Do auditors mitigate earnings management during economic crisis?

Revista de Contabilidad Spanish Accounting Review 22 (1) (2019) 6-20 REVISTA DE CONTABILIDAD SPANISH ACCOUNTING REVIEW revistas.um.es/rcsar Do auditors mitigate earnings management during economic crisis? Nuria Reguera-Alvarado*, Pilar de Fuentes, Joaquina Laffarga Department of Accounting and Financial Economics, University of Seville, Seville, Spain ARTICLE INFO Article history: Received 21 February 2018 Accepted 9 July 2018 Available online 31 December 2018 JEL classification: Keywords: Abnormal accruals Earnings management Economic crisis Audit quality Corporate governance ABSTRACT This paper addresses earnings management from an international perspective. This study particularly analyzes whether external auditing reduces earnings managements and, if so, whether this relationship is affected by the economic crisis. The study employs a cross-country approach. The sample has 3,830 observations from listed firms from the United States, the United Kingdom, Japan, Italy, France and Spain during the period 2005–2009. Panel data are used to verify the research hypotheses. The findings show that external auditing is an important mechanism to minimize earnings management. In this sense, the Big Four and auditor specialization helped to reduce earnings managements during the economic crisis. Moreover, long audit tenure does not contribute to mitigating discretionary accruals. The study includes implications for academia, practitioners, and policymakers. It provides the literature with complementary evidence on the external auditing and earnings management relationship and how this is influenced by the economic crisis. The article benefits practitioners and policymakers by highlighting the most important characteristics of audit firms related to the mitigation of earnings managements. In addition, this study considers the recent economic crisis as an important economic factor. Given that the crisis affected most countries around the world and caused structural changes within companies, information about the consequences of the crisis and how to deal with them is an important empirical question. ©2018 ASEPUC. Published by EDITUM - Universidad de Murcia. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Códigos JEL: Palabras clave: Ajustes por devengo discrecionales Gestión del resultado Crisis económica Calidad del auditor Gobierno corporativo Intervención de los auditores en la gestión de resultados durante la crisis económica RESUMEN Este artículo aborda la gestión del resultado desde una perspectiva internacional. Concretamente, el estudio analiza si la auditoría externa reduce la gestión del resultado y, si fuera así, si esto se vio afectado por la crisis económica. Para ello se ha utilizado una muestra formada compuesta 3,830 observaciones de empresas cotizadas en Estados Unidos, El Reino Unido, Japón, Italia, Francia y España durante el período 2005-2009. Los resultados muestran que la auditoría externa es un mecanismo de gran repercusión para minimizar la gestión del resultado. En este sentido, las cuatro mayores empresas auditoras y la especialización del auditor ayudaron a reducir la gestión de beneficios durante la crisis económica. Asimismo, un mayor número de años auditando a un determinado cliente consigue que la firma de auditoría no mitigue la gestión del resultado. El estudio incluye implicaciones para académicos, profesionales, y gestores políticos. Se ofrece evidencia complementaria para la literatura sobre auditoría externa y gestión del resultado y cómo se ve influenciada ésta por la crisis económica. El artículo resulta interesante desde el punto de vista de los profesionales y gestores políticos porque resalta las características más importantes de las empresas auditoras relativas a la mitigación de la gestión del resultado. Asimismo, el estudio considera la reciente crisis económica como un importante factor económico. Dado que la crisis afectó a la mayoría de los países y causó cambios a nivel estructural dentro de las empresas, la información acerca de las consecuencias de la crisis y como tratar con ella ha sido considerada una cuestión importante desde el punto de vista empírico. ©2018 ASEPUC. Publicado por EDITUM - Universidad de Murcia. Este es un artículo Open Access bajo la licencia CC BY-NC-ND (http://creativecommons.org/licenses/by-nc-nd/4.0/). * Corresponding author. E-mail address: [email protected] (N. Reguera-Alvarado). https://www.doi.org/10.6018/rc-sar.22.1.354271 1988-4672 (online) /1138-4891 (printed) /©2018 ASEPUC. Published by EDITUM - Universidad de Murcia. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 7 Introduction As one of the main factors that triggers firm performance misrepresentation (Dichev et al., 2013), earnings management affects the decision-making process of all firm stakeholders, including investors, regulators, and analysts (Krishnan, 2003a). Accordingly, prior research has analyzed earnings management from different views, such as corporate governance (Garcia-Meca and Sanchez-Ballesta, 2009), capital markets (Capalbo, et al., 2014), and the institutional framework (Leuz et al., 2003;Doupnik, 2008). Recent research has also examined the auditor’s role as a mechanism to reduce earnings management (Kim et al, 2003;Chen et al., 2008;Gul, Fung, and Jaggi 2009;Boone, Khurana, and Raman 2010;Choi et al., 2010;Ahsen, 2011). Previous studies find that earnings management is affected by auditor size (e.g., DeAngelo, 1981;Francis and Yu, 2009;Choi et al., 2010;Ahsen, 2011) and auditor specialization (e.g., Lim and Tan, 2008;Carson, 2009;Gul et al., 2009;Cahan et al., 2011; Capalbo, et al. 2014). On the other hand, business cycles have a fundamental impact on firms’ reporting quality and market perception of reported earnings (Jiang, et al., 2015). Johnson (1999) shows that the quality of financial information is sensitive to business cycles. Indeed, the macroeconomic context largely determines earnings management (Kousenidis et al., 2013; Quinglu, 2005; Lin and Shih, 2003) as relevant changes in the economic environment represent an incentive for it (Filip and Rattourner, 2014). Nonetheless, although the literature highlights the prominent role of environmental factors, including the recent global financial crisis in accounting reliability across countries (Huang et al., 2013;Dedman and Kausar, 2012;Davis-Friday et al., 2006;Leuz et al., 2003; Garcia-Lara et al., 2005), to the best of the authors’ knowledge, no prior studies examine the role that the economic crisis plays in the relationship between earnings management and auditor quality. Therefore, the recent crisis offers a unique opportunity to study the effect of business cycles on earnings management. In this context, we wonder whether the global economic crisis has modified the managerial incentives for manager earnings, and what the role of audit firms has been. These schemes are in line with Essen et al. (2013), who support the argument that firm good governance prescriptions designed to assure managerial oversight do not hold in a financial