The Impact of Auditor Switch on the Association between Litigation Risk and Audit Quality

Transcription

1 The Impact of Auditor Switch on the Association between Litigation Risk and Audit Quality Presented by Dr Szu-fan Chen Assistant Professor Hong Kong University of Science and Technology #2017/18-06 The views and opinions expressed in this working paper are those of the author(s) and not necessarily those of the School of Accountancy, Singapore Management University.

2 The Impact of Auditor Switch on the Association between Litigation Risk and Audit Quality * Kevin C W Chen The Hong Kong University of Science and Technology acchen@ust.hk Szu-fan Chen The Hong Kong University of Science and Technology sf.chen@ust.hk May 19, 2017 * We thank workshop participants at the Hong Kong University of Science and Technology, the National Taiwan University, Xiamen University, and Tsinghua University for their helpful comments.

3 The Impact of Auditor Switch on the Association between Litigation Risk and Audit Quality Abstract We show that the relationship between litigation risk and audit quality is conditional on whether there is an auditor switch upon an increase in litigation risk. Auditors who took over a client before its imminent litigation allow more income-increasing accruals, have shorter audit lags, and issue fewer going concern opinions than those who stay with risky clients. Moreover, the fees of new auditors reflect a risk premium as these fees are positively related to ex post auditor litigation. In contrast, auditors who stay with risky clients charge higher audit fees and reduce non-audit services, and the audit fees reflect effort as they are negatively related to auditor litigation. In summary, we find that auditor switch leads to different relation between litigation risk and audit quality. 1

4 1. Introduction While theories predict that litigation should motivate auditors to provide higherquality audits, the empirical evidence is mixed. Some studies show a positive relationship between litigation risk and audit quality. For example, Geiger et al. (2006) show that when legislation reduces auditors exposure to litigation, they issue fewer going-concern modified opinions for clients entering bankruptcy. In contrast, some other studies document various strategies that auditors can exploit to mitigate litigation risk may lead to lower audit quality. Specifically, Shu (2000) shows that heightened litigation risk prompts auditor resignation and that the successor auditors are smaller than the predecessors. This result is consistent with the prediction of deep pocket theory in that larger audit firms are more vulnerable to litigation. To the extent that large auditors provide better quality audits than small auditors, Shu (2000) s results suggest that increasing litigation risk reduces audit quality. Simunic and Stein (1996) also argue that increasing litigation risk may lead clients to shift their demand from high quality auditors to low quality auditors. Thus, how litigation risk affects audit quality remains an unresolved issue. 1 We address this question by examining audit firms reaction when their clients face a real increase in litigation risk, i.e., when their clients are sued for providing misleading financial information. If high quality auditors tend to prefer low-risk clients as predicted by deep pocket theory, an increase in litigation risk would lead to a switch from high quality auditors to low quality auditors, i.e., a negative relation between litigation risk and audit quality. On the contrary, if high quality auditors are willing to continue their engagements 1 The conclusions from extant research are further muddled by the fact that the measures of audit quality used in previous studies, such as accrual quality and audit fees, are often subject to alternative explanations. As DeFond and Zhang (2014) discuss, prior research uses variables such as audit fees, going-concern opinions (GCs), and discretionary accruals (DAC) to measure audit quality. The association between litigation risk and GCs as well as DAC may be due to auditor s excessive conservatism, which might not enhance audit quality. The association between litigation risk and audit fees may reflect premium rather than efforts (e.g, Seetharaman et al., 2002; Johnstone and Bedard, 2004; Badertscher et al., 2014). 1

5 with risky clients even in the face of increased litigation risk, we would observe a positive relation between litigation risk and audit quality. High quality auditors may be willing to do so since high quality audits can prevent lawsuits against auditors (e.g., Fuerman, 1997; Kedia et al., 2014). Thus, we first examine the impact of litigation risk on the type of auditors who sever or who do not sever their engagements with risky clients. Our cross-sectional analyses show that, when there is an increase in litigation risk, large auditors are more likely to stay with risky clients than small auditors. Nonetheless, after controlling for audit firm size, our measures of auditor quality are not related to auditor switch during the period of heightened litigation risk. These results suggest that, even with high litigation costs, many high quality auditors still find it profitable to retain risky clients. The willingness of high quality auditors to retain risky clients allows for the possibility that litigation risk would increase audit quality. We next examine how the impact of litigation risk on audit quality differs by whether or not there is an auditor switch. Prior studies suggest that there may be two reasons why litigation risk would reduce audit quality following an auditor switch. The first reason is based on the pecking order of auditors responses to client risk, which predicts that auditor resignation occurs when the litigation costs have grown beyond the incumbent auditor s acceptable level (e.g., Elder et al., 2009; Krishnan et al., 2013). This pecking order implies that only auditors with lower quality than the predecessors would find the engagement profitable as they have lower litigation costs than their predecessors. The second reason why high quality auditors may be less willing to accept risky clients than low quality auditors is that audit efforts cannot effectively increase audit quality in the first year of an engagement (Bell et al., 2015). These reasons lead to our prediction that, when increasing litigation risk prompts auditor switch, the subsequent audit quality would decline. In contrast, we have a different prediction of the impact of litigation risk on audit quality when there is no auditor 2

