Discovering the Concealed Benefits of Auditor-Provided Tax Services

Transcription

1 Discovering the Concealed Benefits of Auditor-Provided Tax Services Working Paper - April Stephan Burggraef * stephan.burggraef@wiwi.uni-muenster.de Christoph Watrin * christoph.watrin@wiwi.uni-muenster.de Falko Weiß * falko.weiss@wiwi.uni-muenster.de Abstract: The paper investigates the effect of APTS on tax planning and earnings quality using a German sample. Our findings differ from previous US studies which we attribute to the tighter restrictions on APTS in Germany. We find a negative association between APTS and tax planning suggesting that knowledge spillover to the tax department might not result in aggressive tax strategies and therefore might not drive the results in this case. Also, we find a positive relation between the level of APTS and the sustainability of tax strategies in client firms. Turning to the accounting perspective we find that APTS are positively related to accrual quality. This is consistent with the presence of beneficial knowledge spillover to the audit department while a loss of professional skepticism does not prevail. However, it appears that earnings persistence decreases with APTS. Altogether, our study suggests the effectiveness and the advantages of the former German legislation. Therefore, the purpose of the recent reform is questionable. We conclude that the 2016 reform which restricted APTS even further might not be necessary according to our findings. JEL Classification: H26, H73, K23, L84, K23, L84, M41, M42, M48 Keywords: Auditor-Provided Tax Services, Regulation, Tax Planning, Tax Sustainability, Accrual Quality, Earnings Persistence We thank the participants of the Workshop on Empirical Tax Research at the Centre for European Economic Research (ZEW) 2016, of the 3 rd Doctoral Research Seminar at Vienna University of Economics and Business, and of the AAA International Accounting Section Midyear Meeting 2017 for their helpful comments. We especially appreciate the useful suggestions by Gopal Krishnan. * University of Münster, Institute of Accounting and Taxation, Universitätsstrasse 14-16, Münster, Germany

2 1 Introduction The association between auditor provided tax services (APTS) and both tax avoidance and earnings quality has been discussed for some time. However, most evidence is based on an US- American background which is different from the European one. Even though the results in prior literature were not unambiguous they point towards a deteriorating effect of APTS on tax payments and earnings quality. Supposedly having those findings in mind, regulators in the EU and Germany recently tightened the regulations on the provision of APTS. The goal of this paper is to investigate whether APTS has a positive or negative effect in an EU setting and whether the regulations prior to the recent reform were sufficient to maintain auditor independence and avoid negative outcomes. In our German sample 95% of the companies buy non-audit services, and 60% obtain APTS. These numbers indicate the high relevance of these services also in Germany. By auditing the financial statement the auditor1 gains a tremendous insight into the firm. The auditor can use this knowledge to give advice on tax planning2 and potentially can help to introduce aggressive tax strategies (e.g. Hogan and Noga [2012]). The knowledge spillover to the tax department makes the auditor especially suited to consult on tax planning (e.g. Gleason and Mills [2011]). However, aggressive tax planning increases the risk of tax disputes. More tax risks can result in financial statement restatements and auditor liability.3 Therefore, auditors that provide tax services might also prefer less tax planning. In contrast, Kinney et al. [2004] even find a negative association between APTS and restatements. It is thus an empirical question whether APTS increase or decrease tax planning. Our finding is consistent with a negative association between APTS and tax planning. There is also evidence that the tax planning level of a firm influences the amount of APTS obtained. The reason for the negative association might be that tax aggressive firms buy tax services from third parties. Another facet of tax strategies is tax sustainability. As pointed out by McGuire et al. [2013] the tax sustainability must not be mistaken for the level of tax planning. Instead it shows whether 1 If not pronounced otherwise the audit firm is meant by auditor. 2 There are different perceptions of the terms tax avoidance, tax planning or tax aggressiveness. Hanlon and Heitzman (2010) see tax avoidance as a continuum of tax planning strategies ranging from perfectly legal and relatively harmless measures to more dubious efforts such as tax sheltering. Although it appears that the legislator in this case aims at the more aggressive end of the continuum, in this paper we follow the broader definition by Hanlon and Heitzman (2010), because our focus is generally on firms actions to influence their tax payments and not on the specific type of action taken. Therefore, we employ the term tax planning because we believe it better represents the broader and more neutral approach. 3 See for example Laux and Newman [2012] for an analysis of the relation between auditor liability regulations and the decision of the auditor to take on clients depending on their riskiness. 2

