Effect of European audit firms on cost of debt and earnings management in private clients ’ audit market segment

This research studies the relation between audit firm choice and benefits that companies could gain in terms of lower cost of debt and earnings management. It focuses on private clients and the non-Big4 audit market segment, where the main driver of auditor choice has not to date been satisfactorily identified. This study identifies and tests a new criterion for auditor choice in private firms based on audit market boundaries (European vs. Domestic audit firms). Using a propensity score matched sample of private companies audited by non-Big4 audit firms in the period 2010 to 2014; this research finds that the choice of a European audit firm is negatively associated with cost of debt and earnings management. Private firms that choose audit firms operating at European level, as consequence, have lower cost of debt and earnings management, mitigate the agency conflicts between lenders and owner/manager, and improve their corporate governance mechanisms.


INTRODUCTION
The non-Big4 private clients' audit market segment is an interesting topic: the Green paper (European Commission, 2010) for example, is against the concentration of audit market and aims to favor the development of non-Big4 audit firms: "The Commission recognizes that continuity in the provision of audit services to large companies is critical to financial stability.To this extent, options such as the ramping up of the capacities of non-systemic firms and exploring the pros and cons of "downsizing" or "restructuring" systemic firms should be further examined.The Commission would also like to explore the possibilities to reduce existing barriers to entry into the audit market, including a debate on existing ownership rules and the partnership model employed by most audit firms." The aim of this study is to explore the benefits in term of cost of debt and earning management of a new criterion (European audit firm vs.Domestic audit firm) to choose the auditors in private clients' audit market segment.
This research contributes to the literature identifying an original audit firm choice criterion that, coherently with the framework of DeFond and Zhang (2014), suggest useful instruments for the evaluation of audit quality from the point of view of auditor supply, using auditor competences, reputation, and litigation risk.Given the gap in the previous literature that show that the current criterium to choose an auditor based on size is not sufficient among non-Big4, this study suggests a criterion based on the European boundaries of the audit market, showing its effectiveness in the reduction of CoD and EM, as an opportunity for clients to mitigate the agency conflict between lenders and managers in private firms through the choice of an EAF.The higher audit quality offered by EAF reduces risks related to earnings management and allows lenders to accept lower level of interests with benefits for all stakeholders.
Audit firm choice is a significant decision that may affect agency conflicts.Literature has widely analyzed the effects of audit firm choice, finding several benefits associated with Big4, such as lower Cost of Debt (CoD), Earnings Management (EM) and agency costs.These benefits are usually connected with high reputation auditors that reduce the litigation risks.However, these results are mainly related to Big4 of public clients, while for private firms and non-Big4 segment findings are mixed and it is an empirical question, which are effective criteria for the selection of audit firms.In countries with competitive audit market of private firms, effective audit firm choice criteria among non-Big4 have not been clearly identified yet.
Literature also analyzes and finds mixed results about difference between second-tier and third-tier, classifying audit firms based on market share as defined by Public Company Accounting Oversight Board (PCAOB).However, in U.S. they are analyzed under the same regulations, reputation and litigation environment while in Europe the new classification here introduced is based on different environments for audit firms.
This research tests the effects on CoD and EM of the choice of European audit firms (EAF) instead of domestic audit firms (DAF).In private firms, CoD is one of the most important drivers of managers' choices, given that debt is usually a significant financial resource and that the main agency conflict is between lenders and managers/owners.On the other hand, agency conflict between lenders and owners/management can also create EM incentives (Watts and Zimmerman, 1986;Li, 2017).
Italy is an interesting setting to investigate because: a) the non-BigN audit market share is significant in the private company segment (around 40%); b) auditors are liable to third parties (Giudici, 2012).
i Investigating agency conflict between lenders and owners/manager is important because lenders care about audit quality and have the power to sue auditors.Competitive audit markets with auditor liability towards third parties occur also in Sweden, Belgium, Denmark and Finland, and are analyzed in the robustness test.In all these countries, creditors can sue auditors, and the non-BigN market share in private firms is respectively 18, 54, 70 and 55%.O'Sullivan (1993) discusses the extension of liability to third party in the United Kingdom.Anantharaman et al. (2016) explore the extent to which auditors can be held liable by third parties for negligence and find that auditors are more likely to issue a modified going-concern report to financially distressed clients from high-liability states than to those from low-liability states.
Considering the endogeneity issue in the research about auditor choice, raised for example, by DeFond and Zhang (2014), this study uses a propensity-score matched sample of Italian companies audited by non-Big4 in the period 2010-2014.As expected, clients of EAF are associated with lower CoD and lower EM than clients of DAF.A battery of robustness tests run on alternative measures of CoD, EM, PSM, size, accounting standards, other countries with high third-party liability confirm our main results.

LITERATURE REVIEW
Literature review is based on the framework of DeFond and Zhang (2014) and we develop our hypothesis in the big picture of audit quality demand, supply and regulatory intervention.

