Non-GAAP Earnings and Auditors Going Concern Opinions

Transcription

1 Non-GAAP Earnings and Auditors Going Concern Opinions Anne Albrecht, Jeff Zeyun Chen, Karen K. Nelson Neeley School of Business, Texas Christian University November 2018 Abstract We examine the role of non-gaap earnings in auditors going concern assessments, focusing on firms with a GAAP loss but a non-gaap profit (i.e., a non-gaap switch to profit). Consistent with non-gaap switches being informative about future loss reversals (Leung and Veenman 2018), auditors are less likely to issue a going concern opinion to financially-distressed firms following a non-gaap earnings switch. Separating non-gaap switches into those reported by managers, analysts, or both, we find a significant result for both analyst-only and confirmatory switches but not for manager-only switches, suggesting that auditors react skeptically to managers non-gaap switches unless corroborated by analysts. The weight auditors place on non-gaap switches in their going concern assessments is generally consistent with the weight implied by a bankruptcy prediction model. Overall, our evidence indicates that auditors incorporate credible signals of non-gaap switches into their going-concern assessments. Keywords: Non-GAAP earnings; going concern opinions; auditor skepticism; bankruptcy We thank Bentley et al. (2018) for generously providing their manager-reported non-gaap data. The paper has benefitted from the insights of John Bizjak, Ted Christensen, Kurt Gee, Nicholas Hallman, Patrick Kielty, and workshop participants at the Public Company Accounting Oversight Board, Ohio State University and Texas Christian University. 1

2 I. INTRODUCTION Over the past decade, it has become nearly as common to measure and evaluate firm performance using non-gaap earnings as GAAP earnings. For example, evidence suggests that 88% of the S&P 500 reported non-gaap earnings in 2015 (Golden 2017), up from 53% in 2009 (Black et al. 2017). Alarmed by the implications of this trend for investors, the SEC has directed increased attention and scrutiny to how companies measure and disclose non-gaap earnings (SEC 2018). At the same time, the FASB is considering how the GAAP financial performance reporting model may itself contribute to the proliferation of alternative earnings measures (Golden 2017; FASB 2018) and the PCAOB is evaluating the auditor s role in non-gaap reporting (PCAOB 2018). Despite this interest, there is little evidence in the academic literature on auditors and non-gaap reporting (Black et al. 2018). Our paper addresses the role of non- GAAP earnings in auditors going concern assessments. Non-GAAP earnings are typically reported outside of the financial statements (e.g., earnings announcements, MD&A, investor calls, analyst reports) and, as the name implies, are not prepared in accordance with generally accepted accounting principles. As a result, current auditing standards do not require non-gaap earnings to be audited (AS 1001, PCAOB 2015a), and the audit opinion does not provide assurance on this information. 1 However, auditing standards do provide guidance on using information outside the audited financial statements to assess audit risk (AS 2110, PCAOB 2015c), fraud risk (AS 2401, PCAOB 2015d); and other information contained in the 10-K (AS 2710, PCAOB 2015f). Non-GAAP earnings may be particularly useful in the context of the auditor s evaluation of a firm s ability to continue as a going concern to the extent that these metrics exclude GAAP items that are less informative 1 Further, the internal controls around non-gaap reporting typically do not fall under internal control over financial reporting (ICFR), and therefore are excluded from the auditor s ICFR opinion (CAQ 2016). 2

3 about ongoing operating performance (Bierstaker et al. 2013). Moreover, loss firms non-gaap earnings are especially predictive of future performance and offset the low predictive ability of GAAP losses (Leung and Veenman 2018). We focus our analysis on firms with a GAAP loss but a non-gaap profit (i.e. a non-gaap switch to profit) to enable us to disentangle the signals provided by GAAP and non-gaap earnings. If both metrics signal financial distress it is difficult to determine if the non-gaap loss, even if less severe than the GAAP loss, provides auditors with meaningful incremental information regarding the firm s ability to continue as a going concern. However, when non- GAAP earnings signal profitability, the financial distress signal in the GAAP loss is potentially dampened. Consistent with this view, Leung and Veenman (2018) find that non-gaap switch firms have significantly better future performance and are more likely to become profitable in the future than loss firms that do not report a non-gaap switch. If auditors consider non-gaap switches as informative about future loss reversals, then they will be less likely to issue a going concern modified audit opinion (hereafter GCO ) relative to other financially distressed firms. We identify non-gaap switches using both manager-reported non-gaap earnings and analyst-reported non-gaap earnings. Prior research suggests that analysts non-gaap earnings are generally more informative than managers non-gaap earnings (Black et al. 2018). However, investors view managers non-gaap earnings to be more credible when they are corroborated by analysts (Bentley et al. 2018). Further, auditing standards indicate that information should be considered more reliable when two sources of audit evidence are consistent (AS 1105, PCAOB 2015b). Therefore, we expect that auditors will be more likely to consider non-gaap switches in their going concern assessments when both managers and analysts provide a non-gaap switch (a confirmatory non-gaap switch). When only one party 3

