volatilities fluctuate too much to be constant, an iterative procedure is used to find a converged value of volatility and firm asset value. The probability of default then is calculated using observables and the converged estimates, as follows: DD = ln(v F )+(μ 0.5σ V 2 )T σ V T and π = N( DD) (7) 3.3 Control variables and monitoring proxies For control variables, we include four firm-year level variables that have been found to affect discretionary accruals: firm size, market to book, leverage, and return on assets (See for example, Dichev (1998), Fang, Huang, and Karpoff (2016), or Kothari, Leone, and Wasley (2005)). 12

13 Finally, we get monitoring proxies as follows: 1. Institutional ownership, defined as shares held (from Thomson Reuters 13f Filings) divided by shares outstanding (from CRSP) for a firm i over fiscal year t as the average over the four quarters 2. Short interest (from COMPUSTAT), defined as shares held short divided by number of shares outstanding for a firm i over fiscal year t 3. Analyst coverage (from IBES), computed as the natural log of 1 plus a firm i s 12-month average number of earnings forecasts in fiscal year t 3.4 Summary statistics Table 1 presents descriptive statistics for our measures. For our baseline study, we require a firm-year observation to have earnings management measures, probability of default, and all the control variables available. The full sample has 112,576 observations. In our sample, the total accruals tend to be slightly negative at -4.4% of total assets of the previous year, while the discretionary accruals and modified discretionary accruals are slightly positive at 2.4%. Though not statistically different from zero, it is suggestive that managers have a slight tendency overall to inflate earnings. The real activities measures come in 10.2%, -3.5%, and -22.5% of total assets of the previous year, for abnormal cash flow, abnormal production, and abnormal discretionary expenses, respectively. Again, it is suggestive that managers tend to inflate earnings through cash flow and discretionary expense management. Finally, probability of default averages 12.4%. This estimate closely matches that of the literature (For example, Bharath and Shumway (2008) find an average of 10.95%). 13

14 4. Main Results Table 2 presents our first result. In panel A, total accruals are down significantly at 1% level by an average of 3.4 percentage points of total assets of last year when firms distress probability goes up by 1 percentage point. The evidence that total accruals goes down with an increase in the probability of bankruptcy is consistent with economic intuition; that is, when a firm has a downturn in business performance leading to higher chances of bankruptcy, then reported earnings follows suit by going down. Both discretionary accruals and modified discretionary accruals are also down significantly by 1.3 and 2.2 percentage points on average when firms distress probability increases by 1 percentage point. Therefore, in comparison with the 3.4 figure decrease, we find a large portion (approximately 38% or 65%) of the decrease in total accruals due to an increase in distress risk is discretionary on the part of the manager. To get an impression of the magnitude of the effect (without considering the direction of the earnings management), we report in panel B the absolute value of the total and discretionary accruals. In panel B, both the absolute value of discretionary accruals and modified discretionary accruals increase significantly by 1.4 and 1.5 percentage points on average, respectively, when firms distress probability increases by 1 percentage point, while the absolute value of the total accruals decreases significantly by 2.5 percentages points of total assets of last year, when firms distress probability increases by 1 percentage point. Here then, the discretionary portion of the total accruals sensitive to an increase in distress risk is approximately 56% or 60%. Given that managers seem to manipulate earnings overall, with the direction being downward, when distress probability increases, next, we separate the discretionary accruals by positive and negative. The reason we do so is because positive and negative discretionary accruals have the interpretation of managerial inflation and deflation of earnings respectively. 14

15 Alternatively, we can interpret such separation as the final outcome of a confluence of incentives. In this view, conditional on a variety of incentives which matter for managers, some select into deciding to inflate earnings while the rest select into deciding to deflate earnings. As a result, two different populations of earnings management behavior emerge. So, in Table 3, the regressions designed in Table 2 is run, in which panel A contains the group with positive discretionary and modified discretionary accruals and panel B contains the group with negative discretionary and modified discretionary accruals. In panel A, we find that if discretionary accruals are positive, then a 1 percentage point increase in the probability of distress is associated with a 2.1 or 1.5 percentage point increase in discretionary accruals, or modified discretionary accruals, respectively. Conversely, in panel B, we find that if discretionary accruals are negative, then a 1 percentage point increase in the probability of distress is associated with a 3.8 or 4.1 percentage point decrease in discretionary accruals, or modified discretionary accruals, respectively. Taken together, we find that if managers decide to deflate (inflate) discretionary accruals, then distress risk magnifies that decision through further deflations (inflations). One potential mechanical issue with the method is that approximately 38% of the firmyears in the sample have zero probability of default because Merton s option model views firms with no debt as having no probability of default. As having zero probability of default ensures that a marginal increase in probability of default is still very close to or effectively interpreted as zero, what is being associated with distress probability may not actually be so. When a manager is sensitive to increases in financial distress, we expect that his decision-making will only be affected when the resulting level of financial distress from an increase in distress risk is non-trivial. Therefore, in Table 4, we repeat our previous analysis from Table 3, replacing on the left hand side of our specification in Table 3 with the negative and positive discretionary accruals after 15

