Abstract
CHINA JOURNAL OF ACCOUNTING STUDIES 2019, VOL. 7, NO. 2, 214–244 https://doi.org/10.1080/21697213.2019.1676066 ARTICLE a b Youfu Yao and Shuang Xue a b School of accountancy, Shanghai University of Finance and Economics, Shanghai, China; Institute of Accounting and Finance & School of Accountancy, Shanghai University of Finance and Economics, Shanghai, China ABSTRACT KEYWORDS Comment letters; internal Taking Chinese A-share listed firms from 2015 to 2017 as our samples, control opinion shopping; this paper investigates the supervision effect of comment letters on headquarters and branch; internal control opinion shopping at audit firm level, branch level and signature partners signing partner level respectively. The empirical results show that the pressure from comment letters can significantly decrease the probabil- ity of internal control opinion shopping at both audit firm level and signing partner level. This supervision effectismorepronouncedwhen there are internal-control-related questions in comment letters or the firm’s annual auditor is required to give opinions on some questions in comment letters. As for the branch level of audit firms, it is found that the supervision effect of comment letters on internal control opinion shopping is more pronounced in the firms which switch from head- quarters to branches. We also find comment letters have spillover effects on firms in the same industry, in the same province or with thesameauditors. 1. Introduction The report of 19th National Congress of the Communist Party of China (CPC) points out that it is necessary to transform the functions of the government and to reform methods of supervision. Since the report of 19th National Congress of CPC, the China Securities Regulatory Commission (CSRC) has committed to deepening the reform of the capital market regulatory system. Since 2019, the CSRC has launched a pilot project of registra- tion system of IPO in SSE STAR market. Comment letters mechanism is becoming one of the most important supervision methods in the registration era. As an innovative supervision method for information disclosure, comment letters have attracted great attention from regulators and academic. Prior literatures find that com- ment letters have information content (Chen, Dend, & Li, 2018a), improve the quality of information disclosure (Bozanic, Dietrich, & Johnson, 2017; Brown, Tian, & Jennifer, 2018; Wang, 2016), decrease accrual-based earnings management (Chen, Dend, & Li, 2019; Cunningham, Johnson, Johnson, & Lisic, 2019), improve audit quality (Chen, Dend, & Li, 2018b) and mitigate stock price crash risk (Zhang & Tang, 2018). Comment letters CONTACT Shuang Xue xuesh@mail.shufe.edu.cn Institute of Accounting and Finance & School of Accountancy, Shanghai University of Finance and Economics, Shanghai, China Paper accepted by Kangtao Ye. This article has been republished with minor changes. These changes do not impact the academic content of the article. © 2019 Accounting Society of China CHINA JOURNAL OF ACCOUNTING STUDIES 215 mechanism is a double-edged sword. For example, comment letters can induce managers to do strategic insider trading (Dechow, Lawrence, & Ryans, 2016) or motivate managers to switch from accrual-based earnings management to real-activities-based earnings management (Cunningham et al., 2019). Under Section 408 of the Sarbanes-Oxley Act of 2002 (SOX), the SEC is required to review the financial statements of publicly listed firms at least once every three years, comment letters and firms’ response letters are not required to be disclosed at the first time (Dechow et al., 2016). The comment letters mechanism of the Shanghai Stock Exchange (SHSE) and Shenzhen Stock Exchange (SZSE) is different from that of SEC in several aspects. First, it is a kind of inquiry of ‘getting to the root’. Second, firms’ auditors may be required to issue opinions on some questions in comment letters. Third, both comment letters and firms’ responses should be disclosed timely. This paper examines the impact of comment letters on internal control opinion shopping in China. Internal control plays an important role in guaranteeing high-quality financial reporting and improving operational efficiency (Cheng, Goh, & Kim, 2018;Zhang &He, 2013). On the one hand, there exists the motivation of internal control opinion shopping. Prior study finds that firms successfully engage in internal control opinion shopping at the level of audit firms during the post-SOX period (Newton, Persellin, Wang, & Wilkins, 2016). The behaviour of audit opinion shopping is become more and more concealed (Li & Zhao, 2014)and theefficiency of decision-making of accounting information users is decreased (DeFond & Zhang, 2014;Wu, 2005). Therefore, internal control opinion shopping has been one of the most important issues for academics and regulators (Newton et al., 2016). The Ministry of Finance of China published Notice on Implementing Internal Control System by Classifications and Batches of Listed Firms on Main Board in 2012, by which listed firms are required to disclose audit reports on internal control. This institutional background provides us a research opportunity for investigating the supervision effect of comment letters on internal control opinion shopping in Chinese capital market. On the other hand, regulators care more about internal control problems than accounting problems (Newton et al., 2016). Internal control is also one of the most concerned issues in comment letters from the Shanghai and Shenzhen Stock Exchanges. In our samples, the proportion of internal-control -related comment letters is nearly one quarter. Firms with low quality of internal control will be required to replenish and rectify internal control system. The pressure from comment letters will push the management to take activities to improve firm’s internal control. Given the improved internal control, the gains of internal control opinion shopping become smaller. A comment letter is also a signal of audit risk, which affects auditor’s decision- making behaviour Therefore, we expect that the comment letters will have a supervision effect on the behaviour of internal control opinion shopping. Characteristics of comment letters may impact their supervision effect. Specifically, the supervision effect of comment letters on internal control opinion shopping is more pronounced when there are questions on internal control in comment letters (internal-control-related comment letters) or the annual auditors are required to verify or issue their opinions on some questions (auditor- verified comment letters). There is a question ‘what do you think about the pressure of comment letters on the auditor’ in a survey conducted by the project of the Ministry of Finance in August 2018. In the 169 valid questionnaires, there are 38.46% of the auditors think the pressure of comment letters is large, and 39.64% of the auditors think the pressure of comment letters is very large. The sum is 78.1%.. 216 Y. YAO AND S. XUE Internal control opinion shopping is affected by the independence of the provider of audit service. Evidences on internal control opinion shopping are generally provided at the level of audit firmssuchasNewtonetal. (2016). Similarly, the first dimension of this study is also at the level of audit firm.Besides,wealsooffer evidences from the other two dimensions. One is direction of auditor switches between the headquarters and branch of audit firm. A given client is usually more important to a branch than to the headquarters of an audit firm, so the auditor independence at the branch level is lower than that at the headquarters level. The internal control of a branch office is generally weaker than that of headquarters. This leads to a comparatively lower audit quality at branch office (Francis & Michas, 2013;Wu,Wang,&Li, 2018). Given this, we expect that comment letters can play a stronger supervision role at the level of audit branch. As we all know, changing audit firm is more costly than changing signing partners, so it is more feasible to switch signature partners to engage in opinion shopping (Chen et al. 2016). The risk perception from comment letters is more specificand direct at the level of signing partners. So the supervision effect of comment letters at the level of signing partner is our test of the third dimension. Based on the characteristics of Chinese-style comment letters and the mandatory dis- closure of internal control audit reports, we take Chinese A-share listed firms from 2015 to 2017 as our samples and employ the model of audit opinion shopping in Lennox (2000)to investigate the supervision effect of comment letters on internal control opinion shopping from above-mentioned three dimensions. The empirical results show that the pressure from comment letters can significantly decrease the probability of internal control opinion shopping and this supervision effect is more pronounced for internal-control-related com- ment letters or auditor-verified comment letters. These findings are robust when a propensity score matching and difference-in-differences (PSM-DID) method is used or when the samples with opinion shopping on financial reports are excluded. Supervision effect of comment letters is more pronounced in the samples of headquarters-to-branch switching and branch-to-branch offices switching. The supervision effect of comment letters on internal control opinion shopping also exists at signing partner level. Further testing shows that the supervision effect exists only in a short run. We also find the supervision effect of comment letters on internal control opinion shopping has spillover effects on firms in the same industry, in the same province or with the same auditors. Our study has several theoretical and practical contributions. First, this paper constructs a supervision framework of comment letters on the behaviour of internal control opinion shopping from three dimensions which include audit firm level, branch level and signing partner level. Our findings largely enrich the literature on the economic consequences of comment letters. Chen et al. (2018b) examine the improvement of audit quality through comment letters. They investigate the effect of comment letters on financial reporting at audit firm level. This study is different from Chen et al. (2018b) since we focus on the behaviour of internal control opinion shopping. And more importantly, this paper goes deep into the branch level and signing partner level of audit firm, which can offer more specific references to regulators to improve the quality of comment letters. Second, we distinguish the switching directions between headquarters and branch, and investigate the effect of comment letters on them, respectively. The findings of this paper extend the study on internal control opinion shopping (Newton et al., 2016) and provide the interesting governance evidence on the auditor switching between head- quarters and branch level, and enrich the empirical research on the headquarters-branch CHINA JOURNAL OF ACCOUNTING STUDIES 217 offices (e.g. Chen, Francis, & Hou, 2019; Francis & Michas, 2013; Wang, Wu, & Zeng, 2016; Wu et al., 2018). Third, our study enriches the effect of comment letters on internal control opinion shopping at the individual level of signing partners. Taking large U.S. non-profit organisa- tions as samples, Fitzgerald, Omer, and Thompson (2018) fail to find the correlation between audit partner switch and internal control quality. Our findings on internal control opinion shopping at the individual level of signing partner fill the blank for profit-making organisations. Last, we find comment letters are effective only in the short run. This implies that regulators need to pay more attention to the timing of auditor switching, so as to improve the efficiency of inquiry mechanism. What’s more, this paper provides the spillover effect of comment letters on firms in the same industry, in the same province and with the common auditors, which provides a deep insight into the effectiveness of comment letters. The remainder of this paper is organised as follows. Section 2 briefly reviews related literature and section 3 provides the theory background and develops the hypotheses. Section 4 introduces the research designs and section 5 presents the main empirical results. Section 6 performs further analyses and robustness tests, and then section 7 concludes the paper. 