6 switch. Auditors staying with risky clients can manage litigation risk by increasing efforts, asking for more risk premiums, or doing both. High quality auditors may exert greater efforts to the extent that their cost of litigation is greater than their cost of efforts, whereas low quality auditors may increase premiums as they are incapable of increasing efforts to mitigate litigation risk. As our first set of empirical tests show that increasing litigation risk does not expel high quality auditors, we expect that auditors staying with risky clients would respond to the increasing litigation risk by increasing audit efforts. To test how auditor switch affects the relation between litigation risk and audit quality, we first obtain a list of public firms involved in securities class action lawsuits for accounting manipulations and then identify if there is an auditor switch during the fiscal year right before these firms filed the first misstated annual reports with the Securities and Exchange Commission (SEC). We use the class action period to identify the fiscal years in which the annual reports were misstated, and define the start date of the class action period as our event date. Once auditors sign off on the first misstated annual reports, they would not be able to escape litigation by resignation. This process presumes that auditors can become aware of managers push for more aggressive accounting treatments during the process of audits before the misstated annual reports are filed. We use several proxies for the quality of audit output: income-increasing accruals, audit lag (i.e., the number of days between the end of fiscal year end and the auditor sign-off date), and the frequency of the issuance of going concern opinions. As audit fees can reflect efforts and/or premiums, we do not directly use audit fees as a proxy for audit quality. To distinguish between audit effort and risk premium, we follow prior studies (e.g., Kaplan and Williams, 2013) by using a simultaneous equation approach to examine the relation between audit fees and ex post litigation risk against auditors. A positive relation would suggest that audit fees reflect premiums, whereas a negative relation would suggest that audit fees reflect efforts. 3

7 We find evidence consistent with our prediction that the impact of litigation risk on audit quality is conditional on whether or not there is an auditor switch. Specifically, we find that auditors staying with risky clients ( stay sample ) allow less income-increasing accruals, increase the audit lag, and issue more going concern opinions in the post-event period than in the pre-event period. We find the opposite results for the sample in which there is an auditor switch ( switch sample ). We also find that auditors in the stay sample adjust non-audit services downward after the event date so as to increase perceived independence, and their post-event audit fees are negatively associated with ex post litigation risk, i.e., the increasing fees reflect greater efforts. In contrast, auditors in the switch sample do not change perceived independence, and their post-event audit fees are positively associated with ex post litigation risk, i.e., the fees reflect risk premiums. Taken together, our results suggest that auditors client portfolio management has a significant impact on how litigation risk affects audit quality. Our study contributes to the literature in the following ways. First, we add to the literature of the relation between auditor tenure and audit quality. Finding a positive relation between auditor tenure and audit quality, prior studies suggest that mandatory auditor rotation may hurt audit quality (e.g., Ghosh and Moon, 2005; Gul et al., 2007; Bell et al., 2015). Our study complements these studies by showing that auditor switch leads to a negative association between litigation risk and audit quality because only low quality auditors would accept risky clients. An important implication of this finding is that a regulation that increases the frequency of auditor switch may reduce the overall audit quality for risky clients, who are the very companies in need of high quality audits. Second, we add to the literature of the relation between litigation risk and audit quality. We show that, as auditors can effectively prevent litigation by increasing efforts, litigation risk does not always prompt high quality auditors to shed risky clients. High quality auditors willingness to stay with risky clients in 4

8 turn allows a positive association between litigation risk and audit quality. This finding helps resolve the debate on the empirical relation between litigation risk and audit quality (DeFond and Zhang, 2014). Last, we show that the interpretation of audit fees can be complicated by auditor switch, a risk management strategy not yet accounted for in studies on litigation risk and audit fees (e.g., Seetharaman et al., 2002; Geiger et al., 2006; Badertscher et al., 2014). Specifically, we find that audit fees reflect a premium that is positively related to ex post litigation risk against auditors when there is an auditor switch; absent an auditor switch, audit fees reflect efforts that are negatively related to ex post litigation risk. Our findings should be useful for future research on the distinction between the premium and the effort components of audit fees. Section 2 of the paper reviews previous research and develop hypotheses. Section 3 explains the sampling process and provides descriptive statistics. Section 4 reports the results of empirical analysis. Section 5 describes sensitivity analysis. Section 6 concludes. 2. Background and Hypotheses 2.1.The determinants of auditor switch in response to increased litigation risk This paper posits that the association between litigation risk and audit quality is conditional on whether an audit firm continues the engagement or is replaced by another firm. Thus, the starting point is to understand the type of auditors who are likely to continue or discontinue. Auditors assess the client risk and determine whether or not to retain the client every year. When there is an increasing likelihood of financial misstatements, auditors can refuse to sign the financial statements and resign. Once auditors sign off on the first misleading financial statements, they risk being implicated in the accounting manipulation. The release of the first misleading financial information marks the start of the class action period in a private securities class action lawsuit. For example, in the securities class action lawsuit against Countrywide Financial Corporation in 2007, investors alleged that the 5

