Familiarity Bias and Auditor Judgments on Going Concern Modifications, Schemes and Mind Maps of Accounting

Download Familiarity Bias and Auditor Judgments on Going Concern Modifications and more Schemes and Mind Maps Accounting in PDF only on Docsity! 1 Familiarity Bias and the Propensity to Issue Going Concern Opinions Abstract This study examines the association between familiarity bias and audit firm judgements using going concern opinion modifications. We conjecture that familiarity bias is highest when an audit firm is least familiar with its clients. We measure familiarity bias using auditor changes because familiarity bias is likely to be the highest on the initial year of an audit relationship. We find the propensity to issue a going concern modification to be much higher during this initial period, even after controlling for a multitude of factors known to be determinants of going concern modifications. Our findings also suggest the phenomenon appears to slowly dissipate over a five- year auditor client tenure. Supporting our theoretical prediction, further results find that as the auditor is more familiar with the client’s industry (e.g. industry expertise) the main effect is moderated. Our results are also strongest when the successor auditor is located in a different locale than the previous auditor, situations leading an even more pronounced lack of familiarity. We also examine SOX 404 internal controls impact, finding auditors are more likely to deem internal controls ineffective when they are unfamiliar with clients. Keywords: Familiarity bias, going concern, internal control weaknesses, auditor judgment 2 I. Introduction We examine familiarity bias in relation to audit firm judgements using going concern opinion modifications. We define familiarity bias as a disproportionate weight in favor of those clients who auditors have greater knowledge of their business. Thus, familiarity bias is highest when an audit firm is least familiar with its clients. We measure familiarity bias using auditor changes because familiarity bias is likely to be the highest in the initial stages of an audit relationship. Familiarity bias has been shown to influence financial market participants’ investment choices (Kang and Stulz 1997; Coval and Moskowitz 1999; Huberman 2001; Li 2004; Ivkovic and Weisbenner 2005; Massa and Simonov 2006; Nofsinger and Varma 2002; Riff and Yagil 2016; and Schumacher 2017). To date, there has been little research regarding whether familiarity bias influences financial market monitors such as auditors. We conjecture that familiarity bias is more likely reflected in auditor’s areas regarding judgment such as issuing going concern opinion modifications. Going concern opinion modifications are important to financial statement users. Going concern opinion modifications signal a company’s operating uncertainties over the next fiscal year (PCAOB 2003). Auditors’ evaluations of these uncertainties are made based on knowledge obtained from audit procedures and knowledge of conditions and events existing at or prior to the completion of fieldwork. The PCAOB Standard Advisory Group Meeting in May 2012 introduced discussion points for going concern communication and evaluation, including the definition of thresholds for substantial doubt, additional details relating to the auditor’s conclusion, and required procedures when evaluating going concerns (PCAOB 2012). While regulators consider whether current going concern guidance and standards adequately disclose relevant company uncertainties (FASB 2013; PCAOB 2012; Tysiac 2014; PCAOB 2016; PCAOB 2017), this study contributes to that discussion because going concern opinion modifications require auditors to process complex information, which represents a setting where familiarity bias may emerge. Prior literature suggests that the auditor client relationship is a determinant for auditors when issuing going concern opinion modifications (Carson, Fargher, Geiger, Lennox, Raghunandun, and Willekens 2013). We use changes in auditor as our familiarity bias measure as we conjecture that familiarity bias is highest during the first year of an auditor client relationship. 5 firm concludes there is substantial doubt that companies will survive the next financial year. In order to determine whether there is substantial doubt regarding survival, auditors use judgment when evaluating both financial and non-financial information, such as strategic initiatives and mitigating factors. Although the going concern opinion signals uncertainty, questions continue to exist about whether audit firms and companies adequately disclose information about operating uncertainties (PCAOB 2016; PCAOB 2012; FASB 2013; PCAOB 2015a; FASB 2016). Recent enhancements to management’s disclosures as well as the PCAOB’s continued discussion about going concern requirements suggest that management and audit firm communications with investors do not adequately inform investors about the uncertainty surrounding companies operations. The ability to accurately identify a company with a going concern issue is critical for external stakeholders evaluating companies’ financial condition (U.S. House of Representatives 2002). B. Familiarity Bias Familiarity bias is a common heuristic individuals use to help categorize data and make decisions when analyzing large amounts of complex information. Familiarity is evidenced to lead individuals to assign greater value to more familiar options (Fox and Levav 2000). Fox and Levav (2000) find that individuals are biased to view less familiar events as less likely to occur. Based on these finding, we suggest familiarity bias may lead auditors to issue going concern audit opinion modifications in the year clients switch to new auditors. Prior research suggests familiarity bias is associated with binary choice inferences. Specifically, individuals are predicted to infer that the more familiar object in a pair has a higher criterion value on the to be judged dimension (Honda, Abe, Matsuka, Yamagishi 2010). Extending these results to our study suggests that auditors who are more familiar with their clients are more likely to not issue going concern audit opinion modifications. Thus, when auditors assess going concern opinion modifications in the initial year the client switched auditors, familiarity bias may lead to a higher probability of issuing going concern opinion modifications. The majority of familiarity bias research within capital markets has focused on investment choices (Huberman 2001; Ivkovic and Weisbenner 2005; Nofsinger and Varma 2002; Riff and Yagil 2016; and Schumacher 2017). Huberman (2001) finds that investors are more likely to hold shares in the phone company that services them, and Ivkovic and Weisbenner (2005) 6 suggest that investors disproportionately hold companies that are headquartered within 250 miles of their home. Further research examined whether professional fund managers would be less prone to exhibit familiarity bias. The findings provide evidence that sophisticated investors also exhibit familiarity bias (Schumacher 2017, Riff and