2. Literature review 2.1. Literature on comment letters Comment letters system is also called inquiry system, and it is a preventive mechanism of information disclosure. Research on comment letters has just started in recent years and mainly focuses on the factors which induce the comment letters issuing and economic consequences of comment letters. Peter and Zhang (2018) find that SEC staff members exhibit personal styles, their styles shape the differences of firms’ financial reporting and female staff members are generally tougher reviewers. Firms with lower profitability, higher business complexity, lower audit quality, or weak internal control have a higher probability to receive a comment letter (Cassell, Dreher, & Myers, 2013). Also, firms with political connections are more likely to receive commentletters (Chen, Deng,Jin,Lou,&Zhang, 2018; Heese, Khan, & Ramanna, 2017). The economic consequences of comment letters can be summarised into reducing information asymmetry, affecting the opportunistic behaviour of management and impacting on the decisions of other stakeholders. Comment letters can reduce information asymmetry. The announcement of comment letters has information content (Chen et al., 2018a). Both information asymmetry and litiga- tion risk are reduced after firms receive comment letters from SEC (Bozanic et al., 2017). Johnston and Petacchi (2017) show that bid-ask spread tends to decline and earnings response coefficients (ERCs) increase after comment letters are issued. Bens, Cheng, and Monica (2016) find that fair-value-related comment letters can reduce the uncertainty of fair value estimates. Zhang and Tang (2018) argue that comment letters can mitigate future stock price crash risk and this effect is more pronounced if a firm’s financial report is more opaque. The comment letters can affect the opportunistic behaviour of management. By theoretical analysis, Li and Liu (2017) conclude that comment letters may constrain 218 Y. YAO AND S. XUE opportunistic earnings management behaviour by improving disclosure and reducing information asymmetry during the IPO period. Cunningham et al. (2019) find that com- ment letters from SEC decrease accrual-based earnings management but induce more real-activities-based earnings management. Chen et al. (2019) also find that comment letters can effectively decrease accrual-based earnings management in Chinese capital market. In terms of insider trading, Dechow et al. (2016) document that insider trading is significantly higher than normal level prior to the public disclosure of SEC comment letters relating to revenue recognition. That is, the insiders have the time-selecting behaviours that they would take the opportunity of their information advantage during the comment letters’ response period to conduct transactions (Li, Zhang, & Xia, 2017). Kubick, Lynch, Mayberry, and Omer (2016) find that firms receiving a tax-related SEC comment letter can decrease their future tax avoidance behaviour. The comment letters also impact the behaviour of firm’s other stakeholders. Auditors tend to perceive higher audit risk and require higher audit fees (Gietzmann & Pettinicchio, 2014)or improve auditquality (Chenetal., 2018b). From the perspective of analysts, Wang (2016) shows that firms which revised the information of business segments after receipt of SEC comment letters can decrease analysts’ forecast errors, reduce forecast optimistic bias and forecast dispersions. Analyst coverage increases after the improvements to firms’ disclosures following a comment letter review (Bozanic et al., 2017). Banks charge higher interest rates for firms receiving comment letters (Cunningham, Schmardebeck, & Wang, 2017). The comment letters have spillover effects. Brown et al. (2018) show that firms without receiving comment letters also tend to provide more firm-specific information in the subsequent year if the industry leader or more industry peers firms have receive comment letters in the current year. Kubick et al. (2016)offer additional evidence of industrial spillover effect for comment letters involving tax issues. 2.2. Internal control opinions shopping Previous studies mainly focus on the existence of audit opinion shopping on financial reports and there is still no a consensus conclusion. Some studies find that firms can successfully engage in audit opinion shopping on financial reports (Lennox, 2000; Chen et al., 2016; Wu, Wang, & Lu, 2013; Li & Zhao, 2014; Wu et al., 2018), while the others fail to find evidences (Krishnan & Stephens, 1995; Lu & Tong, 2003; Newton et al., 2016). Newton et al. (2016) find that some clients are successful in internal control opinion shopping at audit firm level. Chen et al. (2016) extend opinions shopping from audit firm level to opinion shopping at partner level. On the supervision of audit opinion shopping, DeFond and Zhang (2014) argue that both the rotation of auditors and the appointment of the successors should be paid more attention. External supervision makes it more expensive and difficult for firms to engage in audit opinion shopping (Lu & Tong, 2003) and can improve the professional conserva- tism of successive auditors (Huang & Zhang, 2010; Wu, 2005). Chen et al. (2016) show that clients are less likely to be successful in audit opinion shopping if the audit firm is formed as a partnership rather than a corporation. To sum up, comment letters on annual reports have information content, and play an important supervision role in reducing self-interest behaviour of management and increase the conservatism of auditors. In 2015, comment letters mechanism was adopted CHINA JOURNAL OF ACCOUNTING STUDIES 219 by the Shanghai and Shenzhen Stock Exchanges and both comment letters and firms’ responses are required to be disclosed publicly. So far research on comment letters in Chinese capital market is quite few, and the supervision effect of comment letters on internal control opinion shopping is still blank. 3. Theory analysis and hypothesis development 3.1. Hypothesis development Internal control opinion shopping could directly damage auditors’ independence (Newton et al., 2016). As an external supervision mechanism, the comment letters on annual reports are expected to reduce the behaviour of internal control opinions shop- ping. We will explain this expectation from the perspectives of regulators, managers and auditors, respectively. First, comment letters mechanism is an important part of supervision on capital market. In the public interest theory, regulation comes from investors’objective demand for correct- ing market failure (Peltzman, 1976; Stigler, 1971). The government regulation could be a substitute for the law to protect investors’ interests when the legal system is not perfect enough (Chen, Zhang, & Li, 2008; Glaeser & Shleifer, 2001;Peltzman, 1976;Stigler, 1971). To protect the interests of the investors, the CSRC has committed to maintain the healthy operation of the capital market and to correct market failure by establishing the supervision mechanism of information disclosure. As a preventive supervision mechanism, comment letters can make up for the insufficiency of information disclosure. At the same time, the two exchanges declare that whether a firm can truthfully reply the exchange’s comment letters within the given time will be an important factor when they rank the quality of firms’ information disclosure. What’s more, the supervision effect of comment letters can also be realised by arousing the attention of other government departments. For example, ‘30 billion financial fraud by Kangmei Pharmaceutical Co., Ltd ’has been repeatedly ques- tioned by the Shanghai Stock Exchange On 30 May 2019. It not only triggered the Ministry of Finance to intervene and inspect on Kangmei, but also induce a thorough inspection on accounting information quality of the whole pharmaceutical industry. That is, though issued by exchanges, the comment letters can be a clue of CSRC’s investigations and bring a higher risk of internal control opinions shopping. Second, comment letters can bring pressure to firms’ management. When the cost of internal control opinions shopping is greater than the private benefits of management, the management will reduce the behaviour of internal control opinions shopping. On the one hand, the disclosure of comment letters can increase the cost of management fraud. Being questioned means that the firm fails to comply with relevant regulations or its information disclosure is inadequate, which has aroused the awareness of the exchanges. Mandatory public disclosure of comment letters and firms’ responses can bring higher costs and lower benefits of internal control opinions shopping. On the other hand, inquired by exchanges may damage the market reputation of management. The reputation acts as an effective market supervision mechanism. Zhang (2002) points out that reputation can maintain the order of market transactions at a lower cost than the legal system. Market attention and media reports arisen by comment letters have a negative effect on the reputation of 220 Y. YAO AND S. XUE management. When the cost of reputation is greater than the private benefits, the reputa- tion mechanism will play a supervisory role and reduce internal control opinion shopping. Third, auditors may perceive higher audit risk to firms which receive comment letters. Successful opinion shopping is the result of auditor’s concession, which is negatively related to auditor independence (Newton et al., 2016). Comment letters will put pressure to auditors and enhance their independence and awareness of supervision risk since the probability of being punished due to the issuance of inappropriate internal control opinions will increase for comment letter receivers. Gietzmann and Pettinicchio (2014)shows that auditors reassess the reputation risk and litigation risk for comment letters receivers. When firms are issued comment letters, their auditors may make more efforts, expand the scope of audit tests or implement additional internal control audit procedures to obtain more sufficient audit evidences. When internal control issues are involved directly in comment letters, the cost of internal control opinions shopping by management is much higher. Under the pressure of comment letters, the managers have to rectify the weakness and improve the internal control. Given these improvement, the motivation of internal control opinions shopping is largely decreased. Auditors also perceive higher internal control audit risk and will adopt more effective audit procedures to test the internal control weakness. So for the receivers of internal-control-related comment letters, auditors will be less likely to issue loose internal control opinions. In summary, comment letters can increase the supervision risk, both the management and auditor will perceive a high pressure from regulators and investors. The cost of internal control opinions shopping increases, the benefit decreases. This supervision effect of comment letters on internal control opinion shopping will be larger when there are questions on internal control in comment letters. Thus, we propose Hypothesis 1 and 2 as follows. H1: Other things being equal, probability of internal control opinion shopping is lower in firms with comment letters. H2: Compared with receivers of non-internal-control-related comment letters, the super- vision effect of comment letters on internal control opinion shopping is larger for receivers of internal-control-related comment letters. In order to improve the efficiency of inquiry mechanism, some comment letters require firms’ auditors to verify and issue their opinions on some questions (auditor-verified comment letter). On the one hand, when the auditors are required to verify some issues in the comment letters management will face the dual pressure from the exchanges and auditors. On the other hand, auditors need to be more diligent and conscientious in issuing their opinions on issues specified by exchanges, which strengthen auditors’ responsibilities. The perceived higher audit risk and responsibility improve auditors’ independence. Therefore, we expect that when the comment letters require auditors to verify and provide opinions on specific issues, the behaviour of internal control opinion shopping can be significantly reduced: CHINA JOURNAL OF ACCOUNTING STUDIES 221 H3: Compared with firms with non-auditor-verified comment letters, the supervision effect of comment letters on internal control opinion shopping is larger for firms with auditor-verified comment letters. 