Auditor Expertise and Bank Failure: Do Going Concern Opinions Predict Bank Closure?

Abstract

This study investigates how the quality of engagement auditors, assessed using the auditor’s industry expertise and size at both national and state levels, influences the likelihood of going concern opinion (GCO) issuance for U.S. banks from 2002 to 2023. We also examine how auditor quality affects the accuracy of GCOs, specifically regarding Type I (false positive) and Type II (false negative) errors in GCO issuance. Using a dataset of 4992 bank-year observations from 414 unique banks, we analyze the correlations between auditor characteristics and these error types. We find that state-level audit industry experts issue significantly more accurate GCOs, demonstrating lower rates of both Type I and Type II errors compared to their counterparts. National-level experts and larger audit firms primarily show a reduced likelihood of Type II errors, indicating a more conservative approach. Our findings underscore the importance of localized auditor expertise in assessing bank financial health and suggest that enhanced collaboration between auditors and regulators could improve the predictive power of GCOs. These results offer important implications for regulatory policy and emphasize the need for improved audit standards to bolster financial system stability.

1. Introduction

An audit comprises a methodical review and objective examination of an entity’s financial statements. The goal of the auditor’s examination is to express an opinion (in the form of audit report) on the financial statements and then evaluate, as of the opinion date, if there is substantial doubt about the ability of a company to continue as a going concern. The FASB (Financial Accounting Standard Board) defines substantial doubt using a threshold of “probable” that a company will not be able to pay debts as they come due during the next 12 months.

Maintaining investors’ confidence in the financial statements is very important for stabilizing the capital market. The recent failure of Silicon Valley Bank (SVB) exemplifies the potential magnitude of losses when investors fear a company, in this case SVB, might not meet its debt obligations due to capital shortages. This fear arose from a sudden decline in the value of SVB’s held-to-maturity securities portfolio, triggered by rapid interest rate increases starting in 2021 and continuing into 2022. SVB’s losses were substantial, potentially rendering its equity negative in 2022 if unrealized security investment losses were incorporated into comprehensive income, as detailed in “SVB and Beyond: The Banking Stress of 2023” (by Acharya et al., 2023). SVB depositors, fearing deposit losses, initiated mass withdrawals, leading to the bank’s closure on 10 March 2023. The Federal Deposit Insurance Corporation (FDIC) reopened SVB as a Bridge Bank on 13 March 2023, with the closure costing the FDIC approximately $20 billion. This bank failure has led many to question the role and effectiveness of the auditor.

Marty Baumann, former chief auditor of the Public Company Accounting Oversight Board (PCAOB) and adjunct professor at Washington Campus, noted: “I suspect that may cause some people to question the value of an audit, but I believe the audit is incredibly valuable to investors”1. Mr. Baumann questioned the adequacy of bank management’s disclosure regarding duration mismatch risk between assets and liabilities. For instance, SVB’s deposits were primarily from high-tech companies, creating potential spillover effects. If one depositor doubted the safety of their SVB deposits and withdrew funds, it could trigger similar actions by others within the same industry, potentially accelerating SVB’s collapse.

Mr. Baumann’s concerns about the adequacy of current going concern and risk disclosure standards are valid and important. We also question whether SVB’s auditor conducted a sufficiently thorough audit to identify and disclose going concern risks prior to its failure. If a GCO was not issued before SVB’s collapse, we further question whether a higher quality engagement auditor would have issued a GCO. Given the critical importance of maintaining financial stability, any contribution in this area is significant. Therefore, this study examines the association between engagement auditor quality and the propensity to issue GCOs, as well as Type I and Type II errors in GCOs issued to banks.

The auditing literature commonly suggests that higher quality auditors tend to be more conservative in their reports (e.g., Reichelt & Wang, 2010; Minutti-Meza, 2013; M. DeFond & Zhang, 2014; M. Liu, 2020; John & Liu, 2021). This conservatism might manifest as a greater likelihood of issuing GCOs and a higher rate of Type I error (or/and lower rate of Type II error), while controlling for factors influencing audit outcomes (e.g., issuance of GCO and accuracy of GCO issued to auditees). A Type I error in a GCO occurs when a bank receives a GCO but does not fail within 12 months. Conversely, a Type II error occurs when a bank fails within 12 months after receiving a clean audit opinion (no GCO). In simpler terms, a Type I error is a (false positive) false alarm of potential bank failure, while a Type II error is a (false negative) missed signal of a bank’s unhealthy financial status.

Prior research employing engagement auditor’s industry expertise and size as proxies for audit quality has yielded mixed empirical evidence regarding the relationship between and GCO issuance and errors (both Type I and Type II) in GCOs issued to banks. For example, Reichelt and Wang (2010) found audit industry specialists more likely to issue GCOs than non-specialists. However, Minutti-Meza (2013) found no empirical evidence on the relationship between auditor’s industry expertise and the tendency to issue GCOs. Some studies suggest that the audit fees charged to the auditees might be an indicator of audit quality such as that an auditor charging the auditees a higher audit fee is more likely to issue a GCO (e.g., Reichelt & Wang, 2010).

Therefore, we hypothesize that audit quality is related to the propensity of issuing GCOs and the occurrence of Type I and Type II errors in the GCOs issued to banks. However, due to the conflicting evidence in prior literature, we do not predict the direction of this relationship.

To investigate the impact of engagement auditor quality, measured by the auditor’s industry expertise and size at national and state levels, on GCO issuance to banks, we use a sample of publicly traded U.S. banks2. The dataset includes 4492 bank-year observations from 414 unique banks spanning 2002–2023. We measure engagement auditor industry expertise at both national and state levels, based on the auditor’s market share and client portfolio within specific industries and fiscal years. We also construct binary variables, GCO, TypeI, and TypeII, to indicate GCO issuance and the occurrence of Type I and Type II errors, respectively.

We then regress these GCO outcome indicators (GCO, TypeI, and TypeII) on six empirical proxies for engagement auditor quality, controlling for other factors that may influence these outcomes. The results of multivariate regressions show that (1) the issuance of GCOs and Type I and Type II errors in the GCOs issued to the banks are statistically significantly negatively related to empirical proxies for the audit industry expert at the state level, indicating that the audit industry expert at the state level is less likely to over-issue GCOs and issues a more accurate GCO (i.e., lower rate of Type I/II errors in the GCOs issued to banks) than the counterparts; (2) the empirical proxy for the audit industry expert at the national level (constructed based on the engagement auditor’s client portfolio) is statistically significantly negatively associated with the Type II error in GCOs issued to banks; (3) the empirical proxies for the size of the engagement auditor at the national and state level are statistically and significantly negatively associated with the Type II error in the GCOs issued to banks, but not with issuance of GCOs or Type I error in the GCOs issued to banks.

Our findings reveal novel insights regarding non-performing loans (NPLs) and regulatory intervention. Specifically, NPLs are statistically and significantly positively associated with GCO issuance and both Type I and Type II errors. Furthermore, consistent with expectations, higher bank profitability and adequate capital are statistically and significantly negatively associated with GCO issuance and Type II errors. This confirms the relevance of traditional factors used by the prior literature and regulatory bodies in assessing bank closure risk, which are also significantly considered by auditors when issuing GCOs. These findings are not only novel and meaningful but also provide useful guidance for regulators. It is not surprising that a high volume of non-performing loans serves as a red flag for auditors, prompting GCO issuance and potentially leading to a higher incidence of both Type I and Type II errors.

This paper contributes to the literature in several ways. First, it provides novel empirical evidence on the extent to which the engagement auditor’s industry expertise at the state level influences GCO issuance and Type I/II errors for banks. Specifically, we demonstrate that state-level audit industry experts are less prone to over-issuing GCOs and issue more accurate GCOs (lower Type I and Type II error rates) compared to their counterparts. In contrast, national-level audit experts and larger-sized auditors primarily exhibit conservatism in GCO issuance. While, using U.S. public firm data (excluding financial services), the prior literature presents mixed evidence on the relationship between auditor’s industry expertise and GCO issuance, our study (employing banks and bank holding companies) provides additional empirical evidence and clarifies the effect of auditor’s industry expertise on both the accuracy and conservatism of GCOs in the banking sector. This adds novel empirical evidence to the banking literature. Although both this paper and Albrecht et al. (2020) identify high rates of Type I and Type II errors in GCOs issued to banks, our findings are more generalizable due to a larger sample size (4992 bank-year observations) and longer sample period (2002–2023) compared to Albrecht et al. (2020), while addressing different research questions.

Second, our results suggest that audits provide value to investors. Engaging a state-level audit industry expert may protect investor financial interests by ensuring more accurate GCOs for banks. This aligns with regulators’ intentions to protect investors by strengthening the auditor’s “watchdog” role. Banks can mitigate potential harm from Type I GCO errors by hiring state-level audit industry experts. Given the negative sentiment and potential for bank runs triggered by GCOs, even if ultimately false alarms, audits provide a valuable function, and engaging industry specialists may benefit both investors and banks.

Our findings also offer practical implications. The statistical significance of NPLs in predicting GCOs and both error types indicates that non-performing loans are a critical red flag for auditors evaluating going concern. However, auditors should avoid overreacting to NPLs, as a lower return rate on loans may simply reflect a bank’s conservative investment strategy in lower-risk projects. Banks with substantial NPL holdings may exhibit lower default risk, although this strategy might reduce profits compared to lending to higher-risk borrowers. However, lower returns on NPLs could jeopardize bank survivability (e.g., Hotchkiss et al., 2008; S. Liu, 2024). Therefore, the effect of NPLs on bank going concern is complex and requires careful consideration when assessing bank closure risk. Given the high error rates in GCOs documented in this and prior studies (e.g., Albrecht et al., 2020), practitioners should consider employing improved models for predicting bank closure probability to issue GCOs timelier and accurately.

Third, both false alarms (Type I errors) and missed warnings (Type II errors) in auditor reports can negatively impact financial system stability. The striking finding that Type I and Type II error rates for GCOs are 65.2% and 89.7%, respectively, should concern regulators. Our results indicate that most banks receiving GCOs do not fail within 12 months, while most failing banks that should have received a GCO did not receive one prior to closure. Although bank failures are relatively infrequent (3.1% in our sample), any failure can trigger market panic and contagion, potentially leading to catastrophic systemic failure. Preventing such failures is therefore critical.

The high Type I/II error rates in GCOs should prompt attention from regulators like the PCAOB and bank regulators. The PCAOB may need to revise GCO issuance requirements and encourage audit firms to incorporate relevant academic research to improve GCO outcomes. Our results also provide valuable insights for bank regulators. Given their privilege to conduct stress tests, regulators should enhance communication with auditors regarding bank financial status. This improved information sharing can help auditors refine GCO issuance practices and reduce Type I/II error rates. We therefore recommend closer collaboration between bank auditors and regulators to identify effective strategies for preventing bank failures and providing more accurate warnings to banks and investors.