9 company started manipulating earnings at the end of Countrywide Financial announced that it would switch auditors from Grant Thornton to KPMG on January 1, 2004, but Grant Thornton still signed off on the company s 2003 annual report. This annual report was filed on March 12, 2004, and investors alleged it as the first misleading statement Countrywide Financial provided to the public. Thus, the class action period started from March 12, Both Grant Thornton and KPMG were named as defendants in the securities class action lawsuit. Grant Thornton could have avoided the lawsuit altogether had it refused to sign the 2003 annual report. Being able to gather information about the changes in a client s litigation risk and respond quickly is therefore very important to auditors. While there is a rich literature examining the determinants of auditor switch, the focus of most studies is to explain the dichotomous outcomes of resignation (auditor-initiated switch) vis-à-vis dismissal (client-initiated switch). Bockus and Gigler (1998) s theoretical model predicts that increasing litigation risk leads to auditor resignation for larger audit firms due to the deep pocket concern since they are more vulnerable to legal liability. This prediction is supported by empirical evidence that auditors are more likely to resign from clients with high litigation risk (e.g., Krishnan and Krishnan, 1997; Shu, 2000). Shu (2000) further shows that resigned auditors are more likely than those dismissed to be succeeded by small auditors (neither Big N nor national firms), and this tendency is stronger when there is a larger increase in litigation risk. The prior research, however, is largely silent on why a large proportion of auditors would continue engagements when their clients face an imminent litigation risk. For example, although bankruptcy risk is positively associated with litigation risk, Schwartz and Menon (1985) find that only 35 out of 132 firms that eventually went bankrupt replaced their auditors in the four years before bankruptcy. Another example event associated with high litigation risk is financial restatements. Huang and Scholz show that only 19 percent 6

10 of the companies that restated financial reports replace auditors during the five-quarter period around the restatement announcement. As auditors involved in these example events are likely able to detect the increasing litigation risk before the exposure of client misconducts (Lyon and Maher, 2005), the majority of them must find that the profits from the engagement outweigh the costs of litigation. This can be due to either their low quality with low litigation cost or their high quality with the cability to mitigate litigation risk by increasing audit quality. Since only the second scenario can lead to a positive relation between litigation risk and audit quality as documented in prior studies (e.g., Seetharaman et al., 2002; Venkataraman et al., 2008), we discuss why high quality auditors may retain risky clients below. DeFond and Zhang (2014) note two countervailing effects of litigation risk on high quality auditors portfolio management decisions. On one hand, high litigation costs could motivate high quality auditors to avoid risky clients to a greater extent than low quality auditors; on the other hand, high quality auditors are more able to mitigate litigation risk by increasing audit quality. 2 Moreover, high quality auditors tend to be large. Larger auditors have a greater capacity to diversify client risk than smaller auditors. They also have a greater bargaining power than smaller auditors, which allows them to raise audit fees to postpone resignation according to the pecking order of auditor responses to client risk (e.g., Elder et al., 2009; Krishnan et al., 2013). Thus, high quality or large auditors may not strictly prefer less risky clients. 2 Anecdotal evidence appears to support the notion that auditors providing higher audit quality are less likely to be sued. Specifically, auditors are sued much less often than their clients. Fuerman (1997) finds that only about 18% of the private securities lawsuits related to financial disclosure from 1992 to 1997 also named the auditor as defendant and Kedia et al. (2014) find that the number is 27% from 1996 to Despite the low litigation risk, our untabulated results suggest that, in our sample, auditors are added as a defendant at a much later stage in the litigation than their clients. Moreover, lawsuits against auditors are much less likely to be dismissed by the court than those against their clients. These findings suggest that, in the majority of the cases, auditors are sued only when there is strong evidence of auditor negligence. 7

11 Since the decision of continuing an engagement is jointly made by auditors and their clients, we also note that there are similar countervailing effects of litigation risk on clients demand for high quality or large auditors. Clients intended to manipulate financial reports may be inclined to switch from high quality to low quality auditors (or from large to small auditors) to avoid exposing their misconducts. Such auditor switch, however, could lead to a negative market reaction and defeat the purpose of accounting manipulation. Therefore, regardless of whether or not it is the auditor or the client that initiates the auditor switch, it is an empirical question how auditor switch would be related to auditor size and quality. This leads to our first set hypotheses, both stated in null forms as follows. H1a. Large and small audit firms are equally likely to continue the engagement when their clients face an increase in litigation risk. H1b. High and low quality auditors are equally likely to continue the engagement when their clients face an increase in litigation risk. Although high quality auditors tend to be large, audit firm size does not necessarily capture audit quality due to the self-selection problem: large auditors are more able to be selective in accepting clients than small firms. Moreover, large high quality auditors are more able to have a diversified client portfolio and to ask for audit fee increases than small high quality auditors. Thus, we separately examine auditor size and auditor quality, using industry specialization and an office-level quality measure as our proxies for auditor quality. Prior studies suggest that auditors industry expertise captures auditor quality above and beyond auditor firm size as it is negatively related to restatement likelihood even after controlling for the impact of auditor firm size (Chin and Chi, 2009). We also follow Aobdia et al. (2015) by constructing a quality measure at the office level that is independent of auditor firm size. These proxies for auditor quality are time-invariant because the main focus of our first set of 8