4 reports a non-gaap switch (a manager-only or analyst-only non-gaap switch), auditors may deem the information too unreliable to incorporate into their going concern assessments. Our sample consists of firm-years from 2004 to 2015 that appear from a GAAP perspective to be financially distressed (i.e., the firm reports either negative GAAP earnings or negative operating cash flows). 2 The sample period begins after the effective date of Regulation G to ensure that all observations are under the same non-gaap regulatory reporting regime and ends prior to the implementation of Accounting Standard Update No to ensure that all observations are under the same going concern reporting regime. 3 Our final sample consists of 6,861 financially distressed firm-year observations, of which approximately one-quarter are non- GAAP switch firms. As predicted, we find that auditors are significantly less likely to issue a GCO when there is a non-gaap earnings switch. This result is incremental to several client- and auditor-specific determinants of GCOs, including GAAP-based measures of financial performance and special items which are likely sources of non-gaap exclusions. Thus, the results suggest that non- GAAP switches provide an important signal to auditors about the firm s ability to continue as a going concern that is incremental to the information captured in the GAAP financial statements. Separating non-gaap switches into confirmatory switches and manager-only or analyst-only switches reveals that the likelihood of a GCO is negatively associated with confirmatory and analyst-only switches. Manager-only non-gaap switches do not significantly reduce auditors 2 Our inferences are unchanged if we define financial distress using only negative GAAP earnings. 3 Regulation G requires companies to reconcile non-gaap measures to the most directly comparable GAAP measure (SEC 2002). Our sample includes fiscal years beginning after March 28, 2003, the effective date of Regulation G. Accounting Standards Update No requires management to evaluate and disclose going concern uncertainties for annual reporting periods ending after December 15, 2016 (FASB 2014). 4

5 propensity to issue a GCO. Thus, auditors appear to react skeptically to manager-reported non- GAAP switches unless confirmed by analysts. Our results are consistent with auditors incorporating non-gaap switches into their going concern assessments because they are informative about the firm s future prospects. However, it is also possible that non-gaap exclusions opportunistically switch a loss to a profit (Black et al. 2018), potentially misleading auditors about the firm s ability to continue as a going concern. To distinguish between these alternative explanations, we compare how auditors weight non-gaap switches in their going concern assessments to the weight implied by a bankruptcy prediction model. If non-gaap switches are informative about future loss reversals, auditors should weight them consistently with the weight implied by the bankruptcy prediction model. However, if auditors do not detect opportunistic non-gaap switches, then they will overweight the switch in their going concern assessment. We find that non-gaap switch firms are significantly less likely to experience bankruptcy in the subsequent year. Moreover, the weight auditors apply to non-gaap switches in their going concern assessments is consistent with the weight implied in the bankruptcy prediction model. There is no evidence that auditors overweight switches; if anything, there is some evidence consistent with auditors general tendency towards reporting conservatism (e.g., Lennox 1999; Francis and Krishnan 2002; Francis 2004; Francis 2011) as auditors underweight non-gaap switches relative to the weight implied by the bankruptcy prediction model. Separating non-gaap switches into confirmatory, manager-only, and analyst-only switches reveals that there are no manager-only switch observations that declare bankruptcy in year t+1. Therefore, we perform two alternative tests. First, we expand the definition of subsequent bankruptcy to include two years, t+1 and t+2 (Nogler 1995; Geiger et al. 1998; Louwers et al. 5

6 1999). Consistent with our main results, we find that confirmatory and analyst-only non-gaap switches are negatively associated with subsequent bankruptcy but there is no association between manager-only non-gaap switches and subsequent bankruptcy. Further, auditors weight the significant confirmatory and analyst-only non-gaap switches in their GCOs consistent with the weights implied by the bankruptcy prediction model. Second, we examine three alternative GAAP measures of financial distress in year t+1 (excluding observations that declared bankruptcy in that year): (i) bottom-line net loss; (ii) large loss, defined as a loss below the median net loss; or (iii) operating loss. Consistent with the bankruptcy results, there is a significant negative relation between future losses and confirmatory and analyst-only non-gaap switches. Manager-only non-gaap switches are also significantly less likely to experience losses in the subsequent year, but the magnitude of this association is significantly lower than that for either the confirmatory or analyst-only switches. Further, the coefficient estimates on the non-gaap switch variables in the going concern and future loss models are not significantly different. Taken together, our results suggest that auditors consider non-gaap earnings measures in their going concern assessments when they are reliable indicators of future loss reversals. Our paper is relevant to the ongoing regulatory and standard-setting debate on non-gaap earnings. In particular, our results inform the PCAOB s current agenda which separately considers the effectiveness of going concern reporting and the auditor s role regarding non-gaap performance measures. Our paper examines the confluence of these two agenda items and suggests that auditors going concern assessments appear to capture information in non-gaap earnings that signal a switch to future profitability. 6