The remainder of this paper is structured as follows: Section 2 reviews the related literature; Section 3 introduces the banking industry background and develops hypotheses; Section 4 describes the research design, sample, and statistics; Section 5 presents the results; Section 6 discusses the strengths and weakness of using propensity score matching (PSM); and Section 7 concludes.

2. Related Literature

Early audit research establishes that an auditor’s experience enhances the auditor’s knowledge of audit engagements, improving audit quality3. For example, Tubbs (1992) experimentally demonstrated that the auditor’s experience affects knowledge structure, leading to better error detection and understanding. Increased experience enhances auditors’ awareness of various error types and atypical accounting entries, improving accuracy.

Studies also show a positive correlation between auditor-specific knowledge and audit outcomes. Prior research indicates that the task-specific knowledge could improve the auditor’s performance (e.g., Bonner, 1990; Brown & Solomon, 1991; Nelson et al., 1995; Tan & Libby, 1997). Bonner (1990) used experimental methods to document that task-specific knowledge improves experienced auditors’ clue selection and weighting only during analytical risk assessment. Brown and Solomon (1991) found domain-specific knowledge beneficial for the auditor to process the information to assess the account of misstatement risk. Nelson et al. (1995) used experimental approaches to show that mismatches between auditor’s knowledge structures and audit planning tasks negatively impact the ability to access and use prior error frequency experiences when making conditional probability judgments and planning decisions. Nelson et al. (1995) concluded that such mismatches can hinder the use of prior experience and reduce audit planning efficiency. Tan and Libby (1997) test and find that the auditor’s problem-solving ability and technical (general accounting and auditing) knowledge increase with increases in the auditor’s expertise for the subject matter.

Increased experience in specific audit matters enhances the auditor’s task-specific knowledge, subsequently improving the auditor’s performance. Thus, a positive association between the auditor’s experience in a particular audit task and audit quality is expected. Consistent with this expectation, prior studies document that, with the increase in the auditor’s experiences, the auditor is more likely to detect errors in the financial statements and has less biased judgement on the information provided by the auditees (e.g., Libby & Libby, 1989; Libby & Frederick, 1990; Hammersley, 2006; Kaplan et al., 2008; Tan, 1995).

Libby and Frederick (1990) used experimental methods to examine whether and how the auditor’s experience in financial statement errors affects the effectiveness and efficiency of the auditor’s decisions. They found more experienced auditors more likely to detect errors and accurately report them. Experienced auditors are also better at identifying uncommon errors in the financial statements than their counterparts. Hammersley (2006) experimentally examined whether industry-specialized auditors more efficiently and effectively identify the pattern of financial statement misstatements. She found that matched specialists, who are working in the industry in which the specialists have industry-specific training and experiences, more effectively utilize clues and audit procedures to identify misstatements than their counterparts. Using management assessments as information sources, Kaplan et al. (2008) find that the more experienced auditors are less influenced by management’s (potentially biased) assessments than their counterparts. The results of Kaplan et al. (2008) indicate that the more experienced auditor could land in a more objective judgement on the audit engagements when drawing inference from the same information source (e.g., management’s assessment).

Previous studies also show that an increase in the auditor’s accountability could improve the audit quality (Carcello & Li, 2013; John & Liu, 2021; John et al., 2025). Increased accountability may enhance audit task effectiveness, although not necessarily efficiency, such as in the issuance of going concern opinions. For example, Brazel et al. (2004) experimentally examined how workpaper review methods (face-to-face, electronic) impact the auditor’s effectiveness and efficiency in going concern assessments. Brazel et al. (2004) measured efficiency as the time to complete the assessment and effectiveness by “work-paper effectiveness” (proper supporting evidence and conclusions) and “judgment quality” (comparing subjects’ conclusions to experts’). They found face-to-face review more effective but less efficient than electronic review and less susceptible to prior year workpaper bias. Face-to-face review preparers reported higher self-perceived accountability, interpreted by Brazel et al. (2004) as driving more effective audit outcomes.

Some studies suggest that audit industry experts face greater pressure to maintain reputation and thus have higher accountability than non-specialists. Therefore, a positive association between audit industry expertise and audit quality (e.g., going concern opinions) is expected. Empirical evidence supports this expectation. For example, prior studies find that the auditees of the audit industry specialists have a lower level of abnormal accruals compared to auditees of the non-audit industry specialists (e.g., Balsam et al., 2003; Krishnan, 2003). The audit industry specialists are more likely to issue a going concern audit opinion than non-audit industry specialists (e.g., Francis et al., 2005; Lim & Tan, 2008; Reichelt & Wang, 2010). The quality of financial statements of auditees of audit industry specialists is higher than that of non-audit industry specialists (e.g., Dunn & Mayhew, 2004; John & Liu, 2025; S. Liu & Ronen, 2024).

Furthermore, some studies argue that the audit industry specialists charge higher audit fees, which implies that the audit industry specialists produce a higher quality of the audit (e.g., Craswell et al., 1995; M. L. DeFond et al., 2000; Ferguson et al., 2003; Mayhew & Wilkins, 2003; Casterella et al., 2004; Francis et al., 2005; Carson, 2009; Li, 2009; Reichelt & Wang, 2010; Cahan et al., 2011). Therefore, in our empirical design, we also measure the quality of the engagement auditor as the auditor’s size (e.g., the natural logarithm value of the audit fees earned from the clients for the industry in the fiscal year) in Section 4.

3. Banking Industry Background and Hypotheses Development

Banks and bank holding companies in the U.S. are regulated by the Board of Governors of the Federal Reserve System (“Federal Reserve” or “the Fed”), the central bank of the United States. U.S. banks are subject to federal and state regulations, differing from those for other industries. Banks can be chartered nationally or by individual states. National banks are primarily regulated by the Office of the Comptroller of the Currency (OCC), an independent bureau within the U.S. Department of the Treasury. State-chartered banks are chartered by the Federal Reserve System and supervised by the Federal Reserve and their respective state banking systems if they are members of the Federal Reserve System. The Federal Reserve’s 12 regional banks ensure member banks comply with regulations. State-chartered banks not in the Federal Reserve System are regulated by the FDIC (federal deposit guarantor) and their respective state banking systems. The third type includes banks owned by bank holding companies (also financial holding companies), engaging in traditional banking activities like lending and deposit acceptance. Financial holding companies can engage in broader business activities beyond traditional banking, such as insurance underwriting and merchant banking through nonbank subsidiaries.

The Federal Reserve supervises banks to maintain financial system stability and regulatory compliance through periodic on-site examinations and continuous off-site monitoring of financial condition, risks, and reporting (e.g., Hotchkiss et al., 2008; Hirtle et al., 2020; Hirtle & Kovner, 2022; Gauri & Desai, 2020). Bank auditor examinations substantially overlap with Federal Reserve supervision. The Federal Reserve prioritizes bank closure risk due to its significant systemic impact and oversees financial health, accounting systems, and internal controls, similar to audit engagements. Banks and bank holding companies (BHCs) are required to file “Call Reports” (Consolidated Reports of Condition and Income) quarterly for monitoring, and BHCs must also file SEC-style financial statements.

The external (or independent) engagement auditor is responsible for evaluating the aggregate audit and the overall quality of the audit engagement, including determining whether there is any substantiable doubt about the clients’ ability to continue as a going concern for a reasonable period, which should not beyond one year from the date of the financial statements being audited. If there is substantial doubt, the auditor would state his/her concerns in an explanatory paragraph following the opinion of the unqualified report.

The independent auditors of banks are required to periodically meet with bank regulators to communicate about the soundness of the financial condition, profitability, internal control, and business risk of the audited banks. In accordance with the required communication with regulators, the bank auditors should have examined the auditees’ business risk, financial reports, internal controls, etc., and communicated with the banks’ management before the required communication with the regulators. The communication with management of the banks and regulators should in turn well inform the auditors about the banks’ business risk and audit risk, which might have insightful and useful input for auditors providing their audit evaluations on audit engagement to the bank regulators (such as the strength of internal control and opinion of going concern).

Banks face various risks, including interest rate, capital, economic environment, operating, credit, legal, and reputation risks (e.g., John, 1993; Hotchkiss et al., 2008; Balakrishnan et al., 2021; S. Liu, 2024; Penman, 2017). Among the variety of the risks of the banks, the Federal Reserve highly prioritizes the risk of bank closure because of its potential large negative impact on financial system stability (Basel Committee on Banking Supervision, 2014; Balakrishnan et al., 2021). Given the paramount importance of closure risk faced by banks, this paper focuses on variations of the bank auditor’s assessment on the risk of the banks’ going concern.

As noted in Section 2, audit quality influences auditor risk assessments. Audit quality, measured by auditor’s size and industry expertise (e.g., M. DeFond & Zhang, 2014), is the focus of this study’s effect on GCO issuance for the following reasons.

As discussed in previous studies on audit quality in Section 2, the audit quality could impact on the variation of the auditor’s assessment on the risk of the banks’ going concern. Audit quality could be measured as the auditor’s size and industry expertise (e.g., M. DeFond & Zhang, 2014). In this study, we focus on the effect of audit industry expertise (proxied for audit quality) on the issuance of GCOs based on the following reasons.

First, this study centers on the banking industry. We would like to control for any other factors that may impact on the audit outcome when we examine whether and the extent to which the audit quality influences on the auditor’s going concern opinion. Prior studies suggest industry specialists benefit from network synergies (e.g., Bental & Spiegel, 1995; Katz & Shapiro, 1985; Reichelt & Wang, 2010). Reichelt and Wang (2010) attribute these synergies to knowledge sharing within networks, including best practices, industry-specific audit programs, and personal knowledge spillover.

Second, audit industry specialists learn more about clients through professional interactions and gain deeper client knowledge than non-specialists. Because the bank auditors are required to periodically communicate with the regulators, which may increase the knowledge of the engaged auditors, therefore, we control for the impact of regulations and regulators on the outcomes in our empirical tests in Section 4.

Therefore, auditors specializing in banking are expected to be more knowledgeable about the banking industry than non-specialists, leading to more timely and more accurate detection of bank closure risk.