12 hypotheses is on the type of auditors rather than on the quality of their audit services that can vary over time. 2.2.Auditor switch and the change in audit quality Our first set of hypotheses predicts that litigation risk does not always induce a systematic auditor-client realignment from high to low or from low to high quality auditors. The subsequent change in the quality of audits, therefore, would depend on how successor auditors behave relative to their predecessor in the case of auditor switch and how incumbent auditors change their behavior after the increase in litigation risk in the case of no auditor switch. We first discuss the case of auditor switch and then the case absent auditor switch. We presume that auditors can detect the increasing likelihood of accounting manipulation during their audit. The increasing misstatement risk would trigger auditor switch if the risk goes beyond the incumbent auditors level of risk tolerance. 3 Because small auditors are subject to lower litigation costs, both Raghunandan and Rama (1999) and Shu (2000) find that they are more likely to succeed a large auditor who just resign from the engagement. Other than size, we postulate that the quality of auditors also affects their client acceptance decisions after an auditor switch. As auditor switch sends a negative signal to the market about a client s risk (Nichols and Smith, 1983), the client may be only able to engage an auditor with lower quality than its predecessor, leading to a reduction in audit quality. This leads to our next hypothesis, stated in alternative form as follows. H2a. When litigation risk prompts auditor switches, the successor auditors would provide lower quality audits than their predecessors. 3 If the auditor disagrees with the management and refuses to sign off on the financial statements, there may be an auditor dismissal instead of auditor resignation. Alternatively, the client may try to phrase the auditor resignation as dismissal. Regardless of whether it is resignation or dismissal, our story holds as long as it is the increasing fraud risk that prompts auditor switch. 9

13 Litigation risk would not trigger auditor switch if auditors are unaware of the increasing risk or if the cost of terminating the client engagement is higher than the cost of managing the client risk using other strategies. If auditors retaining risky clients are on average unable to detect an increase in client risk, we would not observe any change in auditor behavior after the issuance of misleading financial statements. In contrast, if they are aware of the increasing misstatement risk, then they should be able to increase audit quality, charge a risk premium, or do both as it would be too late to quit once they sign the misleading statements. Since charging a premium cannot deter litigation and the subsequent reputation damage, we expect these auditors to increase audit quality to mitigate litigation risk to the extent that they are competent enough to do so. This leads to the following hypothesis (in alternative form): H2b. After an increase in litigation risk, auditors who continue engagements with risky clients would increase audit quality. We use three measures of audit quality: signed discretionary accruals, audit lag, and going concern opinions. Extant studies (e.g., Heninger, 2001) suggest that auditor litigation risk is positively associated with signed discretionary accruals, i.e., income-increasing accruals. Moreover, restatements of overstated earnings are more frequent and are perceived as more serious than those of understated earnings (Palmrose et al., 2004). Hence, we expect that auditors suspecting an increasing likelihood of misstatement should reduce incomeincreasing accruals. An increase in income-increasing accruals would therefore suggest deteriorating audit quality. Second, we use audit lag as a proxy for audit investments following Ghosh and Tang (2015). Bell et al. (2015) suggest that auditors may invest more time and efforts in the first years of audits than in subsequent years. Finding a shorter audit 10

14 lag after auditor switch would therefore indicate declining audit quality 4. Third, we use the frequency of issuing going concern opinions as a proxy for audit quality. Although firms manipulating financial statements do not necessarily experience financial distress, increasing litigation risk may increase audit quality by motivating auditors to issue more going concern opinions for financially distressed clients (Geiger et al., 2006; Kaplan and Williams, 2013). Thus, finding a reduction in the likelihood of going concern opinions when auditors are facing imminent client risk would also indicate a decline in audit quality. Last, as Schmidt show that auditors can manage litigation risk by increasing perceived auditor independence, we use non-audit services as a proxy for the perceived audit quality. 2.3.Auditor switch and the change in audit effort Increasing litigation risk may affect both inputs and outputs from audit process. Our earlier discussion centers on output-based measures of audit quality, so next we turn to an important measure of audit inputs: audit efforts. Although the audit fees variable is often used as a measure of audit efforts (e.g., Gul, 2006; Hribar et al., 2014), it is not an appropriate measure in our setting for two reasons. First, audit fees may reflect effort or a risk premium. Although it has been well documented that litigation risk leads to higher audit fees, prior studies show that the increase in fees can be due to greater effort, a higher risk premium, or both (e.g., Bell et al., 2008; Venkataraman et al., 2008). Second, extant studies suggest that auditor switch may lead to lower audit fees (e.g., Huang et al., 2009), i.e., low balling, but does not necessarily reduce audit effort (Bell et al., 2015). Hence, we cannot follow prior studies by measuring auditor efforts with audit fees. 4 Alternatively, auditors may begin the process earlier and end earlier, while providing the same or better quality audits. Since we do not have data of the beginning date of audit process, we cannot directly test the length of the actual audit process. Nonetheless, if this is the case, we should not find an increase in income-increasing accruals or a decrease in the likelihood of issuing going concern opinions. 11