Yagil 2016, Nofsinger and Varma 2012). Thus, as a whole, financial market participants seem to assign greater (less) value to those firms with which they are familiar (unfamiliar). Additionally, credit rating agencies are evidenced to exhibit familiarity bias. Ayres and Dolvin (2019) suggest first time credit ratings are lower because the credit rating agencies are less familiar with companies during the initial rating process. We extend the familiarity bias literature stream specific to capital markets by examining familiarity bias in relation to financial markets’ monitors, auditors. C. Auditor Judgment and Decisions Auditor judgment is critical for decisions regarding uncertainty that lead to going concern opinion modifications. Prior literature documents that long-term memory on engagements that involve consideration of further evidence to make decisions impacts auditors’ judgment (Plumlee 1985, Moeckel and Plumlee 1989). This finding relies on the premise that auditors use long-term memory to store evidence gathered for particular clients. Tan (1995) finds that auditor changes reduce the tendency to focus on more consistent facts arising from repeat engagements. For initial audit engagements, the resulting absence of this long-term memory from repeat engagements gives rise to a lack of familiarity and the bias that ultimately accompanies it. Prior research documents that auditors’ judgments are affected by biases (Trotman Tan Ang 2011). Peecher and Piercey (2008) consider the evaluation of audit quality when adverse outcomes exist and find adverse outcomes could bias individual judgments. Going concern opinion modifications are adverse audit outcome and an audit quality measure (PCAOB 2015b). We extend the literature to examine whether familiarity bias are associated with uncertainty judgments related to issuing going concern opinion modifications. D. Determinants of Going Concern Opinions Prior research documents associations between auditor switching and going concern opinion modifications. The prevailing evidence documents a relationship between auditors issuing going concern opinion modifications and clients switching auditors (e.g., Chow and Rice 1982; Smith 7 1986; Geiger et al. 1998; Lennox 2000; Carcello and Neal 2003; Vanstraelen 2003; Chan et al. 2006). However, Hoitash and Hoitash (2009) findings support fewer clients dismiss their auditors following going concern opinion modifications in the post-SOX era. Our study differs from this stream of research because we are examining the propensity to issue a going concern opinion modification in the year a client changes auditors. An unresolved question in the literature is whether switching auditors is successful in terms of removing the going concern opinion modification, which is often referred to as opinion shopping. Early research establishes no association between switching auditors and subsequent improvements in audit opinions (Chow and Rice 1982; Krishnan 1994; Krishnan and Stephens 1995; Krishnan and Krishnan 1996; Geiger et al. 1998). Lennox (2000) does not examine actual auditor switch decisions; instead he analyzes opinion shopping by predicting the opinions that clients would have received if they made switch decisions opposite to those that actually occur. Lennox (2000) suggests that clients would have received less favorable audit opinions if clients made switch decisions opposite to those actually observed. Our measure of familiarity bias is clients that choose to switch auditors. We examine whether familiarity bias is associated to going concern opinion modifications in the initial year that clients make the decision to switch auditors. Our study contributes to the inconclusive evidence between auditor switching and going concern opinion modifications. A related stream of literature examines auditor-client tenure as a determinant for issuing going concern opinion modifications. Geiger and Raghunandan (2002) find a positive association between auditor tenure and the propensity to issue going concern opinion modifications prior to bankruptcy. Read and Yezegel (2016) examine bankrupt clients and find no association between going concern opinion modifications and auditor tenure for Big 4 audit firms. Their results document non-Big 4 firms are less likely to issue going concern opinion modifications in initial audit years. Given competing arguments suggesting whether auditor tenure is associated with the likelihood of issuing going concern opinion modifications, we state the following hypothesis in the null: H1: Audit firms’ familiarity bias is not associated with the likelihood of issuing going- concern opinion modifications. III. Methodology and Results 10 (Carcello and Nagy 2004; Carey and Simnett 2006; DeFond et al. 2002; Lim and Tan 2008). The age of the audit client is measured using the natural log of the years (plus one) since the firm first went public. The size of the audit client is measured using the natural log of the book value of its assets as of the end of the fiscal year. Both audit and non-audit fees have long been hypothesized to influence audit outcomes (Basioudis et al. 2008; Blay and Geiger 2013; Lim and Tan 2008; Robinson 2008); we thus incorporate two fees related variables. The first is ABNORMAL_FEES_PCTit and it measures the total actual client fees as a percentage of expected total fees. It is produced by regressing the natural log of total fees upon the other covariates in the econometric model and then predicting an outcome. Dividing the actual amount of fees by the predicted level of fees gives our measurement. This is done because if fees do influence reporting outcomes, it is most likely to manifest for the most lucrative clientele.6 The second fees related variable is CLIENT_IMPORTANCEit. It is measured as the observation’s total fees as a percentage of the auditor’s total fees for that given calendar year. Similar to abnormal fees, auditor judgement and independence may break down as a client becomes more important to the audit firm (Li 2009). This may change going concern reporting behavior. We also include a bevy of financial performance related controls as the going concern opinion is heavily influence by financial outcomes. The first of these is LEVERAGEit and measures the audit client’s exposure to financial leverage; financial leverage increases the risk of bankruptcy for a firm. It is measured as the ratio of total liabilities to total assets (DeFond et al. 2002). The second of these variables is LOSSit (Bruynseels and Cardinaels 2014). Suffering a net loss, through the correlated loss of cash flow, increases bankruptcy risk. It is measured as a binary variable equal to one if the observation had negative net income for the year ended. The third financial control variable is ROAit. The higher a firm’s return on assets, the lower its bankruptcy risk, ceteris paribus. This variable is measured as net income before special items as a percentage of total assets. The fourth financial control variable is INT_COVERAGEit. It measures a firm’s ability to meet its interest obligations as this ability reduces bankruptcy risk. It 6 This is also done out of statistical practicalities. Total fees or the natural log thereof are highly related to both the size and the age of a firm, two variables already included in the model. This technique effectively orthogonalizes fees to the other variables in the model, reducing complications that might arise from multicollinearity. 