Do Global Audit Firm Networks Apply Consistent Audit Methodologies Across Jurisdictions? Evidence from Financial Reporting Comparability

Transcription

1 Do Global Audit Firm Networks Apply Consistent Audit Methodologies Across Jurisdictions? Evidence from Financial Reporting Comparability Matthew Ege Texas A&M University Young Hoon Kim Texas A&M University Dechun Wang Texas A&M University May 2018 Abstract: Brand name audit firms are actually global networks of local audit firms. These global networks claim to enforce consistent audit methodologies across their member firms. We test this claim and find that clients from different countries have more comparable accruals when they are audited by local audit firms from the same global network as compared to different global networks. To improve identification, we use audit firm switches induced by the failure of Andersen, which serves as an exogenous shock. We also find more (less) comparable accruals when analyzing client-pairs where the pair have recently switched auditors such that they have auditors from the same (different) global network. In falsification tests, having auditors from the same global network is unassociated with differences in cash flows and special items. Lastly, tests suggest that the role of global network methodologies in global financial reporting comparability is more pronounced across stronger investor protection jurisdictions. Key Words: Audit Firm Networks, Comparability JEL Codes: M41, M42 The authors acknowledge helpful comments from Richard Hanus, John McNamara, Joe Schroeder and workshop participants at Texas A&M University. All authors acknowledge support from the Mays Business School.

2 I. Introduction As business has become more global so have audit firms. The largest audit firms have formed international networks, such that they appear as one global audit firm despite being a network of legally independent, individual firms. The role of global audit firm networks is to coordinate the development and enforcement of global strategies, standards, policies, and governance. For example, each affiliate of PricewaterhouseCoopers International Limited (PwCIL) confirms annually that they are in compliance with PwCIL standards. PwCIL states that the network s common methodologies, technologies, and materials are designed to help member firms, partners and staff perform their work more consistently, and support their compliance with the way PwC does business. 1 In this study, we examine whether global audit firm networks (hereafter global networks ) successfully enforce their unique global auditing methodologies and related standards across their global affiliates. Enforcing a consistent auditing methodology across borders is increasingly more important to the US equity market because foreign affiliates of global networks are regularly used by their US counterparts on audits of US multinationals and also are the primary auditor for many SEC registrants. Through its inspection of foreign affiliates, the Public Company Accounting Oversight Board (PCAOB) has identified many instances of insufficient auditing. For example, the PCAOB fined Deloitte Brazil and Deloitte Mexico for various issues that stem from not properly executing audits (Scannell 2016). The PCAOB s concern of inconsistent audit quality within global networks partially drove the recent requirement to disclose the participation of foreign affiliates in audits of SEC registrants (PCAOB 2015). International markets have also experienced auditing that is not in compliance with global network standards such as in the corporate failures at Parmalat in Italy 1 See 1

3 and Satyam in India (Carson 2014). Thus, our study should be of interest to auditing regulators and standard setters around the globe. While Francis et al. (2014) provide evidence consistent with the US Big 4 implementing unique audit methodologies (i.e., unique auditor styles ), such evidence may not apply to the international setting. In the US, an accounting firm is a legal entity that provides for, among other things, shared liability among partners and similar incentives for governance and execution of high-quality audits. Also, all US Big 4 are regulated by the PCAOB. However, a global network is comprised by legally independent domestic audit firms, which do not share liability, could have differing incentives for governance and execution of high-quality audits, and are subject to various national regulators. Thus, despite having global audit methodologies and systems of controls to enforce the methodologies, global networks may be unable to ensure consistent application of their unique audit methodologies across their affiliates. Challenges to consistent application of global audit methodologies include jurisdiction-specific incentives, differences in culture, and network coordination and cooperation problems (Lenz and James 2007; Bik and Hooghiemstra 2018). We focus our study on the global networks of BDO, Deloitte, EY, Grant Thornton, KPMG, and PwC (the Big 6) because (1) each describes their global reach and global audit methodology on their respective websites and (2) regulators such as the PCAOB identify these six audit firm networks as global networks. 2 These global networks have their own unique audit methodology, which is supported by resources such as global knowledge management databases and common industry-specific work programs and training (Carson 2009, 356). Also, these global networks have formal monitoring mechanisms to enforce the consistent application of their audit methodologies across global affiliates. For example, EY (2017a, 5) has a Global Audit Quality 2 See 2

4 Committee [that] monitors and oversees quality globally. The goal of these efforts is to provide high quality audits in all jurisdictions through the consistent application of global audit methodologies. If global networks enforce a unique global audit methodology, client financials of affiliates within a global network should be more comparable compared to client financials of affiliates of different global networks. We follow Francis et al. (2014, 606) and define comparability as the closeness of two firms reported earnings due to the consistency with which rules are applied across firms. If audit firms consistently follow their global audit methodology, then audit firms that are part of the same global network should enforce similar accounting treatments for similar economic transactions across jurisdictions. In other words, in the spirit of Francis et al. (2014), we expect there to be unique auditor styles for each global network. Auditors have more influence over the accounting for accruals compared to cash flows, given the judgement required in the accruals processes (Nelson et al. 2003; Jones 1991; Dechow and Dichev 2002). Therefore, we test whether the Big 6 networks enforce global audit methodologies by using three accruals-based measures of comparability. Specifically, we use differences in total accruals along with two estimations of differences in current accruals, as current accruals are more susceptible to manipulation in the short run (Healy 1985) and therefore more likely affected by auditors. We compare the differences in these accrual measures between cross-jurisdictional client-pairs that use audit firms from the same versus different global networks. Our sample comes from the Compustat (Global Compustat for international companies) and Capital IQ databases and includes 84,365 client-years from 94 countries for years 2003 to This results in 2,165,943 client-year pairs of which 437,923 have auditors that are affiliates of the same global network. 3