3.2. Hypothesis tests: three dimension analysis Above hypothesis will be tested from three dimensions which include auditor switch at firm level, auditor switch at office or branch level and auditor switch at signing partner level. Due to the limitation of data, auditor switch at firm level is the most commonly used in the study of audit opinion shopping. In this paper, we will expand the research from firm level to branch level and partner level. Wallman (1996) and Francis and Yu (2009) argue that, as the basic decision-making unit, the branch or local office is more appropriate to be analysed and evaluated than the whole audit firms. The analysis of organisational hierarchy is embedded in the theory of organisational structure. Jensen and Meckling (1992) point out that the optimal allocation of decision-making power within an organisation is related agency cost and information asymmetry. Information asymmetry and agency problems exist between the headquar- ters and branches of an audit firm. Branches or local offices mainly enjoy the specific knowledge of clients and local market. Meanwhile, as non-independent legal persons, the branches do not bear legal liabilities independently. That is, the potential reputation loss and legal liabilities caused by a branch are borne by the whole audit firm. So branches or local offices have motivation to issue loose audit opinions. Wang et al. (2016) find that in the audit firm with poor internal control, the audit branches are more likely to issue a clean audit opinion to its new client which comes from other audit firm and has received a modified audit opinion in the previous year. Therefore, we expect that internal control opinions shopping is more likely to occur at the level of branches, and the supervision effect of comment letter will be pronounced when firm switches its auditor from a headquarters to a branch office. As signing partner has signature responsibility in issuing internal control audit reports, the success of internal control opinion shopping depends on the signing partner. In addition, when auditors are required to verify and issue their opinions on some issues in comment letters, the signing partners have to bear additional legal responsibility. Wu (2008) points out that there exists a phenomenon of ‘individual partners are punished more than audit firms’. The signing partner is the person who signs the internal control audit reports and takes legal responsibility. So the evidence at the level of signing partner is the reinforcement and confirmation of the evidence at the level of audit firm. Therefore, the pressure of comment letters can play a positive governance role in internal control opinion shopping at the level of signing partner. 4. Research design 4.1. Sample selection Comment letters and firms’ responses are disclosed since 2015. We take Chinese A-share listed firms during 2015 to 2017 as our initial sample, and then exclude the samples as 222 Y. YAO AND S. XUE follows: (1) observations in the financial industry; (2) the observations with missing variables. The final sample consists of 7,544 firm-year observations. We obtain financial data from the China Stock Market and Accounting Research Database (CSMAR) and WIND database. The comment letters on annual reports are manually collected from the Shanghai and Shenzhen Stock Exchanges, Juchao informa- tion website (cninfo.com.cn) and Baidu news website (news.baidu.com) We manually collect the information of headquarters and branches of audit firms from CSMAR based on two of signing partners, supplemented by the website of CICPA (the Chinese Institute of Certified Public Accountants). To mitigate the effect of outliers, all continuous variables are winsorised at 1% and 99% levels. 4.2. Variable definition 4.2.1. Internal control opinion shopping To define internal control opinion shopping, one should compare the opinion that a firm would likely receive if there is an internal control auditor switch with the opinion the same firm would likely receive if no internal control auditor switch occurs (Lennox, 2000). Following the opinion shopping models of Lennox (2000) and Newton et al. (2016), we first examine whether any scope for internal control opinion shopping exists, and, if so, whether firms exploit this scope to obtain more favourable internal control opinions. Specifically, we begin by estimating a probit model to generate predicted probabilities that a firm will receive a modified internal control opinion (InMAO) with and without an internal control auditor switch. We then incorporate the predictions from the first model into a second model that investigates the relationship between scope for internal control opinion shopping and internal control auditor switching. The general form of the first model is as follows: InMAO ¼ α þ α InSwitch þ α InSwitch InMAO þ α InMAO i;t 0 1 i;t 2 i;t i;t1 3 i;t1 þα X þα InSwitch X þ Industry þ Year þ ε ð1Þ (1) m i;t n i;t i;t i;t In Equation (1), InMAO is a dummy variable that equals 1 if firm i receives a modified i,t internal control opinion in year t and 0 otherwise, InMAO is a dummy variable that i,t-1 equals 1 if firm i receives a modified internal control opinion in year t-1 and 0 otherwise. InSwitch is a dummy for an internal control auditor switch, which takes i,t the value of 1 if firm i is audited by a new internal control auditor in year t, and 0 otherwise; X is a set of control variables. Following Newton et al. (2016), we control firm size (LnSize ), thepresenceofanet loss(Loss ), the announcement of i,t i,t a restatement (Restatement ), bankruptcy risk (BankRisk ), growth rate of revenue i,t i,t (Growth ), firm age (LnAge ), acquisitions (MA ), big4 audit firms (Big4 ), institutional i,t i,t i,t i,t ownership (Inshold ), firms that have changes in CEO or CFO (CEFchange ), number of i,t i,t business segments (Bsegment ), and foreign sales (Export ). We also control industry i,t i,t and year fixed effect. We use equation (1) to calculate the firm’s probability of receiving an InMAO in year t with and without an internal control auditor switch, respectively, and measure the The website of Juchao information (cninfo.com.cn) is an official media outlet designated by the CSRC to provide listed firms data. Baidu News Network (news.baidu.com) is the largest Chinese news platform.. CHINA JOURNAL OF ACCOUNTING STUDIES 223 motivation for internal control opinion shopping. Specifically, the conditional probability that a firm receives an InMAO is denoted Pr(InMAO ), where superscript n denotes the i,t decision of internal control audit firm switch. Pr(InMAO ) denotes the probability of i,t receiving an InMAO for firm i if an internal control audit firm switch occurs in year t, and Pr(InMAO° ) denotes that if there is no internal control audit firm switch. We use the i,t difference in the probability of receiving an InMAO between the two situations (i.e. [Pr (InMAO =1)-Pr(InMAO° = 1)]) to measure the motivation for companies’ internal i,t i,t control opinion shopping through changing internal control audit firm, denoted InShop . If the value of InShop is negative, it suggests that the probability of obtaining i,t i,t a modified internal control opinion through switching internal control audit firm is less than the probability of no internal control audit firm switch, the firms have a motive to change auditors. The second step is to bring the test variable of internal control opinion shopping (InShop ) into the following model (2): i,t InSwitch ¼ β þβ InShop þβ Z þ Industry þ Year þ μ (2) i;t 0 1 i;t m i;t i;t In Equation (2), InSwitch is a dummy for an internal control audit firm switch, InShop is the i,t i,t test variable of internal control opinion shopping that calculates from the mode(1). Asignificantly negative β will imply that the firm can realise internal control opinion shopping through changing internal control audit firm. In Equation(2), following Newton et al. (2016), the controls represented by Z include i,t firm size (LnSize ), financial leverage (Lev ), return on total assets (ROA ), state owned i,t i,t i,t enterprises (SOE ), the announcement of a restatement (Restatement ), bankruptcy risk i,t i,t (BankRisk ), growth rate of revenue (Growth ), the absolute value of discretionary i,t i,t accruals (Abs_DA ), acquisitions (MA ), the proportion of shares owned by management i,t i,t team (Dirs ), the proportion of shares owned by the largest shareholder (Fshare ), the i,t i,t ratio of inventory and receivables to total assets (InvRec ), modified financial reports audit i,t opinion in the year t-1(GC ), big4 audit firms (Big4 ), a firm hires a new CEO or CFO i,t-1 i,t (CEFchange ), a firm is punished by the exchanges or CSRC in the year t-1 (Punishment ). i,t i,t-1 We also control year and industry fixed effects. Similarly, considering the situation of signing partner switching, the estimation method of the test variable of internal control opinion shopping (SigInShop ) is consistent i,t with the above steps, the only difference is that SigInShop is computed at the level of i,t signing partner. 4.2.2. Auditor switching between headquarters and branch First, the data of internal control signing partners are collected from internal control audit reports, manually supplemented by the website of CICPA. We further identify whether the internal control signing partners belong to the headquarter office or a branch. The recognition results include four categories: (1) The two signing CPAs of internal control audit report are all come from the headquarters; (2) The two signing CPAs of internal control audit report are all come from the same branch; (3) The two signing CPAs of internal control audit report are come from headquarters and branch offices, respectively; (4) The two signing CPAs of internal control audit report are come from different branches. In order to make the study more precise, following Wang and Xin (2010) and Wang et al. (2016), we define the first category as headquarters auditor and the second category as branch auditor. 224 Y. YAO AND S. XUE Second, there are four directions for auditor switch between headquarters and branch: (1) from headquarters to a branch (TFInSwitch); (2) from a branch to a branch (FFInSwitch); (3) from a branch to headquarters (FTInSwitch) ; (4) from headquarters to headquarters (TTInSwitch). Last, we define the multivariate variable of MInSwitch (= 0, 1, 2, 3, 4): MInSwitch =1 i,t i,t when TFInSwitch =1; MInSwitch = 2 when FFInSwitch =1; MInSwitch = 3 when i,t i,t i,t i,t FTInSwitch =1; MInSwitch = 4 when TTInSwitch = 1. If there is no internal control i,t i,t i,t auditor switch, MInSwitch =0. i,t 4.2.3. Signing partner switch SigInSwitch is a proxy of signing partner switch. It is a dummy variable that equals 1 if two i,t signing partners switch in year t, and 0 otherwise. 