crisis. Accordingly, this paper’s main objective is to analyze the effect of some audit characteristics, proxies for audit quality, on earnings management by considering the impact of the economic crisis on this relationship. As such, we consider three main auditing attributes: (i) auditor firm size, (ii) auditor tenure, and (iii) auditor specialization. We employ an international sample of 989 listed firms from six developed countries (the United States, the United Kingdom, Japan, Italy, France, and Spain) during the period 2005–2009. This time period includes both crisis and non-crisis years. Our results are robust for various models and are scientifically and economically relevant across different economic, social, and institutional environments. By selecting this variety of countries, our study is in line with the current trend that suggests the need and opportunity for cross-country accounting research projects and strengthens the international dimension of the findings (Gordon, et al., 2013). Regarding the measure of earnings management, Francis et al. (2004) and Dechow et al. (2010) show that financial reporting quality can be measured through various indicators such as, among others, persistence, discretionary accruals, smoothness, timeliness, loss avoidance and investor responsiveness. This study, therefore, uses discretionary accruals as a proxy of earnings management. Our results emphasize the important effect of the crisis on earnings management. We find that the economic crisis led to an increase in earnings management and reversed the profile of most earnings management firms. We also find that the audit quality does not only improve the quality of earnings but also plays a more relevant role during the crisis. This study contributes to the existing literature on earnings management in two ways. First, we address the impact of the economic crisis on discretionary accruals. To the best of our knowledge, this study is one of the few that adopts an international point of view. Second, auditor industry specialization continues to attract considerable attention in the literature (e.g., Lim and Tan, 2008;Carson 2009;Gul et al., 2009; Cahan et al., 2011). This study adds to the literature on industry expertise and specialization, which is increasingly more important to auditor clients.80 percent of industry companies consider these factors to be important when choosing an auditor (GAO 2003, 2008). The remainder of the paper proceeds as follows. Section 2 provides an overview of the most recent studies on earnings management in relation to audit quality, and the hypothesis development. Section 3 describes the data and the method. Section 4 summarizes the main empirical results. Section 5 concludes with a summary of the findings and the main contributions of the study. Theoretical background and research hypotheses Recent research has investigated the determinants and consequences of earnings management. However, these studies commonly focus on the individual incentives of managers and pay little attention to the firm’s macroeconomic environment . As a result, previous research does not determine whether a good or bad economic environment affects earnings management because the decision to engage in earnings management is based on investors’ perspectives. Iatridis (2010) suggests four situations in which earnings management is used: (i) to transfer earnings from “good” years to “bad” years; (ii) to postpone income recognition to reduce the tax burden; (iii) to reveal positive results, correlated with the trend of postponing negative results; and (iv) to increase the current or future compensation of managers by using discretionary accounting policies to benefit from stock options or bonus schemes. Although there is not much literature that considers the influence of the recent financial crisis on earnings management, some studies have investigated other crises, such as the Persian Gulf crisis of 1990 (Han and Wang, 1998), the Mexican crisis of 1994 (Davis-Friday and Gordon, 2005), and the Asian crisis of 1997 (Shrieves and Dahl, 2003), as exogenous shocks, and a few studies have considered the global economic crisis of 2008 (Lu, 2012;Habib et al, 2013;Filip and Raffournier, 2014). Upward earnings management should be higher during a period of economic crisis. Most firms are likely to have lower earnings, which can motivate managers to engage in income-increasing earnings management to compensate for the decrease in operational performance (Ahmad-Zaluki et al., 2011). In particular, managers of the firms most strongly affected by the crisis may manipulate earnings upward to avoid a large decline of the firm’s stock price, which would negatively impact their compensation (Charitou et al., 2007). In addition, because debt covenants are based in part on earnings (Dichev and Skinner, 2002), 8 N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 income-increasing manipulations can reduce the probability of violations (DeFond and Jiambalvo, 1994;Iatridis and Kadorinis, 2009;Saleh and Ahmed, 2005;Sweeney, 1994). Conversely, some firms may have incentives to manage earnings downward during an economic crisis. For example, firms that must undertake debt restructuring due to debt covenant violation or failure to meet a debt repayment (Filip and Raffournier, 2014) may want to suppress earnings. DeAngelo et al. (1994) note that by reporting losses managers portray the firm as seriously troubled, which may be useful to extract concessions from employees who otherwise would doubt the existence and persistence of the firm’s difficulties. The search for political advantage can also motivate the reduction of earnings as governments are likely to provide support to firms in financial distress due to an economic crisis (Peltzman, 1976). The previous discussion suggests that periods of economic downturn are associated with a higher level of earnings management, although researchers do not agree on the sign of these manipulations. Nevertheless, crises are likely to be less favorable to earnings management than expansion periods. During crises, firms are subject to increased monitoring from auditors, creditors and other stakeholders, which can result in managers having less discretion to manage earnings (Chia et al., 2007). In line with agency theory (Jensen and Meckling, 1976), external auditing acts as a control tool that minimizes or eliminates managers’ opportunistic practices, such as earnings management, to guarantee a reasonable level of reliability of the financial statements (Watts and Zimmerman, 1986). Chia et al. (2007) find that firms demand better audit services when more controls of accounting reporting exist, as is the case during an economic crisis. Auditor competence and auditor independence are the central concepts of traditional research on audit quality (Gonthier-Besacier, et al., 2012). The auditing literature commonly attributes two roles to auditors through which they guarantee the quality of firms’ financial statements (Kim et al., 2011;Cano and Sanchez, 2012). First, auditing helps to reduce asymmetric information