3 a company is able to maintain its level of tax planning over time. In contrast to a previous finding by Francis et al. [2015], we find that the volatility of the effective tax rates decreases with more APTS, i.e. the tax strategies are more sustainable. Our finding is in line with anecdotal evidence suggesting that advisory and audit companies actively promote sustainable tax strategies (e.g. Deloitte [2013]). If the auditor provides non-audit services to a client, his dependence on the client increases. Economic bonding could potentially endanger auditor independence in fact or in appearance. Despite of several studies highlighting the negative aspects of economic bonding (e.g. Srinidhi and Gul [2007]) other studies such as Cook and Omer [2013] or Choudhary et al. [2016] have already cleared economic bonding of concerns regarding the quality of the financial statement (earnings quality or audit quality in particular). However, Choudhary et al. [2016] find evidence for reduced professional skepticism in the presence of APTS. Though, this effect is said to be due to the cognitive bias in-group identification instead. In contrast to these considerations, knowledge spillover to the audit department from APTS might also strengthen audit quality because of synergy effects. (Chung and Kullapur [2007] or Robinson [2008]). Our results are mainly consistent with the view of knowledge spillovers to the audit department. Earnings persistence is often viewed as an indicator of earnings quality (Dechow et al. [2010]). However, if earnings persistence is the result of earnings smoothing it might be less desirable in order to give a true and fair view of the real performance. Although an investor is interested in less volatile earnings, it is important to face unsmoothed results to accurately evaluate the company s performance. Following this idea, persistent earnings due to earnings management may not be desirable. Depending on the different interpretations of earnings persistence, a possible relation between APTS and earnings persistent may have another interpretation of earnings quality. Accordingly, we also investigate the influence of APTS on earnings persistence and find evidence for a negative relation following a positive association with audit quality. In all aspects the legal background is important because it determines to what extent an auditor can provide non-audit services at all. On the one hand, very strict restrictions may ensure a high degree of auditor independence and could constrain economic bonding between the auditor and his clients thereby avoiding negative effects such as high tax aggressiveness or low earnings quality. However, a side-effect could be an inhibition of a reasonable amount of knowledge spillovers resulting in inefficiencies and possibly even lower earnings quality. On the other hand, a weak regulation of APTS may enable a potentially positive high degree of knowledge spillover. Nevertheless, auditor independence and tax planning activities could be impaired. 3

4 We contribute to prior literature in at least two aspects. First, we present evidence from Germany which is one of the important EU member states. The German setting varies widely from the US-American background. Due to this setting our results also mostly differ from previous studies. Our data on auditor fees are hand-collected giving our sample a unique character. Second, we are able to show that APTS can have positive influences with respect to reduced tax planning and an increasing earnings quality. Since this finding occurs in the German legal background we infer that moderate restrictions on APTS are helpful in avoiding the negative aspects without preventing the positive aspects. This could be a role model for the United States where very loose restrictions appear to be accompanied with mostly negative outcomes. However, our results are also not in favor of further tightening the law as it recently happened in Europe. It remains an area for further research to actually evaluate the reform in a few years. The paper proceeds as follows: Chapter 2 gives an overview of the legal situation in Europe and Germany in particular. Chapter 3 presents prior research while chapter 4 builds on this and develops our hypotheses. In chapter 5, we present the research design and the empirical results for the tax perspective of APTS. We proceed analogously in chapter 6 for the accounting perspective of APTS. Chapter 7 concludes and gives perspectives for future developments and fields of research. 2 EU and German Regulations of APTS This paper is motivated by the reform of the German regulations on the provision of tax services by the auditor. The Abschlussprüfungsreformgesetz (AReG) [Law on the reform of the audit process]4 is an implementation of EU Directive No. 2014/56/EU and gives guidelines for the application of EU Regulation No. 537/2014. It changed Sec. 319a I No. 2 of the German Commercial Code effective June 17 th, Before the reform, an audit firm was precluded from offering audit services to a client if it also provides tax services or legal services in the same year, which have a substantial influence on the financial statement and go beyond a presentation of different tax planning alternatives. In the amended version after the reform, the critical tax services are specified referring to EU Regulation No. 537/2014. These services include the preparation of tax forms, the identification of public subsidies and tax incentives, support regarding tax inspections, calculation of taxes and deferred taxes, and the provision of tax advice in general. They remain possible but in a more restricted way. First, the substantial influence is defined. According to the new law, 4 See BT-Drs. 18/7902 [German Bundestag Printed Paper No. 18/7902]. 4