Demand for audit quality
Clients have incentive to increase audit quality in order to lower agency costs.Literature on agency conflict in private firms finds that as the demand for financial reporting and for external audits mainly arises from the need for debt contracting with banks and other private lenders (Lennox, 2005), principals are typically lenders (Peek et al., 2010;Power, 1997;Vander Bauwhede and Willekens, 2004).A bank may place more trust in client financial reporting and reduce the CoD when a high quality auditor assures it.Previous old studies (Kelly and Mohrweis, 1989;Libby, 1979a, b;Strawser, 1994) as well as recently studies (Baylis et al., 2017;Robin et al., 2017;Chen et al., 2016) show that banks tend to form different perceptions according to the level of audit firm quality.Unlike public companies where internal corporate governance mechanism or surveillance of market authorities may mitigate agency costs, in private firms, audit quality may be the only available instrument to mitigate them (Cano-Rodríguez and Alegría, 2012).Moreover, Gul et al. (2013), analyzing data from several countries in the period 1994 to 2006, find that Big4 choice is related to lower CoD only in countries with stronger 2) Given that audit firm size, among non-Big4 segment, is not effective, we suggest a new audit firm choice criterion (European audit firm vs Domestic audit firms).
3) Clients have incentives to increase audit quality to reduce agency costs and agency conflicts between lenders and manager.4) We expect that European audit firm, through higher audit quality, is associated with lower cost of debt, earnings management and agency costs.Source: Adapted from DeFond and Zhang (2014).
investor protection.
Agency conflicts between lenders and owners/management can also create EM incentive, enhanced in the case of earnings-based debt covenants (Watts and Zimmerman, 1986;DeFond and Jiambalvo, 1994;Sweeney, 1994;Dichev and Skinner, 2002;Gao et al., 2017;Li, 2016).Note also that, especially after the Basel accords, the stability of the banking and financial system has been found to critically depend on client financial reporting transparency (Bushman and Landsman, 2010), making earnings an attribute of crucial importance.Vander Bauwhede et al. (2003) show that in Belgium, BigN constrain EM more than non-BigN only when the company manages earnings opportunistically to have earnings above the benchmark target of prior-year earnings, or where there is incentive to smooth earnings downwards.In other circumstances, BigN do not place any more constraint on EM than non-BigN.Vander Bauwhede and Willekens (2004) use different proxies to measure audit size (auditor market share, number of audit firm clients, number of partners in the audit firm, total assets and operating profit of the audit firm) and again find no significant reduction of EM in Belgian private companies when the audit firm is a BigN firm.Van Tendeloo and Vanstraelen (2008) examine the impact of audit quality on earnings quality in private firms in six European countries.They argue that in countries with a close alignment between tax accounting and financial reporting, financial statements are scrutinized more closely by the tax authorities, which makes the detection of audit failure more likely.They find that Big4 auditors constrain EM more than non-Big4 auditors in private firms, but only in countries with a high tax alignment (Belgium, Finland, France and Spain) compared to low tax alignment countries (The Netherlands, UK).They also categorize non-Big4 auditors into Second-tier and small auditors, but find no indication that the Second-tier auditors constrain EM more than small auditors.
The research proxies the agency costs with CoD and EM and tests how they are affected by auditor choice in private firms and in the non-Big4 audit market segment.
ii Figure 1 show how the demand for audit quality is investigated through CoD and EM and how it is related to the supply of audit quality from EAF vs DAF.

Supply of audit quality
Among the several factors that affect audit quality, the paper focuses on auditor choice criteria among non-Big4 in private firms.These criteria are usually based on audit firm size, auditors reputation and litigation risks.
Previous literature typically compares BigN and non-BigN and, in public firms, find several benefits associated with BigN and their public clients.BigN provide higherquality audits in order to protect brand name reputation from legal exposure (DeAngelo, 1981;Francis and Wilson, 1988;Simunic and Stein, 1987;Firth, 1999;Lennox, 1999;Tomczyk, 1996).Some of benefits gained when audited by a Big4 are lower CoD (Gul et al., 2013;Pittman and Fortin, 2004;Mansi et al., 2004;Causholli and Knechel, 2012) and higher EQ (Becker et al., 1998;Francis et al., 1999a;Teoh and Wong, 1993;Nelson et al., 2002;Kim et al., 2003;Gaver and Paterson, 2001; (3)  Gerayli et al., 2011;Francis et al., 2009;Tsipouridou and Spathis, 2012;Porte et al., 2015).Specifically, DeAngelo (1981) agency-based framework suggests that large audit firms with large numbers of clients entail higher reputation costs as collateral against poor-quality audits.Large clients, particularly those with multinational operations, demand consistent auditing throughout the world, for example from a global audit firm network (Carson, 2009): he argues that global audit firm networks have competitive advantages not available to domestic audit firms.These advantages include knowledge of diverse business practices, an ability to operate across multiple business environments, expertise developed from servicing similar clients in different locations, robust and efficient audit methodology and processes, knowledgeable and expert professional staff, the ability to develop specific industry training and protocols as competences, and superior brand image as well as reputation.Competitive advantages attract clients seeking higher quality audits.
Firm size advantages have been studied also outside auditing.Larger firms interact with a greater number and variety of stakeholders, which would influence the complexity and multidimensionality of any formalized policy (Hart and Sharma, 2004).Larger firms presumably have more resources in the form of human and financial capital (Gallo and Christensen, 2011).Due to functional differentiation, specialization, and decentralization (Damanpour, 1987;Moch, 1976) larger firms have more specialized staff, more evolved administrative processes, and have more sophisticated internal systems to deal with business issues (Damanpour, 1996;Baumann-Pauly et al., 2013).Moreover, taking the perspective of legitimacy theory, some earlier studies were inspired by the argument that firms may increase the quality to hedge reputational risks and to prevent or to react to attacks from powerful stakeholder groups, such as customer pressure groups, and the media (Bansal and Clelland, 2004;Chatterji and Toffel, 2010;Schreck and Raithel, 2015).
The literature also analyzes Second-tier and/or Thirdtier audit firms, based on market share as defined by Public Company Accounting Oversight Board (PCAOB), but finds mixed results, especially in private firms: for example, prior research (Chang et al., 2010;Cassell et al., 2013;Wang and Fan, 2014;Jenkins and Velury, 2011;Weber and Willenborg, 2003) finds a significantly higher audit quality for Second-tier while others do not (Van Tendeloo and Vanstraelen, 2008;Geiger and Rama, 2006).
Previous literature in short shows that size is a significant audit firm choice criterion in public companies.However, in private firms and the non-Big4 segment, it appears to be not sufficient (Lawrence et al., 2011) to differentiate the capacity of audit firms to reduce the agency conflicts.This capacity implies greater resources to invest in training professionals to detect errors.Moreover, auditor size is sensitive to macro-economic effect (Fleischer and Goettsche, 2012).Hodgdon and Hughes (2016) also discuss the dishomogeneity of disclosure quality when audited by one Big4 versus the other Big4.Empirical research is required to identify criteria used by private firms in choosing audit firm, among non-Big4.