5 Using multiple regression analysis, we find consistent evidence that two companies from different jurisdictions have more comparable accruals when audited by audit firms from the same global network. From an economic perspective, accrual differences for client-pairs that have affiliate auditors from the same global network are 1.3 to 7.8 percent smaller than the median accrual difference. This corresponds to a 2.4 to 4.2 percent change in median ROA. To provide evidence on whether our results are consistent with a causal relation, we next examine accrual differences before and after auditor switches. First, to help rule out an alternative explanation that our results are due to clients selecting auditors based upon audit methodology (i.e., a client-auditor matching story), we use the failure of Andersen US as an exogenous shock. Upon the failure of Andersen US, foreign Andersen affiliates had to join other global networks, resulting in their clients being subject to a new global audit methodology. We find that differences in accruals between Andersen clients and clients of the audit firm that audits the Andersen clients after failure become smaller after the Andersen failure. Second, for a broader sample of switches where client-pairs come to have affiliate auditors from the same (different) global network, we find that differences in accruals become smaller (larger) after the audit firm change compared to before the change. These results are consistent with Big 6 global networks affecting client financial reporting comparability by consistently executing their unique global audit methodologies. Next, as falsification tests, we examine whether client-pairs having auditors from the same global network are associated with differences in special items or cash flows. Global audit methodologies are less likely to include specific procedures for one-off items such as special items, and the audit of cash compared to accruals requires much less judgment and is relatively straightforward. Thus, global audit methodologies should have little, if any, effect on the comparability of special items and cash flows across client-pairs. We find no effect of client-pairs 4

6 having auditors from the same global network on differences in special items or cash flows. In additional tests, we find our results are robust to controlling for gross domestic product and client-pairs having headquarters located on the same continent. Inferences remain unchanged if we exclude all US observations, only include client-pairs from European countries after 2004, limit our analysis to client-pairs of the Big 4 global networks, exclude industry specialist audit firms, or use market-based comparability measures. Inferences also remain the same when using global network fixed effects to test for the enforcement of unique global audit methodologies. As a final test, we consider whether investor protection strength affects the magnitude of the effect of global audit methodologies on client financial reporting comparability. In stronger investor protection regimes, regulators and other investor protection forces potentially shape earnings quality such that the audit methodologies of global networks have less of a chance to shape the accounting treatment reported in client financials. However, the repercussions to audit firms for not following audit methodologies are more severe in stronger investor protection regimes. Thus, ex ante it is unclear whether investor protection moderates the effect of global audit methodologies on client financial reporting comparability. In seven of 15 specifications, we find evidence consistent with global network methodologies playing more of a role in financial reporting comparability across high investor protection jurisdictions than low investor protection jurisdictions. We contribute to the literature by providing robust evidence that suggests that global networks enforce a unique global audit methodology, which results in consistent financial reporting practices across jurisdictions within their clientele. Therefore, our study extends the literature on the auditors role in international financial reporting quality. Prior studies focus on how audit quality affects financial reporting quality in terms of faithful representativeness and 5

7 verifiability across countries (e.g., Choi et al. 2008; Francis and Wang 2008). Our study extends this literature by examining the role of auditors in a different dimension of financial reporting quality: comparability. Additionally, this paper contributes to the literature on cross-country comparability (e.g., Fang et al. 2015; Barth et al. 2012) by providing evidence of an economic agent affecting client financial statements in similar ways across jurisdictions. Unlike the existing literature that focuses on the similarity of accounting standards and the monitoring role of economic agents across countries, we examine an economic agent that directly participates in the production of financial information. As such, we also contribute to US-based evidence of the auditor s role in comparability (Francis et al. 2014). Our findings are useful to regulators and standard setters. In particular, while the PCAOB and the European Commission (EC 2003) are concerned about audit quality of global network affiliates, our results show that global networks are able to enforce consistent methodologies across affiliated firms, but the consistency of application varies with the country-level legal environment. Additionally, given the lack of a global auditing regulator, our findings also imply that the global networks are a quasi-enforcer of International Standards on Auditing (ISAs), consistent with the discussion in Carson (2014, 28-30). II. Institutional Background and Hypothesis Development A. Global Audit Firm Networks The largest six global audit firm networks include: BDO, Deloitte, EY, Grant Thornton, KPMG, and PwC. 3 The global reach of these networks expanded significantly during the 1980s 3 A global audit network is defined as a contractual cooperation between legally and economically autonomous national audit firms, which are organized based on partnership principles under the strategic leadership of one or more member firms for the joint fulfilment of international client needs (Lenz and James 2007). 6