Audit industry specialists are widely recognized and reputable. This, in turn, incentivizes them to maintain their reputation and increases their likelihood of issuing GCOs when closure risk is detected. Prior studies (e.g., Francis et al., 2005; Lim & Tan, 2008; Reichelt & Wang, 2010) document a positive association between GCO issuance and the auditor’s industry expertise, suggesting that the audit industry specialists are more conservative in issuing GCOs. Additionally, industry specialists charge higher fees (e.g., Francis et al., 2005; Reichelt & Wang, 2010), implying the auditor’s size may also positively correlate with GCO issuance tendency. Thus, we expect the audit industry specialists and/or larger-sized auditors to be more likely to issue GCOs than non-specialists. However, conflicting evidence exists, with some studies (e.g., Minutti-Meza, 2013) finding no relationship between industry specialization and GCO issuance. Building upon the previous literature, we articulate our first testable hypothesis in its alternative form as follows4:

H1: 

Engagement auditor quality significantly affects the likelihood of issuing a going concern opinion (GCO) in bank audit reports.

Furthermore, more conservative auditors tend to issue GCOs more frequently, including in cases where the client that does not file for bankruptcy within 12 months subsequently after the date of receiving a GCO. This is called the Type I error in the GCO issued to the auditee. Consequently, the higher tendency of issuing GCOs could reduce the Type II error in GCOs, meaning that the high-quality auditor would not miss issuing GCOs to the banks that subsequently are closed. Therefore, a higher quality engagement auditor might issue GCO with a higher (or/and lower) rate of Type I (II) error in the going concern opinion than the counterparts.

Albrecht et al. (2020) found bank auditors have lower Type I error rates for systemically risky banks, regardless of industry specialization, arguing this is due to bank auditors’ lower likelihood of issuing GCOs, reducing Type I errors without increasing Type II errors, and thus producing more accurate GCOs. Per prior discussion, Minutti-Meza (2013) finds no relationship between the audit industry specialists and the tendency of issuance of GCOs. Therefore, it is possible that there is no difference in the rate of Type I (II) errors of GCOs issued by a high-quality engagement auditor and the counterparts in our setting—the banking industry. Although the previous literature documents inconsistent empirical evidence on the relationship between the quality of the engagement auditor and the Type I or/and Type II error rate of GCOs issued to banks, we expect that the quality of the engagement auditor(s) would impact the Type I or/and Type II rate in GCO issued to banks. Our second testable hypothesis is also formulated in its alternative form:

H2: 

Type I and Type II errors in going concern opinions (GCOs) issued by engaging audit firms are significantly influenced by the quality of the engaging audit firm.

4. Empirical Design

4.1. Empirical Model

Adapting the model in previous studies (e.g., Reichelt & Wang, 2010; Minutti-Meza, 2013; Albrecht et al., 2020), we use the following Equation (1) to test our H1 and H2. Subscripts j (bank) and t (year) are suppressed for simplicity:

DVⁱ = α₀ + α₁AQ^k + α₂Size + α₃STDEarn + α₃ROA + α₄Leverage + α₅Loss + α₆MB + α₇AltmanZ + α₈Fin_New + α₉Big4 + α₁₀Tenure + α₁₁CapAsset + α₁₂LoanAsset + α₁₃Provision + α₁₄NPL + α₁₅Restate + α₁₆Interven + α₁₇IC_W + α₁₈Year_FE + α₁₉Industry_FE + ε

(1)

where DVⁱ is the ith dependent variable for bank j in fiscal year t, representing GCO and TypeI/TypeII when testing H1 and H2, respectively. AQ^k represents bank j’s value of the kth proxy for the quality of engagement auditor in the fiscal year t. Per our discussion on the audit quality literature in Section 2, we use two sets of proxies for AQ^k: (i) the engagement auditor’s industry expertise [which is measured on the nation-(state-)wide market/client share of the engagement auditor in a particular industry in fiscal year t: represented by Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S] and (ii) the engagement auditor’s size in a particular industry in fiscal year t (at (1) nationwide, which is presented as Audsize_N; and (2) statewide, which is represented by Audsize_S). Our main discussion on the empirical tests would concentrate on the impact of the engagement auditor’s industry expertise on the GCO outcomes because of the industry-specific of our research setting, while we employ auditor’s size in the industry as the proxy for the audit quality in our robust tests. We detail the construction of the variables used in the analyses as follows.

GCO = 1 if the auditor issues a going concern opinion for the bank-year, and 0 otherwise.

TypeI = 1 if the bank-year receiving a GCO but not subsequently being closed within 12 months following receiving GCO, and 0 otherwise. This variable captures Type I error in the issuance of GCO.

TypeII = 1 if the bank-year is closed within 12 months subsequentially following receiving the clean audit opinion, and 0 otherwise. This variable captures Type II error in the issuance of GCO.

Exp_MKT_N = 1 if the auditor’s market share for a given industry is the largest in the nation (U.S.) in the year t, and 0 otherwise.

Exp_MKT_S = 1 if the auditor’s market share for a given industry is the largest in a particular state of the U.S., in which the company (i.e., bank or/and bank holding company) is located, in the year t, and 0 otherwise.

Exp_NumInd_N = 1 if the auditor has the largest number of clients in a given industry in year t in the nation (U.S.), and 0 otherwise.

Exp_NumInd_S = 1 if the auditor has the largest number of clients in a given industry in year t in a particular state of the U.S., in which the bank is located, and 0 otherwise.

Audsize_N = natural logarithm value of (1 + audit fees of the auditor earned from a given industry nationwide in fiscal year t).

Audsize_S = natural logarithm value of (1 + audit fees of the auditor earned from a given industry statewide, in the state where the auditee[s] is[are] located, in fiscal year t).

Size = natural logarithm value of total assets for bank j in the year t.

STDEarn = standard deviation of the income before extraordinary items over the past four years (t − 4, t − 3, t − 2, t − 1) given t is the current year.

ROA = net income scaled by the average of total assets over year t and t − 1.

Leverage = total liabilities scaled by average total assets.

Loss = 1 if net income for the bank-year is negative, and 0 otherwise.

MB = ratio of market value to book value of bank j in year t.

AltmanZ = Altman Z score computed as noted in Altman (1968)5.

Fin_New = 1 for bank-year issuing new debt or equity in the subsequent fiscal year, and 0 otherwise.

Big4 = 1 if the bank-year is audited by a Big4 audit firm, and 0 otherwise.

Tenure = the number of years that the auditor has audited bank j.

CapAsset = ratio of shareholder equity to total assets.

LoanAsset = ratio of total loans to total assets.

Provision = ratio of loan loss provision to total loans (net of total allowance).

NPL = ratio of non-performing loans to total loans.

Restate = 1 if the bank-year financial statement was restated, and 0 otherwise.

Interven = 1 if the bank-year was intervened by the regulator(s) (i.e., the involvement of SEC, DOJ, FINRA, or/and bank regulator bodies in the restatement process is noted), 0 otherwise.

IC_W = 1 if the bank-year was reported by the external auditor for having internal control weakness, and 0 otherwise.

Year_FE = fiscal year fixed effect variable.

Industry_FE = industry fixed effect variable.

The tested variable in the regression Equation (1) is AQ^k (i.e., Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, and Audsize_S). When testing H1, a positive (negative) sign on coefficient of AQ^k in regression Equation (1) could indicate that the audit industry expert (and large-sized auditor) is more (less) likely to issue GCOs than the counterparts, when controlling for the other factors that may influence the auditor’s tendency of issuing GCOs. When testing H2, a positive (negative) sign on the coefficient of AQ^k in regression Equation (1) could indicate that the GCO issued by the audit industry expert (or large-sized auditor) to the auditee is more (less) likely to have Type I/II error in the GCO than the counterparts, when controlling for the other factors that may influence Type I/II error in the GCO issued to the auditee(s).

Per our Section 3 discussion on the conflicting empirical results on the direction of how the quality of the engagement auditor influences the GCO outcomes, we do not predict the sign on the coefficient of each of these six tested variables in Equation (1).

We control for other factors that may influence the decision of the auditor to express a GCO on the audit report in Equation (1), as noted in previous studies (e.g., Ohlson, 1980; Reichelt & Wang, 2010; Calomiris & Nissim, 2014; Minutti-Meza, 2013; Albrecht et al., 2020; Gauri & Desai, 2020; S. Liu, 2024; S. Liu & Ronen, 2024; John & Liu, 2025). A negative sign on the coefficient of Size, ROA, and CapAsset could be expected in Equation (1) because a bank that is large-sized, more profitable, and/or capital adequate is unlikely to fall. We do not predict the sign on the coefficient of StdEarn, because the direction of impact of StdEarn on the GCO outcomes is unclear. Specifically, more fluctuated earnings could indicate an unsustainable future earnings stream, which could negatively impact the business operation and the survivorship of the bank. Here, the auditor may consider a higher possibility of closure for a bank with a large-valued STDEarn. Oppositely, the volatile earnings could indicate less earnings management, reflecting the real economic earnings of the bank. Therefore, the auditor may consider the bank with a large-valued STDEarn as a good sign of continuing to survive in the industry. A bank [with a higher degree debt (i.e., large-valued Leverage), an operation loss (Loss), a perspective of growth in the market (a large-valued MB), a need for extra external fund (Fin_New), and more lending to the customers (LoanAsset), which could make the bank have a higher chance of exposure to riskier loans,] may face a higher likelihood of being closed and a greater tendency for the auditor to issue a GCO. However, given the dynamical change in the beliefs of the impact of these factors on the business operation of the bank over time, for example, the degree of leverage and growth expectation (proxied by MB) could impact the banks in both negative and positive directions, we do not predict the sign on coefficients of these variables.

A positive Provision coefficient is expected, as higher loan loss estimates indicate greater business operation risks and closure likelihood, leading to GCO issuance (Albrecht et al., 2020; S. Liu, 2024). A positive NPL coefficient is also expected for similar reasons. We do not predict signs for Big4 and Tenure coefficients due to inconsistent prior literature.

A positive sign on the coefficient of NPL in Equation (1) would be expected. A higher portion of non-performing loans that do not earn profit for the bank could indicate a higher chance of operating loss, a higher chance of being closed in the future, and, consequently, a higher chance of receiving a GCO from its auditor (e.g., Albrecht et al., 2020; S. Liu, 2024). We do not predict the sign on the coefficient(s) of Big4 (and Tenure) for the inconsistent argument and empirical evidence in the previous literature.

We also control for the incidence of restating financial statements, the involvement of regulators (e.g., SEC, DOJ, FINRA or/and bank regulator bodies in the restatement process), and internal control weaknesses identified by the external auditor. We would argue that the bank intervened by the regulator(s) maybe more likely to receive a GCO from the auditor. Lastly, but not least importantly, we also control for the year and industry fixed effect to mitigate the influence of the macroeconomy (and unobservable factors) on the auditor’s tendency of issuing GCOs and Type I/II errors in the GCO issued to the auditees.

The aforementioned analyses on control variables used for testing H1 are also applicable to the analyses for the control variables when testing H2 in Equation (1) because the rationales used by the auditor to issue GCOs may also cause the auditor to commit a higher rate of error (Type I/II) in the GCOs issued to the auditees.