5 substantial influence is assumed where aggressive tax planning strategies are pursued, i.e. there is an extensive decrease in the domestic tax base or a significant shift of taxable income to foreign countries. Second, the new Sec. 319a Ia Commercial Code directly refers to the ruling in EU Regulation No. 537/2014, which restricts non-audit services to 70% of the average audit fees in the last three years. In addition, the EU Regulation introduces a black list of services which are prohibited in any case. This list includes services with regard to payroll taxes and custom duties. Under the new legislation APTS are highly limited or even forbidden. The intuition of the new law appears to be that APTS have a detrimental influence on tax planning, because knowledge spillover to the tax department and economic bonding might lead to more tax avoidance. The new restriction on APTS is in line with past regulations of audit services. Sec. 319a Commercial Code was first introduced in 2004 by the Bilanzrechtsreformgesetz (BilReG) and there were further changes in this area by the Bilanzrechtsmodernisierungsgesetz (BilMoG) in These reforms were mainly intended to improve financial statement quality and to strengthen auditor independence. In the US, APTS are far less restricted even when looking at the German situation before the recent reform. If the audit committee agrees on APTS in advance, there are little further restrictions on the provision of these services in the US.5 Since the experience with largely unregulated APTS in the US setting is mostly negative or at least mixed, it is interesting to see what effects the previously moderate German legislation had. Therefore, we investigate whether APTS are as destructive as the European legislator suspects it to be. Since we employ data before 2016, only the former legislation is subject of our analyses. Our results give insights into whether the current further tightening of APTS is really necessary. 3 Prior Research When the auditor provides both tax planning and auditing services, knowledge spillover effects are likely, to the tax department as well as to the audit department. Several studies find a positive effect of APTS on tax avoidance. Cook and Omer [2013] observe a decrease in tax avoidance after the termination or substantial decrease of APTS. According to Hogan and Noga 5 SEC Release No , File No. PCAOB (April 19 th, 2006); SEC Release No , File No. PCAOB (May 2 nd, 2014)]; Rule 3524 of the PCAOB, available at: Also see SEC Release No (January 28 th, 2003); Sarbanes-Oxley Act, Section 201. Contingent-fee arrangements, confidential or overly aggressive tax marketing, planning, or advice, and tax services to certain corporate managers and their family are prohibited (Krishnan et al. [2013], Laffie [2006]). 5