11 is measured as interest expense as a percentage of net income before interest and taxes.7 The fifth financial control variable is GROWTHit. It measures the percentage change in revenue for the client firm from the previous year.8 Growth has an impact on financial performance and going concern reporting (Johnson et al. 2002). The sixth financial control variable is CURRENT_RATIOit and is the ratio of current assets to current liabilities. This essentially measures the liquidity of the client firm and liquidity is often associated with a diminished chance of bankruptcy, especially in the short term. The seventh financial control is MKTBKit (Johnson et al. 2002). It is measured as the ratio of the market value of equity and liabilities to the book value of assets (e.g., equity and liabilities). Ceteris paribus, firms with higher market values typically are thought to have brighter futures while depressed values are potential signals of future financial difficulties. The eighth financial control is ALTMANit (Altman 1968) and is used to measure financial distressed. Financial distress measures such as Altman and Zmijewski (Zmijewski 1984) have been prominent aspect of modeling going concern behavior (Bhaskar et al. 2017; Carcello and Neal 2000, 2003; DeFond et al. 2002; Johnson et al. 2002). The ninth financial control is INVESTMENTSit. It is measured as the ratio of cash and short – term investments as a percentage of total assets. Similar to the current ratio, it is another measure of liquidity and a proxy for risk of bankruptcy. We also control for some auditor characteristics as these might also impact the going concern decision. The first of these is BIGNit. This is a binary variable equal to one if the current year auditor is one of the “big four” accounting firms. The size of the auditor has been associated with the propensity to issue a going concern (Boone et al. 2010; Kaplan and Williams 2012). The second auditor characteristic control is IC_INEFFECTIVEit. This is also a binary variable equal to a one if the auditor deemed the client’s internal control to be ineffective in its opinion about the operating effectiveness of internal control. Such instances have been documented to more likely result in going concern opinions (Goh et al. 2013; Hammersley et al. 2012). The final control for auditor characteristics is LN_IND_CLIENTSit. It is measured as the natural log 7 This measurement is opposite of conventional ways to measure this construct. However, by placing interest expense in the numerator, we avoid missing observations that might arise from have a zero denominator observation for firms that do not have interest expense. It is important to keep these types of firms in the sample as they are usually among the healthiest and are less likely to receive a going concern opinion. 8 Revenues are employed here instead of profits to avoid the complications that arise from computing growth for negative numbers. 12 of the number of clients the audit firm has within that particular two digit SIC code for the given calendar year. Industry specialization and industry expertise have been shown to impact audit outcomes (Reichelt and Wang 2010). C. Descriptive Statistics Table 1 summarizes the descriptive statistics for our primary sample of 35,188 firm-years. Columns 1, 2, and 3 detail the mean, median, and standard deviation of each variable for the entire sample. Going concern opinions occur in approximately 9.2% of all observations. This is slightly higher than the rate (8%) noted by DeFond et al. (2002), but our period incorporates the financial crisis while theirs does not. Changes in auditors occur in approximately 8.3% of all observations. Approximately 56% of all observations are audited by one of the “big four” auditors and 2.8% of the observations are deemed to have some aspect of ineffective internal controls. The average client is 4.2% of the audit firms public client base, but the median value is only 0.01, which reflects the vast differences in client portfolios between the “big four” audit firms and the smaller auditors who undertake public company audits. INSERT TABLE 1 HERE Columns 4 and 5 divide the sample into two groups. The first group (Column 4) is the observations for which no change in auditor occurred. The second group (Column 5) are the observations for which a change in auditor did occur. Both columns display the average value for each variable and Column 6 computes the difference between the two sub samples. Column 7 applies difference in means t-tests to determine if the two sub-samples are statistically different from one another. With the exception of CURRENT_RATIOit, all variables exhibit statistical differences between the two sub-samples. Most pertinent of these is the difference in going concerns between the two sub-samples; client firms incurring a switch in auditor receive going concern opinions in 22% of the observations compared to only 8.1% of client firms not incurring a change in auditor. This difference is statistically different at the p-value < 0.01 level. This univariate analysis does provide some support for our main hypothesis. However, as the other variables indicate, substantial differences exist between the two subsamples; thus, a more rigorous multivariate analysis is warranted to control for these other factors that may influence the likelihood of a going concern opinion. This is especially important for the nine financial control variables included in the analysis. 15 receiving a going concern opinion (Chen and Church 1996; Kaplan and Williams 2013; Willenborg and McKeown 2000; Allen D. Blay et al. 2011)12, these results are economically meaningful. We continue our analysis in Table 4. If the phenomenon we are documenting in Table 3 is real, then we should expect to see the effect moderate over time. Our econometric model in Table 3 is too coarse to detect any tapering off since a single binary variable is the variable of interest. To do this, modify the analysis in Table 3 (Columns 2 through 5) and replace SWITCHit with four binary variables to capture a more nuanced analysis. These are FIRST_YEARit, SECOND_YEARit, THIRD_FOURTH_YEARit, and FIFTH_SIXTH_YEARit. FIRST_YEARit is essentially the same measurement as SWITCHit, it is equal to one if the observation is the first year for the auditor-client relationship. SECOND_YEARit is equal to one if it is the second year of the relationship, THIRD_FOURTH_YEARit is equal to one if it is the third or fourth year and FIFTH_SIXTH_YEARit is equal to one if it is the fifth or sixth year. All other observations fall into the intercept. INSERT TABLE 4 HERE To the extent that familiarity bias decreases with time, we should see our main effect moderate through each of the