15 We approach the question of how litigation risk affects audit effort by exploiting a fundamental difference between effort and a risk premium. Specifically, efforts and risk premia are associated with the ex post incidence of lawsuits in different ways. Lobo and Zhao (2013) show that audit efforts reduce the ex post likelihood of financial restatements, suggesting that audit efforts should be negatively related to ex post litigation risk. In contrast, auditors charge a risk premium to manage residual risk that cannot be mitigated through greater efforts, suggesting that risk premia should be positively related to ex post litigation risk. Our prediction of the impact of litigation risk on audit inputs differs by whether or not there is an auditor switch, just like our predictions regarding audit outputs. If risky clients cannot engage higher quality auditors following auditor switches, the successor auditors would be more likely to charge a risk premium than to increase effort to manage litigation risk. This leads to our next hypothesis, stated in the alternative form, of a positive relation between audit fees and ex post litigation risk against the successor auditors. H3a. When litigation risk prompts auditor switches, audit fees charged by the successor auditors reflect a premium which is positively associated with ex post litigation risk. For auditors retaining risky clients, we expect that their improving audit quality would correspond to an increase in effort. Nonetheless, this prediction does not rule out the possibility that these auditors may also charge a risk premium. The relation between audit fees and ex post litigation risk thus depends on which fee component dominates: we would observe a negative relation between audit fees and ex post litigation risk if fees reflect efforts more than risk premia, and a positive relation if fees reflect more premia than efforts. Again, since effort can prevent litigation whereas premium cannot, we predict that the effort effect 12

16 would dominate the premium effect. This leads to our last hypothesis, stated in the alternative form. H3b. After an increase in litigation risk, auditors who continue engagements with risky clients would raise audit fees that reflect more efforts than risk premia, so their audit fees are negatively associated with ex post litigation risk. 3. Sample and Descriptive Statistics Our study presumes that auditors can mitigate their own litigation exposure by expending effort even though their clients may still be sued. In other words, auditors may have a different litigation exposure than their clients. Hence, we first show descriptive statistics that demonstrate the difference between auditors and their clients litigation risk. We start with a sample of U.S. companies listed on one of the national stock exchanges that allegedly violated Rule 10b-5 in private securities class action lawsuits from 2001 to We obtain the litigation data from the Stanford Securities Class Action Clearinghouse and manually identify whether each lawsuit against a listed company also names its auditor as a secondary defendant. We then merge this litigation data with COMPUSTAT, CRSP, IBES, Audit Analytics, and Thompson Reuters Insiders Data and Institutional Holdings data. We use COMPUSTAT to identify if there is an auditor switch in any specific year. The final sample contains 25,469 firm-year observations not involved in securities class action lawsuits and 1,447 firm-year observations involved in securities class action lawsuits, 188 of which also included auditors as defendants. Table 1 shows the descriptive statistics of auditors and their clients involved in securities class action lawsuits. Detailed descriptions of the variables are included in the appendix. Panel A compares clients involved and those not involved in private securities 13

17 litigation. The comparison shows results very similar to those of Kim and Skinner compared with firms not involved in private securities lawsuits, those involved in lawsuits are more likely to be in highly litigious industries; be growth firms with greater total assets, greater working capital, higher sales growth, and lower fixed assets; be more financially distressed; have more negative, left-skewed, and volatile stock returns; have a higher stock turnover; have greater institutional ownership; have greater external financing needs; and have more insider sales. We also compare their differences in auditor characteristics. Interestingly, clients sued appear to have better quality auditors than those not sued, as their auditors are more likely to be big-4 auditors and to have higher office-level quality as measured by office fixed effect and industry expertise. The univariate results suggest that higher quality auditors in general have riskier clients. Nonetheless, these results do not necessarily suggest that high quality auditors are also more likely to be sued than low quality auditors. Panel B compares auditors sued and those not sued, conditional on their clients being sued. The differences of most client-based variables between the litigation sample and nonlitigation sample either lose statistical significance or change signs. For example, auditors with clients in highly litigious industries, clients that are financially distressed, and clients recently issued equity are less likely to be sued than auditors with clients not associated with these features. One possible explanation is that as these risk factors emerge, auditors become more careful or exert greater efforts to avoid being implicated in litigation. Auditors sued have a lower quality as measured by office fixed effect. Although the mean differences of other measures of audit quality are not statistically significant, the results suggest that high quality auditors are somewhat less likely to be sued. Overall, Table 1 suggests that retaining risky clients does not necessarily lead to more lawsuits for high quality auditors. We next explore formally how litigation risk affects auditor switch for high quality auditors. 14