16 truncating these zero default probability firm-years. Doing so, we see that the association between distress probability and discretionary accruals is unchanged in the negative side but is reduced in magnitude and weakened in significance in the positive side. If a manager decides to inflate earnings, then his discretionary accruals is less sensitive to the point of being unaffected by nontrivial increases in the probability of default. If a manager decides to deflate earnings, then his discretionary accruals remains very sensitive to non-trivial increases in probability of default. Therefore, our evidence indicates that when probability of default increases trivially from zero, managers preserve their motivation to inflate earnings. Effectively, when trivial increases from zero probability of default happen, the firm still appears to be healthy, and managers act as if in healthy times. When probability of default is non-trivially zero and managers want to deflate earnings, an increase in probability of default is associated with a significant deflation in earnings on average. 5. Cross-sectional Analysis Given that we find evidence that managers tend to deflate earnings associated an increase in distress probability, we use cross-sectional analysis to help distinguish among the three explanations of income-deflating earnings management given in the hypothesis section. Our first hypothesis is that distress risk has a more negative effect on a firm s discretionary accruals if the firm has previously built up positive accruals. Hence, if a manager has previously upwardly managed earnings, then to alleviate the likelihood of discovery during distress, the manager needs to downwardly manage earnings. The reason is that, once a firm has distress probability, concerns about the firm s survival and the manager s future contracted compensation may force the manager to truthfully report depressed earnings. To test this proposition, we classify 16

17 firms with an indicator, inflateit, if in the past 1 or 1 and 2 years they have accumulated positive total accruals. From table 5 panel A, the coefficient of interest is that of the interaction of inflateit and probability of default. When a firm has positive accruals in the previous year, the marginal interaction effect of a 1 percentage point increase in distress probability decreases the discretionary and modified discretionary accruals by a significant 4.9 and 4.6 percentage points, respectively. In panel B, conditional on accumulated accruals being positive in the previous two years, the marginal association of a 1 percentage point increase in distress probability is even higher. Thus, we conclude that there is evidence for the reversal story of earnings management. Managers may choose the optimal usage between accrual-based earnings management and real activities earnings management based on an evaluation of the relative costs between the two. Though we do not have an explicit figure of relative cost, we adopt the strategy of Zang (2012) which compares a balanced sample of firm-years based on the availability of calculating discretionary accruals and real activities measures. Zang (2012) argues that when firms are in distress, managers prefer using accrual-based manipulation to using real activities earnings management because the cost of deviating from regular operation is high when firms have poor financial health. However, it is possible that the increase in level of scrutiny firms face when they are in distress makes it costlier to use accrual-based manipulation. In this case, managers will shift to using real activities earnings management. In our analysis, we construct the balanced sample by requiring the sample to have data on both the accruals and the real-activities measures; the requirement ensures that the associations of distress risk on accruals and real activities earnings management concurrently shows up at the same time and over the same set of firms. We predict that because of the increased monitoring by stakeholders during distress, managers will prioritize 17

18 real activities manipulation over accrual-based earnings management, as posited in the hypothesis section. So in table 6, panels A and B, we observe that in the balanced sample, when firms undergo an increase in the distress probability, overall, discretionary accruals go down while real activities abnormal cash flows, and abnormal production go up, and abnormal expenses go down. As a result, we have suggestive evidence that there is a change in equilibrium between accrual-based and real activities earnings management during distress. The crux of our hypothesis is that an increase in the probability of bankruptcy leads to an increase in the monitoring activities by various market participants. For example, debtholders monitor due to the risk that the firm cannot pay back debt, or auditors monitor out of legal obligation and reputational concerns. Such monitoring forces a manager to reverse previously accumulated positive accruals and trades-off to manipulate earnings through real activities. If a firm ex-ante receives significant levels of monitoring by other market participants, then debtholders do not need to directly monitor managers. We report three sources of monitoring in the capital markets that have been shown to decrease earnings management. First, Chung, Firth, and Kim (2002) find that the presence of large institutional shareholdings inhibit managers from changing reported profits to a manager s desired level of profits. Effectively, large institutional holders often place directors onto the board or have indirect channels to management to allow for disciplining of managers. Second, Fang, Huang, and Karpoff (2016) find that the potential of short selling decreases earnings management because short sellers tend to be sophisticated investors such as hedge funds, who can use technical expertise, or industry channels to uncover earnings manipulation. Third, Yu (2008) finds that firms followed by more analysts manage their earnings less. Analysts with financial training, and firm background knowledge are able to monitor the firm 18

