Earnings Management to Avoid Debt Covenant Violations and Future Performance

Abstract

In this study, we examine the trade-offs between earnings management (both accruals and real) and covenant violations by examining how they are associated with future accounting and stock market performance. We analyze a matched-pair sample of covenant violation firms with non-violation firms that have a similar risk of a covenant violation. We have three main findings. First, our evidence indicates that covenant violations are costly events for shareholders as lenders appear to use their control rights in ways that increase the likelihood of loan repayment but impose costs for shareholders. Second, there is limited evidence indicating covenant-related accrual-earnings management activities impose significant costs on shareholders, but we find shareholders are worse off following unsuccessful real earnings management. Third, our evidence indicates that, on average, shareholders at high violation risk firms are better off when their firms successfully engage in accruals earnings management to avoid a violation compared to shareholders at firms that violate a covenant but do not manage earnings. Thus, covenant-related earnings management may be in the best interests of shareholders and is not necessarily evidence of shareholder-manager agency conflicts.

Keywords:

• Earnings management
• Debt covenants violations
• Future firm performance

Acknowledgement

We thank Arizona State University, Duke University, IESE, and INSEAD for financial support. Fernando Penalva acknowledges financial assistance from research projects ECO2016-77579-C3-1-P and PID2019-111143GB-C31 funded by the Spanish Ministry of Economics, Industry and Competitiveness, and the Ministry of Science and Innovation, respectively. We thank Daniel Beneish, Stephen Brown, Katherine Drake, Beatriz García Osma, Joachim Gassen, Leslie Hodder, Helena Isidro (discussant), Andy Leone, Sugata Roychowdhury, Lakshmanan Shivakumar, Jim Whalen, Paul Zarowin (associate editor), two anonymous reviewers, and seminar participants at Arizona State University, Indiana University, European Accounting Association Annual Congress, American Accounting Association Annual Meeting, and IX Workshop on Empirical Research in Financial Accounting for their comments and suggestions.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, doi:10.1080/09638180.2020.1826337.

An online Supplement with additional information and analyzes can be accessed at the journal’s Taylor and Francis website.

Notes

1 We refer to covenant-related earnings management as earning management whose primary purpose is to reduce the likelihood of a covenant violation.

2 Prior research indicates managers experience personal costs following covenant violations, for example through higher forced CEO turnover (Nini et al., 2012; Ozelge & Saunders, 2012). These personal costs potentially exacerbate shareholder-manager agency conflicts prior to covenant violations, and hence, induce managers to make decisions that are detrimental to shareholders.

3 We focus on summarizing the results w.r.t. accruals earnings management in the Introduction. The real earnings management results are generally similar, but are not identical. We discuss both sets of results in more detail in Section 5.

4 Our results for the accounting-based performance measures and capital expenditures are generally consistent with those in Nini et al. (2012). In their analyses of the performance effects of covenant violations, they do not condition on earnings management activities. We discuss how our study differs from theirs below.

5 The corresponding results for real earnings management show that violation firms that did not engage in real earnings management experience significantly higher future ROA growth and significantly lower growth in capital expenditures compared to for non-violation firms that engaged in real earnings management. However, while the non-violation firms experience lower abnormal returns, the difference in not significant.

6 Given these differences with Nini et al. (2012), we reconcile our results with theirs at the end of Section 5. These results indicate the differences depend on firms’ earnings management activities during the pre-violation period and are not due to sample differences based on accounting data and matching requirements.

7 Our results do not provide evidence on the performance consequences of earnings management at either firms with a low risk of a covenant violation or earnings management made for non-covenant related reasons.

8 Private conversations with several commercial bankers confirm that banks use their loan acceleration and termination rights to impose operational changes that enhance the likelihood of repayment.

9 The abnormal returns estimates for violation firms during the post-violation period are consistently positive across the various specifications they examine. However, whether or not the estimates are significant depends on the specific test specification and the time period over which returns are computed.

10 An alternative explanation is that if covenant violations are beneficial for shareholders (Nini et al., 2012), then managers’ efforts to avoid violations are manifestations of agency costs. We discuss this alternative below.