Hypothesis development
Non-Big4 has a significant audit market share in Italy (nearly 40%) and in several other European countries (e.g.Belgium, Denmark and Finland) in private firms.The research looks for a new audit firm choice criterion that assure the same benefits in terms of lower CoD and EM that previous literature found in public clients audited by Big4.Following previous literature, it developed our new criterion based on reputation, competences and litigation risks.Finally, it includes this criterion in the category of supply in the framework of DeFond and Zhang (2014).
The research analyzes the boundaries of the audit market addressed by non-Big4.Given that European Union Directives (European Parliament, 1984Parliament, , 2006a) ) allow audit firms to operate in all member countries, it develops our hypothesis suggesting the classification of audit firms into two groups: 1) European Audit Firms (EAF) that work at European level and 2) Domestic Audit Firms (DAF) that work only in Italy.
The paper investigates differences in the quality of audit firms with clients located in European Union (EAF) and Domestic audit firms with clients located only in one country (DAF).EAF can be viewed as an extension of DeAngelo (1981) arguments where the creation of EAF with high competences and reputation is one way to manage the provision of high-quality audit services to clients.These advantages can be the same for different EAF but may not be available for DAF.The capacity to satisfy clients operating at European level requires legal, fiscal, social and environmental expertise of the country of operation.Demartini and Trucco (2016) have shown how auditor's experience is perceived important from surveys to partners.EAF, moreover are facing additional mandatory competence requirements.A domestic audit firm wishing to perform an audit in another European Union country needs to have a partner, which has passed an aptitude knowledge test of the legislation of that country iii .Thus, the research expects that the choice of hiring an EAF with more competences and reputation than a DAF is associated with lower CoD and EM.
Higher expected quality from EAF is also a result of stricter audit environment stemming from the higher enforcement and litigation risk present in different European countries, given that firms enter in the audit environment of each state where they want to operate.Audit firms that operate in more than one country have to adapt to different enforcement regulations.A stricter audit environment and more enforcement regulations promote audit quality.Maijoor and Vanstraelen (2006) find that a stricter audit environment in a European member state lowers EM compared to other member states.Van Buuren et al. (2014) find that enforcement by audit supervisory authorities is one of the important factors explaining the use of business risk perspectives.Willekens and Simunic (2007) study the joint liability between directors and auditors and the relation on audit effort.Kleinman et al. (2014) argue that it is important to investigate the auditing regulatory regimes in different nations around the world, as well as the nature of crossborder audit inspections and their effect on AQ.There are different auditor liability regimes in the EU, such as the capped versus uncapped liability regimes, and this different litigation risk has a different potential effect on audit quality (EC DG, 2006).
The counterargument is that DAF are more specialized in the country where they operate.Following Francis et al. (1999b) and Ferguson et al. (2003), Francis and Yu (2009) argue that accounting professionals are typically based in specific practice offices and audit clients in the same geographic location.This decentralization reduces information asymmetry and enables auditors to develop better knowledge of existing and potential clients in a particular location.Clients, in turn, have greater knowledge of and confidence in the expertise of locally based personnel who actually perform audits (Carcello et al., 1992).The same argument could be made for DAF: through the specialization in one country they may have better knowledge in a particular location.Moreover, Vera-Muñoz et al. (2006) point out that firm-wide knowledge sharing has practical limitations, and for this reason, it is an open empirical question as to what extent these firmwide mechanisms can effectively increase the hypothesized European effect.
The paper developed our hypothesis in private clients and non-Big4 audit firms.The effect of auditor choice is largely unknown for non-Big4.Competence acquired in operating at European level could have higher marginal value.In the U.S., non-Big4 have been mainly analyzed dividing them into Second-tier and Third-tier audit firms, or into internationalnationallocal audit firms (Beattie and Fearnley, 1995).It introduces the category of EAF (similar to national level) and DAF (similar to local level).The main difference between local and national audit firms in U.S. is related to the number of clients.However, local and national audit firms in the U.S. are under the same regulations and therefore the same reputation and litigation environment.In Europe, the environment is different for EAF and DAF and the paper contributes to the literature testing this audit firm choice criterion.Given previous literature results on reputation, competences and litigation risk, we decide to develop the analysis in the form of a directional hypothesis, with two multivariate regression models respectively for CoD and earnings management: Hp1: Private clients of European Audit Firms have lower

METHODOLOGY
Here presents the sample selection, the audit firms classification and the data collection strategies to identify EAF and DAF; the statistical regression models for CoD and earnings management used to test the hypothesis as well as the propensity score matching model to solve the problem of endogeneity.Prior literature found specific determinants for CoD (quick ratio, ROA, tangible, negative equity, loan maturity) and for earnings management (loss, sales growth, cash flow from operation and its variability), thus we decided to use different regression models.

Sample selection
The sample includes all 1149 Italian companies audited by non-Big4 audit firms (firms with two or more individual owners) with more than one client per year, appearing in Bureau Van Dijck database (Table 1).
We firstly drop public companies because they cannot choose among the different types of audit firms here analyzed, leaving a sample of 895 firms.iv The number of firm-year observations for the period 2010 -2014 for these is 4435.In the period analyzed in this research (2010 -2014), Italian auditors used national auditing standards.These standards are similar to International Standards of Audit (ISA), and meanwhile Italy is moving towards their implementation v .This database includes only the name of the last audit firm engaged and the year of its engagement.Two downloads, one in 2012 and one in 2014, thus supplied the name of the firm that audited the list of clients in our sample at the end of 2012 and at the end of 2014.For each of the audit firms we have the starting year of the engagement.We include only the years for which we know that the audit firm was auditing a specific client, resulting in a sample period different for each firm (unbalanced sample).All the firms in our sample voluntarily choose an external audit firm vi .The problem of self-selection of the sample is lower than in prior studies because the comparison is not with firms that do not undergo audit, but between the types of audit firm that they engage.All the firms in the sample undergo audit.
Secondly, we compute the CoD and we drop observations with missing values for this variable.The final sample used in the logistic regression of the auditor choice model consists of 1798 observations.PSM yields a sample of 1206 observations to be used in the main analysis (Panel A, Table 1).
Thirdly, we compute abnormal accruals and we drop observations with missing values for this variable.The final sample used in the logistic regression of the auditor choice model consists of 1162 observations.PSM yields a sample of 950 observations to be used in the main analysis (Panel B, Table 1).
The industry composition vii of our sample of private firms reflects the industry composition of firms in Italy, with a higher percentage of professional, technical and scientific services, construction activities, wholesale and retail trading; transport and storing activities; lodging and catering services; real estate; hiring services and travel agencies.Other industries represented are manufacturing, electric energy and gas supply; water supply and garbage disposal activities; information and communications.Percentages are lower for entertainment and sport activities; other services, agriculture, forestry and fishing; and minerals extraction (untabulated).