4.2.4. Comment letters (1) Comment letters (CL ) : a dummy variable that equals 1 if the firm is issued i,t-1 a comment letter in the year t-1, and 0 otherwise. (2) Internal-control-related comment letters (InterCL ). A comment letter will be i,t-1 defined as internal-control-related when it includes one of more following items “internal control rectification”, “internal control defects”, “internal control major defects”, “internal control information disclosure fraud/imperfection”, “internal con- trol self-evaluation system is imperfect/absent”, “the effectiveness of internal control” and “internal control management is not standardized”. If a comment letter contains at least one of these items InterCL takes the value of 1, 0 otherwise. Similarly, if i,t-1 a comment letter contains none of these items NonInterCL takes the value of 1, 0 i,t-1 otherwise. (3) Audit-verified comment letters (AuditCKCL ), a dummy variable that equals 1 if i,t-1 auditor is required to issue professional opinions on some issues in the comment letter, and 0 otherwise. Similarly, we define NonAuditCKCL . i,t-1 4.3. Models To test H1, as Chen et al. (2016), we add the lag period of comment letters (CL ) on the i,t-1 base of model (2), the model is as follows: InSwitch ¼ β þβ InShop þβ InShop CL þβ CL þ β Z þ Industry i;t i;t i;t1 i;t1 i;t 0 1 2 i;t 3 m þYear þ μ (3) i;t In model (3), if comment letters can significantly decrease the behaviour of internal control opinion shopping, the coefficient of interaction terms (InShop ×CL ), β , should i,t i,t-1 2 be significantly positive. To test H2, as Cunningham et al. (2019), we construct model (4): InSwitch ¼ β þβ InShop þβ InShop InterCL þβ InShop NonInterCL i;t i;t i;t1 i;t1 0 1 2 i;t 3 i;t þβ InterCL þ β NonInterCL þ β Z þ Industry þ Year þ μ (4) i;t1 i;t1 i;t 4 5 m i;t CHINA JOURNAL OF ACCOUNTING STUDIES 225 According to H2, we expect β is significant larger than β , it indicates that the supervision 2 3 effect of comment letters on internal control opinion shopping is more pronounced in the samples of internal-control-related comment letters. To test H3, we use model (5): InSwitch ¼ β þβ InShop þβ InShop AuditCKCL þβ InShop NonAuditCKCL i;t i;t i;t1 i;t1 0 1 2 i;t 3 i;t þβ AuditCKCL þ β NonAuditCKCL þ β Z þ Industry þ Year þ μ (5) 4 i;t1 5 i;t1 m i;t i;t According to H3, we expect β is significant larger than β , it shows that when the 2 3 comment letters require auditors to issue their professional opinions, the supervision effect will be stronger. In model (3)–(5), Z is a set of control variables, which are the same as those for model i,t (2). We also control year and industry fixed effects. Model (3)–(5) is based on internal control audit firm switches. When considering the direction of auditor switches between headquarters and branches, we replace InSwitch with MInSwitch in the model (3)–(5), and adopt the multinomial non-ordinal logit model to test. In the multinomial non-ordinal logit model, the benchmark group is no auditor switch (MInSwitch = 0). According to i,t theoretical expectation, we expect that the supervision effect of comment letters on internal control opinion shopping is more pronounced in the group of headquarters- to-branch switches (MInSwitch =1). i,t When hypothesis are tested at the level of signing partner, we use SigInSwitch to i,t substitute for InSwitch in the model (3)–(5). The tested independent variable InShop i,t i,t should also be replaced by SigInShop . i,t The detailed definitions of all variables are shown in Table 1. Table 1. Definition of variables. Variables Definition InSwitch A dummy variable that equals 1 if a firm is audited by a new internal control audit firm in year t, and i,t 0 otherwise TFInSwitch A dummy variable that equals 1 if a internal control audit firm switches from a headquarters to i,t a branch office, and 0 otherwise FFInSwitch A dummy variable that equals 1 if a internal control audit firm switches from a branch office to i,t another branch office, and 0 otherwise FTInSwitch A dummy variable that equals 1 if a internal control audit firm switchs from a branch office to i,t a headquarters, and 0 otherwise TTInSwitch A dummy variable that equals 1 if a internal control audit firm switches from a headquarters to i,t another headquarters, and 0 otherwise MInSwitch MInSwitch =0,1, 2,3,4.Ifa firm switches its internal control auditor form a headquarters to i,t i,t a branch, MInSwitch =1;Ifa firm switches its internal control auditor form a branch to a branch, i,t MInSwitch =2;If a firm switches its internal control auditor form a branch to a headquarters, i,t MInSwitch =3;Ifa firm switches its internal control auditor form a headquarters to another i,t headquarter, MInSwitch = 4; If no auditor switch, MInSwitch =0. i,t i,t SigInSwitch A dummy variable that equals 1 if two signing CPAs of internal control audit report switch, and 0 i,t otherwise. InMAO A dummy variable that equals 1 if a firm receives a modified internal control opinion in year t, and 0 i,t otherwise InMAO A dummy variable that equals 1 if a firm receives a modified internal control opinion in year t-1, and i,t-1 0 otherwise CL A dummy variable that equals 1 if the firm received a annual report comment letter in the year t-1 i,t-1 and zero otherwise. (Continued) 226 Y. YAO AND S. XUE Table 1. (Continued). Variables Definition InterCL A dummy variable that equals 1 if a firm receives an internal control-related comment letter in i,t-1 the year t-1, and 0 otherwise. NonInterCL A dummy variable that equals 1 if a firm receives a non-internal control-related comment letter in i,t-1 the year t-1, and 0 otherwise. AuditCKCL A dummy variable that equals 1 if a firm receives an audit-verified comment letter in the year t-1, i,t-1 and 0 otherwise. NonAuditCKCL A dummy variable that equals 1 if a firm receives a non-audit-verified comment letter in the year t-1, i,t-1 and 0 otherwise. CLQnum Numbers of questions in comment letters in the year t-1 i,t-1 NumCL The cumulative number of comment letters within the year t-1 i,t-1 CNumCL The cumulative number of years in which a firm receives continuously comment letters by the end i,t-1 of year t-1 FWordsCL Ln (1+ total number of words in comment letters) in the year t-1 i,t-1 RWordsCL Ln (1+ total number of words in comment letters responses) in the year t-1 i,t-1 LnSize Firm size, measured as the natural logarithm of total assets in the year t i,t Lev Financial leverage, measured as total liabilities divided by total assets in the year t i,t ROA Return on assets, measured as net income divided by average total assets in the year t i,t SOE A dummy variable that equals 1 if a firm is a state owned enterprise, and 0 otherwise i,t Restatement A dummy variable that equals 1 if a firm has restatements in the year t, and 0 otherwise i,t BankRisk Based on the Z-Value Model of Altman (1968) to estimate the bankruptcy risk of the firm in the year t i,t Growth Growth rate, measured as percentage change in annual sales from year t-1 to year t i,t Abs_DA The absolute value of discretionary accruals, measured from the performance-matched Jones model i,t (Kothari, Leone, & Wasley, 2005) MA A dummy variable that equals 1if a firm has the behaviour of acquisitions or mergers, and 0 i,t otherwise in the year t Dirs the proportion of shares owned by the management team in the year t i,t Fshare Shareholding concentration, measured as the proportion of shares owned by the largest i,t shareholder in the year t InvRec Inventory plus receivables divided by total assets in the year t i,t GC A dummy variable that equals 1 if a firm receives a modified audit opinion for its financial reports i,t-1 in year t-1, and 0 otherwise Big4 A dummy variable indicating an international Big 4 auditor in the year t i,t CEFchange A dummy variable that equals 1 if a firm hires a new CEO or CFO in year t i,t Punishment A dummy variable that equals 1 if a firm is punished by the exchanges or CSRC in year t-1, and 0 i,t-1 otherwise Year Year effects Industry Industry effects 5. Empirical results 5.1. Descriptive statistics Panel A and B in Table 2 report the year and industry distributions of comment letters, respectively. Panel A shows that comment letters are gradually increasing from 2015 to 2017 in Chinese capital market. Panel B shows the distribution of comment letters by industry and about 68.89% (952/1382) of comment letters are issued to manufacturing firms. Panel C of Table 2 presents the characteristics of comment letters. The proportion of internal- control-related comment letters is 22.14% (306/1382). Most firms receive one comment letter in the same year, 95 firms receive two comment letters and 3 firms receive three comment letters within one year. 297 firms receive comment letters for two consecutive years and 85 firms receive comment letters for three consecutive years. Table 3 provides the summary statistics for main variables. As shown, the mean of InMAO is 0.034, that is, about 3.4% listed firmsinour samplesreceive modified internal control opinions. CHINA JOURNAL OF ACCOUNTING STUDIES 227 The mean of InSwitch is 0.077, suggesting that there are about 7.7% samples change their internal control auditors, in which the mean of TFInSwitch (headquarters-to-branch switch) is 0.033, the mean of FFInSwitch (branch-to-branch switch) is 0.027, the mean of FTInSwitch t t (branch-to-headquarters switch) is 0.011, and the mean of TTInSwitch (headquarters-to- headquarters switch) is 0.006. That is, headquarter-to-branch switch is the most common. The mean of SigInSwitch is 0.190, indicating that there are 19% samples change their internal control signing partner. The mean of CL is 0.183, indicating that 18.3% samples are inquired by the t-1 Shanghai or Shenzhen Stock Exchanges. The mean of InterCL is 0.041, indicating t-1 that 4.1% listed firms receive internal-control-related comment letters. The mean of AuditCKCL is 0.075, suggesting that 7.5% samples receive auditor-verified com- t-1 ment letters. The largest number of comment letter questions (CLQnum ) is 37. t-1 Table 2. Distribution of comment letters. Panel A : Distribution of comment letters by year Year N 2015 286 2016 531 2017 565 Total 1382 Panel B: Distribution of comment letters by industry Industries A. Agriculture, forestry, livestock farming and fishery 47 B. Mining 35 C. Manufacturing 952 D. Electricity, heat, gas and water production and supply 22 E. Constructions 49 F. Wholesale and retail businesses 57 G. Transportation, warehousing and postal services 16 H. Hotels and catering services 3 I. Information transmission,software and information technology service 89 K. Real estate 44 L. Leasing and business services 18 M. Scientific research and technology service 6 N. Water conservancy, environment and public facilities management 14 P. Education 0 Q. Health and social work 4 R. Culture, sports and entertainment 16 S. Comprehensive 10 Total 1382 Panel C: Characteristics of comment letters (1) Internal control related or not Internal control-related comment letters 306 Non-internal control-related comment letters 1076 Total 1382 (2) The number of comment letters within one year One comment letter 1284 Two comment letters 95 There comment letters 3 Total 1382 (3) Continuity of comment letters Comment letters received only in one year 1000 Comment letters for two consecutive years 297 Comment letters for three consecutive years 85 Total 1382 Following Chen et al. (2016), when studying internal control opinion shopping at the level of signing partner, samples of audit firm switch (580 observations) and mandatory rotation of signing partners (58 observations) should be excluded. 