7 We extend the literature on non-gaap reporting along several dimensions. Prior research suggests that several stakeholders use non-gaap performance metrics, including analysts, compensation committees, creditors, and short sellers, but research is only now beginning to explore auditors use of non-gaap earnings information (e.g., Chen et al. 2012; Hallman et al. 2017). We investigate a previously unexplored reporting context auditors going concern opinions where we expect non-gaap switches by loss firms to be an informative and relevant signal about future performance. We also consider both managers and analysts non-gaap reporting choices. Leung and Veenman (2018) posit that confirmatory non-gaap switches could be particularly informative in assessing future performance of loss firms but they do not test this assertion. Our paper thus answers the call for more research examining the role of manager and analyst non-gaap earnings (Bentley et al. 2018; Black et al. 2018). Our paper also contributes to the literature on auditors reporting of going concern uncertainties. Carson et al. (2013) note that relatively little is known about how mitigating client information and information outside the financial statements more generally bears on auditors going concern assessments. Our paper addresses both of these issues by providing evidence that non-gaap switches reduce the likelihood of financially distressed firms receiving a GCO. Moreover, we document the important role of confirmatory evidence (e.g., non-gaap switches by analysts) in lending credibility to the mitigating information provided by managers. In a setting similar to ours, Feng and Li (2014) find that the likelihood of a GCO is decreasing in the magnitude of managements earnings forecasts. However, they do not include analysts earnings forecasts in their model, and thus it is not possible to determine how the two information sources separately or jointly factor into auditors going concern assessments. We find that auditors 7

8 appear to rely on manager-reported earnings signals in their going concern assessments only when there is confirmatory evidence provided by analysts. II. BACKGROUND AND EMPIRICAL PREDICTIONS Non-GAAP Earnings Non-GAAP earnings typically start with GAAP earnings and then eliminate various components to create an alternative metric for evaluating company performance. The debate over non-gaap reporting centers on the extent to which the resulting performance metrics are informative to financial statement users or opportunistic and potentially misleading. The SEC has historically adopted a cautious view, alerting investors to the potential dangers of non-gaap information (SEC 2001), regulating non-gaap disclosures (SEC 2002), and issuing various Compliance and Disclosure Interpretations. 4 A former chief accountant of the SEC s Division of Enforcement labelled non-gaap measures, particularly those that indicate a profit while the income statement shows a loss, as a fraud risk factor (Leone 2010), and these concerns are reflected in an increasing number of comment letters (Audit Analytics 2017). Consistent with this view, research finds that firms frequently exclude from non-gaap earnings recurring items that are associated with future operating performance (e.g., Doyle et al. 2003; Landsman et al. 2007; Bentley et al. 2018). 5 A key motivation for these exclusions appears to be misleading investors into believing that company performance meets or beats a strategic earnings target (e.g. Bhattacharya et al. 2003; McVay 2006; Black and Christensen 2009; Doyle et al. 2013; Bradshaw et al. 2018). Other studies, however, suggest that non-gaap reporting is generally informative. By excluding one-time items, non-gaap earnings provides a more accurate depiction of core performance (e.g., Bhattacharya et al. 2003; Lougee and Marquardt 4 (updated April 4, 2018). 5 See Black et al. (2018) for a review of the non-gaap literature. 8

9 2004; Curtis et al. 2014). Further, investors appear to respond more to non-gaap metrics than to GAAP earnings (e.g. Brown and Sivakumar 2003; Johnson and Schwartz 2005; Marques 2006; Wieland et al. 2013; Bradshaw et al. 2018), and generally are not misled by non-gaap reporting, especially in the post-reg G era (e.g., Chen 2010; Jennings and Marques 2011; Zhang and Zheng 2011; Huang and Skantz 2016; Whipple 2016). Non-GAAP earnings may be particularly relevant for loss firms which are the focus of our study. Because losses create increased uncertainty about future earnings, managers of loss firms have incentives to disclose informative non-gaap measures to help investors better understand the underlying cash flows (Leung and Veenman 2018). 6 Consistent with this view, CFOs of loss firms rank non-gaap earnings significantly higher than CFOs of profit firms as an important performance measure (Graham et al. 2005). Moreover, non-gaap earnings are more predictive of future performance for loss firms than for profit firms (Leung and Veenman 2018). Most of the non-gaap literature focuses on manager, analyst, and investor use of non-gaap earnings, with little consideration given to auditors. The auditing research that does exist supports the opportunistic view of non-gaap reporting. Chen et al. (2012) find that auditors charge higher audit fees and resign more frequently if client firms report opportunistic non- GAAP earnings, suggesting that auditors are aware of managerial opportunism in non-gaap reporting. Hallman et al. (2018) find that client reporting of non-gaap earnings in the U.K. is associated with higher materiality thresholds and lower audit quality. The potential implications of non-gaap earnings for auditors in our setting differ from these papers in several important ways. First, we focus on loss firms, which, as discussed above, are 6 Indeed, Li (2011) finds that investors have difficulty understanding the persistence of GAAP losses. Specifically, investors appear to treat all GAAP losses as transitory, and hence are surprised by future announcements of negative earnings for firms with persistent GAAP losses. 9

10 more likely to have informative non-gaap disclosures. Second, we examine a reporting context, going concern assessments, where non-gaap earnings may be particularly useful to auditors (Bierstaker et al. 2013). Finally, we consider non-gaap reporting by both managers and analysts, and thus provide evidence on whether analysts non-gaap switches play a confirmatory role to boost the credibility of managers non-gaap switches. Going Concern Opinions Unless liquidation is imminent, financial statements prepared under U.S. GAAP presume that the reporting entity will continue as a going concern. Because of this going concern presumption, AS 2415 (PCAOB 2015e) requires auditors to evaluate a firm s ability to continue as a going concern for a reasonable period of time not to exceed one year beyond the date of the financial statements. Auditors base their going concern assessments on knowledge of circumstances or events obtained during the audit that indicate substantial doubt exists about a firm s ability to continue as a going concern. Such circumstances and events include, but are not limited to, negative trends and other indications of financial difficulty, internal matters (e.g., labor strikes), and external matters (e.g., lawsuits or a significant decrease in customer base). If substantial doubt about the client firm s ability to continue as a going concern exists and management does not have sufficient plans in place to mitigate that doubt, then auditors will modify their opinion to disclose such going concern uncertainties (i.e., issue a GCO). Assessing the likelihood that the client firm will continue as a going concern is challenging because of the uncertainty surrounding future conditions and events. Nonetheless, auditors are frequently criticized for failing to issue a GCO when it becomes apparent ex post that one was warranted. This criticism escalated following the financial crisis when most firms that failed had not previously received a GCO. In response to calls for enhanced going concern disclosures, the 10