18 4. Empirical Results 4.1. Auditor quality and auditor switch amid increasing litigation risk We identify the event of increasing litigation risk first by using the ex post occurrence of lawsuits and then by estimating the change of the ex ante likelihood of litigation risk. For the ex post occurrence of lawsuits, we obtain the data from the Stanford database. The data selection procedure is explained in the previous section. Since securities class action lawsuits define a class action period as the period during which the defendants issued misleading financial statements, we use class action periods to identify which financial reports are allegedly misstated. Then we identify the fiscal year of the first misstated annual report as the year in which there is a surge of litigation risk. If the auditor in this fiscal year is not the same as the auditor one year ago, the observation is included in the switch sample; otherwise, it is classified as a stay. Table 2 shows the characteristics of auditors who switch and those who stay at the end of the fiscal year right before there is a surge of litigation risk. Auditors in the switch sample are less likely to be a big-4 auditor, have shorter tenure, and engage smaller clients than those in the stay sample. Neither one of our two measures of auditor quality that is independent of audit firm size is different between the two samples. The univariate results suggest that large auditors are more likely to retain risky clients while auditor quality is not associated with the likelihood of auditor switch when there is an increase in litigation risk. We conduct a multivariate regression to examine the probability of auditor switch as follows: P Switch α β BIG4 β SPECIALIST_OFF β OFFICE_FE β ACOMPETE β LNTENURE β CLIENT_SIZE ε (1) 15

19 The regression results are presented in Table 3, Column 1. Similar to our univariate results, Column 1 of Table 3 shows that big-4 auditors and auditors with longer engagement with clients are less likely to switch. Since big-4 auditors are also more likely to engage large clients, we find that the relation between client size and the likelihood of auditor switch vanishes after we control for auditor firm size. Our findings reject the null hypothesis of H1a that larger auditors and smaller auditors are equally likely to retain risky clients. The results suggest that large auditors may tend to retain risky clients because they have greater bargaining power or because they are more capable of diversifying client risk. Nonetheless, our findings do not reject the null hypothesis of H1b, suggesting that high quality auditors do not strictly prefer less risky clients. This finding allows for the possibility that high quality auditors would retain risky clients and increase audit quality in response to increasing litigation risk. Since the ex post occurrence of lawsuits may be subject to the survival bias, we also examine auditor switch based on the change of ex ante litigation risk. Following prior studies of securities litigation risk (e.g., Shu, 2000; Kim and Skinner, 2012), we estimate a logistic model of litigation risk by regressing the probability of a client being sued on a set of client innate characteristics that determine client litigation risk: CLIENT_SIZE, FPS, WC, SGROWTH, PPE, MB, RETURN, STDDEV, SKEWNESS, TURNOVER, INST, EPROCEEDS, DPROCEEDS, ITRADING, and IHOLDING. We then take the yearly difference of this estimated probability as a proxy for changes in client litigation risk. Since our focus is on cases in which there is a substantial increase in litigation risk, we keep observations where the change of this ex ante litigation risk is greater than the third quartile of the entire distribution, which is around We then estimate the same multivariate regression for auditor switch using this sample. The results are reported in Table 3, column 2. We continue to find that big-4 auditors are less likely to switch and our audit quality 16

20 measures independent of audit firm size are not associated with auditor switch. Overall, our findings suggest that larger auditors are more likely to retain risky clients. 5 To the extent that larger auditors are more capable of assess client risk changes more accurately than smaller auditors, our results rule out the alternative hypothesis that auditors stay because they are unable to detect an increase in litigation risk. We next examine the implications of these findings with regard to the relation between litigation risk and audit quality The impact of increasing litigation risk on audit quality To examine the impact of increasing litigation risk on audit quality, we focus on the ex post occurrence of lawsuits and exclude those not involving violations of the Generally Accepted Accounting Principles (GAAP). For each lawsuit, we include two years 6 before and two years after the disclosure of the first financial statements, i.e., our event date, in the regressions reported in Tables 4 to 5. Next, we compare the change in audit quality between the switch sample and the stay sample (H2a and H2b) by estimating the following regression model: Audit Quality α β β β (2) As discussed in Section 2, we use three measures of audit quality and one measure of perceived auditor independence. The first measure of audit quality is SIGN_ACCRUALS, signed discretionary accruals estimated from the modified Jones Model, which represents earnings management through income-increasing accruals. The second is AUDIT_LAG, 5 In untabulated analyses we find that the proportion of firms experiencing auditor switch is higher in the first year of misstatements than in the years beforehand. The finding suggests that our method is reasonably accurate in capturing the increase in litigation risk. We also compare the resolution of lawsuits and the length of the class action period between the switch and the stay samples. The difference is not statistically significant at the conventional level. Hence, this result does not suggest that auditor switch is related to the severity of fraud. 6 We use two years because the average class action period in our sample is around 400 days. 17