19 well, and analysts also interact with management and ask tough questions in earnings conferences to keep management more honest. In table 7, the coefficients of interest are the interaction terms between probability of default and the three monitoring proxies. We find statistically significant associations that the various monitoring measures oppose the effect of probability of default on the absolute value of discretionary accruals in panel A. Furthermore, in panel B, we find that the tendency by managers to deflate earnings is reversed by monitoring by institutional owners, short sellers, and analysts following the firm. In some cases, the sum of the main effect of probability of default and the interaction term reverses the overall direction of the effect. As a result, we find evidence that with monitoring, managers may tend to upwardly manage earnings, possibly to prevent bad earnings news from hitting the market. Overall, we interpret our cross-sectional evidence as more suggestive of an increased monitoring hypothesis resulting in trade-offs between accrual-based and real activities earnings management and reversals of previously accumulated positive earnings management. Therefore, this provides evidence against the explanation that managers deflate earnings to get concessions to make it easier to exit distress. The logic is that if monitoring is increased then managers do not need to release a signal to outside stakeholders because the monitors themselves are highly likely to release this information market-wide. 6. Robustness We conduct a robustness check against the debt-covenant hypothesis as one may argue that the main incentive by managers is to beat debt covenants. As mentioned previously, the literature points out that debt covenant violations often occur outside of financial distress and that lenders 19

20 do not regard these technical defaults as very serious. We complement these findings by controlling for actual and potential violations directly in our main specification. To do so, we use a publically available dataset of all debt covenant violations over 1996 to 2008, graciously released by Prof. Amir Sufi on his website. We define actual violations in a given firm-year if in that firmyear a debt covenant violation occurred in at least one of the quarters, and potential violations in a given firm-year if in the future three years, there was at least one debt covenant violation in at least one quarter. In Table 8, panels A and B, we see both probability of default and debt covenant violations as both motivating earnings management. However, in the interaction of probability of default and violations, we find an insignificant association. As a result, our evidence based on probability of default is a distinct effect from that of previous studies of debt covenant violation and earnings management. 7. Conclusion We provide evidence that managers downwardly manage earnings when ex-ante distress risk non-trivially increases. Such an association is ameliorated by sources of market monitoring such as institutional holders, short sellers, and analysts following the firm. Consequently, our evidence is most supportive of a monitoring-based explanation of distress risk. As distress risk increases, managers are under more monitoring by debt holders seeking surer payment. Therefore, managers tend to reverse previously accumulated positive earnings manipulation and shift to a harder-to-uncover real activities earnings management. Overall, we provide stronger evidence on the subject as well as clarify previous explanations through a larger sample and more selection bias robust design. 20

21 References: Chung, Richard, Michael Firth, and Jeong-Bon Kim. "Institutional monitoring and opportunistic earnings management." Journal of Corporate Finance 8.1 (2002): DeAngelo, Harry, Linda DeAngelo, and Douglas J. Skinner. "Accounting choice in troubled companies." Journal of Accounting and Economics (1994): Dichev, Ilia D. "Is the risk of bankruptcy a systematic risk?." The Journal of Finance 53.3 (1998): Dechow, Patricia M., Richard G. Sloan, and Amy P. Sweeney. "Detecting earnings management." The Accounting Review (1995): DeFond, Mark L., and James Jiambalvo. "Debt covenant violation and manipulation of accruals." Journal of Accounting and Economics (1994): Dichev, Ilia D., and Douglas J. Skinner. "Large sample evidence on the debt covenant hypothesis." Journal of Accounting Research 40.4 (2002): Fang, Vivian W., Allen H. Huang, and Jonathan M. Karpoff. "Short selling and earnings management: A controlled experiment." The Journal of Finance (2015). Gopalakrishnan, V., and Mohinder Parkash. "Borrower and lender perceptions of accounting information in." Accounting Horizons 9.1 (1995): 13. Habib, Ahsan, Borhan Uddin Bhuiyan, and Ainul Islam. "Financial distress, earnings management and market pricing of accruals during the global financial crisis." Managerial Finance 39.2 (2013): Hotchkiss, Edith S., et al. "Bankruptcy and the resolution of financial distress." Available at SSRN (2008). Jaggi, Bikki, and Picheng Lee. "Earnings management response to debt covenant violations and debt restructuring." Journal of Accounting, Auditing & Finance 17.4 (2002): Jha, Anand. "Earnings management around debt-covenant violations An empirical investigation using a large sample of quarterly data." Journal of Accounting, Auditing & Finance 28.4 (2013): Kothari, Sagar P., Andrew J. Leone, and Charles E. Wasley. "Performance matched discretionary accrual measures." Journal of Accounting and Economics 39.1 (2005): Lasfer, M. Ameziane, Puliyur S. Sudarsanam, and Richard J. Taffler. "Financial distress, asset sales, and lender monitoring." Financial Management (1996):