6 [2012] a reduction in APTS leads to higher tax payments in the long run and vice versa. McGuire et al. [2012] add that APTS increases tax avoidance if the audit firm is a tax expert. Dhaliwal et al. [2013] provide evidence that the positive effect of APTS is driven by tax planning fees in contrast to tax compliance fees. However, the effect is more pronounced if the auditor provides both tax planning and tax compliance. All these findings suggest knowledge spillover effects. Besides, Donohoe and Knechel [2014] find that due to knowledge spillover a potential fee premium can be offset, which the auditor may demand for tax aggressive behavior of the client. Our second research question is related to tax sustainability. Francis et al. [2015] find that the volatility of the ETR increases with an increase in APTS. McGuire et al. [2013] take a closer look at tax sustainability. They discover that firms exhibit more persistent earnings if they pursue sustainable tax strategies. Guenther et al. [2013] find that firms with a higher volatility in Cash ETR, i.e. higher tax risk, are subject to a higher total firm risk. Therefore, it appears that sustainable tax strategies are also value relevant. Prior literature documents some negative effects of APTS on accounting and audit quality as well as analyst s forecasts. The study by Frankel et al. [2002] finds that the amount of discretionary accruals is positively associated with non-audit fees, i.e. this points to a lower audit quality due to a loss of auditor independence. On the other hand, Ashbaugh et al. [2003] find no statistically significant relation between non-audit fees and the quality of financial statements. This finding is reinforced for example by Chung and Kallapur [2003] and Huang et al. [2007] with regard to non-audit fees in general. However, more closely related to our study, the latter ones find a very weak hint that biased financial reporting is less likely when there are higher values of APTS. In this line is the finding of Robinson [2008] that audit quality improves through knowledge spillover if the auditor provides higher levels of tax services. Krishnan et al. [2013] conclude that investors perceive the improved financial reporting quality through knowledge spillover to be more important than a potentially lower auditor independence. The study by Larcker and Richardson [2004] provides mixed evidence. They find that the absolute value of accruals is positively related to the ratio of non-audit fees to total fees but it is negatively related to the level of all (non-audit and audit) fees. In a study more closely related to ours in terms of research design Srinidhi and Gul [2007] show that the level of non-audit fees negatively affects the quality of accruals. In a German setting but with a smaller sample Sattler [2011] gives evidence that the amount of APTS is generally not relevant for the level of earnings management. 6

7 Francis et al. [2015] identify a positive relation between the level of APTS and analysts forecast error for both pre-tax earnings and tax expense. Their result is in line with Sattler [2011] who identifies an independence violation assumption by external analysts. The main finding of Choudhary et al. [2016] is a positive relation between the estimation error in the tax accounts and APTS. If firms purchase a significant amount of APTS, the estimation error is nearly 10 % higher compared to firms not buying APTS. In contrast, Ashbaugh et al. [2003] do not find systematic evidence for decreasing independence of the auditor in the presence of APTS and challenge the findings of Frankel et al. [2002]. Despite the potentially positive effect of knowledge spillover the mixed evidence regarding the effect of APTS on accounting and audit quality hints at possible economic bonding effects which impair auditor independence. However, reputation effects might restrain auditors to allow clients to engage in unusual accrual choices. With regard to earnings persistence and APTS Francis et al. [2015] find a negative association. Furthermore, some studies investigate the association between APTS and the earnings response coefficient (ERC). For example, in a rather different setting Teoh and Wong [1993] find that the audit quality (proxied by the size of the audit company) has an effect on the ERC. Gul et al. [2006] discover that non-audit services have a negative influence on the value relevance of earnings, also proxied by ERC. In terms of our measure of earnings persistence, we refer to the common idea that future earnings are to a certain extent related to current earnings. This has been often used in prior literature such as Sloan [1996], Richardson et al. [2005], Hanlon [2005] or McGuire et al. [2013]. 4 Hypotheses Development An audit firm that provides non-audit services as well as audit services to a client, makes more revenue with the client. This could result in economic bonding and impair auditor independence. Knowledge spillover to the tax department puts the audit firm in a good position to consult on aggressive tax planning techniques if it has tax expertise (Dhaliwal et al. [2013]). The client might demand advice on tax planning in order to reduce cash outflows to the tax authorities. The increased importance of the client because of APTS makes it more difficult for the auditor to resist any demands of the client. Economic bonding might thus lead to a positive effect of APTS on tax avoidance. However, the auditor must weight economic bonding against the reputational costs that might result from a tax dispute. Further, the audit firm has to consider the risk of a restatement and a 7