8 and 1990s when numerous local audit firms joined international audit firm networks (Nobes and Parker 2008). 4 The member firms (also known as affiliate firms) of these global networks are locally owned and operated and maintain their legal and economic independence (Lenz and James 2007). For example, Deloitte US and Deloitte Norway are member firms of Deloitte Touche Tohmatsu Limited (DTTL). All the member firms of DTTL are legally separate and independent entities, which cannot obligate each other. DTTL and each DTTL member firm are liable only for their own acts and omissions (Deloitte 2017). Member firms have obligations to adopt the network policies and abide by network governance rules and codes of conduct. 5 Each of the Big 6 networks have global audit methodologies and have a system of quality control to monitor adherence to global audit methodologies. For example, each of these global networks are members of the Forum of Firms and have committed to promote the consistent application of high-quality audit practices and standards worldwide and maintain quality control standards in accordance with the International Standards on Quality Control (see 6 As part of their global quality control standards, each global network executes internal inspections of affiliates for compliance with global network policies and procedures. 7 Through efforts such as inspection of affiliates, 4 There were also several mergers between international networks that occurred during this timeframe, as well as the failure of Andersen, which has resulted in the six networks noted here. 5 Lenz and James (2007) discuss in detail the responsibilities of member firms as they relate to the global network. 6 The Big 6 networks were also the Founder Members of the Forum of Firms (IFAC 2015; footnote 15 of Carson 2009). 7 Global networks monitor compliance with numerous network-wide policies. For example, in its 2017 annual review report, PwC states that each PwC member firm s Territory Senior Partner signs an annual confirmation of compliance with PwC's standards. These confirmations cover a range of areas, including independence, ethics and business conduct, enterprise risk management, governance, anti-corruption, anti-money laundering, anti-trust, insider trading and information protection ( 7

9 consistent use of technology, and common training, each global network seeks to ensure the consistent application of the firm s global audit methodology. The global audit methodologies of these networks contain different steps [that] are narrowly defined and detailed prescriptions for the approach to auditing; for instance, descriptions of the planning process, the process of evaluating risks, the approach to testing (Nobes and Parker 2008, 487). 8 The global networks are where accounting practices emerge, become standardized and regulated, where accounting rules and standards are translated into practice, and where professional identities are mediated, formed and transformed (Cooper and Robson 2006, 416). Thus, the goal of these methodologies is to facilitate the delivery of consistent, high-quality audits across all member firms. 9 B. Related Literature and Hypotheses Development Our research question is primarily related to two streams of literature. The first stream explores how affiliates of large global audit firms are associated with financial reporting quality measured by faithful representation of earnings (e.g., misstatements and accruals quality). Broadly speaking, Big N firms deliver higher audit quality compared to smaller firms (see DeFond and Zhang 2014). 8 There are different auditing standards throughout the world. The ISAs are the most widely adopted, with 125 countries currently committed to their use (based on as of April 2, 2018). Other standards exist such as the standards for US public company audits from the PCAOB. Global networks create their global audit methodologies based on ISAs and supplement the methodologies with requirements from local standards where applicable (e.g., PCAOB internal control testing for US public companies). While it is possible that audit teams have to adapt global audit methodologies for specific audits, our understanding is that this is a very rare event that does not affect clients in the vast majority of markets. One reason for this is that either ISAs or PCAOB standards are used in every major economy such that global networks can essentially base their audit methodologies on these two sets of standards. 9 As a firm-specific example of the goal of consistency, EY has developed a Sustainable Audit Quality program that is implemented across all member firms and helps...drive consistency in execution (EY 2017b, 22). EY states that the program includes technology, tools and training, accountability metrics, evaluating and rewarding our people. Additionally, within 100 days of the merger of Coopers & Lybrand and Price Waterhouse all 60,000 Assurance professionals in 150 countries were trained on the [PwC Audit Approach] through a three-day training course delivered at the local-office (Winograd et al. 2000, 176). 8

10 The second line of literature examines the determinants of financial reporting comparability, and has generally focused on accounting standards and monitoring of economic agents. For example, the adoption of IFRS by non-us companies increases financial reporting comparability with US companies (Barth et al. 2012), and IFRS adoption increases financial reporting comparability within European Union countries (Yip and Young 2012). Fang et al. (2015) find that foreign institutional ownership helps improve financial reporting comparability in emerging markets as it relates to US companies. Moreover, they provide evidence that suggests hiring Big 4 auditors is the specific mechanism through which financial reporting becomes more comparable. Francis et al. (2014) find evidence consistent with the US Big 4 each enforcing their unique audit methodologies (the authors call this auditor style ). Specifically, client-pairs from the same US Big 4 auditor have more comparable earnings compared to client-pairs from different US Big 4 auditors. We expect global networks contribute to the improvement of global financial reporting comparability for several reasons. First, as stated earlier, global networks provide access to an expanded set of resources such as training, experts, and tools, which become part of the each network s global audit methodology. For example, Carson (2009) finds that industry expertise in a global network is positively associated with an audit fee premium. This is consistent with expertise being embedded within global network audit methodologies, which then results in affiliate firms being able to charge higher fees for this expertise. The Big 6 also market a consistent level of global service quality as a benefit of using audit firms that are part of a global network (PwC 2012). Thus, we expect expertise related to how to account for and audit specific transactions to transfer across a global network via global audit methodologies, such that audits executed by affiliate firms would audit similar transactions in similar ways. 9