Audit firms classification
Most of the U.S. literature (Francis et al., 1999b  Ireland, Belgium, Netherland, and Luxembourg.viii We checked premises and offices on their websites, to ensure that they actually operate there.We thus defined our sample of audit firms on the basis of the number of clients in more than one country (reputation and litigation risk) and on qualification requirements (competences) required for auditing in the countries selected.Table 2 shows the number of EAF, and names are shown in Appendix A.

Multivariate regressions models
Our model tests the effect of EAF on CoD and EM in private firms.

The Cost of Debt (CoD) model
The CoD model is the following Equation (1): CoD is the average cost of financial debts for firm i and year t, which is the financial cost disclosed in the income statement following Generally Accepted Accounting Principles (GAAP) in Italy, scaled by the total amount of financial debts.Independent control variables were selected on the basis of numerous prior studies on CoD (Kim et al., 2011;Aobdia et al., 2015;Chin et al., 2014;Petersen and Rajan, 1994;Bharath et al., 2008;Karjalainen, 2011;Graham et al., 2008;Lai, 2011;Pittman and Fortin, 2004).The literature on cross-sectional determinants of loan pricing, in general, finds that firm SIZE is inversely related to credit risk.Agency theory predicts that the risk of agency conflicts, such as risk shifting and underinvestment, between a firm's insider and outside lenders increases with financial leverage and leverage maturity structure.To control for this, we include LEVERAGE (Kim et al., 2011;Bharath et al., 2008;Graham et al., 2008;Aobdia et al., 2015;Karjalainen, 2011;Pittman and Fortin, 2004).QUICK or current ratios have been used in prior studies as a proxy of financial risk.Firms with a low value of this ratio may be suffering from liquidity problems, and they may be forced to use more expensive credit (Bharath et al., 2008;Aobdia et al., 2015).It is important to control for profitability through ROA; banks and other private lenders are likely to charge lower interest rates to firms that are more profitable because such firms are better able to service their debt (Kim et al., 2011;Graham et al., 2008;Aobdia et al., 2015).We include TANGIBLE in order to have a measure of asset composition as determinant of CoD.The loan pricing literature suggests that owning tangible assets is inversely related to credit risk, given that they can work as collateral and, thus, the interest rate that lenders charge (Bharath et al., 2008;Aobdia et al., 2015;Graham et al., 2008;Kim et al., 2011;Karjalainen, 2011;Pittman and Fortin, 2004).We include the ALTMAN score of bankruptcy because debt holders may demand higher interest to cover this higher risk (Lai, 2011;Bharath et al., 2008;Graham et al., 2008;Aobdia et al., 2015).Lower values indicate more financial distress, so that a negative association is expected with accrual.Because about 2.8% of private Italian companies in our sample experienced negative equity during the sample period, we include the NEGATIVE EQUITY dummy variable as an additional control for credit risk.Firms with negative equity are more risky financially, and the debt holder may charge them higher interest as compensation (Kim et al., 2011;Karjalainen, 2011;Pittman and Fortin, 2004).We include LOAN MATURITY because the lender requires a liquidity premium for longer-term debt and this liquidity premium translates into a higher loan spread (Bharath et al., 2008;Aobdia et al., 2015;Graham et al., 2008;Lai, 2011;Karjalainen, 2011).ix Because agency conflicts between concentrated ownership and minority shareholders are a frequent problem in Italy, we control also for the OWNERSHIP STRUCTURE.The Italian capital market consists of a relatively large proportion of firms that have concentrated ownership (La Porta et al., 1999;Lins et al., 2013;Gomez-Meija and Nunez-Nickel, 2001;Schulze et al., 2001;Blanco-Mazagatos et al., 2007;Prencipe et al., 2011) x The higher the percentage of total shares held by the largest owner, the less likely a high-quality auditor will be chosen (Lin and Liu, 2009).

The Earnings Management (EM) model
The EM model is the following Equation (2): For discretionary accruals (DACC), we use a linear expectation model following Francis and Wang (2008).This method is preferred in research using a small sample because it does not require a minimum number of observations for each industry.This minimum number is required on the other hand by the cross-sectional Jones (1991) model and its later versions.EAF is defined as before.Independent control variables are selected on the wide of prior numerous studies on EM (Francis and Wang, 2008).We control for SIZE, motivated by the political visibility hypothesis.This predicts that large firms will make incomedecreasing accounting method choices in response to greater political/regulatory scrutiny or when motivated by other underlying constructs (e.g., information environment, capital market pressure, or financial resources) that predict a negative association between size and EM (Dechow et al., 2010).We control for LEVERAGE, because a higher total debt to asset ratio indicates a higher possibility of debt covenant violation, which creates an incentive to increase reported earnings through accruals-based earnings management (e.g., Francis and Wang, 2008;Dechow et al., 2010;DeFond and Jiambalvo, 1994;Francis and Yu, 2009).We control for LOSS given that the evidence that weak performance provides incentives for EM is well-established (Dechow et al., 2010).We control for GROWTH given that it can affect yearly accruals if the relation between accruals and the accruals drivers (sales and gross PPE) is nonlinear (e.g., Francis and Wang, 2008).To have a well specified model, it has been shown that it is important to control for CFO because they vary inversely to discretionary accruals (Dechow et al., 1995) and for their STANDARD DEVIATION.Standard deviation is considered a relatively nondiscretionary driver of accrual variance in resolving problems arising because measures of absolute discretionary accruals are a function of the dispersion in signed discretionary accruals (Hribar and Nichols, 2007).To control for financial distress we include the firm's probability of bankruptcy, estimated using ALTMAN'S score.Lower values indicate more financial distress, so that a negative association is expected with accrual.This is because financially distressed companies have higher incentive to use accruals to increase earnings to avoid revealing problems and possibly affect prices (Reynolds and Francis, 2000;Francis and Yu, 2009).Given the nature of the Italian market, we control also for the OWNERSHIP STRUCTURE.