coefficients for these variables. The results of Table 4 appear to support this notion. For instance, in Columns 1, the coefficients for these variable begin at 0.126 for FIRST_YEARit and proceed to fall in a monotonic sense to 0.087, 0.044 and -0.038 for SECOND_YEARit, THIRD_FOURTH_YEARit, and FIFTH_SIXTH_YEARit, respectively. A very similar pattern emerges in Column 2 with the inclusion of industry and time period fixed effects. Since these two columns are the result of a probit analysis, direct comparisons of the coefficients is extremely difficult since probit coefficients are not cardinal in nature. As a result, we compute both the average marginal effects for these results as well as statistically test the differences between the coefficients. The average marginal effects for Column 1 are 0.009 (FIRST_YEARit), 0.006 (SECOND_YEARit), 0.003 (THIRD_FOURTH_YEARit) and -0.003 12 Such economic implications include 1) substantial investor price reactions to a going concern opinion or a lessened reaction to a bankruptcy announcement after receiving a going concern opinion, 2) changed investor behavior to how they use the financial statements and what information they deem valuable, 3) the delisting of an initial public offering, and 4) reduced exposure to litigation costs for the auditor. 16 (FIFTH_SIXTH_YEARit). Average marginal effects for Column 2 are similar, 0.013, 0.010, 0.005, and -0.001 respectively. The results of Columns 3 and 4 also support the results of Columns 1 and 2. Overall, this supports a pronounced monotonic decline in the probability of a going concern as the auditor becomes more familiar with the client firm. Statistical tests of the differences between these coefficients also support this finding. A chi- squared test of FIRST_YEARit = SECOND_YEARit does not yield statistical differences for either column (chi-squared = 0.98, 0.93), but the FIRST_YEARit coefficients are statistically different from the THIRD_FOURTH_YEARit (chi-squared = 3.47, 5.48) and the FIFTH_SIXTH_YEARit (chi-squared = 9.05, 13.37) coefficients. The SECOND_YEARit coefficient is not statistically different from the THIRD_FOURTH_YEARit coefficient (chi-squared = 1.10, 2.49) but is different from the FIFTH_SIXTH_YEARit (chi-squared = 5.31, 8.84) coefficient. Also, the THIRD_FOURTH_YEARit coefficient is statistically different (chi-squared = 3.17, 4.52) from the FIFTH_SIXTH_YEARit coefficient. Overall, these results support the notion of gradual changes in familiarity bias as the length of the auditor / client relationship increases over time. If familiarity bias is a root cause of the results in Table 3, it is likely stronger in certain settings and weaker in others. Tables 5 and 6 obtain insight on such settings. One such situation might be cases where as the audit firm stays the same but perhaps a different office takes on the engagement. Another situation might be where the auditor changes and another auditor from a different city takes on the engagement. We would expect familiarity bias to be lowest when a different office of the same auditor conducts the audit, increasing when a new auditor in the same city takes on the engagement, and being highest when a new auditor from a different city takes on the engagement. The city is important because the geographic locale can convey a bevy of information to the auditor about the client and reduce familiarity bias. In Table 5 we create three new variables to capture these potential outcomes and they displace SWITCHit from Table 3. The first is AUDITOR_SWITCHit . This is a binary variable equal to one if the audit firm changed from the prior year, but the successor audit firm was from the same geographic city, zero otherwise. The second new variable is OFFICE_SWITCHit. It is also a binary variable and is equal to one if the audit firm did not change from the prior year but the office of the audit firm did, zero otherwise. The third new variable is DUAL_SWITCHit. It is a binary variable equal to one only if the audit firm changed from the prior year and the office of 17 the successor firm is from a different city than the prior audit firm.13 Table 5 thus mimics Table 3 but with these three variables replacing SWITCHit. INSERT TABLE 5 HERE The results in Table 5 are interesting and support our notions in regard to familiarity bias being more pronounced in certain settings. In the probit models of Columns 1 and 2, AUDITOR_SWITCHit is positive and significant in Column 2. This suggests going concern opinions are more likely when there is a change in auditor, even if the successor auditor is from the same locale. OFFICE_SWITCHit is also positive but statistically insignificant. As a result, we cannot determine if a change in auditor office has any impact on familiarity bias. The most impactful result lies with DUAL_SWITCHit, it is positive and highly significant across all four specifications. It appears as if familiarity bias is most pronounced when there is a new auditor from a different locale than the original auditor. Interestingly and despite the fact that AUDITOR_SWITCHit is positive in all cases and significant in one of them, DUAL_SWITCHit, aside from being statistically different from zero is also statistically different from AUDITOR_SWITCHit. 14 This suggests that it is the confluence of both an auditor change and a change in the auditor locale that impacts familiarity bias the most. This matches our expectations for the setting with the highest level of familiarity bias. These results and test lend support to the main findings in Table 3. In Table 6, we conduct several cross-sectional tests to determine if additional settings cause familiarity bias on the part of auditors to ebb and flow. We speculate that familiarity bias is diminished when the auditor has extensive prior experience with similar types of client firms. Similar types of client firms would most likely be represented by client firms that are in the same industry. As a result, our initial control variable LN_IND_CLIENTSit is an excellent variable to interact with SWITCHit. If industry exposure reduces familiarity bias, we expect this variable to 13 The formulation of these variables essentially breaks the initial variable of interest, SWITCHit, down into two components, AUDITOR_SWITCHit and DUAL_SWITCHit. In our sample of 35,188 observations, 2,923 are coded as a one for SWITCHit. Of these, 1,644 are where the audit firm was replaced by a successor firm in the same city (AUDITOR_SWITCHit = 1) and 1,279 were instances where the audit firm was replaced by a successor firm in a different city (DUAL_SWITCHit = 1). There are also 1,253 observations whereas the audit firm did not change but the office location did (OFFICE_SWITCHit = 1). 14 Chi-squared tests of Columns 1 and 2 support this with chi-squared values of 14.35 and 12.72, respectively. The same can be said for the OLS regressions in Columns 3 and 4, f-tests of DUAL_SWITCHit = AUDITOR_SWITCHit result in f-statistics of 31.74 and 30.43, respectively. 