11 Moreover, global networks have global quality control procedures (e.g., inspections across affiliates, common training, technology that monitors independence, etc.) to enforce their global audit methodologies. They also have incentives to enforce the consistent execution of high-quality audits across their affiliates. For example, clients of non-us Andersen affiliates experienced negative abnormal returns on two key dates that revealed reputation-damaging news for the US Andersen affiliate (Cahan et al. 2009). This is consistent with reputation damage for one global network affiliate spilling over to other global network affiliates. Therefore, across global network affiliates, we expect not only the transfer of expertise via global audit methodologies, but also the execution of audits in accordance to the global network s audit methodology. Accordingly, we expect affiliate auditors from the same global network to affect client financial statements in similar ways. Our hypothesis is stated in the alternative: H1: A pair of companies from different countries that are audited by affiliates of the same Big 6 global audit firm network will have more comparable financial reporting than a pair of companies from different countries that are audited by affiliates of different Big 6 global audit firm networks. While we expect the Big 6 global networks to enforce a consistent audit methodology across their affiliates, it is possible that the global networks are unable to achieve this goal for various reasons. For example, the governance of global networks can be difficult and complicated (Lenz and James 2007), and local auditors may apply audit procedures through different cultural lenses, resulting in inconsistent application of global audit methodologies (Bik and Hooghiemstra 2018). Moreover, a global network mainly serves as a coordinating platform because the member firms of a global network are legally independent. Although there are network-wide policies, the member firms may have incentives to deviate from these policies. Accordingly, the PCAOB has expressed concerns about the audit quality of member firms of the global networks (Ferguson 2015, PCAOB 2015, Goelzer 2011). For example, in 2016, the PCAOB fined Deloitte Brazil $8 10

12 million for issuing false audit reports, suggesting that Deloitte failed to effectively monitor and enforce its global methodologies within its Brazilian affiliate. III. Research Design A. Measure of Comparability Following Francis et al. (2014, 606), we define comparability as the closeness of two firms reported earnings due to the consistency with which rules are applied across firms. We compare, across jurisdictions, the closeness of reported earnings for similar client-pairs due to the consistency with which rules are applied. We create pairs of two client-years that are from different countries, are in the same industry, have the same fiscal year ends, and are in the same countryyear quintile rank of size. Empirically, we compare the closeness of the level of accruals of client-pairs. We focus on accruals because accruals have been viewed as the most discretionary (not necessarily opportunistic) component of earnings that can be easily affected by accounting policies and estimation errors (e.g., Jones 1991; Dechow and Dichev 2002). In other words, auditors have more opportunity to affect accruals versus cash flows. 10 We measure accruals in three ways. First, following Francis et al. (2014), we measure financial statement comparability using differences in total accruals. DDDDDDDD_TTTTTTTT iiiiii = aaaaaaaaaaaaaaaa vvvvvvvvvv (TTTTTTTT iiii TTTTTTTT jjjj ) (1) where subscripts i and j denote two clients in a pair and t denotes year, respectively. Next, because long-term accruals can be less susceptible to manipulation in the short run (Healy 1985) and therefore less likely to be affected by auditors, we also employ two measures 10 It is possible to measure financial statement comparability using input-based measures or output-based measures. Input-based measures require examining specific accounting choices made by each firm and aggregating them into an overall index at the country-level (e.g., Ashbaugh and Pincus 2001). Because it is practically difficult to determine accounting inputs of global companies, we rely on output-based measures of accounting comparability. 11

13 based on current accruals. For our second measure we follow the basic idea of Dechow and Dichev (2002) and estimate the accounting mapping from past, present, and future operating cash flows to current accruals at the country-industry-auditor level. The logic of the approach outlined below is the same as in Barth et al. (2012). 11 First, we estimate the relation between current accruals and past, current, and future operating cash flows, separately at the country-industry-auditor level, as follows: CCCCCCCC iiii = ββ 0 CCCCCC + ββ 1 CCCCCC CCCCCC ii,tt 1 + ββ 2 CCCCCC CCCCCC iiii + ββ 3 CCCCCC CCCCCC ii,tt+1 + εε iiii CCCCCC (2a) This regression estimates the relation between current accruals and operating cash flows using company-years that are in the same country and industry and have the same auditor. Superscripts denoted C, F, and A refer to country, industry (Fama French 30), and auditor, respectively. 12 We require each country-industry-auditor to have at least 10 company-year observations. Second, for each company in a client-pair, we calculate predicted current accruals using the coefficients from its own country-industry-auditor level estimation and predicted current accruals using the coefficients from the paired company s country-industry-auditor level estimation. For example, for pair i-j, we calculate two predicted current accruals of company i, CCCCCCCC CCCCCC ii CCCCCC iiii and CCAAAAAA jj iiii, using different coefficients ββ CCCCCC ii and ββ CCCCCC jj as follows: CCCCCCCC CCCCCC ii CCCCCC iiii = ii CCCCCC ββ 0 + ii CCCCCC ββ 1 CCCCCC ii,tt 1 + ii CCCCCC ββ 2 CCCCCC iiii + ii ββ 3 CCCCCC ii,tt+1 (2b) CCCCCC CCCCCCCC jj CCCCCC iiii = jj CCCCCC ββ 0 + jj CCCCCC ββ 1 CCCCOO ii,tt 1 + jj CCCCCC ββ 2 CCCCCC iiii + jj ββ 3 CCCCCC ii,tt+1 (2c) 11 See Appendix A in Barth et al. (2012) for details. 12 The potential benefit of using a model based on Dechow and Dichev (2002) is that cash flows are used as proxies of economic events, as indicated in Barth et al. (2012). While De Franco et al. (2011) use returns to capture economic events, we focus on non-market based economic events mainly because market efficiency is likely different across the jurisdictions in our sample. In additional analyses, we show that our results are robust to using market-based comparability measures. 12