problems and assures users of financial information that this information meets their requirements of reliability, integrity and quality (information role). Second, through a revision of the accounting information, auditors act to decrease the opportunistic behavior of managers and control shareholders. Consequently, auditing reduces the agency conflicts between these agents and their respective principals (monitoring role). This study focuses on the most studied auditing attributes in the literature: auditor firm size, auditor tenure, and auditor specialization. Next, we develop a hypothesis for each of the auditing attributes (auditor size, auditor tenure, and auditor specialization) related to earnings management. In addition, because of the recent economic crisis affecting financial reporting, we consider its influence on the relationship between auditing attributes and earnings management. Auditor firm size and earnings management The auditing literature suggests that auditors have incentives to protect their reputation and mitigate expected litigation costs arising from audit failures (Houghton, et al. 2002; Boone et al., 2010;Cano, 2010). The importance of auditor independence stems from the need to give credibility to an organization‘s financial reports (Rifaat and Karim, 1990; Kilgore et al, 2011;Causholli et al. 2014). DeAngelo (1981) was a pioneer in analyzing the relationship between audit quality and auditor size from a theoretical viewpoint. Her main assumption is that large auditors will have more diversified total fees due to their larger customer portfolio. As such, auditing firms’ business performance—their growth or even their survival—is less dependent on each customer, which fosters the auditor’s independence (Cano, 2010). For this reason, DeAngelo suggests a positive relationship between auditor firm size and audit quality. Huguet and Gandía (2016) propose that audited financial statements are perceived to be the most reliable and thus provide higher quality information. Yet they remark the lack of empirical evidence between audited companies and higher quality information. In line with DeAngelo (1981),Palmrose (1988) and Simunic and Stein (1987) find that, given their “deep pockets” and heavy spending on building their brand names, large audit firms have an incentive to lower litigation risk and protect their reputational capital by providing more credible financial reports. In this manner, the previous literature suggests that audit firm size is an important determinant of auditor quality. Specifically, large auditing firms are able to spend heavily on training and audit technology (thereby contributing to their competence). In addition, because they are less dependent on each individual client, they can better resist client pressure to issue a “clean” (unqualified) audit opinion or to restrain earnings management. Based on these findings, the literature widely recognizes that the Big Four1provide higher quality audits and offer greater reliability for their clients’ financial statements than the non–Big (e.g., Kim et al. (2003) or Huguet (2016).Eshleman and Guo (2014) find a negative relationship between the level of abnormal audit fees paid by the client and the likelihood of using discretionary accruals. Becker et al. (1998) observe that companies audited by the Big Six have lower income-increasing discretionary accruals than firms using non–Big Six auditors. Krishnan (2003a) shows that the Big Four are better at constraining clients’ earnings management compared to the non–Big Four auditors because the non–Big Four’s clients have higher levels of discretionary accruals. Also, Ahsen (2011) and Craswell et al. (1995) note that firms audited by Big Four auditors have lower earnings management than that of firms audited by non–Big Four auditors. Similarly, Francis and Yu (2009), and Choi et al. (2010) show that audit office size is a primary determinant of auditor quality. As a result, prior research links audit quality to earnings management and suggests that large audit firms are more effective at constraining earnings manipulation in firms. Despite the evidence that large audit firms are more able to minimize opportunistic behaviors, Francis et al. (1999) argue that firms with high values for discretionary accruals have a greater opportunity to present opportunistic earnings management and have an incentive to hire a Big Four auditor to provide assurance that the earnings are credible. With respect to the market concentration within the Big Four group of accounting firms, Francis, Michas, and Seavey (2013) find some support for the concerns of regulators. Increased concentration within the Big Four auditors is negatively associated with audit quality. Specifically, when one or two of the Big Four auditors have a concentrated, dominated market share-rather than the market share being spread equally across all Big Four firms-Big Four clients have larger accruals and auditors are less likely to report losses and exhibit less timely loss recognition. These results suggest that a coun- 1The number of Big N auditors has changed during the years (Big 8 until 1987, Big 6 until 1998, Big 5 until 2001 and, currently, Big 4). The Big Four accounting firms are Deloitte, Ernst & Young, KPMG, and Pricewaterhouse- Coopers. N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 9 try’s regulator should not necessarily be concerned with the overall Big Four market share relative to non–Big Four auditors. Instead, regulators ought to have misgivings about the dominance of one or two auditors within the Big Four group, which can potentially have an adverse effect on the quality of audited earnings (Francis et al., 2013). Kim et al. (2003) indicate that Big Six auditors are more effective than non- Big Six auditors only when the conflict of reporting incentive exists between the two issues. Consequently, the evidence on the impact of auditor firm size on earnings management is currently not clear. Recent high-profile reports in the United States, the United Kingdom, and the European Union have raised concerns about the Big Four accounting firms’ concentration of supply and the potentially adverse effect that this concentration may have on audit markets and the quality of audits in these legal jurisdictions (Francis et al., 2013;GAO, 2003, 2008;Oxera, 2006, 2007;United States Treasury, 2006, 2008). Given the previously discussed literature that investigates the evolution of earnings management around events with a significant impact on firms (such as an economic crisis), we formulate our first hypothesis as follows: H1: During the recent economic crisis, firms audited by a Big Four auditor engage less in earnings management than firms audited by non–Big Four auditors. Auditor tenure and earnings management A considerable debate exists in the literature regarding whether longer auditor tenure is associated with higher or lower earnings management. Some studies even state that auditor tenure has a mixed effect. The auditing literature shows that auditor tenure is associated with higher earnings management (Geiger and