20 Since ineffective internal controls appear in only 2.8% of the sample, a first year audit raises the unconditional probability significantly at 67.8% (1.9% / 2.8%) and 71.4% (2.0% / 2.8%) for Columns 1 and 2, respectively. The results for Columns 5 and 6 also further support this story; the number of internal control weaknesses also appears to be heavily influenced by an auditor’s familiarity bias. We also perform several untabulated analyses. In rendering an opinion that internal controls are ineffective, the auditor has several routes with which to make that determination. One route would be the judgment that the client simply misapplied the appropriate financial reporting framework. Another would result from an incident of fraud or deliberate misrepresentations on the part of management. A third route would be that internal control is simply not properly designed, resulting in a lack of separation of duties. The Audit Analytics database codes instances where internal control has been ineffective along these three dimensions.18 We thus create three additional binary variables, RULE_FAILUREit, FRAUD_IRREGit,and IC_WEAKit to capture these variables, respectively. We then reperform Columns 1 and 2 of Table 7 with each of these as dependent variables. The results overwhelmingly support the notion that a change in auditor positively impacts the instance of misapplication of the financial reporting framework (RULE_FAILUREit) and a flaw in internal control systems (IC_WEAKit) as all coefficients are positive and highly significant from a statistical standpoint (p-value < 0.01). However, fraud and other financial reporting irregularities (FRAUD_IRREGit) does not attain a result. We interpret these results as supporting our findings in Table 7, the two instances with results in this analysis are most likely the two with the most auditor judgment. Furthermore, an auditor would not likely accuse a client of fraud unless it is essentially clear from the facts before them. The uncertainty of how accounting rules are applied and whether internal control systems are adequate would give enough leeway for auditor familiarity bias to potentially manifest in rendering an opinion that overall internal control is ineffective. IV. Additional Robustness Tests 18 An observation with ineffective internal control could have one of these or all of them as underlying reasons for ineffective control. 21 A. Fixed Effects Models We further explore our results with a heightened level of econometric rigor and introduce auditor fixed effects to our analysis in Table 3. Auditor fixed effects allow us, in theory, to control for invariant auditor traits, including the propensity to issue a going concern report. We first replicate Columns 2 through 5 of Table 3, employing an indicator variable for each audit firm. When this is performed, the results of Table continue to hold. For instance, the coefficients of SWITCHit for Columns 2 and 3 become 0.145 and 0.162, respectively, and remain statistically significant at the p-value <0.01 level.19 The same pattern holds for the replication of Columns 4 and 5 (OLS regressions). Our primary results are thus robust to the inclusion of auditor fixed effects. We also replicate these analyses with the inclusion of audit client firm fixed effects to control for latent invariant traits at the client-firm level. The results for Columns 2 and 3 of Table 3 continue to hold, the coefficients of SWITCHit become 0.170 and 0.179, respectively and retain statistical significance at the p-value <0.10 level.20 Replication of Columns 4 and 5 of Table 3 using audit client firm fixed effects also hold with positive coefficients that are statistically significant at the p-value < 0.05 levels. Overall, these fixed effects analyses support the notion that latent and endogenous traits at either the auditor level or the audit client level are not driving our primary results within this manuscript. B. Lagged Versions of the Dependent Variable Another robustness analysis we conduct is to re-perform our main analysis in Table 3 with the prior year’s value for GCit, as an additional control variable. While this type of econometric specification is not common in research whereas going concern opinion outcomes are the dependent variable, we perform this robustness analysis for a couple of reasons. First of all, financially distressed firms tend to be distressed for long periods of time. Second, the issuance of a going concern opinion in the prior year likely makes it easier for an auditor to issue one in the current year, especially if the current year auditor is a new one. Prior literature has identified some evidence that the issuance of a going concern report may lead to a subsequent change in an auditor 19 These analyses do cause a small drop in sample size as the dependent variable is invariant for some auditors. 20 The sample size drops dramatically for these tests as many audit client firms never experience a going concern report modification. The sample size drops to 4,130 audit client firm-year observations. 22 (Carey et al. 2008). Consecutive going concern opinion modification issuances have been associated with audit firms’ lack of available knowledge and resources (Harris, Omer, and Wong 2019). By controlling for a prior going concern opinions, we control for this potential threat to our research design. This adds an additional layer of rigor above and beyond the extensive amount of controls already in place in our research design to control for the client-firm’s level of financial distress. We replicate Table 3 (untabulated) with the inclusion of the prior year’s going concern opinion. For Columns 2 and 3, SWITCHit continues to display positive coefficients (0.161 and 0.191, respectively) which are both highly significant at the p-value < 0.01 level. The same holds for replication of Columns 4 and 5 of Table 3 (OLS regressions), the coeffcients also remain positive and are also statistically significant at the p-value < 0.01 level. As expected and throughout these four regressions, the coefficient for GCit-1 is positive and highly significant, indicating substantial serial correlation in going concern behavior on the part of auditors.21 An average marginal effects analysis reveals that a going concern opinion in the prior year raises the likelihood of one in the current year by 9.6%, a very substantial effect. C. Drop Initial Observations In our original coding for the dataset, we retained all observations for which it was the first year in the Compustat dataset (i.e., ipo and spinoff firms) but coded these with SWITCHit equaling zero. While many of these firms likely retain their original auditor, it is impossible to determine if that is the case or not. We thus conduct a robustness analysis dropping these observations from the overall dataset. When this is performed, our overall inferences remain unchanged throughout our analyses. D. Two-Way Clustering Our main econometric specification clusters the standard errors at the client-firm level. This was done since the depth of our panel is not very long (Petersen 2009). However, we do replicate our main findings utilizing two way clustering over the time and audit client firm dimensions, both 21 For instance, in a replication of Column 3 of Table 3, the coefficient for GCit-1 is 1.928 and has a z-statistic of 35.49. 