Propensity-Score matching model
To consider the endogeneity issue, we use propensity-score matching models, developed by Rosenbaum and Rubin (1983), to match a range of client characteristics to examine whether the auditor distinction can be attributed to specific client characteristics xi .Propensity-score matching models match observations based on the probability of undergoing a treatment, which in our case is the probability of selecting an EAF.We use logit models, the most frequent approach (Guo and Fraser, 2010) xii .We replace a DAF audit client with an EAF audit client that has the closest predicted value from the following Equation 3, within a maximum distance of 1% xiii : EAF = α + β1 SIZEit + β2 LEVERAGEit + β3 LOSSit + β4 ASSET_TURNOVERit + β5 QUICKit + β6 SIZE SQUAREit + industry fixed effect + year fixed effect+ e (3) Definitions of variables are shown in Appendix A. Independent variables are chosen on the basis of studies on audit firm choice.xiv  We next compute the goodness of the propensity score match using a Bias measure.xv Estimating Equations 1 and 2 we test the multivariate effect on CoD and EQ in the common support sample (2). when the weight is generated.xvi All the Equations are estimated with industry and year fixed-effects, in order to control for systematic differences in audit firm choice, CoD and EQ across industries and years in the sample xvii .For the sake of brevity, industry and year indicator variables are not reported in the tables.

Descriptive statistics and correlation matrix
Table 3 shows the descriptive statistics of CoD and its control variables in Panel A. It shows descriptive statistics of EM and its control variables in Panel B. The mean CoD for financial debts (7.2%) and for bank debts (untabulated) are similar.The mean CoD is consistent with literature (e.g.Minnis, 2011).The mean of abnormal accruals is 20% of total assets, higher than the usual mean of below 10% for public companies (Cameran et al., 2015).
The client size has a mean of about €53 million and €49 million euro respectively in Panel A and B, significantly lower than the mean size of Italian public firms.The test for mean difference in the last four columns of Table 3 shows that client size is very similar for clients of DAF and EAF.This shows that our sample of private firms is balanced for each group.The financial leverage of the companies is relatively high, liabilities are between a minimum mean of 65% (Panel A) and a maximum mean of 68.9% (Panel B) of total assets in the full samples, which is consistent with our expectation that debt financing is important in privately held firms.The percentage of loss is about 30% in all non-Big4 clients showing a slightly lower performance of private clients that choose an audit firm with experience in auditing public clients.However, there are no significant differences between EAF and DAF.Asset turnover shows that revenues are higher in mean than total assets in all non-Big4 clients.In our sample, short-term assets are always higher in mean than short-term debts (quick ratio higher than 1) showing short-term financial equilibrium.
Other common variables between CoD and EM samples are the Altman score and ownership concentration.The Altman score shows the level of the bankruptcy problem, which lies between 1.537 (Panel A) and 1.971 (Panel B), consistent with the literature (Reichelt and Wang, 2010).In ownership concentration, between 3.3% (Panel A) and 6.5% (Panel B) of companies one shareholder controls at least 75% of the company.
In the CoD sample, firms have a low profitability (ROA of about 1%) given that in the period analyzed companies had not recovered yet from the crisis.Our sample firms have a relatively low level of tangible assets (25.3% of total assets).On average, about 2.8% of private companies in our sample have negative equity during the sample period.This high percentage is also probably due to the lasting effects of the crisis.Finally, the loan maturity shows that short-term debts are 78.9% of long-term debts, with a higher percentage for EAF than for DAF clients.In Italy, there are more bank loans than financing from bonds and other forms than in U.S. Mansi et al. (2004) discuss that in the U.S., public debt securities represent a significant portion of the typical corporation's value.
In the EM sample, sales are always decreasing.The standard deviation and the value of cash flow from operations are 0.13 and 0.042 respectively, consistent with the literature (Reichelt and Wang, 2010).
The purpose of PSM is to identify very similar companies, with the sole difference being the auditor chosen, for the purpose of comparison.Descriptive statistics show that there are no statistically significant differences between EAF and DAF for the following variables: size, leverage, loss, asset turnover or quick ratio.This comes to the proper application of PSM.In the univariate test of mean difference for the CoD and EM variables, CoD is statistically significant lower in EAF than DAF.
The correlation matrix (Table 4) does not show substantial problems of multicollinearity.The mean variance inflation factor is under 4. The highest correlation between variables of the same regression is 36.4% between Altman and ROA, showing an acceptable level of correlation.The same is true of Panel B. The highest correlation between variables of the same regression is -38.7% between CFO and loss, showing an acceptable level of correlation.
In this univariate analysis, EAF is negatively correlated with CoD and abnormal accruals, suggesting that it has higher audit quality, which is consistent with our expectation.CoD is also correlated with higher quick ratio, loan maturity and lower size, ROA, tangible, and Altman score, abnormal accruals are correlated with higher sales growth and lower ownership concentration.These univariate correlations are consistent with expectation and with results from the following multivariate analysis.

Endogeneity issue
To consider the endogeneity issue, we perform our analysis on the propensity score matched sample.The first and third model in Table 5 show the model to identify the propensity score sample using a logistic regression for the audit firm choice.The analysis to identify the propensity score matched sample with the logistic regressions xviii (first and third model) confirms the usefulness of PSM to reduce bias and to improve the robustness of the main analysis: from a sample of 1798 observations, the PSM sample is 1206 (603 EAF and 603 DAF) and the mean and median bias is significantly reduced (from 10.5/7.3 in the first model to 3.3/2.2 in the second model and from 10.00/6.2 in the third model to 4.6/3.2 in the fourth model) with a p-value of the bias test that loose its significance as sign of an effective first stage.

Test of hypothesis
The second model in Table 5 shows our findings related to CoD computed in the propensity score matched sample identified.The fourth model shows our results related to Abnormal Accruals computed in the propensity score matched sample identified (Francis and Wang, 2008).In the OLS regression on the matched sample using PSM, both the coefficient on EAF of CoD (second model) and EM (fourth model) are negative and statistical significant.Specifically, results show that: a) private clients of EAF are associated with lower CoD by 1.1% (including interest expenses and commissions), that is, 7.6% of EBIT xix , compared to the clients of DAF; b) private clients of EAF are associated with lower EM of 1.7% of abnormal accruals over total assets.The Adj. R 2 of 4.6 -7.3% of the regression on the PSM sample is comparable to other Cost of Financial Debt models in prior studies [e.g.8.8% in Gul et al. (2013) and 9% in Karjalainen (2011).