Raghunandan, 2002;Gul et al., 2007;Chen et al., 2008). Johnson, Khurana, and Reynolds (2002) report higher unexpected discretionary accruals when the auditor tenure is short (2–3 years), compared to when it is medium in length (4–8 years). Also, Ghosh and Moon (2005) provide evidence that firms with longer auditor tenures are associated with stronger earnings response coefficients. They suggest that investors perceive the earnings management of firms with longer auditor tenures to be better than the earnings management of firms with shorter auditor tenures. These findings are consistent with learning theory (Glaser and Bassok, 1989;Lapre et al., 2000) and the existence of a learning curve in the auditing industry, which suggest that it takes time for auditors to develop client-specific knowledge to perform an effective audit. Other possible explanations for the relationship between auditor tenure and earnings management are that (i) firms with higher earnings management have a tendency to retain the same auditor as reported in the auditorswitching and opinion-shopping literature (e.g., DeFond and Subrahmanyam, 1998;Krishnan, 1994;Lennox, 2000), and (ii) high-quality auditors may drop risky clients—especially clients with large unexpected discretionary accruals—in the first few years of their audit engagements (Gul et al., 2009). In contrast, other studies find that longer auditor tenure is associated with higher levels of earnings management. The justification of these findings is that longer auditor tenure leads to more friendly relationships with the management and can therefore impair auditor independence. Consequently, these authors support mandatory audit firm rotation (Cohen and Zarowin, 2010;Chi et al., 2011). Furthermore, the effect of tenure may be mixed, with tenure affecting the auditor’s objectivity and knowledge of the client favorably in the early years but adversely in later years (Public Oversight Board, 2002;Casterella et al. 2002;Davis et al. 2009). These studies point to a connection between higher auditor tenure and a closer auditor–customer relationship over time. Thus, auditing reports are more likely to be biased toward the customer’s requests with the goal of retaining customers, hence reducing auditor quality. According to these theoretical arguments and assuming that changes in the economic climate have an impact on earnings management, we propose the following research hypothesis : H2: During the economic crisis, the association between auditor tenure and earnings management is non-linear. Auditor industry specialization and earnings management Auditor industry specialization is a recurrent subject in literature (e.g., Lim and Tan, 2008;Carson, 2009;Gul et al., 2009;Cahan et al., 2011). Audit quality research has focused on the role of auditor industry specialization as a key indicator of auditor quality (Reichelt and Wang, 2010). The literature suggests that specialized auditors have additional incentives to ensure fair reporting (DeAngelo, 1981). Because they better understand industry trends and accounting practices, they can isolate potential errors more effectively (Maletta and Wright 1996). In this way, industry-specialized auditors play an important role in monitoring the financial reporting process (Sun and Liu, 2013). Hogan and Jeter (1999) find that measures of specialization have increased in both regulated and unregulated industries, and Craswell et al. (1995) argue that audit firms market themselves in terms of both a general reputation and industry expertise. Previous research shows a negative relationship between earnings management and auditor industry specialization. Bonner and Lewis (1990) note that, on average, more experienced auditors outperform less experienced auditors. Similarly, Bedard and Biggs (1991) observe that auditors with more manufacturing experience are better able to identify errors in a manufacturing client’s data than auditors with less manufacturing experience. Other studies also suggest that industry expert auditors are associated with a lower likelihood of being involved in SEC enforcement actions (Carcello and Nagy, 2004) and lower probabilities of restatements (Romanus et al., 2008). Consistent with these results and according to the possible impact of the macroeconomic environment, we state the next hypothesis: H3: During the economic crisis, firms audited by a specialist auditor engage less in earnings management than firms audited by non-specialist auditors. Research design Sample In order to test our hypotheses, we have combined two databases: financial information has been obtained from Osiris2, whereas the information about auditors has been drawn from Datastream3. Both databases have information 2Osiris is a product of Bureau van Dijk Electronic Publishing and provides standardized annual accounts for companies throughout the world. 3Datastream is a product of Thomson Reuters Publishing and provides standardized annual accounts for companies throughout the world. 10 N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 on listed companies around the world. The information is very detailed and includes more than financial reports. The initial sample included 1,107 firms corresponding to companies quoted on the stock markets over the period 2005-20094 in the US, the UK, Japan, Italy, France and Spain. However, the following filters were applied to the initial sample: (i) Financial firms were eliminated. Their Balance sheet structures and operating income statements are different to those of the other companies in the sample. This type of firms could probably distort the analysis. Thus, 63 banks were eliminated. (ii) Companies that had extreme values were eliminated, since they distort the results. In total 55 companies were removed. So, after applying filters, the final sample, as detailed in Table 1, includes 3,830 observations corresponding to 989 non-financial companies. Table 1 Composition of the sample by countries Tables Table 1. Composition of the sample by countries Country Companies Observations Percentage U.S. 317 1233 32.20% U.K. 169 653 17.05% Japan 187 764 19.95% Italy 113 384 10.03% France 70 299 7.81% Spain 133 496 12.95% Total 989 3,829 100% To guarantee the robustness of our results in different economic, social, and institutional environments, we use six countries (the US, the UK, Japan, Italy, France and Spain), instead of a single country, which is the common practice followed in previous studies. Additionally, by selecting this variety of countries our study is aligned with the current trend that suggests the need and opportunity of designing crosscountry accounting research projects (Gordon et al. 2013). These six countries are selected for three reasons. Firstly, they represent the most relevant economies of three important worldwide economic areas (America, Europe, and Asia).5Secondly, they adopt the two different accounting normative models, IFRS (the United Kingdom, Japan, Italy, France, and Spain) and GAAP (the United States). The prior literature suggests that IFRS standards aim to be ‘highquality’ rules for the preparation of financial reports. This should mean that their adoption decreases the level of earnings management in comparison with other accounting systems, such as GAAP. Thirdly, in these six countries, the two existing investor protection legal systems – common law (the United States and the United Kingdom) and civil law (Japan, Italy, France, and Spain) – are applied.6La Porta et al. (1998) suggest that higher legal protection of investors implies a better application of rules and, consequently, higherquality accounting information and less accounting manipulation. Hence, with the countries considered here, our results are tested in heterogeneous situations. 