14 Following the logic in Barth et al. (2012), we assume that the estimated coefficients capture an auditor s unique methodology. The predicted current accruals for client i (CCCCCCCC CCCCCC ii iiii ) are the expected outcome of the auditor who audits client i's country-industry-auditor group. In CCCCCC comparison, CCCCCCCC jj iiii represents hypothetical expected current accruals had client i been audited by the auditor who audits the country-industry-auditor group of client j. If the methodologies of the auditor who audits client i s country-industry-auditor group and the auditor who audits the country-industry-auditor group of client j are the same, the expected current accruals should be the same (i.e., CCCCCCCC CCCCCC ii CCCCCC iiii = CCCCCCCC jj iiii ). In the same way, we calculate two predicted current accruals of company j, as follows: CCCCCC CCCCCCCC jj CCCCCC jjjj = jj CCCCCC ββ 0 + jj CCCCCC ββ 1 CCCCCC jj,tt 1 + jj CCCCCC ββ 2 CCCCCC jjjj + jj ββ 3 CCCCCC jj,tt+1 (2d) CCCCCCCC CCCCCC ii CCCCCC jjjj = ii CCCCCC ββ 0 + ii CCCCCC ββ 1 CCCCCC jj,tt 1 + ii CCCCCC ββ 2 CCCCCC jjjj + ii ββ 3 CCCCCC jj,tt+1 (2e) Lastly, for each pair, we calculate the absolute value of the difference between the predicted accruals from the previous steps and take the average to get the comparability measure (DDDDDDDD_CCCCCCCC1 iiiiii ) for pair i-j at year t, as follows: CCCCCCCC_dddddddd iiii = CCCCCCCC CCCCCC ii CCCCCC iiii CCCCCCCC jj iiii (2f) CCCCCC CCCCCCCC_dddddddd jjjj = CCCCCCCC jj jjjj CCCCCCCC CCCCCC ii jjjj (2g) DDiiiiii_CCCCCCCC1 iiiiii = 0.5 CCCCCCCC_dddddddd iiii + CCCCCCCC_dddddddd jjjj (2h) Taken together, the above methodology captures the differences between client i's predicted current accruals when audited by their true auditor versus when audited by the auditor of client j. Similarly, it also captures the differences between client j's predicted current accruals when audited by their true auditor versus when audited by the auditor of client i. Another feature of the above approach is that by estimating two predicted current accruals based on coefficients 13

15 from two auditors, we hold the economic events (cash flows) constant, consistent with the definition of accounting comparability in De Franco et al. (2011). For our third measure, we estimate the relation between changes in revenue and current accruals at the country-industry-auditor level following the logic in prior studies (DeFond and Park 2001; Francis and Wang 2008; Ahmed et al. 2016). 13 This approach is different from the previous one in that we do not rely on past, current, and future cash flows to estimate current accruals. Specifically, we follow the same process as equations (2b) through (2h), but regress current accruals on the change in revenue, separately at the country-industry-auditor level, as follows: CCCCCCCC iiii = αα CCCCCC 0 + αα CCCCCC CCCCCC 1 RRRRRRRRRRRRRR iiii + εε iiii (3) For each company in a client-pair, we then calculate predicted current accruals using the coefficients from its own country-industry-auditor level estimation and predicted current accruals using the coefficients from the paired company s country-industry-auditor level estimation. Lastly, for each pair, we calculate the absolute value of the difference between the predicted accruals from the previous steps and take the average to get the comparability measure (DDDDDDDD_CCCCCCCC2 iiiiii ) for pair i-j at year t. 14 B. Primary Research Design To formally test our hypotheses, we estimate the following equation using OLS regressions: DDDDDDDD_AAAAAA iiiiii = ββ 0 + ββ 1 SShaaaaaaaa_AAAAAAAAAAAAAA iiiiii + γγ CCCCCCCCCCCCCCCC iiiiii + εε iiiiii, (4) where subscripts i and j denote two clients in a pair and t denotes year, respectively. Our dependent 13 The three papers treat current accruals or working capital accruals as a linear function of change in revenue. DeFond and Park (2001) and Ahmed et al. (2016) examined only current accruals, while Francis and Wang (2008) also include long-term accruals (i.e., depreciation and amortization). 14 Our primary results are also robust to using the difference in current accruals (not based on estimation) between client-pairs and to using the difference in abnormal accruals between client-pairs following the logic of DeFond and Park (2001) and Francis and Wang (2008). 14