11 FASB issued Accounting Standards Update No requiring management to evaluate and disclose going concern uncertainties for annual reporting periods ending after December 15, 2016 (FASB 2014) and the PCAOB is currently evaluating the auditor going concern reporting framework and how that framework can be improved (PCAOB 2018). Prior research finds that financial distress is an important determinant of an auditor s decision to issue a GCO. 7 Most of this research focuses on measures of financial distress obtained from the financial statements. For example, it is well-documented that auditors are more likely to issue a GCO to small, unprofitable firms with high leverage and low liquidity (e.g. Kida 1980; Mutchler 1985; Dopuch et al. 1987; Menon and Schwartz 1987). There is less empirical evidence about indicators of financial distress from outside the financial statements. The research that does exist focuses on market-based measures, and finds that auditors are more likely to issue a GCO to firms with lower industry-adjusted returns and higher return volatility (e.g. Dopuch et al. 1987; Mutchler and Williams 1990; DeFond et al. 2002). Other research examines mitigating factors outside the financial statements, such as management plans to raise additional capital, with mixed results (e.g., Mutchler 1985; Mutchler et al. 1997; Behn et al. 2001). Focusing on a sample of firms that issue one-year-ahead management earnings forecasts, Feng and Li (2014) find that higher management earnings forecasts decrease the likelihood that the auditor issues a GCO. However, auditors give less weight to earnings forecasts and earnings forecast changes in the top tercile, perhaps because they perceive extreme forecasts to be less credible. Our paper contributes to this literature in several important respects. First, we examine a mitigating factor, non-gaap earnings, which is commonplace relative to the factors examined in 7 See Carson et al. (2013) for a review of the literature on auditors reporting of going concern uncertainties. 11

12 prior work. Second, our focus on non-gaap earnings switches allows us to disentangle the mitigating information in non-gaap earnings from the reported GAAP loss. In contrast, Feng and Li (2014) focus on the magnitude of the management earnings per share forecast. Unless the financially distressed firm is forecasting a profit, however, it is not clear that the forecast provides auditors with information that mitigates the current period reported loss. Third, we include all financially distressed firms in our analysis, which not only expands our sample size but also mitigates potential self-selection issues associated with firms reporting choices (Feng and Li 2014). Finally, we include both manager and analyst non-gaap switches, allowing us to consider the importance of corroboration for assessing the credibility of mitigating information. Empirical Predictions Auditing standards require auditors to consider information reported outside the financial statements. For example, during risk assessment, the auditor must obtain an understanding of the company from relevant information such as company-issued press releases, company-prepared presentation materials for analysts or investor groups, and analyst reports and transcripts of earnings calls and, to the extent publicly available, other meetings with investors or rating agencies (AS 2110). The auditor is also required to identify and understand performance measures that might affect the risks of material misstatement, including those used by external parties, such as analysts (AS 2110). During fraud assessment, the auditor will consider whether excessive pressure exists for management to meet the requirements or expectations of third parties due to profitability or trend level expectations of investment analysts (AS 2401). 8 Lastly, when considering other information disclosed within the 10-K, auditors will read the MD&A section for consistency with the financial statements (AS 2710). 8 Consistent with these standards, Newton (2018) finds that auditors incorporate analyst-provided information into risk assessment and audit planning. 12

13 As part of their going concern assessment, auditors make judgments about a firm s ability to continue to operate over the next year. Non-GAAP earnings are an alternative measure of financial health that the auditor can use in their assessment. For instance, a non-gaap loss may provide additional evidence that going concern uncertainties exist, whereas a non-gaap profit may provide evidence mitigating going concern uncertainties. Therefore, auditors may consider non-gaap earnings to obtain a more complete picture of the firm s financial health. The auditor s decision to issue a going concern opinion is most salient when an event or circumstance indicates that the company is financially distressed, such as when there is a current year GAAP loss (e.g. Multcher 1985; Dopuch et al. 1987; Defond et al. 2002). If both GAAP and non-gaap earnings tell a similar financial distress story (i.e. both GAAP earnings and non- GAAP earnings are negative), non-gaap earnings are unlikely to provide the auditor with new information beyond the signal contained in GAAP earnings. 9 Rather, the non-gaap earnings information will be most relevant when GAAP earnings suggest financial distress (i.e., a reported loss), but non-gaap earnings indicate the company is profitable. 10 If non-gaap profits mitigate going concern uncertainties, we expect that auditors will be less likely to issue a GCO. Thus, we state our first hypothesis as follows: H1: Auditors are less likely to issue a going concern opinion to financially distressed firms when there is a non-gaap switch. 9 A non-gaap loss that is less extreme than the reported GAAP loss may indicate that financial distress is not as severe as the GAAP financial statements suggest. However, prior literature finds that a loss, regardless of the magnitude, is an indicator of financial distress. For example, Dopuch et al. (1987) note that a current year loss has more predictive power than net income measured as a continuous variable. 10 Carson et al. (2013, 358) note that it is difficult to determine whether auditors use measures of financial distress obtained from outside the financial statements because these measures may be simply a different, yet consistent reflection of distress. For our study, the non-gaap switch to profit creates a setting in which we are better able to examine auditors use of other financial distress measures because the non-gaap profit signal counters the GAAP financial distress signal. 13