8 resulting liability to third parties. Recently, restatements related to tax positions have increased in the US (Audit Analytics [2016]). This might induce an audit firm to advice on less risky tax strategies and APTS could have a negative association with tax avoidance. Most studies that find a positive association between APTS and tax avoidance are related to the US market. Even before the recent reform, EU/German regulations on APTS were tighter than the US standards. This is another argument for a negative relation between APTS and tax planning. We conclude that both the association of APTS with tax planning as well as its direction are an empirical question in the EU setting and state: Hypothesis 1: The level of APTS is associated with the level of tax planning. Tax sustainability generally means that a company maintains certain tax outcomes over time. McGuire et al. [2013] gather some evidence that tax sustainability is a primary goal for many tax departments and that analysts may view unexpected changes in tax rates as a signal of poor management. They also highlight a fact, which is especially relevant in our context, namely that advisory and audit firms such as Deloitte actively promote sustainable tax strategies as being relevant for firm value (Deloitte [2013]). Since the vast majority of firms in our sample is audited by big-4 companies, which additionally provide tax services in many cases, this could be a hint to a possible positive relation between APTS and tax sustainability. Moreover, it seems to be intuitively logical, that an advisory firm, which is very familiar with its client s business model and accounting procedures due to the audit, is quite easily able to propose sustainable tax strategies. However, any other tax adviser and the client s own tax department might be as able to consult on a sustainable tax strategy as the auditor. A third party firm might even have an unbiased view and could be more motivated to introduce a new and sustainable tax strategy while for the auditor the tax affairs of its client could only be a side business. If the audit firm provides the tax services, it might prefer to concentrate on tax compliance and not tax planning in order to reduce the risk from auditing the tax positions. The previous considerations have shown that the direction of the relation between APTS and tax sustainability is not quite clear. It even seems imaginable that there is no effect at all. Therefore, we state our next hypothesis as follows: Hypothesis 2: The level of APTS is associated with the level of tax sustainability. Due to the insights into the client s tax strategy provided by the audit firm s tax department the auditor gets a better understanding of the client s business and the tax strategies in place. De Simone et al. [2015] find that APTS enhance the overall quality of the financial reporting by 8

9 improving the internal control quality in the sense that auditors are aware of material transactions at an earlier stage. According to the principle of congruency of tax accounts and financial statements, the understanding of the tax accounts yields an increase of audit quality. Providing APTS, the audit quality rises due to knowledge spillover to the audit department. The rise in audit quality due to the increased understanding of the underlying business would reduce accrual management. The auditor is able to better identify window dressing in the accrual accounts by the management to smooth earnings and meet analysts forecasts. Following this argumentation, APTS is associated with a rise in accrual quality. The quality of accruals is often regarded as a proxy for earnings quality (see for example Dechow et al. [2010]) which is in turn a characteristic of audit quality (e.g. Ghosh and Moon [2005]). The theory of reduced professional skepticism explains why a negative relation may be observed, i.e. audit quality decreases in the presence of APTS. An auditor might rely on the work done by his colleague in the tax department and might check the financial statements with less skepticism. This phenomenon is known as in-group identification. Because of reduced professional skepticism due to in-group identification, the auditor may not identify earnings management conducted by company executives. Therefore, a reduction of professional skepticism is in line with a reduction of accrual quality. Another argument connected to reduced professional skepticism is economic bonding. This means that due to the strong links between the auditor and its client the auditor could be under pressure not to report a discovered issue in the financial statement. Since an explanation for both a positive and a negative relation between APTS and accrual quality exists, we formulate hypothesis 3 as follows: Hypothesis 3: The level of APTS is associated with the level of accrual quality. If a positive relation between APTS and accrual quality can be identified, the earnings management in favor of earnings smoothing may decline. As a result, earnings persistence may decline, too.6 However, if APTS decrease accrual quality, earnings persistence might also rise. We therefore conclude: Hypothesis 4: The level of APTS is associated with the level of earnings persistence. 6 Please note that we are aware of earnings persistence being evaluated as positive by the capital market. However, it is difficult to differentiate between the operative component due to the underlying business and the smoothed part due to earnings management. In contrast to prior literature, we interpret earnings persistence not exclusively positive but also want to give an alternative explanation for a negative association with audit quality. While the firm value may be positively associated with earnings persistence, we want to give an explanation for a negative association with audit quality. 9