Control variables
Significant control variables in the models analyzed show a negative relation between size, Altman score, loan maturity and CoD, and a positive relation between quick ratio and CoD.Size is inversely related to bankruptcy because debt holders demand higher interest to cover this higher risk (Lai, 2011;Bharath et al., 2008;Graham et al., 2008;Aobdia et al., 2015); the lender requires a liquidity premium for longer-term debt, and this liquidity premium translates into higher loan spread (Bharath et al., 2008;Aobdia et al., 2015;Graham et al., 2008;Lai, 2011;Karjalainen, 2011).On the other hand, the quick ratio does not drive the choice of more expensive credit.In the EM analysis, significant control variables show a negative relation between size and abnormal accruals; and a positive relation between growth, standard deviation of cash flow and abnormal accruals.This confirms that information environment, capital market pressure, and higher financial resources for bigger firms decrease EM (Dechow et al., 2010); and that growth and standard deviation of cash flows are important determinants of abnormal accruals (Francis and Wang, 2008;Dechow et al. 1995).

Alternative cost of debt and earnings quality measures
We repeat the analysis using a different proxy of the .This is an interesting measure in Italy where private companies are mainly financed by banks and not by bonds, as shown by the descriptive statistics.
We also repeat the analysis using the credit default risk rating provided by mode Finance.This company provides the Multi Objective Rating Coefficient p-values are two-tailed, based on asymptotic t-statistics using White (1980) standard errors.Pseudo R2 for PSM p-values are two-tailed.Refer to Appendix A for variable definitions.We use DAF in the logistic regression due to the difference in the number of their clients compared to EAF, to be able to perform a matching with replacement.We use EAF in the main analysis for an easier interpretation.
Evaluation (MORE) in order to assess the level of distress of industrial companies.It provides a creditworthiness opinion (Assessment) of risk class on the following ten-point scale: AAA (extremely strong), AA (strong), A (high solvency), BBB (adequate), BB (adequate in the countryindustry), B (vulnerable), CCC (dangerous), CC (high vulnerable), C (pathological situations), D (no capacity to meet financial commitments).The rating can be used for access to loans in negotiations with banks.We use the following regression model based on Li et al. (2010) and Coefficient p-values are two-tailed, based on asymptotic t-statistics using White (1980) standard errors.Refer to Appendix A for variable definitions.Meet or beat benchmark uses a dummy dependent variable related to the meet or beat the threshold of zero earnings (to avoid reporting a loss) and thus, use a logistic multivariate regression, for which we report the marginal effects.Mansi et al. (2004): In addition to the control variables used in the main analysis, we add Bank debt (natural logarithm of bank debt) and Coverage (operating income after depreciation divided by interest expense).We requested the data on this rating for the matched sample used in the CoD analysis, and received data for a sample of observations for the year 2014.Results show that clients of EAF are associated with higher ratings than firms with a lower default risk (Table 6, Model 2).
Given the shortcomings of the measurement of abnormal accruals, we repeated the analysis using another model for EM.We were interested in seeing whether the results were driven by our chosen measurement of EM.The small earnings increase model, computed at the 2% level, is a proxy of EM, interpreted as the meet or beat benchmark.

Propensity score matched sample
PSM can be performed with many specifications.We repeat the analysis with kernel matching, in which all treated units are matched with a weighted average of all control units with weights that are inversely proportional to the distance between the propensity scores of treated units and control units.Calculation of weighting depends on the specific kernel function adopted.We repeat the analysis without replacement, changing the caliper distance at 0.5% and switching from one-to-one to oneto-many matching.We follow D' Attoma and Pacei (2014) in presenting the results for different methods of PSM.
Table 7 reports that after matching, the mean bias for all explanatory variables is reduced to acceptable levels (Harder et al., 2010).It falls from about 10.0/17.6 before matching to about 7.2/2.2after matching.Table 7 also reports that after matching, the p-values of the joint significance of the explanatory variables are not significantly different between the treatment group and the control group.In short, these test statistics suggest that the matching method is appropriate.Results reported in Table 8 confirm the main analysis findings.
To investigate whether a high quality auditor reduces CoD, Coarsened Exact Matching (CEM) is also used (Table 7).CEM overcomes some of the limitations inherent in PSM (King et al., 2011;Iacus et al., 2012).CEM is a more robust matching technique that is not subject to random matching, because it directly matches on a coarsened range of covariates and does not rely on a first-stage propensity score model.DeFond et al. (2016) encourage research to explore the use of CEM in complementing regression analysis for the purpose of providing robust inferences.We use the same variables used in the first stage propensity score to perform the match.CEM shows the same results as PSM.We can therefore conclude that results are not driven by endogeneity.

Similar market share
To check whether the differences are due to the audit firms' characteristics analyzed and not due to the different size, we perform the analysis comparing audit firms of the same size, that is, we look at the lowest market share among the market share of the EAF and we restrict the sample to audit firms with market share higher than this.In our sample we have bigger firms in DAF than in EAF, and can therefore state that size is not the main driver of this study.Thus, we compare the 20 EAF with the 13 DAF with a similar market share (higher than 0.5%) xxii .Results in Table 8 -Model 1 confirm that EAF have a lower CoD and EM than DAF of similar size.

IFRS versus Italian GAAP
Effects would be higher if private clients use the same set of standards as public clients.In general, private firms adopt Italian GAAP and some of them voluntarily adopt IFRS.We repeat the regression adding an interaction between audit firm choice (EAF vs DAF) and a dummy variable that takes value 1 if the firm voluntarily adopts IFRS and 0 otherwise.Results for the interaction in Table 8 -Model 2 show significant negative coefficients for the interaction EAF*IFRS.The externalities are higher when the client adopts the same standards as the public clients that the firm also audits.