14 Next, we consider the two most common sources of non-gaap earnings and whether they differentially affect auditors decisions to issue a GCO. Prior research suggests that analysts non-gaap earnings are, on average, higher quality than managers non-gaap earnings. For example, Bentley et al. (2018) finds that analysts often ignore managers lower quality non- GAAP measures in favor of providing their own higher quality non-gaap measures. This result is perhaps not surprising because of the discretion afforded to managers in non-gaap reporting. Auditing standards require auditors to base their going concern assessments on audit evidence (AS 2415), which is considered more reliable when that evidence is obtained from a source that is independent of the company (AS 1105). Therefore, auditors are likely to place less weight on managers non-gaap earnings switches. 11 However, auditors will likely consider managers non-gaap earnings switches more credible if they are validated by analysts (Marques 2006; Bentley et al. 2018; Leung and Veenman 2018). Auditing standards suggest that information is more reliable when two or more sources of audit evidence are consistent (AS 1105). If managers report a non-gaap earnings switch but analysts do not, auditors will be less likely to incorporate the switch into their going concern assessments. However, if both managers and analysts report a non-gaap earnings switch (a confirmatory switch), auditors will be more confident in basing their going concern assessment on the non-gaap switch. Thus, we state our second hypothesis as follows: H2: Auditors are less likely to issue a going concern opinion to financially distressed firms when both managers and analysts report a non-gaap switch (a confirmatory switch). 11 Managers can provide non-gaap earnings through public disclosures and/or private communications with their auditor. Public disclosure may increase the perceived credibility of non-gaap earnings because of the potential for increased investor and regulatory scrutiny. Nevertheless, following Feng and Li (2014), we expect that public disclosures will be consistent with private communications to avoid confusing the auditor and potentially damaging managers credibility. Thus, we use non-gaap earnings reported externally as a proxy for the information provided by managers to auditors. In sensitivity tests, we show that the mere public disclosure of non-gaap earnings does not affect auditors going concern assessments unless the non-gaap earnings creates a switch to profit. 14

15 Our last hypothesis examines the weight auditors place on non-gaap switches in their going concern assessments relative to the weight implied in a bankruptcy prediction model. Following prior research, we assess auditors weighting of non-gaap earnings using subsequent client bankruptcy as a benchmark. If non-gaap switches for loss firms are informative about future profitability (Leung and Veenman 2018) and auditors incorporate this information in their going concern assessments, we expect a lower likelihood of a GCO and a lower likelihood of future bankruptcy. Moreover, the weight on the non-gaap switch will be similar in the going concern and bankruptcy prediction models. These arguments lead to our final hypothesis as follows: H3: The weight auditors assign to non-gaap switches in their going concern assessments is the same as the weight implied in a bankruptcy prediction model. Two situations may lead to a different weighting on non-gaap switches in the GCO and bankruptcy prediction models. First, because of auditor reporting conservatism, auditors may underweight the switch in their going concern assessment relative to the weight implied by the bankruptcy prediction model. Alternatively, if non-gaap exclusions opportunistically switch a GAAP loss to a non-gaap profit, auditors may mistakenly incorporate the switch in their going concern assessment, and thus overweight the switch relative to the weight implied by the bankruptcy prediction model. Regression Models III. RESEARCH DESIGN AND SAMPLE We test H1 and H2 using the following probit regression model of the probability of an auditor issuing a GCO: GCOt = β0 + β1ng Switch Variablet + β2controlst + Industry FE + Year FE + ε (1) 15

16 GCO is an indicator variable equal to one for firms with a GCO. 12 NG Switch Variable designates alternative specifications of indicator variables for non-gaap switches. First, we define NG Switch equal to one if either manager- or analyst-reported non-gaap earnings are positive and GAAP earnings are negative. The benchmark group in this analysis consists of firms without non-gaap earnings and firms with manager- and analyst-reported non-gaap earnings that exhibit the same sign as GAAP earnings (i.e., firms that do not report a non-gaap earnings switch). A negative coefficient estimate on NG Switch is consistent with the prediction in H1 that auditors are less likely to issue a GCO when there is a non-gaap earnings switch. We also estimate equation (1) separately for manager-reported non-gaap switches (Manager NG Switch) and analyst-reported non-gaap switches (Analyst NG Switch). Managerreported non-gaap earnings may be of lower quality than analyst-reported non-gaap earnings (Bentley et al. 2018). This difference in quality, along with auditing standards that consider information more reliable when obtained from an independent source, suggests that auditors may place less weight on manager-reported non-gaap switches. To isolate whether managers or analysts, or both, report non-gaap switches, we define Confirmatory NG Switch equal to one if both manager- and analyst-reported non-gaap earnings are positive and GAAP earnings are negative. Further, we define Manager Only NG Switch (Analyst Only NG Switch) equal to one if managers (analysts) are the only party reporting a non- GAAP switch. Essentially, this estimation classifies the sample into four groups: (1) our benchmark firms (as previously described); (2) firms with both manager- and analyst-reported non-gaap switches; (3) firms with only manager-reported non-gaap switches; and (4) firms with only analyst-reported non-gaap switches. If auditors consider confirmatory signals more 12 The appendix provides detailed definitions of all variables. Inferences are unchanged replacing GCO with First GCO, an indicator variable equal to one for firms with a GCO in year t but without a GCO in t-1 or t-2. 16