228 Y. YAO AND S. XUE Table 3. Descriptive statistics. Variables N Mean Std. Dev. Min Median Max InMAO 7544 0.034 0.180 0 0 1 InMAO 7544 0.032 0.175 0 0 1 t-1 7544 0.077 0.266 0 0 1 InSwitch TFInSwitch 7544 0.033 0.178 0 0 1 FFInSwitch 7544 0.027 0.163 0 0 1 FTInSwitch 7544 0.011 0.103 0 0 1 7544 0.006 0.080 0 0 1 TTInSwitch MInSwitch 7544 0.145 0.566 0 0 4 SigInSwitch 6906 0.190 0.392 0 0 1 CL 7544 0.183 0.387 0 0 1 t-1 7544 0.041 0.197 0 0 1 InterCL t-1 NonInterCL 7544 0.142 0.350 0 0 1 t-1 AuditCKCL 7544 0.075 0.263 0 0 1 t-1 NonAuditCKCL 7544 0.108 0.311 0 0 1 t-1 7544 1.474 3.556 0 0 37 CLQnum t-1 NumCL 7544 0.197 0.431 0 0 3 t-1 CNumCL 7544 0.112 0.496 0 0 3 t-1 FWordsCL 7544 1.279 2.715 0 0 8.299 t-1 7544 1.629 3.451 0 0 10.385 RWordsCL t-1 LnSize 7544 22.28 1.255 19.87 22.14 26.09 Lev 7544 0.426 0.206 0.059 0.414 0.911 ROA 7544 0.041 0.055 −0.159 0.037 0.206 7544 0.346 0.476 0 0 1 SOE Restatement 7544 0.101 0.300 0 0 1 BankRisk 7544 2.621 1.929 −0.651 2.172 11.170 Growth 7544 0.243 0.610 −0.564 0.121 4.310 7544 0.069 0.066 0.001 0.049 0.336 Abs_DA MA 7544 0.710 0.454 0 1 1 Dirs 7544 0.069 0.133 0 0.002 0.593 Fshare 7544 0.336 0.145 0.083 0.316 0.731 7544 0.264 0.166 0.008 0.243 0.730 InvRec GC 7544 0.030 0.171 0 0 1 t-1 Big4 7544 0.054 0.225 0 0 1 CEFchange 7544 0.250 0.432 0 0 1 7544 0.105 0.306 0 0 1 Punishment t-1 5.2. Multiple regression analysis 5.2.1. Comment letters and internal control opinion shopping (H1) Table 4 reports the impact of comment letters on internal control opinion shopping. First, the comment letters is not considered in column (1), and the regression coefficient on InShop is significantly negative at 1% level, indicating that listed firms have successfully engaged in internal control opinion shopping. This result lays the foundation for the following tests. Second, to test the impact of comment letters on internal control opinion shopping, an interaction term of CL and InShop is added and the regression result is shown in t-1 columns (2) of Table 4.The regression coefficient on InShop×CL is significantly positive t-1 at 1% level, indicating that the pressure from comment letters can significantly decrease internal control opinion shopping. Considering the directions of auditor switches between headquarters and branch, the results of multinomial non-ordinal logit model are shown in columns (3)–(6), the coeffi- cient on InShop×CL is significantly positive in the headquarters-to-branch group t-1 (MInSwitch = 1) and branch-to-branch group (MInSwitch = 2). While the regression i,t i,t coefficients on InShop×CL are not significant in the groups switching to headquarters t-1 (MInSwitch = 3 and MInSwitch = 4).The coefficient of InShop×CL in headquarters-to- i,t i,t t-1 CHINA JOURNAL OF ACCOUNTING STUDIES 229 Table 4. Comment letters and internal control opinion shopping (H1). Audit firm switch Auditors switch between headquarters and branch Signing partner switch InSwitch InSwitch MInSwitch = 1 MInSwitch = 2 MInSwitch = 3 MInSwitch = 4 SigInSwitch SigInSwitch Variables (1) (2) (3) (4) (5) (6) (7) (8) InShop −4.718*** −7.151*** −8.842*** −9.374*** −2.494 −11.390*** (−3.32) (−3.89) (−5.18) (−4.77) (−0.98) (−2.81) SigInShop −1.614** −2.929*** (−2.00) (−2.89) InShop×CL 5.146*** 7.254*** 4.491* 4.718 0.985 t-1 (2.63) (3.18) (1.88) (1.36) (0.23) SigInShop 3.679** ×CL (2.20) t-1 CL 0.446*** 0.611*** 0.361* 0.583* −0.241 0.012 t-1 (3.25) (3.17) (1.72) (1.90) (−0.43) (0.12) LnSize −0.132*** −0.128** −0.109 −0.095 −0.188 0.063 0.007 0.006 (−2.65) (−2.53) (−1.62) (−1.23) (−1.59) (0.44) (0.21) (0.17) Lev −0.179 −0.226 0.124 −0.394 −1.291 0.033 −0.398 −0.415 (−0.40) (−0.50) (0.20) (−0.59) (−1.23) (0.02) (−1.29) (−1.38) ROA 0.151 0.581 2.575* 0.313 0.897 1.496 0.716 0.753 (0.15) (0.58) (1.94) (0.21) (0.44) (0.50) (1.03) (1.06) SOE 0.191* 0.211* 0.353** −0.317* 0.301 0.927** −0.041 −0.048 (1.74) (1.90) (2.20) (−1.73) (1.03) (2.52) (−0.51) (−0.58) Restatement 0.051 0.004 −0.171 0.220 0.207 −0.488 0.111 0.134 (0.35) (0.02) (−0.77) (1.04) (0.62) (−0.88) (1.09) (1.33) BankRisk −0.007 −0.007 −0.000 −0.001 −0.080 0.097 −0.046 −0.047* (−0.17) (−0.16) (−0.01) (−0.02) (−0.72) (0.71) (−1.58) (−1.72) Growth 0.274*** 0.260*** 0.058 0.109** −0.083 0.096 0.103* 0.099* (4.46) (4.07) (1.28) (2.30) (−0.83) (1.16) (1.88) (1.75) Abs_DA 1.719*** 1.649** 2.275** −0.017 4.251*** 4.037** 0.222 0.201 (2.70) (2.58) (2.54) (−0.02) (3.00) (2.18) (0.44) (0.41) MA 0.175* 0.164 0.173 0.050 0.130 0.610* 0.074 0.075 (1.68) (1.56) (1.14) (0.30) (0.48) (1.73) (1.02) (1.04) Dirs −0.936** −0.857* −0.808 −1.394** 0.152 0.379 0.066 0.082 (−2.08) (−1.90) (−1.20) (−2.01) (0.15) (0.27) (0.27) (0.32) Fshare −0.068 −0.058 −0.227 −0.075 −0.797 1.383 0.296 0.281 (−0.20) (−0.17) (−0.46) (−0.14) (−0.90) (1.31) (1.29) (1.23) InvRec 0.211 0.213 −0.501 0.776 1.407* −0.343 0.002 −0.004 (0.62) (0.63) (−1.07) (1.54) (1.86) (−0.33) (0.01) (−0.02) (Continued) 230 Y. YAO AND S. XUE Table 4. (Continued). Audit firm switch Auditors switch between headquarters and branch Signing partner switch InSwitch InSwitch MInSwitch = 1 MInSwitch = 2 MInSwitch = 3 MInSwitch = 4 SigInSwitch SigInSwitch Variables (1) (2) (3) (4) (5) (6) (7) (8) GC 0.745*** 0.687*** 0.509 0.080 1.741*** 0.844 −0.226 −0.239 t-1 (3.07) (2.85) (1.64) (0.22) (4.38) (1.11) (−1.03) (−0.99) Big4 1.151*** 1.189*** 1.314*** 0.361 2.247*** 0.863 −0.244 −0.229 (6.81) (7.02) (5.61) (0.97) (6.21) (1.55) (−1.47) (−1.42) CEFchange 0.438*** 0.425*** 0.372*** 0.500*** 0.497** 0.472 −0.105 −0.107 (4.65) (4.50) (2.65) (3.31) (2.07) (1.50) (−1.41) (−1.44) Punishment 0.307** 0.264** 0.194** 0.371* 0.237 −0.099 0.216** 0.238** t-1 (2.31) (2.00) (1.99) (1.86) (0.72) (−0.21) (2.20) (2.45) Year&Industry Yes Yes Yes Yes Yes Yes Yes Yes Constant −0.406 −0.646 −1.998 −2.203 −15.319 −9.626*** −1.988** −1.947** (−0.37) (−0.57) (−1.33) (−1.27) (−0.01) (−2.96) (−2.50) (−2.41) Wald Chi2 240.703*** 236.271*** 393.834*** 65.370*** 72.221*** Pseudo R 0.059 0.063 0.071 0.010 0.011 N 7544 7544 7544 6906 6906 Difference (p-value) (3) vs (4): Chi2 = 32.25* (p = 0.055) (3) vs (5): Chi2 = 35.65** (p = 0.024) (3) vs (6): Chi2 = 30.35* (p = 0.085) (4) vs (5): Chi2 = 45.33*** (p = 0.001) (4) vs (6): Chi2 = 36.81** (p = 0.018) (5) vs (6): Chi2 = 39.31*** (p = 0.009) Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. CHINA JOURNAL OF ACCOUNTING STUDIES 231 branch group is significantly higher than the other three groups in column (4)–(6). These results show that the supervision effect of comment letters on internal control opinion shopping is more pronounced in the subsamples of headquarters-to-branch switch. Third, the column (7) of Table 4 presents the existence of internal control opinion shopping at the level of signing partner, and the regression coefficient on SigInShop is significantly negative at 5% level, suggesting that listed firms have successfully engaged in internal control opinion shopping at the level of signing partners. These result is similar with Chen et.al (2016). The regression coefficient on SigInShop×CL is significantly t-1 positive at 5% level in column (8), indicating the supervision effect of comment letters on internal control opinion shopping also exists in the level of signing partners. In summary, empirical results at audit firm level, headquarters and branch level and signing partner level are all consistent with expectation in hypothesis 1. 5.2.2. Internal-control-related comment letters and internal control opinion shopping (H2) According to H2, we expect that the supervision effect of comment letters on internal control opinion shopping is more pronounced in the samples of internal-control-related comment letters. Table 5 reports the regression results. Column (1) presents the results at audit firm level, the regression coefficient of interaction term (InShop ×InterCL )issignificantly positive at i,t i,t-1 1% level while the regression coefficient of interaction terms (InShop ×NonInterCL )isnot i,t i,t-1 significant, indicating that when firms receive internal-control-related comment letters, the supervision effect of comment letters on internal control opinion shopping is stronger. When the switch directions between headquarters and branch are considered, the results are shown in column (2)–(5). In headquarters-to-branch group (MInSwitch = 1) and branch-to- branch group (MInSwitch =2), theregressioncoefficients of interaction terms (InShop × i,t InterCL )are both significantly positive at 1% level. The regression coefficient on InShop × i,t-1 i,t InterCL is significant at 10% level in branch-to-headquarters group (MInSwitch =3)and not i,t-1 significant in headquarter-to-headquarter group (MInSwitch =4). When thecoefficients of Table 5. Internal-control-related comment letters and internal control opinion shopping (H2). Audit firm Signing partner switch Auditors switch between headquarters and branch switch MInSwitch = MInSwitch = MInSwitch = MInSwitch = InSwitch 1 2 3 4 SigInSwitch Variables (1) (2) (3) (4) (5) (6) InShop −7.639*** −8.698*** −8.659*** −2.320 −11.998*** (−3.96) (−4.94) (−4.33) (−0.86) (−2.84) SigInShop −3.023*** (−2.95) InShop×InterCL 9.838*** 11.827*** 9.616*** 8.431* −0.225 t-1 (3.71) (3.33) (2.59) (1.92) (−0.03) InShop×NonInterCL 1.960 4.045 0.245 −0.057 0.958 t-1 (0.92) (1.55) (0.09) (−0.01) (0.19) SigInShop×InterCL 5.022** t-1 (2.38) SigInShop 2.791 ×NonInterCL (1.37) t-1 InterCL 0.818*** 0.858*** 0.736** 1.263*** −0.170 0.336** t-1 (3.48) (2.63) (2.14) (2.83) (−0.15) (2.04) NonInterCL 0.270* 0.470** 0.130 0.220 −0.234 −0.084 t-1 (1.82) (2.19) (0.54) (0.60) (−0.38) (−0.76) (Continued) 232 Y. YAO AND S. XUE Table 5. (Continued). Audit firm Signing partner switch Auditors switch between headquarters and branch switch MInSwitch = MInSwitch = MInSwitch = MInSwitch = InSwitch 1 2 3 4 SigInSwitch Variables (1) (2) (3) (4) (5) (6) Controls Yes Yes Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Yes Yes Constant −0.572 −1.517 −1.092 −13.907 −9.875*** −1.940** (−0.51) (−1.01) (−0.63) (−0.01) (−2.96) (−2.40) Wald Chi2 248.345*** 426.219*** 75.940*** Pseudo R 0.065 0.077 0.011 N 7544 7544 6906 Difference (2) vs (3): Chi2 = 34.34* (p = 0.060) (p-value) (2) vs (4): Chi2 = 34.81* (p = 0.054) (2) vs (5): Chi2 = 33.66* (p = 0.070) (3) vs (4): Chi2 = 42.21*** (p = 0.008) (3) vs (5): Chi2 = 41.39** (p = 0.011) (4) vs (5): Chi2 = 39.20** (p = 0.018) Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. interaction terms (InShop × InterCL ) among four groups are compared, the coefficient in i,t i,t-1 headquarters-to-branch group is significantly larger than those in the other three groups. Finally, the result at signing partner level is in column (6), the regression coefficient of interaction terms (SiglnShop × InterCL ) is significantly positive at 5% level. i,t i,t-1 To conclude, Hypothesis 2 is verified at all three dimensions. 