Other countries with high third-party liability
We select other European countries where the statutory auditor liability to any third party mainly arises from a breach of duty in tort xxiii .On the basis of data availability, we select Belgium (De Poorter, 2008), Sweden (Spirkle, 2013), Finland and Denmark.Financial statement data and data on auditor and date of appointment of the auditor was downloaded from Bureau van Dijck.Data aggregating the audit firms in their global audit firm network was prepared, using the same selection criteria earlier presented as shown at the bottom of Table 9.
Table 9 shows the reduction of the mean and median bias using the PSM on these data with the same variables as presented above.The first two columns show that mean and median bias are higher before matching than after matching, and that the respective pvalues becomes less significant.Table 9 also presents the estimate coefficient of EAF with the respective numbers of observations.Analysis is run country by country.Results show at least one negative association between EAF and lower Cod/EM for each country in the two combinations (EAF and CoD; EAF and EM).
The graphic representation (Figure 2) shows the mean differences in COD and abnormal accruals in Italy, Sweden, Belgium, Denmark and Finland.Cost of financial debts and cost of bank debts have a similar value and trend in Italy.Italy has lower values of COD and abnormal accruals while Belgium has higher value for them.However, in all countries, it is possible to see a significant reduction in their average in EAF compared to DAF.

DISCUSSION
Using a PSM sample of private companies audited by non-Big4 in the period 2010 -2014, we find that EAF are associated with lower CoD and EM, contributing to increasing audit quality and reducing agency costs.Differently from the traditional criterion based on size of audit firms (BigN vs non-Big4), that could not be effective in the non-Big4 setting, we find that audit firms that operate at European level allow the lowering of CoD and EM, given the higher reputation and quality of these audit firms compared with DAF.Previous benefits could be justified because EAF have high reputation costs (DeAngelo, 1981) and high competitive advantages in terms of reputation and competence, e.g.ability to operate across multiple business environments, efficient audit methodology, and staff with professionally certified knowledge of national legislation (Carson, 2009).Moreover, the stricter and different audit environment (Maijoor and Vanstraelen, 2006), enforcement (Van Buuren et al., 2014;Kleinman et al., 2014) and litigation risk prevailing in European countries is a further possible explanation for these findings.Our results support the view that additional competences gained by non-Big4 that operate in the European network, has a high marginal value.We reject the counterargument that DAF being more specialized in the country where they operate have lower CoD and EM.The paper argues that decentralization is not the driver of audit quality at country level.The robustness tests confirm all our main results, and supply interesting indications on credit default rating, and international comparison with countries characterized by similar competitiveness and litigation regulation of the audit market.We find that EAF yield benefits in terms of higher ratings.Finally, our results are not limited to Italy, but can be extended to Sweden, Belgium, Denmark and Finland.

Conclusion
Following the framework of DeFond and Zhang (2014), this research analyses the association between audit firm choice criteria (supply of Audit Quality) and CoD and EM (demand of Audit Quality).While several studies in public We believe that in an audit market with high levels of competition, as envisaged by the Green paper (European Commission, 2010), our analysis will be useful to identify a new criterion (based on the European boundaries of the audit market) that positively affect the agency conflicts between lenders and owners/managers in private firms, through lower CoD and EM.Finding suggests that this audit firm choice criterion is useful to explain agency costs: the higher audit quality offered by EAF reduces risks related to earnings management and allows lenders to accept lower level of interests with benefits for all stakeholders.Regulators could benefit from the results of this research as they could become better aware about consequences of policies on audit independence and competitiveness in the audit market.Regulators currently aiming to improve the competitiveness of audit market will find these findings of interest and could evaluate the opportunity to improve non-Big4 audit firm segment, with special emphasis to EAF.EAF in the non-Big4 more competitive audit market segment appear likely to be associated with lower CoD, EM and agency costs of the clients.Auditor quality, and especially audit independence, is of interest of several stakeholders, such as investors, firms and also other stakeholders.Cutting across all publicly traded corporations is the concern that further regulation of the accounting profession may bring additional regulations in other areas such as corporate governance and capital formation (Kinney, 1999;Gerde and White, 2003).
Results are valid to countries characterized by higher audit market competitiveness, like Italy, Belgium, Denmark and Finland, where the non-BigN market share is higher significant.This explorative analysis could be further investigated in future research to confirm our results in other European countries or in other setting characterized by high audit market share for non-Big4 in private companies.
We do not have variables like audit hours, audit fees, audit report lag for private clients.xv Bias measures the similarity of the distributions of the first stage explanatory variables between the treatment group and the control group.It is calculated for each explanatory variable by dividing the difference in the means between the treatment and control groups by the square root of the average sample variances of the two groups (Rosenbaum and Rubin, 1985).xvi The software Stata creates a weight variable automatically.For observations in the treated group, _weight is 1.For observations in the control group it is the number of observations from the treated group for which the observation is a match.If the observation is not a match, weight is missing.xvii To run audit firm fixed effect, the independent variables must change across time for some substantial portion of the individuals.This is not the case in this study, because we know only the current audit firm for each client and the number of years of tenure since its engagement started, but we do not have information on the past audit firm.xviii The analysis is based on DAF because their clients-year observations are lower in number compared with EAF.We find that large firms are expected to choose high quality auditors (negative relation with DAF) because they are better equipped to handle the audit efficiently (Chaney et al., 2004).We find that high leveraged firms tend to choose higher quality auditors (negative relation with DAF) to reduce their higher agency costs (e.g., Chaney et al., 2004;Fortin and Pittman, 2007).We include loss to control for profitability.xix The economic significance is computed as follows.We take this coefficient and multiply it by the mean financial debts (21614) and divide it by the mean earnings before interest and taxes -EBIT (3135).
xx Cost of capital in the audit literature (Khurana and Raman, 2004;Iatridis, 2012;Azizkhani et al., 2013;Cassell et al., 2013;Lawrence et al., 2011;Guedhami et al., 2014;Choi and Lee, 2014) has been measured by exante cost of equity capital (for example with the models of Gebhardt et al., 2001;Claus and Thomas, 2001;Ohlson and Juettner-Nauroth, 2005;Easton, 2004;Gode and Mohanram, 2003).These models imply the use of financial analyst earnings forecasts and stock prices that are not available for private firms.Other studies (Mansi et al., 2004;Fortin and Pittman, 2007;Li et al., 2010) measure the cost of capital with the marginal cost of debt (the yield to maturity at the issuance date for the largest bond the firm issued in year t+1, minus the Treasury bond yield with similar maturity) and the Standard & Poor's senior debt rating in year t.Standard & Poor's rates a firm's debt from AAA (indicating a strong capacity to pay interest and repay principal) to D (indicating actual default).Bond rates are less well-fitted in this context, given that the main source of financing is from banks and not from bondholders.In private firms, bonds are often similar to stock option and they may represent a supplement to shareholder remuneration.The cost of total debt, measured using as denominator the amount of total debts (Pittman and Fortin, 2004;Kim et al., 2011;Lai, 2011;Minnis, 2011;Causholli and Knechel, 2012) has a mean value of about 2% with a standard deviation of 2%, similar to other countries like Korea (about 2% in Kim et al., 2011), lower than U.S. (about 7% in Minnis, 2011).In Italy, cost of total debt is much lower because it includes non-interest-bearing debt.This proxy is therefore excluded from the analysis.
xxi Changing the threshold level, results are qualitatively the same.
xxiii Countries where the statutory auditor"s liability to any third parties mainly arises from a breach of duty in tort are Belgium (De Poorter, 2008), Sweden (Spirkle, 2013), Finland, Denmark, Portugal, Greece and Luxembourg.For United Kingdom, Ireland, Netherlands, Germany, Spain, Austria liability towards third parties is subject to restrictive conditions following European Commission (2001).Other countries joining the European Union after 2004 are excluded from the analysis.