17 credible, they will be less likely to issue a GCO when both managers and analysts paint a more favorable picture of the firm s financial condition than otherwise suggested by GAAP earnings. Accordingly, H2 predicts a negative coefficient on Confirmatory NG Switch. Following prior literature, the regression model controls for several factors that affect auditors decisions to issue a GCO. Specifically, we expect a negative relation between GCO and firm size (LnAssets) as large firms have more negotiation power with lenders in the event of covenant violations and are more likely to survive financial difficulties. The probability the auditor issues a GCO should increase as the firm s financial and stock performance deteriorates, suggesting a negative coefficient on CFO, ROA, and Return and a positive coefficient on Loss and Lag Loss. Because risky firms are more likely to fail, we predict a positive coefficient on Return Volatility and a negative coefficient on Zscore, an inverse measure of the risk of bankruptcy (Altman 1968). The ability to raise new capital reduces the likelihood of failure, suggesting a negative relation between GCO and Finance and ΔLeverage. Liquidity and solvency are important predictors of bankruptcy risk, and thus the coefficient on Liquidity should be negative, whereas the coefficient on Leverage should be positive. To the extent that young firms are more prone to failure, we expect the coefficient on LnAge to be negative. Carcello et al. (1995) find that GCOs are associated with longer reporting delays. Accordingly, we predict a positive relation between GCO and Delay. Several studies document evidence consistent with GCOs reflecting financial reporting quality (e.g., Francis and Krishnan 1999; Bartov et al. 2000; Bradshaw et al. 2001). We measure financial reporting quality using DisAcc and expect its coefficient to be positive. Large auditors are more likely issue GCOs because of reputation and litigation concerns (Mutchler et al. 1997), suggesting a positive coefficient on Big 4. Unique to our study, we include special items reported under GAAP 17

18 (Special Items) as they are likely sources of non-gaap exclusions, and thus could subsume the effect of non-gaap earnings on GCOs. Finally, we include industry and year fixed effects to control for industry-wide and macro-level factors that affect auditors decisions to issue GCOs. Our third hypothesis compares the weight that auditors place on non-gaap switches in their going concern assessments with the weight implied by a bankruptcy prediction model. This test uses the bankruptcy prediction model as a benchmark against which to assess auditors weighting of non-gaap earnings in their going concern assessments. We specify the probit bankruptcy prediction model as follows: Bankruptcy t+1 = α0 + α1ng Switch Variablet + α2controlst + Industry FE + Year FE + ε (2) Bankruptcyt+1 is an indicator variable equal to one if the distressed firm files for bankruptcy in year t +1. Because this is a prediction model, we measure bankruptcy in year t +1 and all other variables in year t. We use the same NG Switch Variable specifications and control variables as defined above for the going concern model. Our interest lies in a comparison of the coefficient estimates on NG Switch Variable in equations (1) and (2). A similar magnitude on these coefficients suggests that auditors incorporate the information in non-gaap switches in their going concern assessments, consistent with H3. A coefficient that is smaller in magnitude in equation (1) suggests that auditors underreact to (or are conservative in interpreting) the information in non-gaap earnings switches for future performance. In contrast, a coefficient that is larger in magnitude suggests that auditors overreact to (or are misled by) opportunistic non-gaap switches. Sample Table 1 describes the sample selection process. We begin with 50,185 firm-year observations during 2004 to 2015 with audit opinions in Audit Analytics and the financial statement data in 18