4Since we have used lagged values for some variables, the dataset covers the period 2004-2009. 5The six countries selected are the five countries with the highest gross domestic product (GDP) around the world (http://databank.worldbank. org/data/download/GDP.pdf). 6Common law offers better legal protection than civil law (La Porta et al., 1998). To improve the reliability of our results, we require at least six country-year-sector observations based on the Standard Industrial Classification (SIC). This enhances the comparability of our results with previous research on earnings management during the crisis (Habib et al., 2013). This method of classification allows the formation of homogeneous industry groups, as shown in Table 2. According to this table, manufacturing is the most representative sector in all the countries. The industry composition is balanced across countries, so that we can rule out any bias due to the over or underrepresentation of sectors. Variables Dependent variable: Discretionary accruals Regulators have expressed their concern about auditors’ work because it is expected that they allow their clients to engage in an aggressive management of earnings (Levitt, 1998). Besides, auditors try to guarantee their independence to protect their reputation and avoid litigation costs, although this is a controversial issue if we consider the flexibility allowed by the standard reporting process (Mayhew et al., 2001). Accordingly, we define earnings management as a proxy for audit quality because if auditors’ independence is compromised earnings management could be an appropriate measure of its impairment (Causholli et al, 2014). Dechow et al. (2010) state that there is no measure of earnings management which is superior for all the decision models. Since all of the proxies for earnings management have at their core the reported accrual-based earnings number, it makes sense to focus on earnings management. This course of action is coherent with the comprehensive survey of Dechow and Skinner (2000). They support the prevalence of earnings management, and provide both the academic and practitioner views on earnings management. Furthermore, McNichols (2000) arguments that the main issue in earnings management is the measure of discretionary accruals. Nonetheless, Huguet and Gandía (2016) found that audited companies have a lower level of absolutely discretionary accruals than the non-audited ones. The literature has developed a number of accruals models to detect the discretionary component of earnings management (De Fond and Jiambalvo 1994;Rees et al., 1996;Teoh et al., 1998). We consider a selection of models on based on Dechow et al. (2010), who conducted a review on earnings management and identified the five models most often used to calculate discretionary accruals: the Jones (1991) model, the modified Jones model (Dechow et al., 1995), the Jones model adjusted to returns on assets (ROA; Kothari et al., 2005), the Dechow and Dichev (2002) model, and the Francis et al. (2005) model. Then, to choose the most suitable model, we reviewed the models cited in Google Scholar published until April 2018. Table 9 provides the results of this review. According to our findings, the literature most often uses the Jones model and its variations. Hence, we adopt the Jones model, the modified Jones model, and the Jones model adjusted to ROA to calculate earnings management. It can be seen that the Jones model (including some variations) is the one most often used by the literature. Furthermore, some authors have also used models developed by Dechow and Dichev (2002) and Kothari et al. (2005). Meanwhile, Dechow et al. (2010) carried out a review about earnings management and found that the models most used to calculate discretionary accruals are the Jones (1991), the modified Jones (Dechow et al., 1995), Kothari et al. (2005), the Dechow and Dichev (2002) and the Francis et al. (2005). N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 11 Table 2 Composition of the sample by countries and sectors Table 2. Composition of the sample by countries and sectors Country Basic materials, energy, and construction Manufacture Transportation, communication s and utilities Consumer goods Real Estate Services Services U.S. 5.52% 57.20% 13.95% 13.38% 2.49% 7.46% U.K. 18.07% 37.83% 15.16% 14.24% 2.30% 12.40% Japan 4.06% 70.16% 11.39% 7.46% 3.01% 3.93% Italy 4.95% 60.94% 16.41% 3.65% 1.04% 13.02% France 5.02% 34.11% 12.04% 10.37% 15.38% 23.08% Spain 11.09% 42.52% 19.15% 1.21% 15.52% 10.48% Table 9 Models of earnings management: citations review Table 9. Models of earnings management: citations review Nº cites MODEL Jones, J.J. (1991). Earnings Management During Import Relief Investigations. Journal of Accounting Research, 29(2), 193-228 7740 Jones Dechow, P.M., Sloan, R.G., and Sweeney, A.P. (1995). Detecting Earnings Management. The Accounting Review, 70(2), 193-225. 7985 Jones modified Peasnell, K., Pope, P.F., and Young. S. (2000). Detecting Earnings Management Using Cross- Sectional Abnormal Accrual Models. Journal of Accounting and Business Research, 30(4), 313- 326. 426 Marginal model Kang, S.H. and Sivaramakrishnan, K. (1995). Issues in Testing Earnings Management and an Instrumental Variable Approach. Journal of Accounting Research, 33(2), 353-367 481 Kang and Sivaramakrishnan Kasznik, R. (1999).On the Association Between Voluntary Disclosure and Earnings Management. Journal of Accounting Research, 37(1), 57-81. 1416 Jones Cash-flow Jeter, D.C., and Shivakumar, L. (1999). Cross-Sectional Estimation of Abnormal Accruals Using Quarterly and Annual Data: Effectiveness in Detecting Event-Specific Earnings Management. Journal of Accounting and Business Research, 29(4), 299-319. 227 Garza-Gómez, X., Okumara, M., and Kunimura, M. (1999). Discretionary Accrual Models and the Accounting Process, SSRN Working Paper Series. 50 Accounting process Kothari, S. P., Leone A. J., and Wasley, C.E. (2005). Performance matched discretionary accrual measures. Journal of Accounting and Economics, 39, 163-197. 4833 Jones model adjusted to ROA Dechow, P. M., and Dichev, I. D. (2002). The Quality of Accruals and Earnings: The Role of Accrual Estimation Errors. The Accounting Review, 77, 35-59. 