Reputational Implications for Partners After a Major Audit Failure: Evidence from China

In a major upside, focusing on a large audit firm generates power for the analysis since the negligence of a single partner threatens each individual partner’s reputation with the risk magnified in larger audit firms that have more partners signing audit reports (Lennox and Li 2012). In other words, a partner’s worries about the degree of care exercised by other partners increases with their number; i.e., partners’ cross-monitoring incentives are stronger in larger audit firms.

It is important to note that litigation against auditors in Hong Kong is an exception (Ferguson and Majid 2003).

El Ghoul et al. (2012) provide equity pricing evidence that subjecting auditors to intense litigation exposure is responsible for the major fault line that distinguishes the US from the rest of the world on the performance gap between Big 4 and non-Big 4 auditors. In fact, prior evidence casts doubt on whether researchers can reliably gage the separate impact of auditors’ incentives with event studies on the collapse of a large US public accounting firm. For example, Menon and Williams (1994) report evidence implying that the negative share price reaction that clients of the public accounting firm Laventhol and Horwath suffered after its bankruptcy in 1991 reflects the insurance explanation, although Baber et al. (1995) find that both the reputation and insurance explanations were responsible. More recently, Chaney and Philipich (2002) attribute the negative stock returns incurred by Arthur Andersen’s clients surrounding when investors learned about its apparent complicity in the Enron scandal to reputational implications. However, Nelson et al. (2008) provide evidence that calls this interpretation into question.

Moreover, China does not permit class action lawsuits that can have a sobering impact on auditors’ incentives to constrain firms against manipulating their financial statements (Ball 2009; Mahoney 2009). As far as we know, there have been no successful civil lawsuits against auditors of listed firms in China. More generally, minority investors in modern China have hardly had any legal recourse against tunneling by insiders, who are frequently politically connected, and security regulators have minimal jurisdiction over controlling entities (e.g., Allen et al. 2005; Fan et al. 2007; Jia et al. 2009; Jiang et al. 2010; Chen et al. 2014).

Studies like ours are frequently labeled “natural experiments” in that we analyze outcome measures for observations in treatment and control groups that have not been randomly assigned. We explain in more detail later in the paper that our treatment group comprised ZTQ partners who were not implicated in the YGX audit failure and our control group comprised partners in other large audit firms. We compare differences in outcomes before and after the ZTQ–YGX events in 2001 between the groups to identify the importance of partners’ incentives to protect their reputations by monitoring their fellow partners. Major advantages of implementing a difference-in-differences design include its simplicity for narrowing the range of plausible competing explanations for the results along with its potential to bypass many of the endogeneity complications that routinely arise when making comparisons between heterogeneous individuals (e.g., Bertrand et al. 2004). Although we assume that the reputations of non-ZTQ partners in the control group are immune to the YGX audit failure, any damage to their reputations (contagion effects) would work against our tests rejecting the null hypotheses.

Guedhami and Pittman (2006) find evidence implying that investor perceptions of audit quality rise when countries impose tough criminal enforcement against auditors. In China, discipline from sanctions leveled by regulators plays a role in the country evident in the two ZTQ partners having their licenses revoked. However, we isolate the reputational impact on the former ZTQ partners who were not implicated in the scandal according to the government; i.e., we specifically exclude from our analysis the two ZTQ partners who were later prohibited from auditing listed companies.

For expositional convenience, we label partners not directly involved in the YGX audit failure as “innocent” or “guiltless” since the country’s regulator did not impose any sanctions on them. However, it is important to stress that under a broader interpretation these partners may shoulder some blame for failing, for example, to closely monitor the two partners implicated in the scandal and to develop rigorous quality control systems (e.g., training programs, policies designed to attract and retain capable personnel, etc.) that are intended to ensure that all partners uniformly supply high-quality audits.

Regulatory events prevent us from extending our sample period beyond the 1996–2009 timeframe. In December 1995, the CICPA issued the first batch of China’s Independent Auditing Standards of the Certified Public Accountants (CIAS’s), which specify the responsibilities of auditors as well as the content and format of audit reports. DeFond et al. (2000) find that auditor independence improved after 1995. The National Development and Reform Commission and the Ministry of Finance constituted “The measures for the administration of Certified Public Accountants service charges” on 27 January 2010. It formulates guidelines on audit fees. In addition, the Ministry of Finance and the State Administration for Industry and Commerce issued “Interim provisions on promoting large scale accounting firms adopt the special general partnership” on 21 July 2010. Under the special general partnership, a complicit partner or partners should bear unlimited liability or unlimited liability jointly and severally if the liability is incurred due to intentional or gross negligence, while the other partners should bear limited liabilities up to their share in the partnership. Under the interim provisions, audit partners not only share reputation, but also the liabilities stemming from audit failure. We narrow our focus to the 1996–2009 period to avoid providing spurious results on the predictions arising from regime shifts distorting our evidence. In addition, it follows that the reputational impact will subside as the scandal becomes more distant, an issue that we discuss in more detail later in the paper. To show how our findings provide implications for future practice and make further contributions to extant research, we extend the sample period to 2013 and add a dummy variable indicating whether an observation was before or after 2009 to the analysis. The results of the perceived audit quality tests show that reputational effects decreased after 2009. However, the results for the audit fee analysis are mixed. It is important to exercise caution in interpreting these results given the regime shift after 2009.