Appendix A defines all variables employed in our analyses.

4.2. Sample Selection and Descriptive Statistics

4.2.1. Sample Selection

Audit-related data are sourced from Audit Analytics, and financial data are sourced from Compustat-Bank. Following prior research (e.g., Francis et al., 2005; Carson, 2009; Li, 2009; Reichelt & Wang, 2010; Minutti-Meza, 2013), we initially use all available audit fee data to compute proxies for auditor industry expertise and size at national and state levels, termed “audit quality data”. Using all available audit fee data minimizes measurement error in empirical proxies. We then merge “audit quality data” with Audit Analytics datasets (SOX_404_Internal_controls, bankruptcy notification, revised_audit_opinions, and financial_restatements).

Subsequently, we merge Audit Analytics and Compustat-Bank data, retaining observations with complete data for all variables. Our final sample (July 2002–December 2023) comprises 4992 bank-year observations (414 distinct banks), including 159 closed bank-year observations (22 distinct banks).

4.2.2. Sample Descriptive Statistics

Tabulated statistics are based on data winsorized at the 1st and 99th percentiles for continuous variables to mitigate outlier effects.

Table 1. Sample statistics.

Variable	Mean	Std Dev	Median	Minimum	Q1	Q3	Maximum
GCO	0.009	0.089	0.000	0.000	0.000	0.000	1.000
TypeI	0.006	0.067	0.000	0.000	0.000	0.000	1.000
TypeII	0.028	0.161	0.000	0.000	0.000	0.000	1.000
Exp_MKT_N	0.134	0.340	0.000	0.000	0.000	0.000	1.000
Exp_MKT_S	0.018	0.132	0.000	0.000	0.000	0.000	1.000
Exp_NumInd_N	0.191	0.393	0.000	0.000	0.000	0.000	1.000
Exp_NumInd_S	0.034	0.182	0.000	0.000	0.000	0.000	1.000
Audsize_N	3.223	2.113	2.718	0.053	1.291	5.653	6.341
Audsize_S	1.145	1.026	0.890	0.033	0.454	1.414	4.609
Size	7.861	1.766	7.498	4.895	6.639	8.692	13.887
StdEarn	60.111	222.776	4.027	0.018	1.214	16.346	1601.340
ROA	0.007	0.009	0.009	−0.042	0.006	0.011	0.020
Leverage	0.109	0.078	0.096	0.000	0.051	0.147	0.391
Loss	0.079	0.269	0.000	0.000	0.000	0.000	1.000
MB	1.266	0.654	1.182	0.000	0.857	1.605	3.499
AltmanZ	0.045	0.013	0.043	0.023	0.035	0.054	0.089
Fin_New	0.753	0.431	1.000	0.000	1.000	1.000	1.000
Big4	0.379	0.485	0.000	0.000	0.000	1.000	1.000
Tenure	11.755	6.659	10.000	1.000	6.000	17.000	23.000
CapAsset	0.100	0.028	0.098	0.032	0.083	0.115	0.203
DepositAsset	0.776	0.082	0.790	0.500	0.732	0.835	0.909
LoanAsset	0.657	0.121	0.672	0.219	0.593	0.742	0.872
Provision	0.005	0.009	0.002	−0.005	0.001	0.005	0.055
NPL	0.019	0.025	0.010	0.000	0.005	0.021	0.145
Restate	0.047	0.211	0.000	0.000	0.000	0.000	1.000
Interven	0.005	0.073	0.000	0.000	0.000	0.000	1.000
IC_W	0.019	0.137	0.000	0.000	0.000	0.000	1.000

Based on our entire sample of 4492 bank-year observations (414 different banks) over the period of 2002–2023, Table 1 reports the descriptive statistics of the variables used in our analyses. Please refer to the variable measurements in Appendix A.

presents descriptive statistics for the 4992 bank-year observations (414 banks) from 2002 to 2023. Mean values for GCO, TypeI, and TypeII are 0.009, 0.006, and 0.028, respectively, indicating low failure, GCO issuance, and error rates, making statistically significant results challenging but potentially informative.

The mean values of the variables of Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, and Audsize_S are 0.134, 0.018, 0.191, 0.034, 3.223, and 1.145, respectively. These results indicate that on average, 13.4% of auditors included in our sample are classified as audit industry experts at the national level based on the auditor’s market share in a particular industry in a fiscal year in the U.S., while on average, only 1.8% of auditors (included in our sample) are classified as audit industry experts at the state level based on the auditor’s market share in a particular industry in a fiscal year in the state of the U.S., in which the bank is located. Similarly, only a very small percentage (3.4%) of auditors are classified as an industry expert based on the number of the banks audited by the auditor in a certain industry in a particular state in a fiscal year. The mean values of auditor’s size at the national and state level indicate that the auditors included in our sample are large-size auditors at both national and state levels.

The mean values of variables of Size, StdEarn, ROA, Leverage, Loss, and MB are 7.861, 60.111, 0.007, 0.109, 0.079, and 1.266, respectively. These results indicate that on average, the sample banks are large in size and have volatile earnings, a lower profitability, a low ratio of debt to assets, a small occurrence of operating loss, and a low expected growth. The mean value of AltmanZ (i.e., Altman Z score as Altman, 1968) is 0.045, which shows that the majority of the sample banks are not subject to the risk of being closed. In contrast, the mean value of AltmanZ for the closed banks from the untabulated descriptive statistics for the closed banks is −1.911, indicating that the closed banks show financial stress as measured by the Altman Z score.

The mean values of the variables of Fin_New, Big4, Tenure, CapAsset, LoanAsset, Provision, and NPL are 0.753, 0.379, 11.755, 0.100, 0.657, 0.005, and 0.019, respectively, indicating frequent new financing, a fair proportion of Big Four audits, long auditor tenure, adequate capitalization, high loan-to-asset ratios, and low estimated loan losses. The mean values of the variables of Restate, Interven, and IC_W are 0.047, 0.005, and 0.019, respectively, indicating low rates of restatements, regulatory intervention, and internal control weaknesses.

5. Results

5.1. Univariate Analyses

5.1.1. Correlations

Table 2. Pearson (above) and Spearman (below) correlation table for all variables used in the analyses.