11 For example, Sweeney (1994) finds that in the nine cases where a firm could have avoided a covenant violation by changing inventory flow assumptions, four firms chose not to do so. She speculates that large negative tax consequences dissuaded them. This evidence suggests firms balance the costs and benefits of covenant-related earnings management.

12 Following Zang (2012), we do not examine abnormal cash flows from operations (AbCFO) because real activities manipulation affects this variable in different directions and the net effect is potentially ambiguous. Furthermore, as Cohen and Zarowin (2010) discuss in their footnote 8, ‘the same activities that lead to abnormally high production costs also lead to abnormally low CFO; thus, adding these two amounts leads to double counting.’

13 Results are qualitatively similar if we either use the continuous measures of the earnings management proxies (DWCA and RM) or indicator variables that equal one if DWCA (RM) is positive, and zero otherwise, respectively.

14 For example, firms with pre-existing relationships with their lenders may less willing to damage this relationship by engaging in covenant-related earnings management. In addition, relationship banks may be less willing to impose costly penalties on firms following a violation because they do not want to risk jeopardizing the relationship (Dichev & Skinner, 2002).

15 To the extent managers have non-covenant related incentives to manage earnings that are related to performance and liquidity, then the accruals regression models will at least partially capture discretionary accruals related to these incentives.

16 The data can be obtained from Amir Sufi’s webpage (http://faculty.chicagobooth.edu/amir.sufi/chronology.html). The online appendix to Nini et al. (2012) contains the details of this sample’s construction and composition.

17 The presence of hidden violation firms in our non-violation sub-samples will reduce our ability to find significant differences between the sub-samples. That is, it will bias against rejecting the null for Hypotheses 1 and 3.

18 We include Debt Rating and Top Auditor in order to capture other corporate governance factors which are likely correlated with both the likelihood of a violation, the likelihood of engaging in covenant-related earnings management, and future performance. We chose not to include additional variables related to corporate governance or executive incentives because their inclusion would lead to a large reduction in sample size. Violation firms tend to be relatively small, and hence, are not tracked by many corporate governance data providers. For example, requiring incentive compensation data from Execucomp results in a roughly 60% decrease in sample size.

19 The evidence in Roberts and Sufi (2009) indicates that firms frequently renegotiate their loan agreements and most renegotiations take place without a covenant violation. Thus, some of our non-violation non-earnings management firms may have avoided a violation by renegotiating their loan agreements. To the extent this type of renegotiation is costly, our ability to find significant differences between the sub-samples will be reduced.

20 We implicitly assume that both groups will experience similar covenant-related earnings management incentives. To the extent that this assumption is not descriptive, then we expect this measurement error will reduce our ability to find significant results and our evidence must be interpreted accordingly.

21 To simplify the exposition, we use expressions like ‘did manage earnings’ (‘did not manage earnings’) or (‘did not) ‘engage in covenant-related earnings management’ as a shorthand for ‘more (less) likely to have managed earnings’ or ‘more (less) likely to have engaged in covenant related earnings management.’

22 The bottom four rows of Table 5 (and 6) provide the sum of the corresponding coefficients from the regression results to obtain the absolute effect for each group of firms. Significance levels are based on F-tests on whether the sum is different from zero. Thus, the absolute 2.207 percent increase in ROA for [V=Yes; AEM=No] firms is significantly different from zero at the 1% level.

23 The results in Table 6 where REM is the proxy for earnings management are generally similar to those in Table 5. There are two main exceptions. First, in the ΔROA regression, the Violation coefficient is positive but not significant (t-statistic = 1.25) despite significant decreases in all four measures of expenditures. Second, in the BHAR regression, the Intercept continues to be positive (4.691) but is no longer significant (t-statistic = 1.39).

24 The REM results in Table 6 are similar but somewhat weaker. Specifically, the absolute magnitudes of Intercept and Violation coefficients are smaller (4.619 vs. 5.593 and -5.917 vs. -11.755, respectively) and only the Violation coefficient is significant (t-statistic = -1.83).