Also, to avoid biasing the analysis toward supporting our predictions, we exclude data for 2001 because it routinely took the former ZTQ partners some time to find a new employer. Moreover, even for those that managed to quickly join another audit firm, they experienced delays in becoming signing partners given the continuous nature of auditing. In China, the fiscal year end is 31 December while audit reports must be completed by 30 April, leaving the former ZTQ partners a short window to become signing partners on 2001 audit engagements for listed clients at their new firm. In fact, only two former ZTQ partners became signing partners in 2001. All of the major events in the ZTQ–YGX audit failure from the initial fraud allegations to the eventual punishment leveled against the audit firm and two of its partners occurred in 2001, ensuring that the surrounding years under study were not contaminated.

In untabulated analysis, all of our core results are materially insensitive to including “Tongren” CPA firm in our sample. Moreover, our evidence persists when we exclude “DH” CPA firm, which had the second most listed clients in 2000 and merged with Ernst and Young in early 2002. Chen et al. (2010) report that many of DH’s clients switched to other audit firms during the 2002–2004 timeframe. Finally, the evidence on our predictions remains when we specify the rest of the Top 8 audit firms as the matching sample after Chen et al. (2011).

This 38 equals the 57 clients of ZTQ in 2000 minus the 6 listed clients which did not disclose their engagement partners’ names, 4 listed companies which did not disclose the names of their engagement partners and the branch name of their audit firm, and 9 listed firms audited by the 2 ZTQ partners whose practicing licenses were later canceled.

Our matching sample is the original partners of the rest of the Top 10 audit firms in 2000. Naturally, the sample size decreases over time for several reasons: some of them no longer audit listed firms, although they still work at audit firms that have the license to audit listed firms; some moved to audit firms that do not have the license to audit listed firms (very small audit firms); and some moved to other industries.

Besides the univariate tests, we also conduct regression analysis. In the first regression, the dependent variable is ZTQ_client, which equals 1 if the listed firm is audited by former ZTQ auditors, and 0 otherwise. In the second regression, the dependent variable is ZT, which equals 1 if the listed firm is audited by former ZT auditors, and 0 if the listed firm is audited by former TQ auditors. In both regressions, we find that the coefficients on POST are negative and statistically significant at the 1 % level, lending additional support to our predictions.

Moreover, additional data inspection reveals that none of the former ZTQ partners and partners from the matching sample became fund managers or investment bankers after they left the audit industry.

This analysis involves 13 years of data with repeated annual observations for each company. However, after applying firm-level clustering to address the lack of independence, the results in Columns (3)–(6) hold at the 1 % level.

All of the evidence reported in Panel B persists at the 1 % level when we narrow the period under study to the 10 years surrounding the ZTQ–YGX events, 1996–2006: in order to reflect that the reputational impact may be concentrated nearer the audit failure—or restrict the matching sample to audit partners that have at least five clients to address potential mean reversion in the number of ZTQ partners per client after the YGX scandal. Similarly, our results are also robust to excluding existing clients (i.e., clients that were with ZTQ before 2001) from the analysis to sharpen the focus to strictly new clients.

The results in Columns (3)–(6) remain statistically significant at the 1 % level when we apply firm-level clustering.

DeFond et al. (2000) and Wang et al. (2008) identify large audit firms according to whether they are among the 10 largest auditors in China based on clients’ assets. Reinforcing their auditor choice specification, Gul et al. (2010) find that Big 4 audit firms increase the amount of firm-specific information impounded into share prices of listed firms in China. More generally, prior research routinely gages audit quality with the presence of a Big 4 audit firm (e.g., Mansi et al. 2004; Kim et al. 2011).

The evidence on the predictions in H₂ in Panel A of Table 2 and H₅ in Panel B of Table 2 remains at the 1 and 10 % levels, respectively, when we shorten the timeframe to the first 5 years, 2002–2006, after the ZTQ–YGX events.