	1	2	3	4	5	6	7
Variable	GCO	TypeI	TypeII	Exp_MKT_N	Exp_MKT_S	Exp_NumInd_N	Exp_NumInd_S
GCO		0.169	0.156	−0.018	−0.012	−0.013	−0.016
TypeI	0.169		−0.010	0.004	0.010	−0.006	−0.012
TypeII	0.156	−0.010		−0.014	−0.021	−0.032	−0.024
Exp_MKT_N	−0.018	0.004	−0.014		−0.020	0.704	0.041
Exp_MKT_S	−0.012	0.010	−0.021	−0.020		−0.036	−0.025
Exp_NumInd_N	−0.013	−0.006	−0.032	0.704	−0.036		0.028
Exp_NumInd_S	−0.016	−0.012	−0.024	0.041	−0.025	0.028
Audsize_N	−0.026	−0.019	−0.086	0.488	−0.042	0.551	−0.029
Audsize_S	−0.016	−0.012	−0.078	0.259	−0.086	0.326	0.177
Size	−0.052	−0.045	−0.069	0.168	0.025	0.228	−0.067
StdEarn	0.047	0.039	0.001	0.122	0.016	0.153	−0.089
ROA	−0.128	−0.070	−0.104	0.085	0.022	0.083	−0.005
Leverage	0.002	−0.025	0.059	0.107	−0.021	0.077	−0.015
Loss	0.242	0.093	0.160	−0.046	−0.007	−0.050	−0.018
MB	−0.113	−0.048	0.002	0.126	−0.017	0.137	0.002
AltmanZ	0.002	−0.009	0.094	0.090	0.000	0.009	−0.040
Fin_New	−0.026	−0.004	−0.027	−0.012	−0.008	0.039	−0.004
Big4	−0.029	−0.030	−0.049	0.506	−0.048	0.498	−0.047
Tenure	−0.036	−0.025	−0.110	0.175	0.014	0.150	−0.018
CapAsset	−0.118	−0.035	−0.146	−0.108	0.005	−0.092	−0.041
DepositAsset	0.066	0.046	0.003	−0.099	0.017	−0.070	0.029
LoanAsset	0.006	−0.016	0.061	−0.097	0.033	−0.035	−0.029
Provision	0.091	0.004	0.135	−0.001	−0.031	−0.026	−0.034
NPL	0.128	0.081	0.084	−0.047	−0.032	−0.086	−0.008
Restate	0.044	0.022	0.035	0.041	−0.006	0.033	0.010
Interven	0.047	0.031	0.048	0.005	−0.010	−0.011	−0.013
IC_W	0.073	0.010	0.025	−0.002	−0.009	0.017	0.030
	8	9	10	11	12	13	14
Variable	Audsize_N	Audsize_S	Size	StdEarn	ROA	Leverage	Loss
GCO	−0.028	−0.016	−0.049	−0.011	−0.330	0.001	0.242
TypeI	−0.024	−0.012	−0.042	−0.007	−0.101	−0.022	0.093
TypeII	−0.077	−0.074	−0.071	−0.033	−0.225	0.073	0.160
Exp_MKT_N	0.488	0.333	0.167	0.102	0.067	0.108	−0.046
Exp_MKT_S	−0.047	−0.074	0.022	0.003	0.017	−0.024	−0.007
Exp_NumInd_N	0.552	0.386	0.184	0.022	0.062	0.072	−0.050
Exp_NumInd_S	−0.036	0.171	−0.063	−0.036	0.012	−0.010	−0.018
Audsize_N		0.617	0.592	0.281	0.098	0.120	−0.065
Audsize_S	0.659		0.547	0.406	0.077	0.080	−0.052
Size	0.604	0.506		0.638	0.148	0.133	−0.098
StdEarn	0.484	0.411	0.798		−0.013	0.123	0.020
ROA	0.119	0.076	0.258	0.100		−0.059	−0.774
Leverage	0.112	0.066	0.123	0.075	−0.102		0.023
Loss	−0.063	−0.043	−0.104	0.156	−0.465	0.024
MB	0.140	0.106	0.127	−0.029	0.488	0.053	−0.274
AltmanZ	−0.078	−0.140	−0.235	−0.137	0.137	0.305	0.049
Fin_New	0.153	0.171	0.191	0.124	−0.077	0.060	−0.033
Big4	0.835	0.503	0.544	0.450	0.136	0.181	−0.066
Tenure	0.301	0.319	0.379	0.314	0.109	0.109	−0.079
CapAsset	0.011	0.033	0.077	0.047	0.192	−0.223	−0.167
DepositAsset	−0.172	−0.141	−0.221	−0.152	0.026	−0.904	0.055
LoanAsset	−0.105	−0.150	−0.171	−0.145	−0.064	−0.028	0.035
Provision	0.014	0.020	0.016	0.150	−0.335	0.132	0.354
NPL	−0.092	−0.082	−0.159	0.037	−0.422	0.030	0.361
Restate	0.075	0.043	0.059	0.090	−0.047	0.062	0.075
Interven	−0.006	0.014	0.028	0.064	−0.032	0.022	0.048
IC_W	0.012	0.032	0.025	0.053	−0.061	0.035	0.092
	15	16	17	18	19	20	21
Variable	MB	AltmanZ	Fin_New	Big4	Tenure	CapAsset	DepositAsset
GCO	−0.112	−0.001	−0.026	−0.029	−0.037	−0.156	0.061
TypeI	−0.051	−0.012	−0.004	−0.030	−0.024	−0.048	0.040
TypeII	0.012	0.090	−0.027	−0.049	−0.112	−0.163	−0.008
Exp_MKT_N	0.134	0.083	−0.012	0.506	0.178	−0.094	−0.102
Exp_MKT_S	−0.011	−0.005	−0.008	−0.048	0.011	0.011	0.025
Exp_NumInd_N	0.136	0.007	0.039	0.498	0.163	−0.084	−0.063
Exp_NumInd_S	−0.002	−0.043	−0.004	−0.047	−0.008	−0.037	0.027
Audsize_N	0.150	−0.039	0.131	0.897	0.359	−0.022	−0.182
Audsize_S	0.115	−0.108	0.143	0.542	0.381	−0.051	−0.166
Size	0.092	−0.185	0.191	0.554	0.433	−0.033	−0.269
StdEarn	−0.015	−0.026	0.096	0.296	0.252	−0.099	−0.255
ROA	0.337	0.062	−0.026	0.101	0.121	0.251	−0.037
Leverage	0.077	0.294	0.020	0.175	0.099	−0.199	−0.919
Loss	−0.243	0.037	−0.033	−0.066	−0.086	−0.175	0.046
MB		0.249	−0.092	0.170	0.063	−0.154	−0.041
AltmanZ	0.224		−0.210	0.064	−0.042	−0.125	−0.246
Fin_New	−0.083	−0.190		0.038	0.092	0.041	−0.052
Big4	0.165	0.060	0.038		0.447	−0.053	−0.226
Tenure	0.060	−0.028	0.099	0.429		0.012	−0.147
CapAsset	−0.138	−0.145	0.055	−0.041	0.046		−0.098
DepositAsset	−0.027	−0.249	−0.078	−0.232	−0.146	−0.085
LoanAsset	−0.075	0.081	0.123	−0.156	−0.140	0.063	0.081
Provision	−0.202	0.306	−0.048	0.052	0.027	−0.161	−0.066
NPL	−0.386	0.129	0.009	−0.108	−0.022	0.027	0.000
Restate	−0.024	0.028	−0.008	0.086	0.020	−0.008	−0.069
Interven	−0.026	0.011	−0.002	−0.004	−0.010	−0.023	−0.023
IC_W	0.015	0.012	0.020	−0.001	−0.041	−0.031	−0.019
	22	23	24	25	26	27
Variable	LoanAsset	Provision	NPL	Restate	Interven	IC_W
GCO	0.012	0.274	0.276	0.044	0.047	0.073
TypeI	−0.010	0.052	0.130	0.022	0.031	0.010
TypeII	0.051	0.201	0.173	0.035	0.048	0.025
Exp_MKT_N	−0.102	−0.020	−0.039	0.041	0.005	−0.002
Exp_MKT_S	0.033	−0.021	−0.029	−0.006	−0.010	−0.009
Exp_NumInd_N	−0.047	−0.032	−0.076	0.033	−0.011	0.017
Exp_NumInd_S	−0.018	−0.030	−0.017	0.010	−0.013	0.030
Audsize_N	−0.132	−0.001	−0.095	0.081	−0.005	0.009
Audsize_S	−0.203	−0.011	−0.100	0.038	0.020	0.013
Size	−0.271	0.006	−0.146	0.046	0.040	0.005
StdEarn	−0.263	0.103	−0.019	0.005	0.047	−0.019
ROA	−0.054	−0.749	−0.629	−0.063	−0.053	−0.098
Leverage	−0.094	0.075	−0.005	0.066	0.029	0.041
Loss	0.043	0.667	0.557	0.075	0.048	0.092
MB	−0.087	−0.219	−0.320	−0.009	−0.025	0.014
AltmanZ	0.086	0.168	0.074	0.030	0.021	0.007
Fin_New	0.111	−0.030	−0.031	−0.008	−0.002	0.020
Big4	−0.165	0.010	−0.100	0.086	−0.004	−0.001
Tenure	−0.172	−0.025	−0.074	0.020	−0.012	−0.040
CapAsset	0.072	−0.214	−0.149	0.001	−0.024	−0.038
DepositAsset	0.152	0.000	0.072	−0.071	−0.033	−0.022
LoanAsset		0.023	−0.022	−0.040	−0.032	0.019
Provision	0.017		0.679	0.080	0.081	0.100
NPL	−0.002	0.455		0.073	0.073	0.069
Restate	−0.026	0.069	0.059		0.292	0.115
Interven	−0.040	0.059	0.054	0.292		0.146
IC_W	0.014	0.054	0.045	0.115	0.146

Table 2 reports correlation statistics for the variables used in our analyses based on the sample of 4492 bank-year observations (414 distinct banks) over the period of 2002–2023. All continuous variables are winsorized at the 1st and 99th percentiles to mitigate the influences of outliers. Values bolded represent statistical significance at the 0.1 level or lower, where the p-values for the test statistics of the estimated correlation coefficients are based on two-sided t-tests. Please refer to the variable measurements in Appendix A.

presents Pearson (above diagonal) and Spearman (below diagonal) correlation matrices of the variables used in our analyses.

The Pearson [Spearman] correlations for the pairs comprising GCO and Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, and Audsize_S are −0.018 (p-value > 0.1) [−0.018 (p-value > 0.1)], −0.012 (p-value > 0.1) [−0.012 (p-value > 0.1)], −0.013 (p-value > 0.1) [−0.013 (p-value > 0.1)], −0.016 (p-value > 0.1) [−0.016 (p-value > 0.1)], −0.028 (p-value < 0.05) [−0.026 (p-value < 0.05)], and −0.016 (p-value > 0.1) [−0.016 (p-value > 0.1)], respectively. The negative correlation result between the proxies for the quality of the engagement auditor and issuance of GCOs indicates that the audit industry experts and large-sized auditors are less likely to issue a going concern opinion to the audited banks.

The Pearson [Spearman] correlations for the pairs comprising TypeI and Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, and Audsize_S are 0.004 (p-value > 0.1) [0.004 (p-value > 0.1)], 0.010 (p-value > 0.1) [0.010 (p-value > 0.1)], −0.006 (p-value > 0.1) [−0.006 (p-value > 0.1)], −0.012 (p-value > 0.1) [−0.012 (p-value > 0.1)], −0.024 (p-value < 0.1) [−0.019 (p-value > 0.1)], and −0.012 (p-value > 0.1) [−0.012 (p-value > 0.1)], respectively. The results of the proxies for audit industry experts having more clients and large-sized auditors are negatively correlated with the likelihood of Type I errors in GCOs issued to audited banks. These results indicate that the engagement auditor having the largest number of clients in the nation and the state and large-sized auditors are less likely to have Type I errors in going concern opinions issued to the banks.

Similar to the analyses for the correlations for the pair of GCO/TypeI and the proxies for the audit industry expert and auditor’s size, the Pearson [Spearman] correlations for the pairs comprising TypeII and Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, and Audsize_S are −0.014 (p-value > 0.1) [−0.014 (p-value > 0.1)], −0.021 (p-value < 0.1) [−0.021 (p-value < 0.1)], −0.032 (p-value < 0.05) [−0.032 (p-value < 0.05)], −0.024 (p-value < 0.1) [−0.024 (p-value < 0.1)], −0.077 (p-value < 0.001) [−0.086 (p-value < 0.001)], and −0.074 (p-value < 0.001) [−0.078 (p-value < 0.001)], respectively. The negative correlations between the proxies for the audit industry expert and auditor’s size and the likelihood of Type II errors in GCO issued to banks indicate that the audit industry expert and large-sized auditors are less likely to have Type II errors in going concern opinions issued to the banks.

The negative correlations for the pairs comprising TypeI/TypeII and proxies for the audit industry expert and large-sized auditors show that both Type I and Type II error rates in the GCOs issued by the industry expert and large-sized auditor are lower than those issued by the counterparts. Moreover, the result that the pairs of correlations between Type II error and proxies for the quality of engagement auditors outnumber the correlations between Type I error and proxies for the quality of engagement auditor indicates that the audit industry expert and large-sized auditors are more conservative on the issuance of GCOs as indicated by a higher rate of Type II error than Type I error. In other words, high-quality auditors issue more accurate GCOs with a tendency for conservatism in the issuance of GCOs. These findings are consistent with the theory and empirical evidence in previous studies (e.g., Reichelt & Wang, 2010; M. DeFond & Zhang, 2014).

We apply the method used to analyze the correlations between the dependent variables and the proxies for the quality of engagement auditors on the analyses of the correlations for the pairs comprising GCO/TypeI/TypeII and control variables. Consistent with our expectation about the impact of the controlled factors on the GCO outcome, auditees’ size, profitability, growth expected by the capital market, new financing activities, audit engagement conducted by the Big4 auditors, the duration of audit engagement conducted by the same auditor, and adequacy of capital are negatively related to the issuance of GCO and the errors in the GCOs issued to the banks. Furthermore, the auditees’ earnings volatility, business operating loss, estimated losses in the loans, non-performing loans, occurrence of restating financial statements, intervention from regulator(s), and internal control weakness identified by the external auditor(s) are positively related to the issuance of GCOs and the errors in the GCOs issued to the auditees.

Overall, univariate tests in Table 2 provide interesting results with informative indications. Consistent with our expectation and the previous literature, the auditor’s industry expertise is positively correlated with the accuracy of issuance of GCOs, while the auditor’s size is positively correlated with conservatism on the issuance of GCOs. Moreover, these results are partially consistent with the results reported by Albrecht et al. (2020) that document a negative relationship between the likelihood of the “over-issue” of GCOs to systemically risky banks, resulting in more accurate GCOs. However, our results are based on a larger sample and a much longer sample period, providing a more generable inference regarding the different users of the empirical results documented in this study (e.g., practitioners, investors, and regulators) than Albrecht et al. (2020).