4308 Dechow and Dichey Table 10. Results specification test In this manner, we have compared three accrual models to calculate earnings management: the Jones model (Jones, 1991), the modified Jones model (Dechow et al. 1995) and the Jones model adjusted to ROA (Kothari et al. 2005). In order to determine which model has fewer margins of error calculating discretionary accruals after doing so based on the three models used (the Jones model, the modified Jones model and the Jones model adjusted to ROA), we apply a specification and a power test. These tests are based on the calculation of the number of times that type I errors and type II errors occur for each estimated model, following the methodology of Brown and Warner (1980, 1985). On the one hand, a specification test refers to the number of times type I errors are committed. Table 10 shows the results of type I errors. The results test Z to calculate the proportion for each accrual model and the number of regressions per country. Our results show that the three models are at optimal levels, which are established as 10%. Nevertheless, for the Jones model in the United States and Japan, this limit was not satisfied. Thus, in order to determine if the percentage of error was lower than 10%, we applied another test. This test confirmed that these values were not higher that 10%. In conclusion, we can confirm that the three models are well specified. Table 10 Results specification test Table 10. Results specification test Country Nº Regressions Jones Jones modified Jones adjusted to ROA ET1* Ho: P=0,1 ET1* Ho: P=0,1 ET1* Ho: P=0,1 U.S. 500 0.08** 0.07 0.09 0.46 0.12 0.18 U.K. 500 0.12 0.14 0.08 0.10 0.10 0.77 Japan 500 0.07** 0.05 0.12 0.23 0.12 0.18 Italy 500 0.11 0.5 0.11 0.55 0.11 0.55 France 500 0.11 0.37 0.12 0.14 0.09 0.30 Spain 500 0.10 0.88 0.09 0.30 0.11 0.37 TOTAL 3000 0.10 0.57 0.10 0.85 0.10 0.42 * Numbers of ET2 are expressed in parts hundred. **In these cases, to reject the null hypothesis, the hypotheses that were less than 0.1 confirmed that the argument was tested. * Numbers of ET2 are expressed in parts hundred. ** In these cases, to reject the null hypothesis, the hypotheses that were less than 0.1 confirmed that the argument was tested. 12 N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 Table 11 shows the results for type II errors. Following Dechow et al. (1995),Jeter and Shivakumar (1999), and Peasnell et al. (2000), we assess the earnings management detection power of the three accrual models using known magnitudes of simulated earnings management. This potency test refers to the number of times type II errors are committed; namely, the capacity of each model to detect discretionary accruals. To calculate type II errors, we manipulate the variables to work out the specification of each model. These variables increased from 1% to 20% in relation to the initial values. The table shows that the Jones model adjusted to ROA produces a lower type II error percentage: The percentage error of this model is 30.3% compared to 35.8% and 37.5% for the Jones model and the modified Jones model, respectively. Table 11 shows the high power of the Jones model adjusted to ROA in detecting discretionary accruals for Spain, producing an error of only 16.4%. However, in Italy, the Jones model adjusted to ROA does not obtain a good approximation to discretionary accruals, having a percentage of error of 58.6%. In the other countries, the percentage of type II errors is around the mean of the Jones model adjusted to ROA (30.3%). Table 11 Results potency test Table 11.Results potency test Country Jones Jones modified Jones adjusted to ROA Nº Regressions ET2* Nº Regressions ET2* Nº Regressions ET2* U.S. 20000 0.086 20000 0.328 30000 0.261 U.K. 20000 0.218 20000 0.239 30000 0.247 Japan 20000 0.499 20000 0.522 30000 0.359 Italy 20000 0.598 20000 0.616 30000 0.586 France 20000 0.537 20000 0.372 30000 0.201 Spain 20000 0.212 20000 0.173 30000 0.164 TOTAL 120000 0.358 120000 0.375 180000 0.303 * ET2Numbers are expressed in parts hundred. * ET2 Numbers are expressed in parts hundred. Given our previous findings, we conclude that the Jones model adjusted to ROA is better to calculate discretionary accruals than the Jones model and the modified Jones model. For this reason, all subsequent calculations use the Jones model adjusted to ROA. Kothari et al. (2005) propose the Jones model adjusted to ROA based on the idea that accruals are correlated with the company’s current and past performance . Specifically, to increase the model specification when firms experience extreme financial performances, Kothari et al. introduce ROA to the Jones model as a control variable. The Jones model adjusted to ROA is represented in Equation (1). TACCt At−1=β0 At−1+β1∆REVt At−1+β2PPEt At−1+β3ROAt+e (1) where, TACCt =total accruals in year t REVt =changes in revenue in year t PPEt =property, plant and equipment in year t ROAt =return on total assets in year t At-1 =the total assets at the beginning of year t e is the error term of the accruals of the accrual model, which represents discretionary accruals, and the t subscripts represent years. Independent variables Related with auditor quality, we define three variables: auditor size, auditor tenure, and auditor industry specialization. In accordance with the previous literature (DeAngelo, 1981;Krishnan, 2003a;Boone et al, 2010;Ahsen, 2011), we consider auditor size as a proxy of auditor quality. Hence, we define a variable, SIZE, which is equal to 1 if the auditor is a Big Four auditor firm (Deloitte, Ernst & Young, KPMG, or PricewaterhouseCoopers), and zero otherwise. Table 3 reports the composition of audit firms, distinguishing two groups according to auditor size. From 2005 to 2007, the representation of the Big Four auditors in the companies was around 75%. However, with the beginning of the economic crisis, this percentage increased to 94%. Thus, the economic crisis motivated firms to contract services from a Big Four auditor to guarantee the quality of accounting statements. Table 3 Composition of the audit firms for the period 2005-2009 by auditor size Table 3. Composition of the audit firms for the period 2005-2009 by auditor size Percentage audit firm by year. EY: Ernst & Young; PWC: Pricewaterhouse Coopers; DT: Deloitte; KPMG: KPMG. Audit firm 2005 2006 2007 2008 2009 Big 4 auditor EY 16.8% 19.5% 18.3% 25.8% 24.4% PWC 22.1% 24.6% 22.3% 25.7% 25.9% DHS 19.1% 21.3% 20.7% 22.7% 23.5% KPMG 13.5% 13.2% 14.5% 19.8% 20.3% Non-Big 4 auditor 28.3% 21.4% 24.3% 6.0% 5.9% Percentage audit firm by year. EY: Ernst & Young; PWC: Pricewaterhouse Coopers; DT: Deloitte; KPMG: KPMG. Following prior studies (Myerset al., 2003;Ghosh and Moon 2005), we measure auditor tenure as the cumulative number of years that the auditor has been employed by the firm. We do not employ a continuous measure for auditor tenure because the relationship between auditor tenure and audit quality may not be linear. Instead, we define two variables to capture the effect of tenure on audit quality. First, we use the number of consecutive years that the firm has retained the auditor (TENURE). Second, we define another variable that represents the cumulative number of years squared that the auditor has been employed by the firm (TENURE2). Regarding auditor industry specialization, the literature