19 Compustat to calculate the control variables. 13 We exclude financial institutions and utilities because these industries have unique financial characteristics that predict bankruptcy and auditors decision to issue a GCO (Geiger and Raghunandan 2002). We also limit the sample to financially distressed firms for which the going concern decision is most salient (Hopwood et al. 1994; Mutchler et al. 1997; Reynolds and Francis 2000; DeFond et al. 2002; Geiger and Raghunandan 2002). We define financially distressed firms as those that either report negative GAAP earnings or negative operating cash flows. We further require stock return data in CRSP and annual actual EPS in IBES which we use as our proxy for analyst-reported non-gaap earnings (e.g., Brown 2001; Bradshaw and Sloan 2002; Abarbanell and Lehavy 2003; Brown and Shivakumar 2003; Doyle et al. 2003; Doyle et al. 2013). Finally, we require the firm to be included in the database compiled by Bentley et al. (2018) which we use to extract quarterly manager-reported non-gaap earnings. 14 If non-gaap earnings is missing in any, but not all, of the quarters during a fiscal year, we fill in GAAP earnings for that quarter. 15 We then calculate annual manager-reported non-gaap earnings by aggregating across quarters in a fiscal year. 16 After these data restrictions, our final sample includes 6,861 firm-year observations. 13 Specifically, our sample includes fiscal years beginning after March 28, 2003, the effective date of Regulation G, and ends on December 31, 2015, prior to the effective date of Accounting Standards Update No We thank Bentley et al. (2018) for access to this data available at The database contains non-gaap earnings for firms with a quarterly 8-K earnings announcement on EDGAR. 15 This approach assumes that the firms included in the database without manager-reported non-gaap earnings in a given quarter implicitly agree with the GAAP quarterly earnings number for that quarter. Inferences are unchanged if we define Manager NG Switch equal to one if there is a manager-reported non-gaap earnings switch in the fourth quarter or if any quarter in the fiscal year has a manager-reported non-gaap earnings switch. 16 The Bentley et al. (2018) data contains non-gaap EPS numbers. Because the number of shares outstanding can change throughout the year, we first multiple the quarterly non-gaap EPS amount by the number of diluted shares used to calculate EPS (Compustat CSHFDQ) to get a raw, quarterly earnings number. We aggregate this measure for each firm fiscal year, and then divide by the annual number of diluted shares used to calculate EPS (Compustat CSHFD) to get to the annual manager-reported non-gaap EPS amount. 19

20 IV. RESULTS Univariate Analysis Table 2, Panel A reports descriptive statistics for the regression variables used in our empirical tests. We find that 5.1 percent of sample firms receive a GCO, but only 1.3 percent file for bankruptcy in the next year, consistent with auditors conservatively reporting going concern uncertainties. Almost one-quarter of the sample reports a non-gaap switch. 17 Confirmatory non-gaap switches (0.145) are most common, followed by analyst-only non-gaap switches (0.077) and manager-only non-gaap switches (0.022). 18 The mean and median of CFO and ROA are negative, and 92 percent of the sample reports a GAAP loss. The mean (median) Return is ( 0.132). These weak financial and market-based performance measures reflect the sample composition of financially distressed firms. The mean and median delay in 10-K filing is approximately 71 days, consistent with longer reporting delays for financially distressed firms. Finally, the Big 4 audits 75.4 percent of the sample. Panel B of Table 2 reports Pearson correlations. As expected, GCO and Bankruptcyt+1 are positively correlated. Consistent with H1, GCO and NG Switch are negatively correlated. Thus, at the univariate level, auditors are less likely to issue GCOs when non-gaap earnings cross the profit threshold. NG Switch is also negatively related to Bankruptcyt+1, consistent with non- GAAP switches indicating a higher likelihood of future profitability (Leung and Veenman 2018). Panel B also reveals that GCOs are positively correlated with GAAP losses, leverage, 17 This figure is lower than the 40.4 percent of the sample in Leung and Veenman (2018) that convert a quarterly GAAP loss to a non-gaap profit. This difference is likely attributable to their focus on firm-quarters compared to our focus on firm-years. In other words, firms may have more than one non-gaap switch quarter per year. 18 Bentley et al. (2018) finds that manager and analyst non-gaap earnings are the same for 63% of their sample. For our sample, 87% of manager non-gaap switch firms are also analyst non-gaap switch firms (calculated as confirmatory switches out of manager switches). The larger overlap in our sample is likely attributable to our focus on the non-gaap profit threshold rather than the precise non-gaap earnings amounts. 20

21 bankruptcy risk, discretionary accruals, and reporting delays, and negatively correlated with firm size, age, performance, liquidity, and the presence of a Big 4 auditor. Table 3, Panel A presents univariate tests of differences in means between non-gaap switch firms (NG Switch = 1) and those without a non-gaap switch (NG Switch = 0). Consistent with H1, mean GCO is significantly lower for non-gaap switch firms. Less than one percent of non- GAAP switch firms receive a GCO compared to 6.6 percent of firms without a non-gaap switch. Subsequent bankruptcy is also less common for non-gaap switch firms, with less than one percent of non-gaap switch firms declaring bankruptcy in year t+1 compared to 1.6 percent of firms without a non-gaap switch. Several firm characteristics also vary with NG Switch. Specifically, non-gaap switch firms are significantly larger and have relatively better financial performance (CFO and ROA) in the current year. This finding is consistent with Leung and Veenman (2018), who find loss firms are more likely to report non-gaap earnings when the underlying cash flows are strong. Although non-gaap switch firms are more likely to have a current year GAAP loss, by definition, they are less likely to report a loss in the prior year. In addition, non-gaap switch firms have higher leverage, lower liquidity, and lower risk of bankruptcy. They are also more mature, more likely to have a Big 4 auditor, and file their 10-K in a more timely fashion. Thus, by some measures the non-gaap switch firms appear less financially distressed while other measures suggest greater financial distress. Nonetheless, the results reported in this panel highlight the importance of controlling for differences in firm characteristics in evaluating the relation between NG Switch and GCO. We compare mean earnings magnitudes for the manager and analyst non-gaap switch samples in Table 3, Panel B. Of the 1,685 non-gaap switch observations in our sample, 1,157 have a manager-reported non-gaap switch, 1,521 have an analyst-reported non-gaap switch, 21