Figure 1 .
Figure1.The association between auditor choice and agency theory in private firms and the non-Big4 segment.Note: 1) Auditors have incentives to increase audit quality to reduce reputation and litigation risk.2) Given that audit firm size, among non-Big4 segment, is not effective, we suggest a new audit firm choice criterion (European audit firm vs Domestic audit firms).3)Clients have incentives to increase audit quality to reduce agency costs and agency conflicts between lenders and manager.4) We expect that European audit firm, through higher audit quality, is associated with lower cost of debt, earnings management and agency costs.Source: Adapted from DeFond and Zhang (2014).

Figure 2 .
Figure 2. Cost of Debt, Cost of Bank Debt and Abnormal accrual in EAF compared with DAF.

Table 1 .
Sample selection.Total number of Italian companies audited by a non-Big4 audit firm with at least 2 clients in the Bureau Van Dijck database in 2014 1149 Less public companies or companies subjected to mandatory audit in 2014 Starting from total number of observations for the period 2010 -2014 4435 Less observations with missing values necessary to compute variables related to abnormal accruals (observations lost mainly for lack of data on cash flows) -3273 Total number of observations in the regression model for auditor choice in Earnings Management analysis 1162 Less observations not matched in Propensity Score Matching model -212 Total number of observations in the matched sample for Earnings Management analysis 950 ; Weber and

Table 2 .
Non-Big4 Audit firm classification.Similarly, in the European Union, excluding Big4 that operate at international level, we analyze EAF in the same way as audit firms operating at national level (within Europe) and DAF in the same way as audit firms that operate at local/regional level (within individual European State).To classify audit firms as EAF or DAF and to see if they are allowed to operate at European level, we check the presence of audit firms belonging to the same network in the registers of the following European countries: France, UK, (Read et al., 2004)iger and Rama, 2006)analyzes audit firms that operate at international level (BigN), at national level (within U.S.) and local/regional level (within individual U.S. State).The three levels are even more important in markets characterized by a lower presence of BigN(Read et al., 2004), like the private company market.
for audit firms that operate in more than one country in Europe with only private clients in Italy, and 0 otherwise.
Li et al. (2010)))05)cludes interest and commission.FollowingFrancis et al. (2005),Karjalainen (2011); Cano-Rodríguez and Alegría (2012);Gul et al. (2013), we choose a measure that includes only interest-bearing debt.Li et al. (2010)support the use of CoD in analyzing the consequences of auditor choice for several reasons: the public debt market is significantly larger than the equity market in some contexts; CoD is relatively well defined with less mis-specification than cost of equity; CoD is not affected by the difference of more or less sophisticated investors given that the information environment in the debt market is characterized by numerous information
*,**,*** is respectively 0.1, 0.05, 0.001 the p-value of the t-test of the difference in the mean between EAF and DAF.Variable definition in Appendix A.

Table 4 .
Correlation matrix.Pearson correlation coefficient.Refer to Appendix A for variable definitions.Significant coefficient at 0.10 are in bold.Variable definition in Appendix A.
dependent variable CoD.We compare the financial costs to different values of the debt, changing the denominator of the variables.We use a more restricted Cost of interest-bearing Debt, including only the Cost of Bank Debt xx

Table 5 .
Multivariate analysis between EAF and DAF within non-Big4.

Table 6 .
Alternative measure of Cost of Debt and of EQ.

Table 7 .
White (1980)estimation of propensity score matching.Coefficient p-values are one-tailed, based on asymptotic t-statistics usingWhite (1980)standard errors and clustered by firms.Pseudo R2 for PSM pvalues are two-tailed.See Appendix A for variable definitions.Results for Kernel (normal) and Kernel (Epanechnikov) are very similar.

Table 8 .
Audit firm market share and International Financial Reporting Standards (IFRS).

audit firms with a market share higher than 0.5% between EAF and DAF Between EAF and DAF Interaction with IFRS
White (1980)p-values are one-tailed, based on asymptotic t-statistics usingWhite (1980)standard errors and clustered by firms.See Appendix A for variable definitions.All the regressions presented are run on the propensity score matched sample.This sample is the output of the first model with dependent variable the auditor choice.For the IFRS analysis the first stage regression includes also a dummy variable of 1 if IFRS and 0 if Italian GAAP, to define the propensity score matched sample.

Table 9 .
Additional analysis: other countries.

PS matching Cost of financial debt Abnormal accruals Mean bias (Median bias) p-value (N) Estimate (N) Mean bias (Median bias) p-value Estimate (N)
White (1980)and F-test p-values are one-tailed, based on asymptotic t-statistics usingWhite (1980)standard errors and clustered by firms.See Appendix A for variable definitions.Year and industry fixed effect included.Because data was not available, the control variable for the ownership concentration is not included.In our sample Belgium has 205 Non-Big4; Finland has 70 Non-Big4; Denmark has 384 Non-Big4; Sweden has 55 Non-Big4; France has 1977 Non-Big4.Within these Non-Big4 in each country, EAF are the same 20 listed in the variable definition table (Appendix A), except that PKF and BKR are not present in Finland, and Morison is not present in Finland or Sweden.The sample includes audit firms with at least 2 clients per year.The number of non-Big audit firms is computed aggregating audit firms with different names into a single audit firm if they are part of the same group.