22 Lee, Sang Whi, and Donald J. Mullineaux. "Monitoring, financial distress, and the structure of commercial lending syndicates." Financial management (2004): Merton, Robert C. "Theory of rational option pricing." The Bell Journal of Economics and Management Science (1973): Rosner, Rebecca L. "Earnings manipulation in failing firms." Contemporary Accounting Research 20.2 (2003): Roychowdhury, Sugata. "Earnings management through real activities manipulation." Journal of Accounting and Economics 42.3 (2006): Bharath, Sreedhar T., and Tyler Shumway. "Forecasting default with the Merton distance to default model." Review of Financial Studies 21.3 (2008): Sweeney, Amy Patricia. "Debt-covenant violations and managers' accounting responses." Journal of Accounting and Economics 17.3 (1994): Tirole, Jean. The Theory of Corporate Finance. Princeton University Press, Yu, Fang Frank. "Analyst coverage and earnings management." Journal of Financial Economics 88.2 (2008): Zang, Amy Y. "Evidence on the trade-off between real activities manipulation and accrual-based earnings management." The Accounting Review 87.2 (2011): Zhang, X. Frank. "Accruals, investment, and the accrual anomaly." The Accounting Review 82.5 (2007):

23 Table 1 Summary Statistics Count Mean Std.Dev Min Q1 Median Q3 Max TA DA MDA acfo aprod adisexp probd size mb lev roa InstOwn ShortInt AnCov Table 1 presents the summary statistics for our sample of firms. TA, DA, and MDA represent total accruals, discretionary accruals (Jones model), and discretionary accruals (modified Jones model), respectively. acfo, aprod, and adisexp are abnormal cash flows from operations, abnormal production, and abnormal discretionary expenses, respectively. These six measures are all in units of percentage points of firm i's previous year total asset value. probd is the probability of default estimated by the procedure outlined in Bharath and Shumway (2008). The control variables are size (log of total assets), mb (market-to-book ratio), lev (firm leverage) and roa (return on assets). The InstOwn variable is defined as the shares held by institutional owners divided by shares outstanding for a firm i over fiscal year t as the average over the four quarters, the ShortInt variable is defined as the shares held short divided by number of shares outstanding for a firm i over fiscal year t, and the AnCov variable is defined as the natural log of 1 plus firm i's 12-month average number of earnings forecasts over fiscal year t. 23

24 Table 2 Panel A Accruals on Probability of Default (1) (2) (3) VARIABLES TA DA MDA probd *** ** *** (0.002) (0.005) (0.005) size *** *** *** (0.000) (0.001) (0.001) mb 0.000*** 0.001*** 0.001*** (0.000) (0.000) (0.000) lev *** 0.025*** 0.023*** (0.003) (0.007) (0.007) roa 0.362*** 0.400*** 0.372*** (0.011) (0.023) (0.025) Constant (0.005) (0.012) (0.013) Observations 112, , ,575 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes Panel B Absolute Value of Accruals on Probability of Default (1) (2) (3) VARIABLES TA_abs DA_abs MDA_abs probd 0.025*** 0.014*** 0.015*** (0.002) (0.004) (0.005) size *** *** *** (0.000) (0.001) (0.001) mb 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) lev 0.029*** 0.052*** 0.052*** (0.002) (0.006) (0.006) roa *** *** *** (0.009) (0.019) (0.020) Constant 0.068*** 0.105*** 0.111*** (0.005) (0.011) (0.010) Observations 112, , ,575 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes 24

25 25 This table reports OLS regression analysis of the association between earnings management and distress risk. In Panel A, the dependent variables in regressions 1-3 are total accruals, discretionary accruals (Jones model) and discretionary accruals (modified Jones model), respectively. In Panel B, the dependent variables in regressions 1-3 are the absolute value of total accruals, discretionary accruals (Jones model) and discretionary accruals (modified Jones model), respectively. probd is the probability of default estimated by the procedure outlined in Bharath and Shumway (2008). The control variables include size (log of total assets), mb (market-to-book ratio), lev (firm leverage) and roa (return on assets). The robust standard errors are reported in the parentheses. ***, **, and * represent significance levels at 0.01, 0.05, and 0.10, respectively.