5.2.3. Auditors-verified comment letters and internal control opinion shopping (H3) Table 6 reports the impact of auditors-verified comment letters on internal control opinion shopping. Results at audit firm level are shown in the column (1) of Table 6. The regression coefficient of interaction terms (InShop × AuditCKCL ) is significantly i,t i,t-1 positive at 1% level while the coefficient of interaction terms (InShop × NonAuditCKCL ) i,t i,t-1 is not significant, suggesting that when the comment letters require auditors to issue their professional opinions, the supervision effect of comment letters is stronger. Then, the switch directions between headquarters and branch are considered in column (2)–(5). In headquarters-to-branch group (MInSwitch = 1) and branch-to-branch group (MInSwitch =2), the regression coefficients of interaction terms (InShop ×AuditCKCL ) i,t i,t-1 are significantly positive at 1% and 5% level, respectively. The regression coefficient is significant at 5% level in branch-to-headquarters group (MInSwitch = 3) and not significant in headquarters-to-headquarters group (MInSwitch = 4). Similarly, with the results in Table 5, when the coefficients of interaction terms (InShop ×InterCL ) among four groups are i,t i,t-1 compared, the coefficient in headquarters-to-branch group is significantly larger than those in the other three groups. Lastly, the result at signing partner level is in column (6), the regression coefficient of interaction term (SiglnShop ×AuditCKCL )issignificantly positive at 5% level. i,t i,t-1 To sum up, hypothesis 3 are proved in all three dimensions. CHINA JOURNAL OF ACCOUNTING STUDIES 233 Table 6. Auditors-verified comment letters and internal control opinion shopping (H3). Audit firm Signing partner switch Auditors switch between headquarters and branch switch MInSwitch = MInSwitch = MInSwitch = MInSwitch = InSwitch 1 2 3 4 SigInSwitch Variables (1) (2) (3) (4) (5) (6) InShop −7.393*** −8.333*** −8.426*** −2.358 −11.947*** (−3.94) (−4.81) (−4.24) (−0.87) (−2.86) SigInShop −2.946*** (−2.90) InShop× 6.948*** 7.561*** 5.423** 4.898** 2.476 AuditCKCL (3.24) (2.86) (1.99) (2.02) (0.51) t-1 InShop× 2.014 5.662 1.039 −4.618 −1.857 NonAuditCKCL (0.75) (1.58) (0.28) (−0.88) (−0.29) t-1 SigInShop× 4.683** AuditCKCL (2.21) t-1 SigInShop× 2.294 NonAuditCKCL (1.05) t-1 AuditCKCL 0.729*** 0.654** 0.634** 1.248*** 0.218 0.054 t-1 (3.82) (2.42) (2.29) (3.38) (0.31) (0.39) NonAuditCKCL 0.190 0.541** 0.057 −0.300 −0.715 −0.030 t-1 (1.11) (2.30) (0.20) (−0.60) (−0.83) (−0.26) Controls Yes Yes Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Yes Yes Constant −0.632 −1.554 −1.177 −14.065 −9.683*** −1.947** (−0.56) (−1.03) (−0.67) (−0.01) (−2.93) (−2.41) Wald Chi2 247.236*** 425.594*** 73.740*** Pseudo R 0.064 0.077 0.011 N 7544 7544 6906 Difference (2) vs (3): Chi2 = 35.09* (p = 0.051) 4 (p-value) (2) vs (4): Chi2 = 39.83** (p = 0.020) (2) vs (5): Chi2 = 30.45* (p = 0.083) (3) vs (4): Chi2 = 44.12***(p = 0.005) (3) vs (5): Chi2 = 40.02** (p = 0.015) (4) vs (5): Chi2 = 39.31** (p = 0.018) Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard rrors are clustered by firms. 6. Additional tests and robustness checks 6.1. The timeliness of comment letters and internal control opinion shopping Since internal control opinion shopping is closely related to internal control auditors switches, we use the timing of auditors switches to investigate the timeliness effect of comment letters on internal control opinion shopping. Most auditors switches (auditor hiring decisions) occur before the end of financial year (31 December), while others occur between the beginning of the next financial year and audit report date. Out of 580 samples of internal control auditor switches, 465 samples belong to the former, and 115 belong to the later. Combining the date of comment letters issuing with the timing of auditors switches, we can measure the timeliness effect of comment letters on internal control opinion shopping. For example, a firm receives a comment letter on 20 May 2016, it is possible for the firm to switch auditor between 20 May 2016 and the annual report date (for example, 30 April 2017). If a firm switches auditor before 31 December 2016, that is, in the same calendar year the firm receives the 234 Y. YAO AND S. XUE comment letter, the supervision effect of comment letters should be greater than that of auditor switch between 31 December 2016 and 30 April 2017. We define a ternary variable (TimInSwitch = 0,1,2) which is set 0 when no internal control auditor switch, 1 when internal control auditor switch happens in the issuing year of comment letters, 2 when internal control auditor switch happens in the next year but before the annual report date. We run multinomial non-ordinal logit model to test the difference between three groups and the benchmark group is TimInSwitch =0. The regression results are reported in columns (1)–(2) of Table 7. In column (1) (TimInSwitch = 1), the regression coefficient on InShop ×CL is significantly positive at i,t i,t-1 1%, while the regression coefficient on InShop ×CL is not significant in the column (2) i,t i,t-1 where TimInSwitch = 2. These results suggest that the supervision effect of comment letters on internal control opinion shopping exists only in the short run. Further, we consider the impact of audit delay (i.e. The calendar days between balance sheet date and audit report date). Audit delay measures the time spent by auditors to complete audit work and it also reflects clients’ audit risk to a certain extent (Pizzini, Lin, & Ziegenfuss, 2015). We compute Ln(1+ the calendar days between balance sheet date and audit report date) to measure audit delay, and divide the samples into long audit delay group and short audit delay group based on its median. Combine timing of auditor’s switch and audit delay we get four groups and then regress model (3) for each group, respectively. Theresults areshown in columns(3)–(6) of Table 7. Out of the four groups, the coefficient on InShop ×CL is significantly positive only in long-audit-delay and earlier-auditor-switch i,t i,t-1 group. Overall, results in Table 7 show that the supervision effect of comment letters exists in a short run and for the high audit risk clients. Table 7. The timeliness of comment letters and internal control opinion shopping. Considering the influence of audit delay Earlier Auditor Later Auditor switch switch Long audit delay Short audit delay TimInSwitch TimInSwitch TimInSwitch TimInSwitch TimInSwitch =1 TimInSwitch =2 =1 =2 =1 =2 Variables (1) (2) (3) (4) (5) (6) InShop −9.198*** −1.259 −6.899*** −1.210 −14.900*** −3.666 (−6.70) (−0.57) (−4.02) (−0.49) (−6.12) (−0.61) InShop×CL 5.894*** 1.737 6.679*** 1.755 3.669 4.534 t-1 (3.41) (0.60) (2.95) (0.55) (1.34) (0.42) CL 0.479*** 0.321 0.624*** 0.351 0.207 0.243 t-1 (3.25) (1.28) (3.14) (1.25) (0.88) (0.36) Controls Yes Yes Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Yes Yes Constant −2.220* 1.644 −2.592 2.023 −3.172* 1.315 (−1.85) (0.75) (−1.52) (0.79) (−1.70) (0.29) Wald Chi2 342.03*** 217.77*** 195.89*** Pseudo R 0.073 0.087 0.092 N 7544 3717 3827 Difference Chi2 = 66.38*** (p = 0.001) Chi2 = 36.74** (p = 0.018) Chi2 = 32.05 (p = 0.656) (p-value) Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. CHINA JOURNAL OF ACCOUNTING STUDIES 235 6.2. The intensity of comment letters on internal control opinion shopping The contents and frequency among comment letters are different. Some include more questions or words than the others. Some firms receive one comment letter while others receive several comment letters within one year. Higher intensity of comment letters implies that there are more defects or serious defects in information disclosure. Higher intensity of comment letters will catches more attention from regulators, investors and auditors. The risk of opinion shopping is higher not only for auditors but also for managers. The higher the intensity of comment letters, the stronger the supervision effect of comment letters on internal control opinion shopping. We measure the intensity of comment letters by five variables: the number of questions in comment letters (CLQnum ), the frequency of comment letters within one year i, t-1 (NumCL ), the number of consecutive years receiving comment letters (CNumCL ), i, t-1 i, t-1 total number of words in comment letters (FWordsCL ) and total number of words in i,t-1 responses (RWordsCL ). For each intensity variable, we set Group_H (Group_L )equals i,t-1 i,t-1 i,t-1 1 when intensity variable is greater than or equal to (less than) its median and 0, otherwise. Take the number of questions in comment letters (CLQnum )asanexample.When i,t-1 CLQnum is greater than or equal to its median, Group_H is equal to one, 0 otherwise. i, t-1 i,t-1 When CLQnum is less than its median, Group_L is equal to one, 0 otherwise. As for the i, t-1 i,t-1 variables of FWordsCL and RWordsCL , which are the same as those for the above i,t-1 i,t-1 definitions. As for the variables of NumCL , when the cumulative number of comment letters i, t-1 within year t-1 is more than one, Group_H is equal to one, 0 otherwise; when the cumulative i,t-1 number of consecutive inquiry year is equal to one, Group_L is set equal to one, 0 otherwise. i,t-1 The Group_H (or Group_L ) variables of CNumCL is thesameas NumCL . i,t-1 i,t-1 i, t-1 i, t-1 Table 8. The intensity of comment letters on internal control opinion shopping. Numbers of questions Numbers of CL within Consecutive years Words in Words in CL of CL one year receiving CL CL responses InSwitch InSwitch InSwitch InSwitch InSwitch Variables (1) (2) (3) (4) (5) InShop −7.297*** −7.382*** −7.121*** −7.626*** −7.371*** (−3.90) (−3.86) (−3.86) (−4.02) (−3.94) InShop×Group_H 6.479*** 13.640*** 7.333*** 7.874*** 7.194*** t-1 (3.10) (3.68) (3.08) (3.73) (3.24) InShop×Group_L 1.240 3.831* 4.022* −2.146 1.818 t-1 (0.39) (1.88) (1.83) (−0.65) (0.69) Group_H 0.606*** 1.320*** 0.710*** 0.800*** 0.786*** t-1 (3.69) (3.77) (3.59) (4.52) (4.53) Group_L 0.165 0.350** 0.334** −0.049 0.042 t-1 (0.82) (2.44) (2.21) (−0.26) (0.22) Controls Yes Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Yes Constant −0.583 −0.772 −0.674 −0.661 −0.503 (−0.52) (−0.69) (−0.60) (−0.59) (−0.44) Wald Chi2 242.201*** 47.219*** 241.288*** 255.658*** 250.769*** PseudoR 0.064 0.066 0.063 0.067 0.065 N 7544 7544 7544 7544 7544 Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. 236 Y. YAO AND S. XUE Table 8 reports the impact of intensity of comment letters on internal control opinion shopping. The results show that the coefficients on InShop×Group_H are positively sig- t-1 nificant in all five regressions and are much larger than those of InShop×Group_L .These t-1 findings suggest that the greater the intensity of comment letters, the stronger the govern- ance effect of comment letters on the behaviour of internal control opinion shopping. 6.3. The spillover effects of comment letters on internal control opinion shopping In previous sections, we investigate the direct supervision effect of comment letters. Does the comment letters have a deterrent spillover effect? We test it for the firms in the same industry, in the same province and with the common auditors. (1) Peers in same industry. Firms in the same industry are natural peers and they face the same economic policies, the same market competitive environment (Kedia, Koh, & Rajgopal, 2015) and the same regulatory shocks. The industrial peers are usually taken as reference or benchmark by regulators. (2) Peers in the same province. Firms located in the same geographic area can also be regarded as peers. Prior studies document that geographic proximity is associated with informational advantages in portfolio decisions (Coval & Moskowitz, 2001). For example, firms in the same area are likely to be covered extensively in local newspapers, making them aware of each other (Kedia et al., 2015). Given these findings, we expect comment letters may have a spillover effect on the behaviour of internal control opinion shopping of peers in the same province. (3) The common auditors. Brown and Knechel (2016) show that the quality of account- ing information with the same auditor are more similar than those firms with different auditors. When one of the clients got comment letters, the perceived audit risk may be transferred to the other clients with the same auditor. Following Kedia et al. (2015) and Xue, Ru, and Dou (2017), We define the exposure of comment letters in the same industry (ExpoInd_CL ) and the exposure of comment t-1 letters in the same province (ExpoProv_CL ): The former is the number of other firms t-1 receiving comment letters in the same industry divided by the total number of firms in the same industry; the latter is the number of other firms receiving comment letters in the same province divided by the total number