16 variable DDDDDDDD_AAAAAA iiiiii is one of the three measures of client-pair accrual differences: difference in total accruals (DDDDDDDD_TTTTTTTT iiiiii ), difference in predicted current accruals based on cash flows (DDDDDDDD_CCCCCCCC1 iiiiii ), or difference in predicted current accruals based on changes in revenue (DDDDDDDD_CCCCCCCC2 iiiiii ). 15 Our independent variable of interest is SShaaaaaaaa_AAAAAAAAAAAAAA iiiiii, which is coded one when two clients i and j at year t have auditors that are affiliates of the same global network. If global networks maintain and enforce consistent audit methodologies and accordingly increase financial statement comparability of their clients across jurisdictions, we expect the coefficient on SShaaaaaaaa_AAAAAAAAAAAAAA iiiiii (ββ 1 ) to be negative. Francis et al. (2014), citing Lang et al. (2010), state that there is no theoretical or empirical guidance for controls for an earnings comparability regression. Thus, we follow both papers and include control variables that capture various pair-specific characteristics, including differences and minimum of size, leverage, market-to-book, operating cash flows, losses, standard deviation of sales, standard deviation of cash flows, and sales growth. All variables are defined in the Appendix. Prior studies have documented evidence that the same and similar accounting standards improve financial reporting comparability across countries (Barth et al. 2012; Lang et al. 2010). Thus, we also control for same accounting standard, Same_Standard, which is coded 1 when two clients in a pair report financial statements under the same accounting standard. For example, if one client uses US GAAP and the other uses IFRS, we classify that pair into a different standard pair (Same_Standard = 0). Furthermore, if one client uses local GAAP, which is neither US GAAP 15 Our dependent variables capture the closeness of accruals between two paired clients and therefore are calculated as the absolute value of the difference. Because our dependent variables range from zero to one, we employ fractional response methodology (using Stata commands fracreg logit/probit) following Papke and Wooldridge (1996) as a robustness check. Inferences remain the same using this methodology (untabulated). 15

17 nor IFRS, then we define all the pairs including that client as using different accounting standards (Same_Standard = 0). Additionally, we include year, industry (Fama French 30), and country fixed effects. Standard errors are clustered at the client-pair level. IV. Sample Selection and Descriptive statistics A. Sample Selection We begin with Compustat companies (Global Compustat for international companies) from 2003 through 2016 where necessary accounting information to calculate control variables is non-missing. We start our sample in 2003 to avoid the impact of the demise of Arthur Andersen. For non-u.s and non-canadian companies, we collect audit firm information from the Capital IQ database because it provides better coverage of auditor information than Global Compustat for non-us and non-canadian companies. For US and Canadian clients, we obtain auditor information from Compustat North America Database. We then retain observations for clients of the Big 6 global audit networks (i.e., BDO, Deloitte, EY, Grant Thornton, KPMG, and PwC). Following Francis et al. (2014), we retain company-years with fiscal year ends in March, June, September, and December, and require at least 50 company-years to exist in a given two-digit SIC industry. We delete companies with names containing Holding and LLP, and also exclude financial institutions. We also require at least 10 company-years to exist for a given country-industryauditor. We remove client-years that do not report quarterly accounting information for at least 8 of the last 16 quarters because some control variables require quarterly data. Finally, we truncate total accruals (TACC) and current accruals (CACC) at the 2 nd percentile and 98 th percentile based on country-year combinations to mitigate effects of outliers in country-year combinations. This results in 84,365 client-years from 94 countries. Table 1, Panel A presents the sample selection 16

18 details and Panel B and C present the number of client-year observations in our study by country and global network. [Insert Table 1] We then create pairs of two clients that are from different countries, are in the same industry, and have the same fiscal year ends. Furthermore, we require the size (total assets) of two clients in a pair to be in the same quintile rank of size. Quintile rank of size is calculated for each country-year combination. Although difference in size (SSSSSSSS_dddddddd) is included as a covariate in the main tests, we match clients by size rank to further control for potential differences in unobservable client characteristics. Prior research suggests that client size captures many underlying client characteristics that are correlated with measures of audit quality (e.g., Lawrence et al. 2011). 16 Thus, creating client-pairs based on a size ranking helps to protect against results being due to unobservable systematic differences between clients. This process results in 2,165,943 total client-year pairs. B. Descriptive Statistics and Univariate Results Table 2 presents descriptive statistics and univariate results. Continuous variables have been winsorized at the 1 st and 99 th percentiles. Out of 2,165,943 client-pairs in our sample, 20.2 percent of pairs are shared auditor pairs (i.e., pairs of clients that are audited by affiliates of the same global network). Average differences in Diff_TACC, Diff_CACC1, and Diff_CACC2 between client-pairs are 8.5 percent, 3.3 percent, and 2.7 percent of total assets, respectively. Univariate analyses reveal that shared auditor pairs have statistically smaller differences in accruals than non-shared auditor pairs, on average. This suggests that financial statements of shared auditor pairs are more comparable than those of non-shared auditor pairs and that global 16 It is also likely that clients select the Big 4 and non-big 4 for particular reasons (DeFond and Zhang 2014). We find consistent results when we limit our analyses to Big 4 client-pairs only. 17