26 Table 3 Panel A Positive Accruals on Probability of Default (1) (2) (3) VARIABLES TA DA MDA probd *** 0.015** (0.003) (0.008) (0.008) size *** *** *** (0.000) (0.001) (0.001) mb 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) lev *** 0.056*** (0.003) (0.009) (0.010) roa 0.169*** * *** (0.014) (0.036) (0.038) Constant 0.054*** 0.099*** 0.112*** (0.008) (0.015) (0.015) Observations 55,003 55,003 55,003 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes Panel B Negative Accruals on Probability of Default (1) (2) (3) VARIABLES TA DA MDA probd *** *** *** (0.003) (0.004) (0.004) size 0.006*** 0.009*** 0.009*** (0.000) (0.001) (0.001) mb *** ** ** (0.000) (0.000) (0.000) lev *** *** *** (0.004) (0.006) (0.006) roa 0.327*** 0.209*** 0.225*** (0.013) (0.017) (0.017) Constant *** *** *** (0.007) (0.016) (0.014) Observations 53,431 53,431 53,431 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes This table reports OLS regression analysis of the association between earnings management and distress risk for fims with income-inflating earnings management and income-deflating earnings management, separately. In Panel A, the sample comprises firms with income-inflating earnings management, defined as firms with positive discretionary accruals (both Jones and modified Jones model) in year t. In Panel B, the 26

27 27 sample comprises firms with income-deflating earnings management, defined as firms with negative discretionary accruals (both Jones and modified Jones model) in year t. In both Panels A and B, the dependent variables in regressions 1-3 are total accruals, discretionary accruals (Jones model) and discretionary accruals (modified Jones model), respectively. probd is the probability of default estimated by the procedure outlined in Bharath and Shumway (2008). The control variables include size (log of total assets), mb (market-to-book ratio), lev (firm leverage) and roa (return on assets). The robust standard errors are reported in the parentheses. ***, **, and * represent significance levels at 0.01, 0.05, and 0.10, respectively.

28 Table 4 Panel A Positive Accruals with Truncated probd (1) (2) (3) VARIABLES TA DA MDA probd ** 0.017** (0.003) (0.008) (0.008) size *** *** *** (0.000) (0.001) (0.001) mb 0.001*** 0.001** 0.001*** (0.000) (0.000) (0.000) lev *** 0.036*** (0.004) (0.012) (0.013) roa 0.165*** (0.017) (0.043) (0.046) Constant 0.045*** 0.093*** 0.104*** (0.008) (0.017) (0.018) Observations 32,188 32,188 32,188 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes Panel B Negative Accruals with Truncated probd (1) (2) (3) VARIABLES TA DA MDA probd *** *** *** (0.003) (0.004) (0.004) size 0.007*** 0.008*** 0.008*** (0.000) (0.001) (0.001) mb *** (0.000) (0.000) (0.000) lev *** *** *** (0.005) (0.007) (0.007) roa 0.350*** 0.221*** 0.233*** (0.016) (0.020) (0.020) Constant *** *** *** (0.010) (0.024) (0.021) Observations 32,822 32,822 32,822 R-squared Industry Fixed Effects Yes Yes Yes Year Fixed Effects Yes Yes Yes Firm Cluster Yes Yes Yes This table reports OLS regression analysis of the association between earnings management and distress risk for fims with income-inflating earnings management and income-deflating earnings management, separately, after truncating the probability of default at 0. In Panel A, the sample comprises firms with income-inflating earnings management, defined as firms with positive discretionary accruals (both Jones and 28

29 29 modified Jones model) in year t. In Panel B, the sample comprises firms with income-deflating earnings management, defined as firms with negative discretionary accruals (both Jones and modified Jones model) in year t. In both Panels A and B, the dependent variables in regressions 1-3 are total accruals, discretionary accruals (Jones model) and discretionary accruals (modified Jones model), respectively. probd is the probability of default estimated by the procedure outlined in Bharath and Shumway (2008). The control variables include size (log of total assets), mb (marketto-book ratio), lev (firm leverage) and roa (return on assets). The robust standard errors are reported in the parentheses. ***, **, and * represent significance levels at 0.01, 0.05, and 0.10, respectively.