of firms in the same province. Common audit firm (ComAudit_CL ) is a dummy variable that equals 1 if a non- t-1 comment-letter firm shares the same audit firm with at least one comment-letter firm in year t, and 0 otherwise. Similarly, common signing partner (ComSigner_CL )is t-1 a dummy variable that equals 1 if a non-comment-letter firm shares the same signing partner with at least one comment-letter firm in year t, and 0 otherwise. Table 9 reports the spillover effects of comment letters on internal control opinion shop- ping. The tests for peers in the same industry and province are present in column (1) and (2) respectively. Both the coefficients on InShop×ExpoInd_CL and InShop×ExpoProv_CL are t-1 t-1 significantly positive at 10% level. Results of spillover effect for peers with common audit firm and signing partners are reported in column (3) and (4) respectively. The coefficient on CHINA JOURNAL OF ACCOUNTING STUDIES 237 Table 9. The spillover effects of comment letters on internal control opinion shopping. Common auditors Same industry Same province Common audit firm Common signing partner InSwitch InSwitch InSwitch InSwitch Variables (1) (2) (3) (4) InShop −5.784*** −5.477*** −10.986*** −9.843*** (−2.83) (−4.95) (−6.39) (−3.10) InShop 13.512* ×ExpoInd_CL (1.77) t-1 InShop 7.722* ×ExpoProv_CL (1.89) t-1 InShop 3.400* ×ComAudit_CL (1.91) t-1 InShop 2.646** ×ComSigner_CL (2.21) t-1 ExpoInd_CL 1.219 t-1 (1.61) ExpoProv_CL 0.452 t-1 (0.73) ComAudit_CL −0.221* t-1 (−1.70) ComSigner_CL −0.037 t-1 (−0.20) Controls Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Constant −0.331 −0.369 −0.828 −0.815 (−0.29) (−0.48) (−0.66) (−0.60) Wald Chi2 65.29*** 179.75*** 206.86*** 180.59*** Pseudo R 0.045 0.057 0.066 0.062 N 6162 6162 6162 6162 Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. The reasons for sample reduction are as follows: the total samples in this section does not include comment-letters firms (7544–1382 = 6162). InShop×ComAudit_CL and InShop×ComSigner_CL are significantly positive at 10% and 5% t-1 t-1 level, respectively. Overall, the supervision effect of comment letters on internal control opinion shopping has spillover effects for peers in the same industry, in the same province or with the common auditor. 6.4. Robustness tests 6.4.1. Addressing the endogeneity issue (1) Difference-in-Differences Analyses To mitigate the endogeneity problem arising from the causal relationship between comment letters and the behaviour of internal control opinion shopping, In the first stage of PSM, the dependent variable is Letter (i.e. Letter equals 1 if the firm receives a comment letter in year t t t, and 0 otherwise). Following Cassell et al. (2013), we control firm size (LnSize ), financial leverage (Lev ), the presence of t t a net loss (Loss ), the announcement of a restatement (Restatement ), litigation status (Litigation ), internal control audit t t t opinion (InMAO ), CFO_BOD (i.e. an indicator variable that equals 1 if the CFO is also on the board of directors, and 0 t t otherwise), bankruptcy risk (BankRisk ), being punished by the exchanges or CSRC (Punishment ), growth rate of revenue t t (Growth ), merger and acquisitions (MA ), firm age (LnAge ), IndepRatio (i.e. the proportion of independent directors on t t t t the board), CEO_Chair (i.e. an indicator variable that equals 1 if the CEO is also the chairman of the board of directors, and 0 otherwise), CEO or CFO change (CEFchange ), internal control audit firm switching (InSwitch ), big4 audit firms t t (Big4 ), year and industry fixed effects. t 238 Y. YAO AND S. XUE we follow previous studies (Cunningham et al., 2019; Johnston & Petacchi, 2017) and employ a difference-in-differences (DID) method to address the potential endogeneity. Specifically, we expand the sample period from 2013 to 2017, which can better compare the changes pre- and post-comment letters between the comment-letter firms and non- comment-letter firms. First, we employ the propensity score matching method based on the nearest neighbour matching principle to construct the experimental group and control group. Second, setting the key explanatory variables in DID model: Treat equals 1 for the treatment firms (i.e. firms that receive comment letters) and 0 for control firms. For both treatment firms and the matched control firms, we define Post, equals to 1 in the year following the receipt of comment letters, and 0 otherwise. Finally, we construct the following DID model: InSwitch ¼ β þβ InShop þβ InShop Post Treat þβ Post Treat i;t i;t i;t i;t i;t i;t 0 1 2 i;t 3 þ β InShop Treat þ β InShop Post þ β Treat þ β Post þ β Z i;t i;t i;t i;t i;t i;t i;t 4 5 6 7 m þ Industry þ Year þ μ i;t (6) Table 10. Comment letters and internal control opinion shopping: control endogeneity. PSM-DID All firm-years during 2013–2017 [−1,+1] around comment-letters year Heckman two-step InSwitch InSwitch InSwitch Variables (1) (2) (3) InShop −1.173* −1.246 −6.973*** (−1.81) (−1.18) (−3.75) InShop×Post×Treat 3.001** 4.086** (2.04) (2.03) Post×Treat 0.026 −0.016 (0.26) (−0.12) InShop×Treat −0.566 −0.349 (−0.67) (−0.29) InShop×Post −3.704*** −4.807*** (−3.48) (−3.38) Treat 0.027 0.062 (0.48) (0.81) Post −0.414*** −0.409*** (−4.04) (−3.74) InShop×CL 4.968** t-1 (2.52) CL 0.445*** t-1 (3.23) IMR −0.476 (−0.96) Controls Yes Yes Yes Year&Industry Yes Yes Yes Constant −1.863*** −1.347** −0.810 (−4.19) (−2.05) (−0.71) Wald Chi2 346.11*** 241.75*** 236.70*** Pseudo R 0.063 0.077 0.062 N 9598 5691 7544 Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. CHINA JOURNAL OF ACCOUNTING STUDIES 239 Control variables (Z ) in model (6) are same as those in model (3). We expected that i,t coefficient β is significantly positive. Table 10 reports the results after controlling endogeneity by mdoel (6). The result in column (1) is regressed on all firm-years during 2013–2017 and the result in column (2) is regressed on the sample which is limited one year before and one year after the receipt of comment-letter year. The coefficients on InShop×Post×Treat are significantly positive at 5% level in both columns. The results are consistent with the main findings in section 5. (2) Heckman two-step model. Listed firms inquired by the exchanges may be selective, which may lead to sample selection bias. Heckman two-step model is used to alleviate the endogeneity caused by the self-selection problem (Heckman, 1979). We conduct the test using Heckman two-step model. In the first stage, we use the probit model to regress CL i,t (Control variables are the same as those for the method of PSM above), and calculate the Inverse Mill’s Ratio (IMR). Then, we add IMR to the second step and the results are shown in column (3) of Table 10. The regression coefficient on InShop×CL is significantly t-1 positive at 5% level. That is, our main conclusions still remains in Heckman model. We also notice the coefficient of IMR is not significant, which implies that the endogeneity arising from self-selection is not serious in our study. Table 11. Comment letters and internal control opinion shopping: additional robustness tests. Control motivation of financial report audit Expanding the period of Alternative indepen- Excluding firms opinion shopping comment letters dent variable with punishment InSwitch InSwitch InSwitch InSwitch InSwitch Variables (1) (2) (3) (4) (5) InShop −19.609*** −6.627*** −6.020*** −9.216*** (−10.83) (−3.55) (−5.10) (−6.09) InShop×CL 5.170** 4.969*** 3.820** 4.483** t-1 (1.96) (2.63) (2.34) (2.28) CL 0.448*** 0.446*** 0.389*** 0.690*** 0.379** t-1 (2.83) (3.27) (3.25) (3.25) (2.51) InShop(raw) −0.510*** (−3.91) InShop(raw)×CL 0.316*** t-1 (2.67) FRShop −0.904 (−1.37) FRShop×CL 0.019 t-1 (0.02) Controls Yes Yes Yes Yes Yes Year&Industry Yes Yes Yes Yes Yes Constant −1.322 −0.592 −2.199** −0.467 −0.959 (−1.13) (−0.54) (−2.41) (−0.43) (−0.82) Wald Chi2 264.58*** 261.09*** 353.96*** 259.65*** 225.80*** Pseudo R 0.072 0.064 0.056 0.072 0.064 N 7201 7544 11,852 7544 6754 Notes: z-statistics are given in brackets; ***, **, * indicate significance at 1%, 5%, and 10% levels, respectively. Standard errors are clustered by firms. 240 Y. YAO AND S. XUE 6.4.2. Exclude alternative explanation Our findings may be driven by financial report audit opinion shopping. To be sure that our findings on internal control opinion shopping are not attributable to audit opinions shopping on financial reports we will take two additional tests. First, excluding modified audit opinions on financial reports. We remove 343 observa- tions where clients have modified audit opinion on financial report in year t and year t-1 and rerun model (3) on the remaining samples. The results are shown in column (1) of Table 11. The coefficient on InShop×CL is significantly positive at 5% level, which is t-1 consistent with the main results of this paper. Second, we control the motivation of audit opinion shopping on financial report. We employ the audit opinion shopping model of Lennox (2000)toconstruct a testing variable of financial report opinion shopping (FRShop )justaswhatwe i,t do to define InShop. The result is shownincolumn(2) of Table 11.The coefficient on InShop×CL is significantly positive at 1% level, suggesting that the result is qualita- t-1 tively unchanged. 6.4.3. Expanding the period of comment letters After the Direct Train of information disclosure of the Shanghai and Shenzhen Stock Exchanges in 2013, and the two exchanges began to issue comment letters but they were not mandatory disclosed before 2015. But some firm voluntarily disclosed the comment letters before 2015. We expand the research period from 2013 to 2017 (including 121 comment letters in 2013 and 148 comment letters in 2014). Column (3) of Table 11 reports the result. The coefficient on InShop×CL is significantly positive at 5% level, indicating t-1 our findings are consistent and robust. 6.4.4. Alternative proxy for internal control opinion shopping Following Lennox (2000) and Newton et al. (2016), we compute an alternative indepen- dent variable for scope of opinion shopping. We define this alternative variable as InShop(raw) based on model (1). Specifically, using the coefficients from Model (1), the i,t predicted value when InSwitch is set to 1 less the predicted value when InSwitch is set to 0, denoted InShop(raw) [InMAO -InMAO° = InShop(raw) ]. The results is shown in column i,t i,t i,t i,t (4) of Table 11 and the coefficient on InShop×CL is significantly positive at 1% level. t-1 6.4.5. Excluding alternative explanation of punishments In order to systematically exclude the alternative explanation of the impact of punishment on internal control opinion shopping, we remove the 790 observations which have been punished in year t-1. The results is presented in column (5) of Table 11 and the main conclusions are still supported. 