25 One possible explanation for the lack of significance in the BHAR regression is that any negative effects of AEM have already been incorporated in stock prices prior to the violation month. Consistent with this explanation, Figure 1 shows that [V=No; AEM=Yes] firms experience some modest deterioration in performance starting seven months prior to the violation.

26 We find no significant relative or absolute difference in future ROA growth for [V=No; REM=Yes] firms compared to [V=No; REM=No] firms. These findings are in contrast to the results in Gunny (2010). She provides evidence that firms engaging in REM to just meet earnings benchmarks have relatively better subsequent accounting performance (ROA) than firms that do not engage in REM and miss or just meet the benchmarks. We note that our respective settings (covenant violations vs. benchmark beating) are very different and likely to have different implications for REM.

27 When managers decide whether or not to engage in covenant-related earnings management, they will take the expected violation costs into account. Thus, our results may suffer from an endogeneity bias. Thus, firms are more likely to engage in earnings management when the expected costs of a violation are relatively high, and vice-versa, choose not to engage in earnings management when expected violation costs are relatively low. In which case, our results underestimate the differences between the two groups of firms. However, without good instrumental variables or a compelling source of exogenous variation, we are unable to address this issue directly. Thus, our results must be interpreted accordingly.

28 We focus on the analyses in Table 10 Panel B because they most clearly show violation firms experience positive abnormal returns during the post-violation period. In contrast, the results in their Table 11 show insignificant abnormal returns for violation firms following violations. As discussed in Nini et al. (2012), these tests are of relatively low power. Nini et al. (2012) also examine abnormal returns during months [+1, +60] and find the mean CAR is not significantly different from zero. Given that we only examine returns through month +24, we do not replicate their analysis over this extended time period.

29 In addition, we replicate Figure 8 in Nini et al. (2012), which shows monthly abnormal returns in event time for the full sample of violation firms that have the required stock market date between months (-12, +24). The results are presented in Figure A3 in the online Supplement which is similar to the corresponding Figure 8 in Nini et al. (2012).

30 We also perform additional analyses where we limit the sample of violation firms used in Panel A to those firms that had the required accounting data in Compustat (i.e., we do not require a match with a comparable non-violation firm. The untabulated results are quantitatively very similar to those reported in Panel B both in terms of coefficient magnitudes and significance levels.

31 The negative returns for [AEM = 1] firms over months [+1, +3] are insignificantly different from zero based on the robust t-statistic, but are significant at the 5% level based on the t-statistic.

32 Notice that a positive (negative) value of CHGAR and CHGINV represents a decrease (increase) in accounts receivable and inventories, while a positive (negative) value of CHGAP, CHGTAX, and CHGOTH represents an increase (decrease) in accounts payable, taxes payable, and other items. These variables are referred to as RECCHY, INVCHY, APALCHY, TAXCHY, and AOLOCHY in the current version of Compustat. Like Collins et al. (2017), we recode missing values of RECCHY, INVCHY, APALCHY, and TAXCHY as zero if there is a non-missing value of AOLOCHY. Conversely, if AOLOCHY is missing but the other items are not missing, then we recode AOLOCHY as zero. We undo the year-to-date nature of these quarterly cash flow statement items and compute the quantities for the quarter under consideration.

33 The traditional modified-Jones model for total accruals includes an additional regressor: gross property, plant and equipment. We drop this term because our dependent variable is working capital accruals, which are not affected by this item. Our inferences are not affected by this choice.

34 Roychowdhury (2006) includes advertising expenses as part of discretionary expenses. We exclude advertising because it is not reported in quarterly Compustat data.

35 Our inferences are unchanged if we include observations with missing values in these analyses.

Additional information

Funding

Fernando Penalva acknowledges financial assistance from research projects ECO2016-77579-C3-1-P and PID2019-111143GB-C31 funded by Spanish Ministry of Economy, Industry and Competitiveness, and the Ministry of Science and Innovation, respectively.