The correlations between control variables and subsequent VIF tests on the control variables in the regressions do not indicate any collinearity concerns.

5.1.2. Contingency Table Analysis

Table 3. Univariate analysis of going concern opinions and subsequent bank closure.

		Bank Closure		Total
		0	1
GCO	0	4806	140	4946
issuance
	1	30	16	46
	Total	4836	156	4992

Based on the sample of 4992 bank-year observations (414 distinct banks) over the period of 2002–2023, Table 3 reports univariate analyses of going concern opinions and subsequent closure of the banks, illustrating how the bank closure status is associated with the status of receiving a GCO before the closure. Please refer to the variable measurements in Appendix A.

presents a univariate analysis of going concern opinions and subsequent failure status. The results of Table 3 are layout based on (1) the whether the bank receives a GCO during the current year, which is indicated by a value of one (zero) denoting the issuance of one (zero) GCO in the first column, and (2) whether the bank fails within 12 months after receiving the GCO in the first row, which is indicated by a value of one (zero) denoting (not) being closed. In our sample, 156 bank-year observations subsequently failed, 140 of which were not issued a GCO within the period of 12 months before the closure, resulting in a 89.7% Type II error rate in GCOs. Of 46 banks receiving GCOs, 30 did not fail, resulting in a 65.2% Type I error rate. Bank failure is defined as Compustat deletion due to bankruptcy, liquidation, leveraged buyout, privatization, or regulatory closure.

Type I error rate (bank closure = 0, GCO = 1): 65.2%

Type II error rate (bank closure = 1, GCO = 0): 89.7%

These univariate analysis results are consistent with our hypotheses. However, because these results are possibly driven by other factors, we further conduct multivariate analyses in the next subsection.

5.2. Multivariate Analyses

Regression Analyses

To establish the causal inference from our regression results, we use the propensity score matching (hereafter, PSM) method to control for the factors that are documented in M. DeFond and Zhang (2014), may cause the auditees to select an audit industry expert.

Table 4. Regression of the issuance of a going concern opinion (GCO) on the proxies for the audit industry expert.

	1	2	3	4
	Exp_MKT_N	Exp_MKT_S	Exp_Numind_N	Exp_Numind_S
AQk	−0.000581	−0.00755 ***	0.00162	−0.0108 ***
	(−0.123)	(−3.347)	(0.512)	(−4.157)
Size	0.00557	−0.00120	−0.000517	−0.00232 ***
	(1.414)	(−1.104)	(−0.502)	(−2.619)
StdEarn	−3.52 × 10−05 *	−1.54 × 10−05 **	−3.02 × 10−06	−6.19 × 10−06
	(−1.737)	(−2.272)	(−0.564)	(−1.214)
ROA	−3.626 **	−2.154 ***	−3.286 ***	−1.724 ***
	(−2.224)	(−3.655)	(−3.215)	(−2.761)
Leverage	0.0715	0.0279 *	−0.00143	−0.00564
	(1.191)	(1.893)	(−0.0806)	(−0.382)
Loss	−0.0460	−0.0111	−0.0155	−0.00827
	(−1.302)	(−0.710)	(−0.538)	(−0.678)
MB	−0.0122	0.00252	0.00426	0.00127
	(−0.835)	(1.080)	(0.847)	(0.349)
AltmanZ	−0.282	0.0489	0.363 **	0.207
	(−0.514)	(0.330)	(2.161)	(0.937)
Fin_New	−0.0286	−0.000483	−0.00618	−0.00334
	(−1.522)	(−0.112)	(−0.955)	(−0.918)
Big4	−0.00433	0.00106	−0.00142	0.00416
	(−0.931)	(0.420)	(−0.468)	(1.495)
Tenure	−0.000251	−3.80e-05	0.000412	0.000260 *
	(−0.555)	(−0.224)	(1.435)	(1.789)
CapAsset	−0.366 ***	−0.214 ***	−0.251 ***	−0.278 ***
	(−2.830)	(−3.865)	(−3.110)	(−3.984)
LoanAsset	−0.0143	0.00598	−0.00222	0.00769
	(−1.283)	(0.705)	(−0.181)	(0.794)
Provision	1.162	0.121	0.454	0.458
	(1.256)	(0.236)	(0.719)	(0.935)
NPL	0.419	0.610 ***	0.495 **	0.0733
	(1.455)	(4.150)	(2.448)	(0.452)
Restate	0.00190	0.00919	0.0176	−0.00638
	(0.301)	(1.302)	(1.081)	(−0.641)
Interven	−0.000158	0.00885	−0.0193	0.0229
	(−0.00781)	(0.249)	(−0.678)	(0.623)
IC_W	−0.0186	−0.0243	0.0134	0.0156
	(−0.670)	(−1.333)	(0.733)	(1.235)
Constant	0.0699 *	0.0345 **	0.0234 **	0.0412 ***
	(1.927)	(2.519)	(1.999)	(2.594)
Year FE	Y	Y	Y	Y
Industry FE	Y	Y	Y	Y
Observations	4992	4992	4992	4992
Adjusted R2	0.243	0.117	0.169	0.085

All regressions include an intercept as well as year-fixed and industry-fixed effects. Table 4 reports OLS coefficient estimates and t-statistics in ( ) based on the PSM (propensity score matching) method to control for the possible estimation bias arisen from selecting an audit industry expert. The statistical significance of the coefficient estimators is based on the robust standard errors corrected for bank-level clustering and White’s heteroskedasticity consistent estimator. Here, *, **, and *** indicate statistical significance in means at the 10%, 5%, and 1% levels (two-tailed), respectively. Please refer to the variable measurements in Appendix A.

reports the results of the PSM regressions of issuance of a going concern opinion (GCO) on the proxies for the audit industry expertise (i.e., Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, Exp_Numind_S) and control variables that may influence the dependent variable (GCO). In Table 4, the estimated coefficients for Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, and Exp_Numind_S in the GCO regression are −0.000581 (p-value > 0.1), −0.00755 (p-value < 0.01), 0.00162 (p-value > 0.1), and −0.0108 (p-value < 0.01), respectively, showing that the issuance of GCOs to banks decreases with the use of an audit industry expert at the state level but not at the national level. These results suggest that the audit industry expert at the state level is less likely to over-issue GCOs, when controlling for other factors that may cause the issuance of GCOs. Among the statistically significant estimated coefficients, the coefficients for the variable of ROA in the regression of GCO on Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, and Exp_Numind_S are −3.626 (p-value < 0.05), −2.154 (p-value < 0.01), −3.286 (p-value < 0.01), and −1.724 (p-value < 0.01), respectively, suggesting that the likelihood of receiving a GCO decreases with the increasing in the profitability of the bank. Similarly, the estimated coefficients for the variable of CapAsset in the GCO regressions show that a bank with adequate capital is less likely to receive a GCO. Oppositely, the coefficients for the variable of NPL in the GCO regressions show that a bank with a higher portion of non-performing loans is more likely to receive a GCO.

Using the sample of 4492 bank-year observations (414 distinct banks) over the period of 2002–2023, Table 4 reports the results of the regressions of GCO (i.e., the indicator variable of issuing going concern opinion) on the proxies for the engagement auditor’s industry expertise (i.e., Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, and Exp_NumInd_S in columns 1, 2, 3, and 4, respectively), controlling for the factors that may impact the issuance of GCOs.

Panel A (B) of

Table 5. Regression of Type I and Type II errors in GCOs on the proxies for the audit industry expert.

(A)
	1	2	3	4
	Exp_MKT_N	Exp_MKT_S	Exp_Numind_N	Exp_Numind_S
AQk	0.00242	−0.00325 ***	0.00274	−0.00536 ***
	(1.131)	(−2.610)	(1.088)	(−3.696)
Size	0.000393	−0.000877 *	1.05 × 10−05	−0.00107 *
	(0.650)	(−1.673)	(0.0216)	(−1.944)
StdEarn	−1.36 × 10−06	1.55 × 10−06	−1.59 × 10−06	2.44 × 10−06
	(−0.405)	(0.446)	(−0.645)	(1.043)
ROA	−0.652	−0.502	−0.836	−0.490
	(−1.472)	(−1.428)	(−1.414)	(−1.280)
Leverage	−0.0243 *	−0.00359	−0.0299 *	−0.0174 *
	(−1.895)	(−0.426)	(−1.793)	(−1.958)
Loss	0.00646	0.0109	0.000710	0.00358
	(0.755)	(1.021)	(0.0635)	(0.481)
MB	−0.000357	0.00151	0.00455	0.000287
	(−0.136)	(0.888)	(1.313)	(0.125)
AltmanZ	0.133	0.0292	0.0223	0.142
	(1.088)	(0.382)	(0.166)	(1.173)
Fin_New	0.000705	0.00159	0.000805	−0.00135
	(0.182)	(0.596)	(0.161)	(−0.593)
Big4	−0.00366 **	7.81 × 10−05	−0.00209 *	0.00238
	(−2.472)	(0.0590)	(−1.711)	(1.459)
Tenure	0.000244	4.74 × 10−06	0.000291	7.66 × 10−05
	(1.356)	(0.0454)	(1.368)	(1.027)
CapAsset	−0.0653	−0.0408	−0.0639	−0.101 *
	(−1.404)	(−0.916)	(−1.389)	(−1.938)
LoanAsset	−0.0113 *	−0.00662	−0.0107	−0.00530
	(−1.751)	(−1.348)	(−1.468)	(−1.143)
Provision	−0.545	−0.532	−0.586	−0.498
	(−0.899)	(−1.259)	(−0.786)	(−1.340)
NPL	0.284	0.312 **	0.472	0.145
	(1.364)	(2.267)	(1.584)	(1.211)
Restate	−0.000927	0.00210	−0.000679	−0.00268
	(−0.233)	(0.511)	(−0.172)	(−0.552)
Interven	0.0121	0.0259	0.0174	0.0341
	(0.581)	(0.690)	(0.611)	(0.875)
IC_W	0.000759	−0.000279	−0.000458	0.00437
	(0.116)	(−0.0497)	(−0.0704)	(0.775)
Constant	0.0115	0.0152 *	0.0117	0.0207 *
	(1.335)	(1.664)	(1.570)	(1.854)
Year FE	Y	Y	Y	Y
Industry FE	Y	Y	Y	Y
Observations	4992	4992	4992	4992
Adjusted R2	0.043	0.039	0.057	0.028
(B)
	1	2	3	4
	Exp_MKT_N	Exp_MKT_S	Exp_Numind_N	Exp_Numind_S
AQk	−0.000533	−0.0276 ***	−0.00748 *	−0.0247 ***
	(−0.0858)	(−7.503)	(−1.920)	(−4.433)
Size	0.0109 ***	0.00229	0.00628 ***	0.00287
	(2.705)	(1.372)	(3.408)	(1.305)
StdEarn	−4.92 × 10−05 **	−1.31 × 10−05 *	−2.19 × 10−05 *	−3.38 × 10−06
	(−2.376)	(−1.751)	(−1.716)	(−0.374)
ROA	−2.539	−1.916 **	−1.265	−1.682 **
	(−1.574)	(−2.561)	(−1.113)	(−2.020)
Leverage	0.167 **	0.0527 *	0.133 **	−0.0167
	(2.409)	(1.822)	(2.414)	(−0.566)
Loss	−0.0524	−0.0232	−0.00649	−0.0159
	(−1.346)	(−1.446)	(−0.207)	(−1.118)
MB	−0.0192	0.00818 ***	−0.00801	0.00514
	(−1.416)	(2.646)	(−1.053)	(0.829)
AltmanZ	−0.379	0.149	0.535	0.226
	(−0.604)	(0.647)	(1.157)	(0.660)
Fin_New	−0.0288	−0.00526	0.000583	−0.00685
	(−1.572)	(−1.035)	(0.0962)	(−1.583)
Big4	−0.0191 ***	−0.0188 ***	−0.00618	−0.0217 **
	(−2.692)	(−3.865)	(−1.047)	(−2.187)
Tenure	−0.00236 ***	−0.00114 ***	−0.00268 ***	−0.00118 ***
	(−4.132)	(−3.920)	(−4.788)	(−3.815)
CapAsset	−0.535 ***	−0.188 ***	−0.555 ***	−0.407 ***
	(−3.632)	(−2.839)	(−3.704)	(−4.286)
LoanAsset	0.00596	0.0547 ***	−0.0179	0.0876 **
	(0.165)	(3.589)	(−0.359)	(2.050)
Provision	0.963	0.811	−0.624	1.089 *
	(0.986)	(1.407)	(−0.819)	(1.881)
NPL	0.874 **	0.562 ***	1.209 ***	−0.112
	(2.287)	(3.371)	(2.644)	(−0.546)
Restate	−0.00266	0.0218 **	−0.00167	0.00671
	(−0.287)	(2.468)	(−0.167)	(0.607)
Interven	0.0412	0.0575	0.0388	0.0714
	(0.970)	(0.915)	(0.729)	(1.119)
IC_W	−0.0118	−0.0238	0.0173	0.000143
	(−0.303)	(−1.191)	(0.563)	(0.00810)
Constant	0.116 **	0.0127	0.0765 **	0.0292
	(2.449)	(0.709)	(2.388)	(0.823)
Year FE	Y	Y	Y	Y
Industry FE	Y	Y	Y	Y
Observations	4992	4992	4992	4992
Adjusted R2	0.138	0.077	0.105	0.070