25 continue a lengthy and robust stream of literature in regard to familiarity bias and how it impacts the economic decisions of individuals. 26 APPENDIX GCit Binary variable equal to one if the auditor client firm-year received a going concern modification to its annual initial audit report, zero otherwise. Obtained from form 10-K, 10-K405, 10KSB, 10KSB40, or 10-KT in the Audit Analytics database (variable = GoingConcern). SWITCHit Binary variable equal to one if the auditor differs from the prior year auditor within the Audit Analytics database (variable = AuditorKey), zero otherwise. Initial observations for a firm-year are coded as a zero since an actual change in auditor is not observed. LN_AGEit Continuous variable equal to the natural log of the difference between the calendar year of the observation and the calendar year of the observation’s initial public offering (Compustat variable ipodate) plus one. LN_ATit Continuous variable equal to the natural log of the observation’s total assets (Compustat variable at). ABNORMAL_FEES_PCTit Continuous variable equal to the difference between total predicted fees (both audit and non-audit fees, Audit Analytics variable TotalFees) and actual total fees, divided by actual total fees. Predicted total fees calculated using linear regression with total fees as the dependent variable and all other covariates as predictor variables. CLIENT_IMPORTANCEit Continuous variable equal to the total fees (Audit Analytics variable TotalFees) for the audit client for the year divided by the sum of the total fees for all audit client firms of the auditor for the calendar year. LEVERAGEit Continuous variable equal to total liabilities (Compustat variable lt) divided by total assets (Compustat variable at). 27 LOSSit Binary variable equal to one if net income (Compustat variable ni) is negative, zero otherwise. ROAit Continuous variable equal to net income before special items (Compustat variables ni and spi) divided by total assets (Compustat variable at). INT_COVERAGEit Continuous variable equal to interest expense (Compustat variable xint) divided by net income before interest and taxes (Compustat variables ni, txt, and xint). GROWTHit Continuous variable equal to the current year revenues (Compustat variable revt) minus the prior year revenues, divided by prior year revenues. CURRENT_RATIOit Continuous variable equal to current assets (Compustat variable act) divided by current liabilities (Compustat variable lct). MKTBKit Continuous variable equal to the market value of assets (Compustat variable mkvalt + Compustat variable lt) divided by the book value of assets (Compustat variable at). ALTMANit Continuous variable equal to the observation’s Altman Z-Score, as promulgated by Altman (1968). INVESTMENTSit Continuous variable equal to the observations cash and short term investments (Compustat variable che) divided by total assets (Compustat variable at). BIGNit Binary variable equal to one if the observation’s auditor is one of the “big four” audit firms (i.e., Deloitte, EY, PWC, or KPMG). Derived from Audit Analytics variable AuditorKey. IC_INEFFECTIVEit Binary variable equal to one if the observation’s auditor deemed internal controls to be ineffective (Audit Analytics variable EffectiveInternalControls), zero otherwise. 30 Table 1: Descriptive Statistics Table 1 presents the descriptive statistics for the full sample of 35,188 firm-year observations. The sample period begins with firm-years ending in calendar year 2008 and ends with all firm-years ending in calendar year 2017. All continuous variables have been winsorized at the 0.01 and 0.99 levels and are defined in the Appendix. (1) (2) (3) (4) (5) (6) SWITCH = 0 SWITCH = 1 Mean Median Std. Dev. Mean Mean Difference GC 0.092 0.000 0.289 0.081 0.220 -0.139 -25.181 *** SWITCH 0.083 0.000 0.276 n/a n/a n/a n/a LN_AGE 2.568 2.773 0.847 2.583 2.401 0.182 11.157 *** LN_AT 6.036 6.340 2.654 6.188 4.251 1.937 36.514 *** ABNORMAL_FEES_PCT 0.157 0.010 0.676 0.153 0.225 -0.072 -5.549 *** CLIENT_IMPORTANCE 0.042 0.001 0.135 0.039 0.071 -0.032 -12.029 *** LEVERAGE 0.835 0.589 1.592 0.805 1.171 -0.366 -11.917 *** LOSS 0.360 0.000 0.480 0.347 0.505 -0.158 -17.050 *** ROA -0.214 0.012 1.066 -0.183 -0.556 0.373 18.183 *** INT_COVERAGE 0.018 0.140 0.696 0.146 0.069 0.077 5.701 *** GROWTH 0.180 0.045 0.859 0.169 0.296 -0.127 -7.637 *** CURRENT_RATIO 1.449 2.370 2.678 2.373 2.34 0.033 0.642 MKTBK 3.053 1.375 7.165 2.907 4.667 -1.76 -12.746 *** ALTMAN -1.138 1.649 24.626 -0.599 -7.089 6.49 13.680 *** INVESTMENTS 0.184 0.088 0.227 0.183 0.193 -0.010 -2.358 ** BIGN 0.560 1.000 0.496 0.593 0.190 0.403 43.096 *** IC_INEFFECTIVE 0.028 0.000 0.166 0.027 0.045 -0.018 -5.791 *** LN_IND_CLIENTS 2.421 2.639 1.514 2.496 1.601 0.895 31.018 *** T-Stat (7) 31 Table 2: Univariate Correlation Matrix Table 2, presents a pairwise correlation coefficient matrix for the full sample of 35,188 firm-year observations within our primary sample. All variables are defined in the Appendix. All statistically significant relations (p-value < 0.10) are denoted in bold and italics. G C SW IT C H LN _A G E LN _A T ABN O RM AL_F EES_ PCT C LIE NT_I M PO RTANC E LEVERAG E LO SS RO A IN T_C O VERAG E G RO W TH C U RREN T_R ATIO M K TBK ALTM AN IN VEST M EN TS BIG N IC _I N EFFECTIV E LN _I N D _C LIE NTS GC 1.000 SWITCH 0.133 1.000 LN_AGE -0.124 -0.059 1.000 LN_AT -0.511 -0.191 0.222 1.000 ABNORMAL_FEES_PCT -0.069 0.030 -0.029 0.004 1.000 CLIENT_IMPORTANCE 0.116 0.064 -0.022 -0.251 0.047 1.000 LEVERAGE 0.495 0.063 -0.054 -0.351 -0.140 0.069 1.000 LOSS 0.378 0.091 -0.200 -0.464 0.011 0.054 0.193 1.000 ROA -0.570 -0.097 0.129 0.485 0.133 -0.073 -0.789 -0.336 1.000 INT_COVERAGE -0.146 -0.030 0.012 0.193 0.007 -0.023 -0.087 -0.196 0.125 1.000 GROWTH 0.080 0.041 -0.172 -0.101 0.024 0.025 0.003 0.070 -0.062 -0.027 1.000 CURRENT_RATIO -0.161 -0.003 -0.050 -0.158 0.116 0.019 -0.228 0.067 0.112 -0.071 0.040 1.000 MKTBK 0.428 0.068 -0.107 -0.433 -0.098 0.052 0.656 0.206 -0.686 -0.109 0.080 -0.063 1.000 ALTMAN -0.522 -0.073 0.034 0.391 0.159 -0.065 -0.841 -0.254 0.819 0.089 0.007 0.261 -0.547 1.000 INVESTMENTS 0.050 0.013 -0.173 -0.360 -0.034 -0.008 -0.027 0.276 -0.123 -0.147 0.093 0.547 0.189 0.001 1.000 BIGN -0.266 -0.222 0.099 0.548 -0.060 -0.341 -0.173 -0.180 0.215 0.060 -0.053 -0.005 -0.158 0.188 -0.006 1.000 IC_INEFFECTIVE -0.010 0.031 0.001 0.024 0.017 -0.016 -0.018 0.035 0.019 -0.008 0.009 -0.007 -0.014 0.018 -0.011 0.021 1.000 LN_IND_CLIENTS -0.280 -0.163 0.014 0.482 -0.048 -0.415 -0.180 -0.161 0.209 0.091 -0.029 0.010 -0.173 0.186 0.058 0.575 0.032 1.000 32 Table 3: Primary Hypothesis Tests Table 3 presents the results of our primary hypothesis test. The variable of interest in this table is SWITCHit, and the dependent variable is GCit. Columns 1 through 3 employ probit regression and columns 4 and 5 employ OLS regression. Column 1 omits SWITCHit to show the results without the variable of interest. Columns 2 and 4 do not include industry and time