7. Conclusions Based on the comment letters of the Shanghai and Shenzhen Stock Exchanges, taking Chinese A-share listed firms from 2015 to 2017 as our samples and using the audit opinion shopping model of Lennox (2000), this paper explores the supervision effect of comment letters on internal control opinion shopping from three dimensions (i.e. audit firm level, branch level and signing partner level). The empirical results show that (i) The CHINA JOURNAL OF ACCOUNTING STUDIES 241 probability of internal control opinion shopping is significantly decreased after com- ment letters are issued. This governance effect is more pronounced when there are internal-control-related questions in comment letters or when the firms’ annual auditors are required to give their professional opinions on some issues in comment letters. These findings are robust when PSM-DID design or Heckman two-step model are used. (ii) The supervision effect of comment letters on internal control opinion shopping is more pronounced when firms switch auditors from headquarters to branch since the independence of branches of audit firms is usually weaker than that of headquarters. The supervision effect of comment letters on internal control opinion shopping also exists at the level of signing partners. (iii) The supervision effect of comment letters on internal control opinion shopping is found only in the issuing year of comment letters which implies that auditors’ perceived pressure from comment letters can only last for a short run. (iv) The supervision effect of comment letters on internal control opinion shopping can spill over to firms in the same industry, in the same province or with the common auditors. Our findings have theoretical contributions and practical implications. First, this is the first study on the supervision effect of comment letters on internal control opinions shopping. We also investigate the impact of specific characteristics of comment letters such as internal- control- related comment letters or auditor-verified comment letters, which extends the study of the economic consequences of comment letters. Second, from the perspective of provider of audit service, we document evidences of internal control opinion shopping from three dimensions and the supervision effect of the comment letters. Our findings provide some clues for the two exchanges to improve the efficiency of inquiry mechanism. For example, out of all auditor switches, headquarter-to-branch switches should be paid more attentions. Third, combined the timing of auditor switch with the timing of comment letters, we document that the supervision effect of comment letters on internal control opinion shopping exists only in the short run. Overall, this paper investigates the supervision effect of comment letters on internal control opinion shopping. The registration reform of IPO is carrying on in Chinese market and our findings offer empirical evidences for regulators to optimise comment letters mechanism and push listed firmstoimprove the firewall of internal control. Acknowledgments We appreciate the helpful comments and suggestions from reviewers and editors. Shuang Xue acknowl- edges financial supports from the National Science Foundation of China (Project No. 71572102 and Project No. 71872107), MOE Project of Key Research Institute of Humanities and Social Science in University (16JJD790038) and Program for Innovative Research Team of SHUFE. Disclosure statement No potential conflict of interest was reported by the authors. 242 Y. YAO AND S. XUE Funding Shuang Xue acknowledges financial supports from the National Science Foundation of China [Project No. 71572102 and Project No. 71872107], MOE Project of Key Research Institute of Humanities and Social Science in University [16JJD790038] and Program for Innovative Research Team of SHUFE. References Altman, E.I. (1968). Financial ratios, discriminant analysis, and the prediction of corporate bankruptcy. Journal of Finance, 23(4), 589–609. Bens, D.A., Cheng, M., & Monica, N. (2016). The impact of SEC disclosure monitoring on the uncertainty of fair value estimates. The Accounting Review, 91(2), 349–375. Bozanic, Z., Dietrich, J.R., & Johnson, B.A. (2017). SEC comment letters and firm disclosure. Journal of Accounting & Public Policy, 36(5), 337–357. Brown, S.V., & Knechel, W.R. (2016). Auditor-client compatibility and audit firm selection. Journal of Accounting Research, 54(3), 725–775. Brown, S.V., Tian, X.L., & Jennifer, W.T. (2018). The spillover effect of SEC comment letters on qualitative corporate disclosure: Evidence from the risk factor disclosure. Contemporary Accounting Research, 35(2), 622–656. Cassell, C.A., Dreher, L.M., & Myers, L.A. (2013). Reviewing the SEC’s review process: 10-K comment letters and the cost of remediation. The Accounting Review, 88(6), 1875–1908. Chen, D.H., Zhang, T.S., & Li, X. (2008). Law environment,government regulation and implicit contract: Empirical evidence from the scandals of Chinese listed companies. Economic Research Journal, 3,60–72. (in Chinese). Chen, F., Francis, J.R., & Hou, Y. (2019). Opinion shopping through same-firm audit office switches. Working Paper. Chen, F., Peng, S.S., Xue, S., Ye, F., & Yang, Z. (2016). Do audit clients successfully engage in opinion shopping? partner-level evidence. Journal Of Accounting Research, 54(1), 79-112. Chen, Y.S., Dend, Y.L., & Li, Z. (2018a). Does the non-penalty regulation have information content? evidence from inquiry letters? Journal of Finance Research, 4, 155–171. (in Chinese). Chen, Y.S., Dend, Y.L., & Li, Z. (2018b). Can non-administrative penalty regulation improve audit quality. Auditing Research, 5,82–88. (in Chinese). Chen, Y.S., Dend, Y.L., & Li, Z. (2019). Effectiveness of the front-line regulation of the Chinese stock exchanges: Evidence from comment letters. Management World, 3, 169–185. (in Chinese). Chen, Y.S., Deng, Y.L., Jin, Y.F., Lou, H.T., & Zhang, X. (2018). Political connection and regulatory scrutiny through comment letters: Evidence from China. International Review of Finance, 9,1–10. Cheng, Q., Goh, B.W., & Kim, J.B. (2018). Internal control and operational efficiency. Contemporary Accounting Research, 35(2), 1102–1139. Coval, J.D., & Moskowitz, T.J. (2001). The geography of investment: Informed trading and asset prices. Journal of Political Economy, 109(4), 811–841. Cunningham, L., Schmardebeck, R., & Wang, W. (2017). SEC comment letters and bank lending. Working Paper. Cunningham, L.M., Johnson, B.A., Johnson, E.S., & Lisic, L.L. (2019). The switch up: An examination of changes in earnings management after receiving SEC comment letters. Contemporary Accounting Research, Forthcoming. Dechow, P.M., Lawrence, A., & Ryans, J.P. (2016). SEC comment letters and insider sales. The Accounting Review, 91(2), 401–439. DeFond, M., & Zhang, J.Y. (2014). A review of archival auditing research. Journal of Accounting and Economics, 58(2/3), 275–326. Fitzgerald, B.C., Omer, T.C., & Thompson, A.M. (2018). Audit partner tenure and internal control reporting quality: U.S. evidence from the Not-For-Profit sector. Contemporary Accounting Research, 35(1), 334–364. CHINA JOURNAL OF ACCOUNTING STUDIES 243 Francis, J.R., & Michas, P.N. (2013). The contagion effect of low-quality audits. The Accounting Review, 88(2), 521–552. Francis, J.R., & Yu, M. (2009). Big 4 office size and audit quality. The Accounting Review, 84(5), 1521–1552. Gietzmann, M.B., & Pettinicchio, A.K. (2014). External auditor reassessment of client business risk following the issuance of a comment letter by the SEC. European Accounting Review, 23(1), 57–85. Glaeser, E.L., & Shleifer, A. (2001). A reason for quantity regulation. American Economic Review, 91(2), 431–435. Heckman, J.J. (1979). Sample selection bias as a specification error. Econometrica, 47(1), 153–161. Heese, J., Khan, M., & Ramanna, K. (2017). Is the SEC captured? evidence from comment-letter reviews. Journal of Accounting and Economics, 64(1), 98–122. Huang, K., & Zhang, L.M. (2010). Audit regulation, auditor switching and successor auditor conservatism. Auditing Research, 1,65–74. (in Chinese). Jensen, M.C., & Meckling, W.H. (1992). Specific and general knowledge, organizational structure in contract economics (pp. 251–274). (L. Werin & H. Wijkander, eds). Oxford: Blackwell. Johnston, R., & Petacchi, R. (2017). Regulatory oversight of financial reporting: Securities and exchange commission comment letters. Contemporary Accounting Research, 34(2), 1128–1155. Kedia, S., Koh, K., & Rajgopal, S. (2015). Evidence on contagionin earnings management. The Accounting Review, 90(6), 2337–2373. Kothari, S.P., Leone, A.J., & Wasley, C.E. (2005). Performance matched discretionary accrual measures. Journal of Accounting and Economics, 39(1), 163–197. Krishnan, J., & Stephens, R.G. (1995). Evidence on opinion shopping from audit opinion conservatism. Journal of Accounting and Public Policy, 14(3), 179–201. Kubick, T.R., Lynch, D.P., Mayberry, M.A., & Omer, T.C. (2016). The effects of regulatory scrutiny on tax avoidance: An examination of SEC comment letters. The Accounting Review, 91(6), 1751–1780. Lennox, C. (2000). Do companies successfully engage in opinion-shopping? evidence from the UK. Journal of Accounting and Economics, 29(3), 321–337. Li, B., & Liu, Z.B. (2017). The oversight role of regulators: Evidence from SEC comment letters in the IPO process. Review of Accounting Studies, 22(3), 1229–1260. Li, L., Zhang, D.L., & Xia, P. (2017). Annual report supervision, reducing holding-shares of the insiders and market responses: A study based on annual report enquiry letters issued by the Shen zhen Stock Exchange. Contemporary Finance & Economics, 12, 108–119. (in Chinese). Li, Q.Y., & Zhao, Y.B. (2014). Empirical study on audit opinion shopping after financial restatement. Auditing Research, 5, 101–107. (in Chinese). Lu, Z.F., & Tong, P. (2003). Audit opinion, auditor change and supervision policy: An empirical study of rule 14. Auditing Research, 3,30–35. (in Chinese). Newton, N.J., Persellin, J.S., Wang, D.C., & Wilkins, M.S. (2016). Internal control opinion shopping and audit market competition. The Accounting Review, 91(2), 603–623. Peltzman, S. (1976). Toward a more general theory of regulation. The Journal of Law and Economics, 19(2), 211–240. Peter, T.D., & Zhang, H. (2018). Styles of regulators: Evidence from the SEC’s comment letters. Working paper. Pizzini, M., Lin, S., & Ziegenfuss, D.E. (2015). The impact of internal audit function quality and contribution on audit delay. Auditing: A Journal of Practice & Theory, 34(1), 25–58. Stigler, G.J. (1971). The theory of economic regulation. The Bell Journal of Economics and Management Science, 2(1), 3–21. Wallman, S.M.H. (1996). The future of accounting, part III: Reliability and auditor independence. Accounting Horizons, 10(4), 76–97. Wang, B., & Xin, Q.Q. (2010). Do office practice of audit firms influence audit quality and fees? Auditing Research, 2,70–76. (in Chinese). Wang, C.F., Wu, X., & Zeng, T.B. (2016). Audit firm governance of branches and office-level acceptance of initial engagement: Evidence from the CICPA survey data. Accounting Research, 3,87–94. (in Chinese). Wang, Q. (2016). Determinants of segment disclosure deficiencies and the effect of the SEC comment letter process. Journal of Accounting & Public Policy, 35(2), 109–133. 244 Y. YAO AND S. XUE Wu, L.N., Wang, C.F., & Lu, Z.F. (2013). Auditor switch in a corporation group and opinion shopping. Auditing Research, 1,70–78. (in Chinese). Wu, L.S. (2005). Audit opinion shopping: Behavior characteristics and regulatory strategy. Economic Research Journal, 7,66–76. (in Chinese). Wu, X. (2008). Consequence of regulatory sanction on an individual auditor without on the audit firm: Empirical evidence. Accounting Research, 8,23–31. (in Chinese). Wu, X., Wang, C.F., & Li, B. (2018). Competition in China’s public accounting service market: Evidence from newly-established branch offices. Accounting Research, 12,12–18. (in Chinese). Xue, J., Ru, Y., & Dou, C. (2017). Punishing one threatens a hundred? the deterrent effect of exposure mechanism on top-executive excess perquisites. Accounting Research, 5,60–66. (in Chinese). Zhang,J.S., & Tang,X.J.(2018). Does the preventive regulation mitigate stock price crash risk? Evidence from comment letters. Journal of Management Sciences in China, 21(10), 112–126.(in Chinese). Zhang,J.X., & He,Y.N.(2013). Thetypeofinternal controlaudit opinion and individual investors’ confidence in unqualified audit opinion on financial statement. Auditing Research, 4,93–100. (in Chinese). Zhang, W.Y. (2002). Reputational foundation of the legal system. Economic Research Journal, 1,3–13. (in Chinese).
Journal
China Journal of Accounting Studies
– Taylor & Francis
Published: Apr 3, 2019
Keywords: Comment letters; internal control opinion shopping; headquarters and branch; signature partners