Panel A: Dependent Variable = TypeI (i.e., Type I error in GCO issued to banks); Panel B: Dependent Variable = TypeII (i.e., Type II error in GCO issued to banks). All regressions include an intercept as well as year-fixed and industry-fixed effects. Table 5 reports OLS coefficient estimates and t-statistics in ( ) based on the PSM (propensity score matching) method to control for the possible estimate bias arisen from selecting of an audit industry expert. The statistical significance of the coefficient estimators is based on the robust standard errors corrected for bank-level clustering and White’s heteroskedasticity consistent estimator. Here, *, **, and *** indicate statistical significance in means at the 10%, 5%, and 1% levels (two-tailed), respectively. Please refer to the variable measurements in Appendix A.

reports the results of the regressions of TypeI (TypeII) on the proxies for the audit industry expert (i.e., Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, Exp_Numind_S) and control variables that may influence the dependent variables (i.e., TypeI/TypeII).

In Panel A of Table 5, the coefficients for Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, and Exp_Numind_S in the TypeI (i.e., Type I error in the GCO issued to banks) regressions are 0.00242 (p-value > 0.1), −0.00325 (p-value < 0.01), 0.00274 (p-value > 0.1), and −0.00536 (p-value < 0.01), respectively, showing that the rate of Type I error in the GCOs issued to banks decreases with the presence of an audit industry expert at the state level but not at the national level. These results suggest that an audit industry expert at the state level is less likely to have a Type I error in GCOs issued to banks, when controlling for other factors that may cause Type I errors in GCOs. In other words, the GCO issued by the audit industry expert at the state level is less likely to be a false positive alarm because the false positive error rate in the GCO issued by the industry expert at the state level is lower than the counterparts. Among the statistically significant coefficients, the coefficients for the variables of Size, Leverage, and Big4 in the TypeI regressions suggest that the rate of Type I errors in the GCOs decreases with the increase in size, the amount of borrowing of the bank, and audits performed by Big4 auditors.

In Panel B of Table 5, the coefficients for Exp_MKT_N, Exp_MKT_S, Exp_Numind_N, and Exp_Numind_S in the TypeII (i.e., Type II error in the GCO issued to banks) regressions are −0.000533 (p-value > 0.1), −0.0276 (p-value < 0.01), −0.00748 (p-value < 0.1), and −0.0247 (p-value < 0.01), respectively, showing that the rate of Type II errors in the GCOs issued to the banks decreases with the presence of an audit industry expert at the national/state level. These results suggest that an audit industry expert, regardless of expert status at the national or state level, is less likely to commit Type II errors in GCOs issued to banks, when controlling for other factors that may cause Type II errors in GCOs. In other words, a bank is more likely to receive a GCO before its failure if it was audited by an audit industry expert. Among the statistically significant coefficients, the coefficients for the variables of Size, Leverage, ROA, Big4, Tenure, CapAsset, and NPL in the TypeII regressions suggest that the rate of Type II errors in the GCOs increases with the increase in the size, the amount of borrowing, and NPLs of the bank, while it decreases with the increase in ROA, the duration of the engagement conducted by the same auditor, adequacy of the capital of the bank, and audits performed by Big4 auditors. These results are interesting. On the one hand, the auditor tends to issue a false negative clean audit opinion to banks with a large size, high amounts of borrowing, and substantial NPLs, and these banks were subsequently closed. This phenomenon might be caused by the common perception of too big to fall and of the safety in the investments in loans with the lower risk and returns. On the other hand, the information embedded in the bank’s profitability and adequacy of the capital, as well as hiring a Big4 auditor and being audited by the same engagement auditor, might reduce the rate of Type II error in GCOs.

The results of the regressions in Table 4 and Table 5 collectively suggest that audit industry expert is less likely to over-issue GCOs. The GCOs issued by the audit industry expert at the state level is more accurate than the counterparts, with lower Type I and Type II error rates in GCOs. Moreover, the audit industry expert at the national level is only associated with a lower rate of Type II error without reducing Type I errors, suggesting that the audit industry experts at the national level are slightly conservative on issuing GCOs. These results align with expectations that state-level audit industry experts have better client knowledge, leading to more accurate GCOs. Consistently, Type I and Type II error rates are reduced by state-level expertise measured by different proxies, while only Type II error rates are marginally reduced by one national-level expertise proxy.

The audit industry expert at the state level could issue more accurate GCO than audit industry expert at the national level might be due to the following reasons. First, audit industry experts at the state level might communicate with regulators in the state more frequently and more effectively than the audit industry expert at the national level, which could in turn improve the accuracy of GCOs issued to the banks. Second, the local auditors could have advantages for visiting the clients’ sites more frequently than national auditors to obtain more useful information about the clients’ business operation and financial-healthy status, therefore, resulting in a better judgement on the going concerns of the banks. Additionally, the greater number of the clients the national auditor has in the industry, the more industry-specific knowledge the auditor has about the clients. However, increasing the industry market share of the national auditor does not improve the auditor’s performance of issuing GCOs.

In a sum, the regression results indicate that the audit industry experts at the state level know their clients (banks) more than audit industry experts at the national level and other counterparts and, therefore, issuing more accurate GCOs to the banks.

Using the sample of 4492 bank-year observations (414 distinct banks) over the period of 2002–2023, Panel A [B] of Table 5 reports the results of the regressions of TypeI [TypeII] (i.e., the indicator variable of incidence of Type I [II] error in GCO issued to banks) on the proxies for the engagement auditor’s industry expertise (i.e., Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, and Exp_NumInd_S in columns 1, 2, 3, and 4, respectively), controlling for the factors that may impact the occurrence of Type I and Type II errors in GCOs issued to banks.

5.3. Additional Tests

5.3.1. Alternative Measurements of Audit Quality

To test the robustness of our results to different auditor quality proxies, we use alternative measurements for the quality of the engagement auditor to conduct robust tests.

First, using the top three-ranked auditors nationally and statewide, we re-run regressions, and the results of which are qualitatively the same as those reported in Table 4 and Table 5.

Next, we employ the auditor’s size at the national and state levels as the proxies for the engagement auditor’s quality to redo our regression analyses reported in Table 4 and Table 5.

Table 6. Regression of GCO and Type I and Type II errors in GCOs on proxies for the auditor’s size.