period fixed effects while Columns 3 and 5 do. All variables are defined in the Appendix. Robust two-tailed z and t-statistics are presented in parentheses below the coefficients. *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively. All standard errors are clustered at the firm level. (1) (2) (3) (4) (5) GC GC GC GC GC SWITCH 0.201*** 0.229*** 0.030*** 0.031*** (5.138) (5.760) (5.380) (5.598) LN_AGE -0.111*** -0.108*** -0.115*** -0.002 -0.003 (-4.541) (-4.429) (-4.583) (-0.721) (-1.528) LN_AT -0.287*** -0.286*** -0.337*** -0.032*** -0.034*** (-16.821) (-16.700) (-17.347) (-22.105) (-21.988) ABNORMAL_FEES_PCT 0.004 -0.000 0.002 -0.002 -0.002 (0.136) (-0.014) (0.055) (-0.710) (-0.710) CLIENT_IMPORTANCE -0.006 0.003 -0.086 -0.004 -0.023 (-0.044) (0.020) (-0.644) (-0.226) (-1.153) LEVERAGE 0.155*** 0.157*** 0.169*** 0.012*** 0.014*** (3.183) (3.211) (3.583) (3.398) (3.866) LOSS 0.914*** 0.911*** 0.956*** 0.083*** 0.075*** (19.278) (19.188) (18.062) (19.823) (17.930) ROA -0.325*** -0.325*** -0.266*** -0.065*** -0.062*** (-5.873) (-5.820) (-5.317) (-13.097) (-12.755) INT_COVERAGE -0.039** -0.038** -0.038** -0.013*** -0.013*** (-2.273) (-2.232) (-2.168) (-5.980) (-5.578) GROWTH 0.024* 0.023* 0.016 0.011*** 0.009*** (1.927) (1.842) (1.247) (4.902) (3.980) CURRENT_RATIO -0.127*** -0.126*** -0.136*** -0.013*** -0.015*** (-6.182) (-6.154) (-6.078) (-13.698) (-14.102) MKTBK -0.008** -0.008** -0.008** 0.001 0.000 (-2.217) (-2.235) (-2.381) (1.305) (0.948) ALTMAN -0.002 -0.002 -0.001 -0.001*** -0.001*** (-1.157) (-1.151) (-0.577) (-3.303) (-2.919) INVESTMENTS -0.571*** -0.562*** -0.608*** -0.084*** -0.086*** (-5.240) (-5.145) (-5.136) (-6.554) (-6.243) BIGN 0.122** 0.145** 0.270*** 0.021*** 0.035*** (2.038) (2.406) (3.834) (5.282) (6.371) IC_INEFFECTIVE 0.303*** 0.292*** 0.319*** -0.007 -0.007 (3.603) (3.473) (3.576) (-0.795) (-0.790) LN_IND_CLIENTS -0.044** -0.043** -0.072*** -0.008*** -0.016*** (-2.296) (-2.224) (-2.794) (-5.883) (-5.533) Model Probit Probit Probit OLS OLS Constant Yes Yes Yes Yes Yes Industry FE's No No Yes No Yes Year FE's No No Yes No Yes N 35,188 35,188 35,188 35,188 35,188 R-squared 0.569 0.570 0.590 0.449 0.460 35 Table 6: Cross-sectional Tests Table 6 conducts several cross sectional tests for situations where familiarity bias would be more or less pronounced. The variables of interest in this table are the interaction terms. Columns 1 through 6 employ probit regression while columns 7 through 12 employ OLS regression. All variables are defined in the Appendix. Robust two-tailed z and t-statistics are presented in parentheses below the coefficients. *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively. All standard errors are clustered at the firm level. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) GC GC GC GC GC GC GC GC GC GC GC GC SWITCH 0.266*** 0.310*** 0.209*** 0.236*** 0.258*** 0.290*** 0.063*** 0.063*** 0.033*** 0.034*** 0.053*** 0.054*** (4.786) (5.490) (5.310) (5.889) (5.113) (5.619) (6.308) (6.341) (5.731) (5.976) (6.594) (6.697) SWITCH_x_LN_IND_CLIENTS -0.054* -0.067** -0.020*** -0.019*** (-1.745) (-2.183) (-5.408) (-5.268) SWITCH_x_SPECIALIST -0.365 -0.447 -0.059*** -0.066*** (-1.025) (-1.208) (-2.939) (-3.187) SWITCH_x_MKT_SHARE -1.239* -1.435* -0.354*** -0.353*** (-1.695) (-1.901) (-5.464) (-5.451) LN_IND_CLIENTS -0.036* -0.063** -0.007*** -0.014*** (-1.782) (-2.365) (-4.731) (-4.876) SPECIALIST 0.158* 0.081 0.007** 0.005 (1.756) (0.785) (1.961) (0.881) MKT_SHARE 0.616* 0.264 0.005 -0.031 (1.690) (0.641) (0.268) (-1.346) Model Probit Probit Probit Probit Probit Probit OLS OLS OLS OLS OLS OLS Constant Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Industry FE's No Yes No Yes No Yes No Yes No Yes No Yes Year FE's No Yes No Yes No Yes No Yes No Yes No Yes Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes N 35,188 35,188 35,188 35,188 35,188 35,188 35,188 35,188 35,188 35,188 35,188 35,188 R-squared 0.570 0.590 0.569 0.589 0.569 0.589 0.450 0.461 0.448 0.458 0.449 0.459 36 Table 7: Analysis of Internal Control Weaknesses Table 7 conducts several tests to determine if internal control reporting mimics going concern reporting. SWTICHit is the variable of interest in these tables. Columns 1 and 2 employ probit regression and Columns 3 and 4 employ OLS regression. Columns 5 and 6 utilize Poisson regression as the dependent variable for those columns takes the form of a count variable. All variables are defined in the Appendix. Robust two-tailed z and t-statistics are presented in parentheses below the coefficients. *, **, and *** indicate significance at the 0.10, 0.05, and 0.01 levels, respectively. All standard errors are clustered at the firm level. (1) (2) (3) (4) (5) (6) IC_INEFFECTIVE IC_INEFFECTIVE IC_INEFFECTIVE IC_INEFFECTIVE IC_WEAKNESS_COUNT IC_WEAKNESS_COUNT SWITCH 0.307*** 0.322*** 0.023*** 0.023*** 0.822*** 0.855*** (6.760) (7.062) (5.488) (5.687) (5.528) (5.657) LN_AGE 0.007 -0.001 0.000 0.000 -0.027 -0.046 (0.339) (-0.032) (0.354) (0.174) (-0.358) (-0.602) LN_AT 0.037*** 0.044*** 0.002*** 0.003*** 0.113*** 0.144*** (3.589) (3.742) (3.555) (3.625) (2.879) (3.236) ABNORMAL_FEES_PCT 0.048* 0.054** 0.003* 0.003* -0.218* -0.206 (1.896) (2.123) (1.780) (1.806) (-1.721) (-1.638) CLIENT_IMPORTANCE -0.239 -0.015 -0.010* 0.003 -0.573 -0.229 (-1.643) (-0.102) (-1.657) (0.389) (-1.437) (-0.566) LEVERAGE -0.023 -0.014 -0.000 -0.000 -0.058 -0.032 (-0.821) (-0.512) (-0.342) (-0.010) (-0.712) (-0.405) LOSS 0.305*** 0.281*** 0.021*** 0.020*** 0.989*** 0.904*** (7.996) (7.127) (7.276) (6.609) (7.417) (6.379) ROA 0.070 0.069 0.003*** 0.003*** 0.085 0.078 (1.602) (1.580) (2.582) (2.651) (0.656) (0.626) INT_COVERAGE -0.023 -0.014 -0.002 -0.001 -0.025 -0.011 (-1.195) (-0.701) (-1.128) (-0.617) (-0.533) (-0.237) GROWTH 0.030** 0.029** 0.002* 0.002* 0.070** 0.079** (2.231) (2.089) (1.853) (1.700) (2.168) (2.383) CURRENT_RATIO -0.006 -0.013 -0.000 -0.001 -0.020 -0.033 (-0.752) (-1.464) (-0.685) (-1.473) (-0.705) (-1.040) MKTBK 0.005 0.005 0.000 0.000 0.017 0.014 (1.461) (1.195) (1.380) (1.335) (1.416) (1.053) ALTMAN 0.000 0.000 0.000 0.000 0.007 0.008 (0.067) (0.112) (0.032) (0.029) (1.589) (1.498) INVESTMENTS -0.120 -0.195 -0.008 -0.013** -0.703* -0.949** (-1.082) (-1.631) (-1.330) (-2.010) (-1.805) (-2.431) BIGN 0.086* -0.076 0.006* -0.004 0.303** -0.042 (1.810) (-1.281) (1.915) (-0.976) (1.964) (-0.208) GC -0.043 -0.042 -0.004 -0.004 0.380 0.418 (-0.549) (-0.517) (-0.795) (-0.789) (1.457) (1.572) LN_IND_CLIENTS -0.012 0.066*** -0.001 0.004*** -0.023 0.091 (-0.819) (2.700) (-0.818) (2.681) (-0.441) (1.100) Model Probit Probit OLS OLS Poisson Poisson Constant Yes Yes Yes Yes Yes Yes Industry FE's No Yes No Yes No Yes Year FE's No Yes No Yes No Yes Controls Yes Yes Yes Yes Yes Yes N 35,188 35,188 35,188 35,188 35,188 35,188 R-Squared 0.021 0.047 0.006 0.012 0.005 0.009 37 References Allen D. Blay, Marshall A. Geiger, and D. S. North. 2011. The Auditor's Going-Concern Opinion as a Communication of Risk. AUDITING: A Journal of Practice & Theory 30 (2):77-102. Altman, E. I. 1968. Financial Ratios, Discriminant Analysis and Prediction of Corporate Bankruptcy. Journal of Finance 23 (4):589-609. Ayres, D., and S. Dolvin. 