19 networks apply a unique audit methodology. [Insert Table 2] V. Test Results A. Main Results Table 3 presents the results of regressions of differences in accruals on shared auditor and other control variables following equation (4). The coefficients on Shared_Auditor are negative and statistically significant at the one percent level across all three columns. From an economic perspective, the coefficients on Shared_Auditor represent between 1.3 and 7.8 percent of the median accrual difference, which corresponds to a 2.4 to 4.2 percent change in median ROA. 17 The signs of the coefficients on the control variables are overall consistent with those in Francis et al. (2014). These initial results are consistent with our hypothesis (H1) that global audit networks have their own unique audit methodology that they apply consistently across their global affiliates, which in turn results in a more comparable accruals structure and smaller differences in accruals. [Insert Table 3] B. The Exogenous Shock of Andersen Audit firms and clients choose to engage each other. Thus, it is possible that clients who prefer certain accounting treatments get matched to audit firms willing to agree with client preferences. Thus, our evidence so far could reflect the matching of client preferences to global networks instead of global networks enforcing unique audit methodologies. We do not believe that this type of self-selection affects inferences because theory suggests that clients choose between 17 The median values of Diff_TACC, Diff_CACC1, and Diff_CACC2 are 0.064, 0.023, and 0.018, respectively. Accordingly, the coefficients on Shared_Auditor per each model represent 1.3% (0.0008/0.064), 5.2% (0.0012/0.023), and 7.8% (0.0014/0.018) of the median accrual difference. Also, the median ROA of companies in our sample is 3.33%. Therefore, the economic effects of having auditors from the same global network represent 2.4% (0.0008/0.0333), 3.6% (0.0012/0.0333), and 4.2% (0.0014/0.0333) of median ROA, respectively. 18

20 Big N and non-big N audit firms, not within Big N audit firms (e.g., DeFond and Zhang 2014; Francis et al. 2014). However, to provide stronger identification, we next utilize the failure of Andersen as an exogenous shock. The US Andersen practice surrendered its license to practice accounting in 2002 due to charges related to accounting fraud at Enron. Subsequently, US Andersen offices and Andersen foreign affiliates joined other global networks. We use these switches to examine whether the financial reporting of Andersen clients became more comparable to the clients of the global network these clients joined after the US Andersen failure. Specifically, we utilize two approaches. First, we utilize a sample of Andersen clients and clients of the global network that the Andersen clients joined subsequent to 2001 (i.e., Andersen clients have new auditors in 2002 due to the Andersen US failure). We compare the comparability of client-pairs of Andersen clients and the new global network clients before and after the event year. 18 For example, if a client of Andersen Germany switched to EY Germany in 2002, we focus on cross-country client-pairs that include this client in Germany and clients of EY in other countries and test whether the financial statements become more comparable after 2002 when the two companies start sharing the same global network auditor. For this subsample, we estimate equation (4) after replacing Shared_Auditor with Post, which equals one for years greater than or equal to 2002, and zero otherwise. In other words, Post captures the difference in comparability for the same client-pairs before and after the switch in auditors due to the exogenous shock. Because it could take time for the effect of the auditor switch to manifest, we examine changes in comparability measures before and after the auditor switch based upon sample periods from [t-1, 18 For this test, we use annual accounting data to measure control variables instead of quarterly data since some countries do not report quarterly data until early 2000s. We continue to create client-pairs across countries based on industry, size quintile, and fiscal year-end. 19

21 t+1] to [t-4, t+4]. Second, because Andersen clients might not have all stayed with the local Andersen affiliate in the move to the new global network, we perform the same test after removing all Andersen clients that did not switch to the global network that received the majority of Andersen clients. 19 We exclude these clients on the chance they specifically chose a new audit firm because of the accounting treatments provided by the audit firm. Table 4, Panel A and B provide results for both samples, respectively. In both samples, we find that the coefficient on Post is negative and significant across all estimations. This provides strong evidence that endogeneity due to clientauditor matching is not a concern for our inferences. [Insert Table 4] As a falsification test, we create cross-country client-pairs using a sample of Andersen clients and clients of the global networks that the Andersen clients did not join subsequent to For example, for this falsification test, if a client of Andersen Germany switched to EY Germany in 2002, we focus on cross-country client-pairs that include this client in Germany and clients of BDO, Deloitte, Grant Thornton, KPMG, and PwC (but not EY) in other countries. The logic behind this test is that Andersen clients should not become more comparable to clients of global networks that they did not join. Results are presented in Table 4, Panel C. As expected, we do not find evidence of increased comparability in this test. Specifically, the coefficient on Post is insignificant (positive) when the dependent variable is Diff_TACC and Diff_CACC1 (Diff_CACC2). These results help to rule out concerns that our results in Panel A and B are driven by poor accrual quality for old Andersen clients (e.g., perhaps old Andersen clients experienced increased financial reporting comparability compared to clients of all other audit firms because 19 For example, 83 percent of Andersen Germany clients in our sample went to EY Germany. In this test we exclude any Andersen Germany clients that did not switch to EY Germany. 20