provides several criteria to measure the degree of auditor industry specialization. First, Yardley et al. (1992) and Know (1996) estimate industry specialization by the proportion of an auditor’s audit fees earned from one industry of all those served. This approach is widely known as auditor portfolio share. Second, Gramling and Stone (2001) and Krishnan (2003b) calculate the auditors’ industry expertise as a proportion of the total audit fees earned by all the auditors who serve that particular industry. This second approach is known as auditor industry market shares. Both approaches use sales or assets as the base to estimate the proportion of audit fees because audit fee information is largely unavailable. Based on the arguments from Krishnan (2001, 2003b), which find that the portfolio share approach captures the effort of the auditors to differentiate their products better than the industry market share approach. So, the portfolio share approach may be a better proxy for auditors’ industry expertise than the industry market share. Consistent with these arguments, we use the auditor portfolio share approach as a proxy for auditor industry specialization. We define auditors with a large industry market share (defined as a two-digit SIC code) as the specialist (SPEC). We consider an auditor as having a large market share in the in- N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 13 dustry if the auditor holds at least 15 percent of that industry (Krishnan, 2003b). Table 4 reports the auditor industry specialization percentage, considering whether the auditor firm specialist is a Big Four auditor. Specialist firms are the most representative and account for around 65% of the observations. This percentage has increased each year except in 2009. What is more, in 2008 and 2009, the percentage of other specialized audit firms decreases considerably from 15.20% in 2007 to 0.49% in 2008. Table 4 Auditor specialization for the period 2005-2009 Table 4. Auditor specialization for the period 2005-2009 Percentage audit firm by year. EY: Ernst & Young; PWC: Pricewaterhouse Coopers; DT: Deloitte; KPMG: KPMG. Audit firm 2005 2006 2007 2008 2009 Specialist EY 12.4% 13.3% 14.7% 22.8% 18.1% PWC 19.7% 18.1% 17.3% 21.5% 23.3% DT 9.5% 13.3% 15.6% 11.9% 10.2% KPMG 6.2% 8.1% 9.6% 14.8% 10.9% OTHERS 15.3% 9.1% 13.8% 0.6% 0.9% TOTAL 63.2% 61.7% 70.9% 71.8% 63.5% Non-Specialist EY 4.6% 3.4% 3.6% 3.0% 6.3% PWC 2.4% 3.0% 5.0% 4.1% 2.6% DT 9.6% 4.9% 5.1% 10.7% 13.3% KPMG 7.3% 3.3% 5.0% 5.0% 9.3% OTHERS 13.0% 23.8% 10.4% 5.4% 4.9% TOTAL 36.8% 38.31% 29.1% 28.3% 36.5% Percentage audit firm by year. EY: Ernst & Young; PWC: Pricewaterhouse Coopers; DT: Deloitte; KPMG: KPMG. The economic crisis had important consequences for the development of economic activity by creating a strong climate of hostility. Given our interest in the impact of the economic crisis on earnings management, determining the crisis time framework is key to our study. While many definitions exist for a crisis, we define this following Guenther and Young (2002). They use two variables: income per capita and the unemployment rate. When both variables evolve in a negative way in two consequent years, that country is defined as being in crisis. Table 5 shows how both variables reacted since the beginning of the economic crisis. The onset of the crisis cannot be set uniformly for all six countries. The starting point of the crisis is 2007 in the United States, 2008 in Japan, Italy, Spain, and France, and 2009 in the United Kingdom. Accordingly, we define a dummy variable, CRISIS, which equals 1 if the country is in crisis, and zero otherwise. Table 5 Evolution of the macroeconomic variables Table 5. Evolution of the macroeconomic variables Country Variable Variation 05-06 Variation 06-07 Variation 07-08 Variation 08-09 Mean variation U.S. Income per capita 0.050 0.039 0.012 -0.027 0.019 % unemployment -0.098 0.000 0.261 0.376 0.135 U.K. Income per capita 0.065 0.143 -0.061 -0.231 -0.021 % unemployment 0.152 -0.019 0.019 0.312 0.116 Japan Income per capita -0.042 0.003 0.115 0.032 0.027 % unemployment -0.068 -0.049 0.026 0.200 0.027 Italy Income per capita 0.042 0.127 0.077 -0.094 0.038 % unemployment -0.117 -0.103 0.098 0.141 0.005 France Income per capita 0.049 0.138 0.090 -0.085 0.048 % unemployment 0.000 -0.091 -0.075 0.187 0.005 Spain Income per capita 0.075 0.148 0.089 -0.097 0.054 % unemployment -0.076 -0.024 0.361 0.372 0.159 Control variables Our model includes a group of control variable to account for certain firm characteristics, namely:, firm size, leverage, and growth opportunities. The size of the firm is defined as the log of total assets (LNTAB), firm leverage is measured with the ratio of total debt to total assets (DTBA) and the growth opportunities are measured as the ratio of equity to market-to-book value (MTB). In addition, to control for some institutional effects, we introduce a dummy variable (LAW), which equals 1 if the firm belongs to a common law country, and zero otherwise. We also control for the sector. We include dummy variables as well to control by the sector. Table 6 provides the definitions of the independent variables. Table 6 Definition of variables Table 6. Definition of variables Abbreviation Variable Definition Expected Sign DACC Discretionary accruals The absolute value of the residuals of the Modified Jones Model. CRISIS Economic crisis Dummy variable: 1 if the country has been affected by the economic crisis; 0 otherwise. +/- SIZE Auditor size Dummy variable: 1 if the audit firm is a Big Auditor; 0 otherwise. - TENURE Auditor tenure Cumulative number of years the auditor has been employed by the firm. - SPEC Auditor specialization Dummy variable: 1 if the audit firm is the specialist in the industry; 0 otherwise. - LNTAB Firm size Total assets (logarithm) - DTBA Leverage Book value of debt /Total asset + MTB Growth opportunities Market to book ratio ROA Profitability Return on total assets + LAW Legal system Dummy variable: 1 if the country belongs to the Anglo- Saxon corporate system; 0 otherwise. - SDACC Sign of discretionary accruals Dummy variable: 1 if the company has negative discretionary accruals; 0 otherwise. +/- Table 7 displays the descriptive statistics of the most characteristic variables. Based on the statistics reported in Table 6, the average DACC of firms equals 0.063 (standard deviation =0.055). According to the variables about audit firms, the mean number of years of auditor tenure is 6. In addition, the variables regarding firm characteristics show big companies according to size, leverage and growth opportunities. Table 8 reports the correlation matrix between the main variables of the model. Although the correlation coefficients are not high, we compute the variance inflation factor (VIF) to test the lack of multicollinearity in our estimations. The VIF is well under 10, which is the threshold value for multicollinearity (Kutner et al., 2005). 