	GCO		TypeI		TypeII
	1	2	3	4	5	6
	Audsize_N	Audsize_S	Audsize_N	Audsize_S	Audsize_N	Audsize_S
AQk	−0.0846	0.0355	0.0784	0.0525	−0.653 ***	−0.689 ***
	(−0.295)	(0.119)	(0.249)	(0.171)	(−5.281)	(−3.222)
Size	−0.813 **	−0.809 **	−0.839 **	−0.837 **	0.304 **	0.240 *
	(−2.327)	(−2.309)	(−2.251)	(−2.241)	(2.248)	(1.748)
StdEarn	0.000927	0.000936	0.00329 **	0.00323 *	−0.0143 **	−0.0117 *
	(0.301)	(0.305)	(1.962)	(1.885)	(−2.087)	(−1.873)
ROA	−57.34 *	−56.57 *	−44.07	−43.69	−49.62 **	−44.23 **
	(−1.940)	(−1.924)	(−1.293)	(−1.279)	(−2.288)	(−2.135)
Leverage	4.923	4.935	−6.252	−6.330	0.886	0.714
	(1.299)	(1.306)	(−1.288)	(−1.304)	(0.613)	(0.492)
Loss	0.511	0.491	0.433	0.460	−0.236	−0.174
	(0.571)	(0.548)	(0.463)	(0.492)	(−0.425)	(−0.321)
MB	−1.247 ***	−1.245 ***	0.0165	0.0245	0.00869	−0.103
	(−2.793)	(−2.755)	(0.0319)	(0.0478)	(0.0482)	(−0.570)
AltmanZ	38.10 *	38.24 *	38.13 *	38.21 *	1.686	−0.757
	(1.706)	(1.720)	(1.668)	(1.669)	(0.186)	(−0.0846)
Fin_New	0.0882	0.0808	0.563	0.585	−0.252	−0.293
	(0.161)	(0.148)	(0.938)	(0.981)	(−0.792)	(−0.936)
Big4	1.504	1.092	0.548	0.835	1.890 ***	−0.347
	(1.047)	(1.515)	(0.351)	(0.993)	(3.359)	(−1.222)
Tenure	−0.00180	0.000636	−0.0272	−0.0314	−0.156 ***	−0.131 ***
	(−0.0361)	(0.0128)	(−0.459)	(−0.541)	(−6.023)	(−5.199)
CapAsset	−70.74 ***	−70.68 ***	−26.61 **	−26.44 **	−28.69 ***	−29.24 ***
	(−5.445)	(−5.458)	(−2.547)	(−2.531)	(−4.709)	(−4.811)
LoanAsset	1.475	1.460	−3.459	−3.470	3.984 ***	3.521 ***
	(0.573)	(0.567)	(−1.368)	(−1.370)	(3.622)	(3.300)
Provision	−10.68	−10.25	−40.22	−40.08	10.92	16.47
	(−0.402)	(−0.385)	(−1.366)	(−1.361)	(0.562)	(0.870)
NPL	16.48 *	17.00 **	17.27 **	17.06 **	18.36 ***	16.26 ***
	(1.925)	(2.020)	(2.108)	(2.079)	(3.019)	(2.805)
Restate	−0.0329	−0.0641	0.411	0.390	0.920 **	0.893 **
	(−0.0417)	(−0.0807)	(0.459)	(0.430)	(2.054)	(1.997)
Interven	1.722	1.836	0.658	0.670	1.587 *	1.811 *
	(1.054)	(1.134)	(0.415)	(0.420)	(1.704)	(1.883)
IC_W	1.679 *	1.673 *	1.586	1.530	−0.368	−0.299
	(1.845)	(1.789)	(1.126)	(1.098)	(−0.610)	(−0.498)
Constant	5.852	5.649	4.669	4.772	−2.957 **	−2.373
	(1.592)	(1.556)	(1.294)	(1.335)	(−1.992)	(−1.616)
Year FE	Y	Y	Y	Y	Y	Y
Industry FE	Y	Y	Y	Y	Y	Y
Observations	4992	4992	4992	4992	4992	4992
Adjusted R2	0.605	0.605	0.328	0.335	0.338	0.323

All regressions include an intercept as well as year-fixed and industry-fixed effects. The table reports logistic regression coefficient estimates and t-statistics in ( ) based on the robust standard errors corrected for bank-level clustering and White’s heteroskedasticity consistent estimator. Here, *, **, and *** indicate statistical significance in means at the 10%, 5%, and 1% levels (two-tailed), respectively. Please refer to the variable measurements in Appendix A.

reports the results of the regressions of GCO/TypeI/TypeII on the proxies for the auditor’s size at the national and state level (denoted as Audsize_N and Audsize_S, respectively) and control variables that may influence the dependent variables. Previous literature suggest that the larger-sized auditors provide better audit quality (e.g., Craswell et al., 2002).

The coefficients for Audsize_N [and Audsize_S] in the regression of GCO, TypeI and TypeII are −0.0846 (p-value > 0.1) [0.0355 (p-value > 0.1)], 0.0784 (p-value > 0.1) [0.0525 (p-value > 0.1)], and −0.653 (p-value < 0.01) [−0.0689 (p-value < 0.01)], respectively. These results show that the larger-sized engagement auditors are less likely to commit Type II errors in GCOs issued to banks but are not statistically-significant associated with issuance of GCOs and Type I errors in GCOs. The result (that the larger-sized engagement auditor commits a lower rate of Type II errors in GCOs issued to banks without reducing the issuance of GCOs) indicates that the larger-sized engagement auditors are more conservative in issuing GCO. For the sake of brevity, we omit the analyses of the statistically significant coefficients for the control variables in the regressions reported in Table 6, which are the same as the analyses of the control variables in the regressions reported in Table 4 and Table 5, which were articulated in Section 5.2. The results of Table 6 also indicate that the size of the engagement auditors captures an aspect of audit quality that differs from that of industry expertise of the engagement auditors.

Using the sample of 4492 bank-year observations (414 distinct banks) over the period of 2002–2023, Table 6 reports the results of the regressions of GCO (in columns 1–2), TypeI (in columns 3–4), and TypeII (in columns 5–6) on the proxies for the engagement auditor’s industry size at the national level (Audsize_N, in columns 1, 3, and 5) and state level (Audsize_S, in columns 2, 4, and 6), controlling for the factors that may impact the issuance of GCOs and the incidence of Type I and Type II errors in GCOs issued to banks.

The results reported in Table 4, Table 5 and Table 6 collectively suggest that the audit industry expert at the state level is more knowledgeable on the clients’ going concern and business operations, thereby, issuing a more accurate issuance of GCOs than the counterparts, while the audit industry expert at the national level and large-sized auditors only show conservatism in the issuance of GCOs.

5.3.2. Additional Control Variables

The amount of deposits to the banks might be a factor of influencing the chance of closing the bank (e.g., S. Liu, 2024) because the deposits in the bank could be considered as a liability for the bank to pay the depositor back. Therefore, this factor could impact the chance of the issuance of a going concern opinion. This is similar to the effect of the variable Leverage on the GCO outcomes. To control for the possible impact of deposits on the going concern opinion, we redo our multivariate analyses on the regression of GCO/TypeI/TypeII on the quality of engagement auditor, controlling for the ratio of deposit to total assets (coded as “DepositAsset”) in addition to the control variables (with and without the variable Leverage) reported in the Table 4, Table 5 and Table 66. The results obtained when controlling for DepositAsset are qualitatively the same as those reported in Table 4, Table 5 and Table 6.

5.3.3. Extension of Duration Between the Date of Issuance of a GCO and Bank Closure

To assess the sensitivity to GCO issuance and closure duration, we redo analyses in Table 2, Table 3, Table 4, Table 5 and Table 6, redefining the variable “TypeI” as false positive GCOs for banks not failing within 18 months and the variable “TypeII” as false negative GCOs for banks failing but not receiving GCOs 18 months prior. The results remain qualitatively similar to those reported in Table 2, Table 3, Table 4, Table 5 and Table 6.

5.3.4. Alternative Regression Method Analyses7

To assess the sensitivity of our results to different econometrical methods, we redo our regression analyses without applying PSM on the regressions of outcome variables (i.e., GCO, TypeI, and TypeII) on the tested variables (i.e., Exp_MKT_N, Exp_MKT_S, Exp_NumInd_N, Exp_NumInd_S, Audsize_N, Audsize_S). The results of using the logistic regression of the outcome variables on the tested variable are qualitatively similar to those reported in Table 4 and Table 5, when controlling for other factors that may impact on the outcomes and varying the control variables, as discussed in Section 5.3.2, and the duration of the date of issuance of a GCO and bank closure, which was analyzed in Section 5.3.3. The results of using alternative regression methods show that the audit industry expert at the state level is more likely to issue GCOs and is associated with the lowest rate of Type I and Type II errors for GCOs issued to banks compared to the counterparts.

6. Discussion

We use the propensity score matching (PSM) method to reduce the possible confounding effect caused by the common characteristics of the client firms (banks) when we examine the effect of the audit industry expert on the GCO outcomes. It is possible there would be an endogenous issue in the regression model (i.e., Equation (1)) because banks (with certain characteristics) might intendedly choose the audit industry expert who subsequently correctly decides to issue a GCO. This could create an endogenous issue in our regression results. Therefore, we first use the PSM method to match the banks by the characteristics driving the banks to select an audit industry expert. We then run the regression. Doing so would reduce the estimation bias caused by the confounding factors (i.e., the characteristics of the banks in our setting), which might otherwise alter the results of the tested effect of the audit industry expert on the GCO outcomes.

Using PSM could benefit our empirical results in (but not limited to) the following ways: (1) reducing estimation bias caused by certain characteristics of the banks, (2) reducing the possible sample selection bias that arises from a non-randomized procedure, and (3) making estimations of the causal impact of the effect of the audit industry expert on the GCO outcomes possible. While benefiting from PSM, this method has some weaknesses such as the following: (1) possibly omitted confounding factor(s) that are not controlled for in the PSM estimation--then, the results might be still biased; and (2) because we applied the PSM method on the banking sector, not other industries, our results might not be generalizable to other industries. Given the possible drawbacks of the PSM method, although our results provide a causal inference about the effect of audit industry experts on the GCO outcomes, we suggest that the readers interpret our results with caution.

7. Conclusions

This study examines the effect of the quality of the engagement auditor on the issuance of going concern opinions (GCOs) and Type I and Type II errors in the GCO issued to banks. We empirically measure the quality of the engagement auditor using proxies for the auditor’s industry expertise at the national and state levels in our main tests and using proxies for the engagement auditor’s size at the national and state level in our additional tests.

Using the sample of 4492 bank-year observations (414 distinct banks), the univariate tests show that the majority of proxies for the quality of the engagement auditors (except the proxy for the audit industry expert at the national level based on the auditor’s market share) are statistically significant and negatively correlated with Type II error in GCOs, while the auditor’s size at the national level is statistically significant and negatively correlated with the issuance of GCOs. The 2 × 2 contingency table shows a rate of 65.2% (89.7%) for Type I (II) errors in GCOs issued to banks for the entire sample. Controlling for the factors that may impact the GCO outcomes, the results of multivariate empirical tests confirm those of the univariate tests. In particular, results of regressions show that the issuance of GCOs and Type I and Type II errors in GCOs issued to banks decrease with the presence of an audit industry expert at the state level, while only Type II errors in GCOs decrease with the use of an audit industry expert at the national level (constructed based on the auditor’s client portfolio) and increase with the engagement auditor’s size at the national and state levels.

The results of this study collectively indicate that the audit industry expert at the state level could improve GCO outcomes for banks (i.e., issuance of GCOs and accuracy of GCOs), while the audit industry expert at the national level and large-sized engagement auditors behave conservatively in the task of expressing the going concern opinion. Our results suggest that (i) the audit industry expert at the state level knows the clients (i.e., banks) better than the counterparts, (ii) the audit has value for the investors and banks, (iii) hiring an audit industry expert at the state level might benefit the banks, and (iv) the relevant regulatory bodies (e.g., bank regulators, PCAOB) might need set up the regulations to improve the GCO outcomes, potentially reducing the bank’s chance of being closed by providing early warning signal(s). Thus, our results from this study not only provide useful inference to the investors, practitioners, and regulators, but also add novel empirical evidence to the literature by showing the effect of an auditor’s industry expertise on a bank’s closure prediction.