We also evaluate whether this evidence persists in a regression framework. In this analysis, we specify the dependent variable as BIG_audit, which equals 1 if an auditor is hired by a “large” audit firm, and 0 otherwise. We rank the audit firms based on the number of listed clients and then classify the Top 10 audit firms or Big 4 as “large” audit firms and other audit firms as “small” audit firms. ZTQ is a dummy variable, which equals 1 if the auditor is a former ZTQ auditor, and 0 otherwise. ZT is a dummy variable, which equals 1 if the auditor is a former ZT auditor, and 0 if the auditor is a former TQ auditor. After He and Ke (2014), we include four control variables in the model to capture individual auditor characteristics: (i) Experience is the natural logarithm of the number of years since the first year when an auditor served as a signing auditor for a publicly listed firm, (ii) Modified is the mean fraction of an auditor’s clients who were issued a non-clean audit opinion over the past 5 years t − 5 to t – 1, (iii) Enforcement is the mean fraction of an auditor’s clients that violated financial reporting and disclosure regulations over the past 5 years t − 5 to t – 1, and (iv) Specialist is the mean value of SPEC over the past 5 years t – 5 to t – 1. SPEC is a dummy variable indicating auditor specialization in one or more economically important industry sectors. An industry sector is considered economically important if it represents at least 1 % of total assets of all Chinese listed companies. An auditor is designated as an industry specialist if the size of her within-industry clientele in terms of audited total assets belongs to the highest decile of its annual distribution (Knechel et al. 2013). Reinforcing our other evidence consistent with the prediction in H₂, we find in successive regressions that the coefficients on ZTQ and ZT are negative and statistically significant at the 1 % level.

The CSRC began to require listed firms to disclose their audit fees in 2001.

In unreported sensitivity analysis, we control for two other variables in the audit fee model: total receivables and inventory, each scaled by total assets. All of our core results hold in these regressions with adjusted R ² increasing trivially to 0.468. Also, we cannot control for the number of subsidiaries and the proportion of subsidiaries that are foreign because of poor data availability (in another study focusing on Chinese firms, Wang et al. 2008 also do not control for these potential determinants). Although the explanatory power of our audit fee model is slightly lower than in prior research on US firms, it exceeds that observed in recent evidence on Chinese firms. For example, the Adj. R ² in Panel C of Table 3 is 0.463, which is higher than the 0.377 that Wang et al. (2008) report.

The sample size of the matching sample is much larger than the ZTQ sample. As a sensitivity test, we specify the rest of the Top 3 audit firms in 2000 as the matching sample (ZTQ was ranked first in terms of the number of listed clients among all audit firms in 2000). The sample size of matching sample is reduced. In both the audit fee and perceived audit quality tests, our core evidence persists.

Given that it is conceivable that the reputational fallout that partners experience will subside as the scandal becomes more distant, we run additional tests to examine this issue. First, for the audit fee analysis, we include a dummy variable POST in the model, which equals 1 if it is in the period 2002–2004, and 0 otherwise. In untabulated results, we find that the coefficient on ZTQ * POST is statistically indistinguishable from zero. Second, for the perceived audit quality analysis, we include two dummy variables, POST₁ and POST₂, in the model. POST₁ is a dummy variable that equals 1 if it is in the period 2002–2004, and 0 otherwise. POST₂ is a dummy variable that equals 1 if it is in the period 2005–2009, and 0 otherwise. In these estimations, we find that the coefficient on UE * POST₁ fails to load, while the coefficient on UE * POST₂ is statistically significant. However, the difference between these two coefficients is not statistically significant. Consequently, there is no clear evidence implying that the reputational impacts dissipate over time during the 8-year period under study. We find similar evidence when we specify 2005 or 2006 as alternative cutoff points for identifying the post periods. Overall, we only find some suggestive supporting that the partner reputational impacts gradually fall over the 2002–2009 timeframe.

After re-estimating the tests on the ZT and TQ subsamples, we find virtually identical evidence supporting our predictions.

We continue to find almost identical evidence on our predictions when we follow Nelson et al. (2008) by removing from all regression observations with studentized residuals exceeding 2 in absolute value, and when we control for the extent of marketization with an index that reflects the depth of advances in market-oriented institutional transformation and economic development in the province (Chen et al. 2011).

Our core inferences in this section hold when we interact each control variable with UE. However, including these interactions engenders serious multicollinearity problems in the small sample. More specifically, we find that the variance inflation factor (VIF) for each regression exceeds 40 after including the interaction terms. In contrast, the VIFs are the vicinity of 3.5 in the specifications without the interactions between the control variables and UE. Accordingly, we follow Nelson et al. (2008) by excluding these interactions from the estimations.

We find that the ERCs of the matching sample decline in the post period. One possible explanation is that the ERCs of all of the firms subside in the post period (from the period 2002 to 2009) since listed firms were required to disclose quarterly reports after 2002. We did not calculate unexpected earnings based on the quarterly ROAs because we did not have access to quarterly earnings data before 2002.

Recent research applies SUEST, which was originally proposed by Zellner (1962) as an econometric technique for testing cross-model hypotheses; e.g., Hayes et al. (2012) and Price et al. (2011). In an important upside, this approach does not assume that the two regression models share the same residual.

We also re-estimate the regressions on the full sample after adding the dummy variable ZTQ (or ZT). In results not reported in tables, we find qualitatively similar evidence.