21 defined as the number of days between the fiscal year end and the auditor signature date. The third measure of audit quality is OPINION, an indicator variable that equals one for going concern opinions and zero otherwise. We estimate an ordinary least squares model for the first two dependent variables as they are continuous variables, and estimate a logistic model for OPINION as it is a dichotomous variable. Our measure of perceived auditor independence is NAS, the natural log of non-audit service fees. We also present ordinary least squares test results using audit fees as dependent variables. As audit fees may reflect effort or premium, we do not use it as an indication of audit quality at this stage. The interpretation of audit fees would be examined in the next section. POST is a dummy variable that equals one for the period after our event date, i.e., the start date of class action period, and zero otherwise. We expect that, if auditors in the stay sample improve audit quality after the event date, the coefficient on POST would be negative when we use SIGN_ACCRUALS as the dependent variable, and positive when we use AUDIT_LAG and OPINION. SWITCH is a dummy variable that equals one for the switch sample and zero for the stay sample. POST*SWITCH represents how auditors in the switch sample change audit quality relative to the stay sample. H2a predicts that the coefficient on POST*SWITCH is positive when we use SIGN_ACCRUALS as the dependent variable, and negative when we use AUDIT_LAG and OPINION. 7 Following prior studies, we use different sets of control variables for different dependent variables. The CONTROL variables in the model for SIGN_ACCRUALS include: SIZE, SGROWTH, NSEG, LOSS, CFO, STDCFO, STDDEV, LEVERAGE, ZSCORE, MB, FPS, BIG4, and ROA. The control variables in the model for AUDIT_LAG and audit fees include: SIZE, NSEG, FOREIGN, MB, OPINION, TENURE, SPECIALIST_OFFICE, CITISIZE, DEC_END, and SCALE. 7 This research design essentially assumes that, had there not been auditor switch, the switch sample would have behaved in the same way as the stay sample. In the robustness analyses we find the same results by using a different benchmark for the switch sample: conditional on there is an auditor switch, how auditors whose clients are eventually involved in litigation behave differently than those whose clients are not sued. 18

22 Almost all control variables in the above two models are included in the model for OPINION. Full descriptions of these control variables are provided in the appendix. Table 4 shows the regression results for model (2). The coefficient on POST is statistically significantly negative when we use SIGN_ACCRUALS as the dependent variable and positive when we use AUDIT_LAG and OPINION. The results suggest that increases in client risk motivate auditors in the stay sample to take more time to audit, to constrain the use of income-increasing accruals, and to issue more going concern opinions for financially distressed clients. These findings are consistent with H2b that, absent auditor switch, litigation risk would increase audit quality. On the other hand, the signs on POST*SWITCH are opposite to the signs of POST in all three tests, suggesting that successor auditors in the switch sample finish audits earlier, allow more income-increasing accruals, and issue fewer going concern opinions than they should have done had there not been an auditor switch. This finding is consistent with H2a and the notion that, in the presence of auditor switch, litigation risk would result in a reduction in the quality of audit service. The last three columns in Table 4 show the regression results for total fees (TF), audit fees (AF), and non-audit services (NAS). We find that the coefficient on POST is significantly positive when we estimate TF and AF, and negative when we estimate NAS. Our findings suggest that auditors in the stay sample increase audit fees and slash non-audit services when facing imminent client risk, perhaps in an attempt to appear more independent. On the other hand, the coefficient on POST*SWITCH is statistically significantly negative for AF, and the sum of POST and POST*SWITCH is not statistically significantly different from zero (0.147+(-0.151) = , p-value of two-tailed F-test is 0.950). Similarly, the coefficient on POST*SWITCH is statistically significantly positive for NAS, and the sum of POST and POST*SWITCH is not statistically different from zero (p-value is 0.685). 19

23 Moreover, the positive coefficient on SWITCH for the NAS regression suggests that auditors in the switch sample provide more NAS and so may be perceived as less independent. Overall, Table 4 suggests that auditors retaining risky clients increase audit quality whereas those accepting risky clients do not. Next we conduct further tests to help understand the different risk management strategies undertaken by auditors in different samples. 4.3.Efforts vis-à-vis risk premium While ex ante litigation risk would increase audit fees, audit fees may either reduce ex post litigation risk if they reflect efforts or increase ex post litigation risk if they reflect a premium. The endogenous relation between audit fees and litigation risk suggests that a simple linear model is not suitable for examining the relation between audit fees and ex post litigation risk. To further distinguish between audit efforts and premiums, we use a 2SLS simultaneous equations approach, following Kaplan and Williams (2013) as follows: AF β β 4 _ _ _ (3) AL β β _ (4) As before, all of the variables are defined in the appendix. At stage 1 we estimate an ordinary least squares model to determine the estimated audit fees, and at stage 2 we use the estimated AF as the instrumental variable, AF[Instrument], to conduct a probit analysis on AL. Kaplan and Williams (2013) argue that TURNOVER (trading volume of the client s shares) is unique to auditor litigation risk and not related to going concern opinions, a proxy for misstatement risk. As other studies do not use turnover 20