22 other audit firms imposed very strict auditing upon Andersen clients). C. Broader Switch Sample We next examine effects within a switching sample broader than the Andersen switches. Specifically, we utilize our primary sample that begins in 2003 and focus on client-pairs where at least one client in the client-pair changes its auditor in year t. For this analysis, we identify those pairs where the pair had auditors from different global networks in year t-1, but have auditors from the same global network in year t. We call these client-pairs S_switch pairs because they have auditors from the same global network in year t. For the S_switch subsample, we test whether comparability of client-pairs increases after clients switch from having different auditors to having the same auditor. We use equation (4) after replacing Shared_Auditor with Post, which equals one for years greater than or equal to the switch year, and zero otherwise. [Insert Table 5] Table 5, Panel A presents the results. We find negative and significant coefficients on Post in one of four, three of four, and four of four specifications using DDDDDDDD_TTTTTTTT, DDDDDDDD_CCCCCCCC1, and DDDDDDDD_CCCCCCCC2 as the dependent variable, respectively. The results are stronger when using current accruals versus total accruals, which could be because auditors have more influence over current accruals. From an economic perspective, the coefficient on Post for DDDDDDDD_CCCCCCCC1 for the [t-4, t+4] period is , which represents approximately 3.0 percent of the median difference in accruals and approximately 2.1 percent of the median ROA. Next, we examine a sample of client-pairs where at least one client switches its auditor resulting in the client-pairs switching from having auditors from the same global network to having auditors from different global networks. We call these client-pairs D_switch pairs and test whether there is a decrease in comparability after the switch. Table 5, Panel B presents the results. 21

23 In the tests of DDDDDDDD_CCCCCCCC1 and DDDDDDDD_CCCCCCCC2, we find evidence of a decrease in comparability after the auditor changes (positive and significant coefficients on Post in all eight specifications), consistent with our expectations. We find a positive, significant coefficient on Post when using Diff_TACC as the dependent variable in one of four specifications. These results suggest that client switches away from a specific global network result in accruals for the switching clients becoming less comparable to accruals for clients remaining with auditors from the global network. In Panel C of Table 5, we test for differences in the coefficients between the S_switch and D_switch samples. In all but two cases (the [t-4, t+4] and [t-3, t+3] sample periods with Diff_TACC as the dependent variable) we find that the coefficients in the S_switch sample are statistically different from the corresponding coefficients in the D_switch sample. Overall, these results suggest that client-pairs increase in comparability after clients come to have auditors from the same global network and decrease in comparability after clients cease having auditors from the same global network. These results combined with the results using Andersen as an exogenous shock strongly suggest that global networks enforce unique audit methodologies, which affect the comparability of their clients financial reporting. VI. Additional Analyses A. Cash Flows and Special Items In this section we consider two falsification tests aimed at examining accounting that is less likely to be uniquely affected by global network audit methodologies. First, we have argued that the application of global audit methodologies should be most observable in accruals compared to cash flows because accruals are subject to significant judgement. In other words, the guidance within global audit methodologies for auditing cash is likely more similar across the Big 6 compared to the guidance for auditing accruals. Second, global audit methodologies are less likely 22

24 to contain specific guidance and procedures for one-off, uncommon transactions. Therefore, we expect to find little, if any, effect of having auditors from the same global network on differences in cash flows and special items. Thus, we re-estimate equation (4) after including the difference in the cash flows and the difference in special items as dependent variables. 20 Results are presented in Table 6. The coefficients on Shared_Auditor are insignificant in both columns, suggesting that our primary results capture the application of global network audit methodologies. Additionally, these results suggest that global audit methodologies affect comparability through auditing of accruals and not through auditing cash or non-routine items. [Insert Table 6] B. Other Country Characteristics It is possible that country-specific factors that are not controlled for in our primary model affect inferences. While we cannot think of country-specific factors that would explain the pattern of results thus far, we next perform several robustness tests aimed at further alleviating such concerns. First, auditors from geographically closer areas enforce similar accounting practices and implement similar auditing techniques. For example, it is possible that European affiliates of a global firm communicate more with and lend personnel more frequently to other European affiliates versus affiliates from other parts of the world. To rule out that geographic proximity drives our results, we control for Same_Region, which is coded 1 when two clients in a pair are 20 We use the difference in total cash flows instead of the difference in cash flows from operations because certain line items such as cash interest payments can be classified as operating or financing activities under different accounting standards. Even within the same accounting standards (i.e., IFRS), clients have flexibility in classifying some cash transactions into different activities. For example, under IFRS, cash interest payments can be classified as operating or financing activities (IAS 7). Thus, it is not appropriate to use operating cash flows for a falsification test. 23