14 N. Reguera-Alvarado et al. /Revista de Contabilidad Spanish Accounting Review 22 (1)(2019) 6-20 Table 7 Descriptive statistics of the main variables Table 7. Descriptive statistics of the main variables Mean, standard deviation, quartile one, median and quartile three of the variables. DACC is the absolute value of discretionary accruals; CRISIS is a dummy variable equal to 1 if the country has been affected by the economic crisis and 0 otherwise; SIZE is a dummy variable equal to 1 if the audit firm is a Big Auditor and 0 otherwise; TENURE is the number of years for which the current auditor has audited the client; SPEC is a dummy variable equal to 1 if the audit firm is the specialist in the industry and 0 otherwise; LNTAB is the size of the firm; DTBA is the financial leverage ratio; MTB is the growth opportunities; ROA is the return on total assets; LAW is a dummy variable equal to 1 if the country belongs to the Anglo-Saxon corporate system and 0 otherwise; SDACC is a dummy variable equal to 1 if the company has negative discretionary accruals and 0 otherwise. Variable Mean Std. Dev. Q1 Median Q3 DACC 0.061 0.053 0.021 0.046 0.086 CRISIS 0.436 0.496 0.000 0.000 1.000 SIZE 0.819 0.385 1.000 1.000 1.000 TENURE 6.383 3.495 3.000 7.000 9.000 SPEC 0.680 0.466 0.000 1.000 1.000 LNTAB 13.491 2.625 11.380 13.769 15.658 DTBA 0.525 0.239 0.350 0.534 0.689 MTB 2.558 1.688 1.360 2.079 3.312 ROA 0.049 0.071 0.021 0.045 0.081 LAW 0.508 0.500 0.000 1.000 1.000 SDACC 0.496 0.500 0.000 0.000 1.000 Mean, standard deviation, quartile one, median and quartile three of the variables. DACC is the absolute value of discretionary accruals; CRISIS is a dummy variable equal to 1 if the country has been affected by the economic crisis and 0 otherwise; SIZE is a dummy variable equal to 1 if the audit firm is a Big Auditor and 0 otherwise; TENURE is the number of years for which the current auditor has audited the client; SPEC is a dummy variable equal to 1 if the audit firm is the specialist in the industry and 0 otherwise; LNTAB is the size of the firm; DTBA is the financial leverage ratio; MTB is the growth opportunities; ROA is the return on total assets; LAW is a dummy variable equal to 1 if the country belongs to the Anglo-Saxon corporate system and 0 otherwise; SDACC is a dummy variable equal to 1 if the company has negative discretionary accruals and 0 otherwise. Methodology As usual in this line of research, we divide our investigation into two stages. First, we estimate total accruals and compute the discretionary component using equation (1). Second, using the absolute value of discretionary accruals, we analyze the relation between the discretionary accruals and our independent variables, as shown in Equation (2) DACCi,t=β0+β1CRISIS +β2SIZE +β3TENURE+ β4TENURE2+β5SPEC +β6LNTABi,t+β7DTBAi,t+ β8MTBi,t+β9ROAi,t+β10LAWi,t+β11SDACCi,t+ 5 ∑ J=1 β12DU M_SECTjt +ϵit (2) where β0is the intercept and βiis the coefficient of each independent variable. The sub-index i identifies the individual and the sub-index t the year; µirepresents the fixed individual effect; and it, the stochastic error. The stochastic error term combines both the measurement errors of any independent variable and the omission of explanatory variables. We use time-series and cross-sectional data to form the panel data. The panel data approach enables us to control for the unobservable constant heterogeneity or fixed effects term Arellano 2003. This term is intended to reflect the firmlevel characteristics and it thereby avoids the omission bias and renders more efficient estimates. The fixed effects term is unobservable and, consequently, is subsumed in the random disturbance. A key element in panel data is the relation between the fixed effects term and the other explanatory variables. This correlation is analyzed with the Hausman test, which examines the null hypothesis of the lack of correlation between the independent variables and the fixed effects term.7Accordingly, we use the Hausman test to choose the most suitable estimation method. Results To check the consistency of our results, we run four different models. Table 12 provides the results. We initially include all the independent variables (column 1). Then, we check our model proposed in Section 3.3., excluding auditor size (column 2). We next include all the independent variables except those related to auditor tenure (column 3). Finally, we use all the independent variables except those concerning auditor specialization (column 4). In all the models, we also test the effect of the crisis, both as an independent variable and interacted with the other independent variables. According to the results obtained from the Hausman test, random effects (RE) have been used in the regression analysis. Our results show a positive effect of the CRISIS variable on the dependent variable. That is, the economic crisis creates incentives for managers to manipulate earnings. This result is in line with Davis and Stobaugh (1995),Lin and Shih (2003), and Kousenidis et al. (2013). Earnings management may occur for different reasons, including the desire to influence stock prices and benefit executive compensation schemes. Another possible motivation is loss-avoidance, so that managers have incentives to modify the income to reach targets set before the crisis. Another possible motivation is “big bath” practices, where managers boost losses in a context of generalized losses in their industry to show a better performance in the coming years. In any case, our results suggest that the economic crisis creates an environment in which managers are under pressure or have incentives to modify earnings. Regarding our first hypothesis on the relation between audit size and earnings management, we find a positive effect of the SIZE variable. This could be explained by the recent high-profile reports in the United States, the United Kingdom and the European Union, which have raised concerns about the Big Four accounting firms’ concentration of market share and the potentially adverse effect that this concentration may have on audit markets and the quality of audits in these legal jurisdictions (Francis, et al 2013). In line with our argument, the majority of sample firms are audited by a Big Auditor (Table 4), and Cahan et al. (2011) suggest that the large market shares of the Big Four are due to a decline in both the quality and the costs of the audit. Notwithstanding, we consider as a dependent variable the absolute value of discretionary accruals and large values for this variable include large positive and negative accruals. Concretely, our sample is made up of large negative discretionary accruals (mean=-0.063; std. dev.=0.052) in relation to large positive accruals (mean=0.059; std. dev.=0.054). This means that the positive association between audit size and earnings management is explained by auditor conservatism, since auditors are more permissive with decreasing earnings (Huguet and Gandía, 2016;Kim et al, 2003). In this sense, research 7This test follows a chi-squared distribution with as many degrees of freedom as estimated coefficients. When the null hypothesis is rejected, the fixed effects term must be eliminated with the within groups technique. Otherwise, the random effects method applies.