24 as an explanatory variable for audit fees, we use it as an exogenous variable for AL. DEC_END (if a client s end of fiscal year is December) and SCALE (the percentile rank of the city-industry number of audit clients for each audit firm) are unique to audit fees. Prior studies do not find that these two measures to be related to auditor litigation risk (e.g., Jha and Chen, 2015). The end month of a company s fiscal year is not systematically related to whether or not an auditor is (perceived as) independent. The economic scale of auditor operation is also unlikely to affect an auditor s independence. Our simultaneous equations results are presented in Table 5. Although our exogenous variables are highly correlated with LAF but not AL when we use both the switch and stay samples before excluding cases not involving GAAP violations, after excluding non-gaap cases they are highly correlated with LAF but not AL only when we use the stay sample. Further analysis shows that excluding non-gaap violation cases reduces the switch sample size and lead to severe multi-collinearity problem. Instead of using different model specifications for different samples, we report the results using the same model both before and after excluding non-gaap violation cases. Columns 1 and 2 show the first- and secondstage results for the switch sample, respectively, before excluding allegations not involving GAAP violations. Columns 3 and 4 show the results excluding allegations not involving GAAP violations. The coefficient for AF[Instrument] is significantly positive regardless of whether or not we exclude those not involving GAAP violations. This suggests that in the switch sample, the audit fees reflect a premium that is positively related to ex post litigation risk against auditors, consistent with H3a. In contrast, audit fees in the stay sample appear to reflect efforts as shown in columns 5 to 8. In both columns 6 and 8, AF[Instrument] is significantly negative, which is consistent with our H3b that higher audit fees is associated with lower ex post litigation risk against auditors in the stay sample. 21

25 To supplement our tests on the relation between audit fees and ex post auditor litigation risk, we conduct a similar analysis on the relation between audit fees and ex post client litigation risk. We first match each firm involved in securities class action lawsuits with a firm not involved in litigation by year, industry (2-digit SIC code), and size. We also match by whether or not a control firm switches auditor at the same time as the litigated firm. We then conduct a 2SLS simultaneous equations analysis for AF and CL using model (3) and the following model: CL β β (5) Similar to our models for AF and AL, TURNOVER is related to CL but not AF, whereas DEC_END and SCALE are unique to AF. We include other client characteristics related to CL in model (5) and use the same first-stage model for AF. Table 6 presents the results for both the switch and stay samples. As shown in column 2, we do not find a statistically significant coefficient for AF[Instrument] in the switch sample. As for the stay sample, column 4 shows that AF[Instrument] is not significant either. Thus, regardless of whether or not there is an auditor switch amid heightened litigation risk, audit fees on average do not reflect more efforts that deter ex post litigation against clients or higher risk premium that predict ex post litigation against clients. These results provide support to our presumption that auditors risk management strategies may affect their own litigation exposure but not necessarily their clients litigation risk Sensitivity analysis In our tests for H2a and H2b, we compare the stay sample and the switch sample, who are similar in terms of increasing litigation risk but different in terms of the occurrence of 22

26 auditor switch. In addition to comparing these two litigation subsamples, we conduct a matched-pairs analysis. We match each firm in the litigation sample with a control firm (selected from firms not involved in private securities litigation in our sample period) by firm size 8, year, the occurrence (or the absence) of auditor switch in the same year, and industry (1-digit SIC code). Hence each of the litigation subsamples becomes a treatment sample with its own matched control sample. In contrast to our earlier tests for H2a and H2b, the treatment and control sample are similar in terms of the occurrence (or the absence) of auditor switch but different in terms of client risk. We then estimate a similar model as model (2) by replacing SWITCH with TREAT, which is a dummy variable that equals one for firms in the litigation sample and 0 for firms in the control sample. The results are shown in Table 7 9. Similar to the results in Table 4 and consistent with our H2a and H2b, we find that auditors in the SWITCH sample allow more income-increasing accruals and reduce audit lag, whereas auditors in the STAY sample allow less income-increasing accruals and increase audit lag. In untabulated sensitivity tests, our results are robust with respect to several alternative interpretations. Our results are not sensitive to including or excluding lawsuits not directly involving violations to the GAAP. Although the technical compliance with GAAP does not preclude the likelihood of misstatement (Securities and Exchange Commission, 2002), almost all lawsuits against auditors in our sample involve some allegations of GAAP violations. Our findings do not appear to be driven by the difference in litigation risk against auditors before and after the Sarbanes-Oxley Act (SOX) as the majority of our sample covers the post-sox period. 8 We require the difference in firm size to be no more than 30% of the size of the treatment firm. 9 We do not show the results for OPINION when using the switch sample, because several variables including TREAT perfectly predicts OPINION. The potential explanation includes the reduced sample size after matching and the inherently difference between the treatment sample and the control sample in terms of the clients financial distress level. 23