2019. Unfamiliarity Breeds Resentment: Familiarity Bias in Initial Credit Ratings. Working paper, Butler University. Basioudis, I. G., E. Papakonstantinou, and M. A. Geiger. 2008. Audit Fees, Non-Audit Fees and Auditor Going-Concern Reporting Decisions in the United Kingdom. Abacus 44 (3):284- 309. Bhaskar, L. S., G. V. Krishnan, and W. Yu. 2017. Debt Covenant Violations, Firm Financial Distress, and Auditor Actions. Contemporary Accounting Research 34 (1):186-215. Blay, A. D., and M. A. Geiger. 2013. Auditor Fees and Auditor Independence: Evidence from Going Concern Reporting Decisions*. Contemporary Accounting Research 30 (2):579- 606. Boone, J. P., I. K. Khurana, and K. K. Raman. 2010. Do the Big 4 and the Second-tier firms provide audits of similar quality? Journal of Accounting and Public Policy 29 (4):330- 352. Bruynseels, L., and E. Cardinaels. 2014. The Audit Committee: Management Watchdog or Personal Friend of the CEO? The Accounting Review 89 (1):113-145. Carcello, J. V., and A. L. Nagy. 2004. Audit Firm Tenure and Fraudulent Financial Reporting. AUDITING: A Journal of Practice & Theory 23 (2):55-69. Carcello, J. V., and T. L. Neal. 2000. Audit Committee Composition and Auditor Reporting. The Accounting Review 75 (4):453-467. ———. 2003. Audit Committee Characteristics and Auditor Dismissals following “New” Going‐Concern Reports. The Accounting Review 78 (1):95-117. Carey, P., and R. Simnett. 2006. Audit Partner Tenure and Audit Quality. The Accounting Review 81 (3):653-676. 40 Kaplan, S. E., and D. D. Williams. 2012. The changing relationship between audit firm size and going concern reporting. Accounting, Organizations and Society 37 (5):322-341. ———. 2013. Do Going Concern Audit Reports Protect Auditors from Litigation? A Simultaneous Equations Approach. The Accounting Review 88 (1):199-232. Knechel, W. R., and A. Vanstraelen. 2007. The Relationship between Auditor Tenure and Audit Quality Implied by Going Concern Opinions. AUDITING: A Journal of Practice & Theory 26 (1):113-131. Krishnan, J. 1994. Auditor switching and conservatism. The Accounting Review 69 (1): 200–215. Krishnan, J., and J. Krishnan. 1996. The role of economic trade-offs in the audit opinion decision: An empirical analysis. Journal of Accounting, Auditing and Finance 11 (4): 565–586. Krishnan, J., and R. Stephens. 1995. Evidence on opinion shopping from audit opinion conservatism. Journal of Accounting and Public Policy 14 (3): 179–201. Lennox, C. 2000. Do companies successfully engage in opinion shopping? Evidence from the U.K. Journal of Accounting and Economics 29 (3): 321–337. Li, C. 2009. Does Client Importance Affect Auditor Independence at the Office Level? Empirical Evidence from Going-Concern Opinions*. Contemporary Accounting Research 26 (1):201-230. Li, K., 2004. Confidence in the Familiar: An International Perspective. Journal of Financial and Quantitative Analysis 39, 47-68. Lim, C.-Y., and H.-T. Tan. 2008. Non-audit Service Fees and Audit Quality: The Impact of Auditor Specialization. Journal of Accounting Research 46 (1):199-246. Massa, M., Simonov, A., 2006. Hedging, Familiarity and Portfolio Choice. The Review of Financial Studies 19, 633-685. Moeckel, C., and R. D. Plumlee, 1989, Auditors’ confidence in recognition of audit evidence, The Accounting Review 64: 653. Nofsinger, J.R., Varma, A., 2012. Individuals and Their Local Utility Stocks: Preference for the Familiar. Financial Review 47, 423-443. Peecher, M. E., and M. D. Piercey, 2008, Judging audit quality in light of adverse outcomes: evidence of outcome bias and reverse outcome bias. Contemporary Accounting Research 25: 243–274. 41 Plumlee, R. D., 1985, The standard of objectivity for internal auditors: memory and bias Effects. Journal of Accounting Research 23: 683–699. Public Company Accounting Oversight Board (PCAOB). 2003. AU Section 341: The auditor's consideration of an entity's ability to continue as a going concern. Available at: http://pcaobus.org/Standards/Auditing/Pages/AU341.aspx. Public Company Accounting Oversight Board (PCAOB). 2012. Standing Advisory Group Meeting: Going concern. Available at: http://pcaobus.org/ News/Events/Documents/05172012_ SAG Meeting/Going_Concern.pdf. Public Company Accounting Oversight Board (PCAOB). 2015a. Standard-setting projects. Available at: http://pcaobus.org/Standards/Documents/201503 _standard_setting_agenda.PDF. Public Company Accounting Oversight Board (PCAOB). 2015b. Release No. 2015-005: Concept release on audit quality indicators. Available at: https://pcaobus.org//Rulemaking/Docket%20041/Release_2015_005.pdf. Public Company Accounting Oversight Board (PCAOB). 2016. Standard-setting agenda. Office of the Chief Auditor. Available at: ttps://pcaobus.org/Standards/Documents/2016Q3- standard-setting-agenda.pdf. Petersen, M. A. 2009. Estimating Standard Errors in Finance Panel Data Sets: Comparing Approaches. Review of Financial Studies 22 (1):435-480. Read, W.J., and Yezegel, A. 2016. Auditor Tenure and Going Concern Opinions for Bankrupt Clients: Additional Evidence. Auditing: A Journal of Practice & Theory 35 (1): 163-179. Reichelt, K., and D. Wang. 2010. National and Office-Specific Measures of Auditor Industry Expertise and Effects on Audit Quality. Journal of Accounting Research 48 (3):647-686. Riff, S., and Y. Yagil. 2016. Behavioral Factors Affecting the Home Bias Phenomenon: Experimental Tests. Journal of Behavioral Finance 17, 267-279. Robinson, D. 2008. Auditor Independence and Auditor-Provided Tax Service: Evidence from Going-Concern Audit Opinions Prior to Bankruptcy Filings. AUDITING: A Journal of Practice & Theory 27 (2):31-54. Schumacher, D., 2017. Home Bias Abroad: Domestic Industries and Foreign Portfolio Choice. The Review of Financial Studies 31, 1654-1706. 42 Smith, D. B. 1986. Auditor ‘‘subject-to’’ opinions, disclaimers and auditor changes. Auditing: A Journal of Practice & Theory 6 (1): 95–108. Tan, H.-T., 1995, Effects of expectations, prior involvement, and review awareness on memory for audit evidence and judgment. Journal of Accounting Research 33: 113–135. Trotman, K.T., H.C. Tan, and N. Ang. 2011. Fifty-year overview of judgment and Decision- Making Research in Accounting. Accounting and Finance 51: 270-360. Tysiac, K. 2014. FASB defines management’s going-concern responsibilities. Journal of Accountancy, American Institute of CPA's. Available at: http://www.journalofaccountancy.com/news/2014/aug/201410841.html U.S. House of Representatives. 2002. The Sarbanes-Oxley Act of 2002. Public Law 107-204 [H. R. 3763]. Washington, DC: Government Printing Office. Vanstraelen, A. 2003. Going-concern opinions, auditor switching, and the self-fulfilling prophecy effect examined in the regulatory context of Belgium. Journal of Accounting, Auditing and Finance 18: 231–253. Willenborg, M., and J. C. McKeown. 2000. Going-concern initial public offerings. Journal of Accounting and Economics 30 (3):279-313.Zmijewski, M. E. 1984. Methodological issues related to the estimation of financial distress prediction models. Journal of Accounting Research 22:59-82.