22 and 993 have both (i.e. Confirmatory NG Switch = 1). By construction, the GAAP EPS mean is negative and the non-gaap EPS and exclusion means are all positive. The mean Manager NG Exclusion and Analyst NG Exclusion is and 2.469, respectively. As these magnitudes are measured on a per share basis, this finding implies that the non-gaap switch firms have significant swings from a GAAP loss to non-gaap earnings, whether reported by managers, analysts, or both. 19 Regression results Table 4 reports the results of estimating equation (1). 20 Model 1 includes NG Switch as the primary variable of interest, while Models 2 and 3 consider the alternative sources of non-gaap switches, Manager NG Switch and Analyst NG Switch, respectively. The coefficient estimate on NG Switch in Model 1 is negative and significant, indicating that auditors are less likely to issue a GCO when there is a non-gaap switch. We also find a significantly negative association between GCOs and both manager-reported (Model 2) and analyst-reported (Model 3) non-gaap switches. Further, the coefficient estimate on Manager NG Switch ( 0.341) is less than that on Analyst NG Switch ( 0.455), although the difference is not statistically significant ( 2 = 0.67; p- value = 0.41). Thus, the results reported in Table 4 support H1, but do not support an inference that auditors place less weight on manager-reported non-gaap switches in their going concern assessments. Results for the control variables are generally consistent with predictions. Across all models, GCO is negatively related to LnAssets, CFO, ROA, Return, Zscore, Liquid, and ΔLeverage, and 19 For example, in fiscal year 2012, HP Inc. reported a GAAP loss of ($6.41), but a non-gaap profit of $4.05, indicating a Manager NG Exclusion of $ HP s 2012 earnings release can be found at: 20 The number of observations is less than the sample size reported in Table 1 due to the use of fixed effects in our probit regression models. Probit models require variation in the dependent variable for each fixed effect group, otherwise the observations in groups without variation are dropped. 22

23 positively related to Loss, Lag Loss and Delay. In other words, larger firms with stronger performance and lower bankruptcy risk are less likely to receive GCOs. In contrast, firms with GAAP losses and longer reporting delays are more likely to receive a GCO. Although the findings in Table 4 suggest that auditors consider both manager and analyst non-gaap switches, these tests do not fully isolate switches reported by only managers, only analysts, or by both. For example, Manager NG Switch indicates a manager-reported non-gaap switch regardless of whether analysts confirm or dispute the switch. Thus, we re-estimate equation (1) using Confirmatory NG Switch, Manager Only NG Switch and Analyst Only NG Switch as our variables of interest. Table 5 reports the results of this analysis, with Model 1 including Confirmatory NG Switch and Model 2 including all three non-gaap switch variables. In both models, we find a significant negative coefficient estimate on Confirmatory NG Switch, indicating that auditors are less likely to issue a GCO when managers and analysts report confirmatory evidence of a non- GAAP switch. This finding supports H2 and is consistent with AS 1105 suggesting that auditors should consider information more reliable when two or more sources of evidence are consistent. In Model 2, we also find a negative and incrementally significant coefficient on Analyst Only NG Switch, implying that auditors consider analyst-reported non-gaap switches as good news for distressed firms regardless of whether managers also report a non-gaap switch. This finding is consistent with the suggestion in AS 1105 that information is more reliable when sourced from outside the company. In contrast, the coefficient on Manager Only NG Switch is insignificant, suggesting that auditors react skeptically to manager-reported non-gaap switches unless corroborated by analysts. Overall, the evidence in Table 5 suggests that auditors incorporate non- GAAP switches in their going concern assessments when this information is perceived to be 23

24 more credible, either because two pieces of corroborating evidence are available or because the evidence is obtained from a source independent of the client. 21 Our analyses so far have not differentiated between firms without non-gaap earnings and firms with non-gaap earnings that do not switch to profit. Ex ante, we have no reason to believe that audit opinions should differ between the two groups. Accordingly, we combine them into one benchmark group. As a sensitivity check, we refine our research design to allow the effect of non-gaap earnings on the probability of auditors issuing GCOs to differ between the two groups. Specifically, we re-estimate the GCO model including a non-gaap no-switch indicator variable equal to one if a non-gaap earnings number is reported (either by managers or analysts), but does not switch the firm from a loss to a profit. In this analysis, the benchmark group contains firms without manager or analyst non-gaap earnings. In untabulated results, we continue to find a significantly negative coefficient on NG Switch, Confirmatory NG Switch, and Analyst Only NG Switch, and an insignificant coefficient on Manager Only NG Switch. The non- GAAP no-switch variable is insignificant. These results suggest that auditors going concern assessments are unaffected by the mere presence of non-gaap earnings unless the non-gaap earnings creates a switch to profit. We next compare the weight auditors place on non-gaap switches in their going concern assessments to the weight implied by the bankruptcy prediction model. Analogous to Table 4, the bankruptcy prediction results in Table 6, Panel A include NG Switch, Manager NG Switch, and Analyst NG Switch as the non-gaap switch variables. In all three estimations, the non-gaap switch variables are negative and significant, indicating that the likelihood of bankruptcy in year 21 Because GCOs are relatively rare events, we re-estimate the models in Tables 4 and 5 using a penalized likelihood model to reduce small sample bias (Firth 1993) with no change in inferences (untabulated). Tests of the bankruptcy prediction model reported below are also robust to this alternative estimation procedure. 24