10 5 The Tax Perspective 5.1 Research Design To analyze hypothesis 1 empirically, this paper employs an OLS regression to investigate the relation between APTS and tax planning. Equation (1) shows the regression model with LRCASHETR being the dependent variable and ATR being the variable of interest. LRCASHETR i,t = β 0 + β 1 ATR i,t + β 2 SIZE i,t + β 3 LEV i,t + β 4 PPE i,t + β 5 ROA i,t + β 6 INTAN i,t +β 7 INTLTY i,t + β 8 Deloitte i,t + β 9 EY i,t + β 10 KPMG i,t (1) + β 11 PwC i,t + Industry Fixed Effects + Year Fixed Effects + ε i,t The dependent variable is a long-run version of the cash ETR and proxies for tax planning. As Dyreng et al. [2008] point out, long-run measures are helpful for eliminating one-year effects from the ETR measures. LRCASHETR is calculated as the sum of income taxes paid from t-4 through t divided by the sum of pretax income from t-4 through t.7 Since all firms in the sample are listed in the German prime standard and report their earnings to the German tax authority, the underlying tax code is the same for all firms. Therefore, we are not dependent on the advantage of GAAP ETR measures 8 advantage of comparability due to a common IFRS code. Additionally, as Hanlon and Heitzman [2010] point out, the advantage of cash ETR is that it also captures tax planning strategies which induce the shifting of taxes over time.9 Therefore, we believe LRCASHETR is the more appropriate proxy. For reasons of completeness, we also run our tests using LRGAAPETR and we also use annual calculations of both versions, i.e. CASHETR (GAAPETR) is computed by dividing income taxes paid (income tax expense) in a period by the pretax income in the same period. In contrast to the potentially higher informative value of LRCASHETR a potential advantage of the annual CASHETR is that we do not lose so many observations due to data restrictions. When using long-run measures we exclude those observations with negative ETR values and those being greater than one. For annual ETRs, observations with negative values for pretax income and for income taxes paid and income tax expense respectively are omitted due to issues of interpretation. A firm is defined to be engaged in tax planning if it pays less taxes to the tax authorities and therefore has a low effective tax 7 Please note that measures of long-run ETRs must not be confused with the coefficient of variation of ETRs as used in the analysis of H2. Although both concepts use data over several years, the intuition is different. Longrun ETRs still look at the level of ETR and thereby proxy for tax avoidance/planning. The coefficient of variation does not specifically look at the level of ETR but captures to what extent a certain level is sustainable over time. 8 GAAP ETR measures use income tax expense instead of income taxes paid in the numerator. 9 This goal cannot be achieved using GAAPETR because it also comprises deferred taxes. 10

11 rate. ATR is our measure of APTS. Specifically, it is the fee ratio of APTS to auditor-provided audit services. According to Sec. 314 I No. 9 of the German Commercial Code firms have to disclose the fees they paid to the auditor of the consolidated financial statement for audit services, other audit related services, tax services (APTS), and other services. The auditor of the consolidated financial statement is usually the local branch of the audit company, e.g. in case of PricewaterhouseCoopers in Germany this would be PwC AG Wirtschaftsprüfungsgesellschaft. The fees represent all services provided by this local entity worldwide. However, we are unable to obtain reliable information about the fees the entire network of an audit company receives because only a few firms report this information. This could be a potential limitation of our inferences.10 We use the fee ratio described above because we are interested in the impact of additional tax services provided by auditors. We scale this measure with the fee for audit services to make the values more comparable across firms.11 A previous study by Sattler [2011] analyzing a German sample shows that the quality of the fee disclosure is sufficiently good and reliable. Besides our variable of interest, we employ several control variables. SIZE is the natural logarithm of total assets, LEV indicates the companies leverage as long-term liabilities scaled by total assets, PPE shows the company s property, plant and equipment scaled by total assets, ROA shows the return on assets, i.e. pretax income scaled by total assets, and INTAN the intangible assets scaled by total assets. INTLTY is a variable indicating the degree of a firm s internationality and is calculated as the number of foreign subsidiaries divided by the total number of subsidiaries for each company in our sample.12 All continuous variables except for CASHETR and INTLY are winsorized at a level of Values of ETR measures are limited between zero and one. Due to the construction, this also applies to INTLTY. Looking at accounting data and the impact of auditors, most researchers apply an indicator variable equal to one if the firm is audited by a big-4 audit firm and zero otherwise. Since we are focusing on the German prime standard, the portion of big-4 audits is quite high (ca. 85%). Nevertheless, there might be a significant influence of the audit firm. We do not differentiate between big-4 audit firm and others but each single big-4 firm. We employ indicator variables as shown in equation 10 We also run our tests with the limited data on network fees. The results are generally consistent. However, the statistical relevance of such tests is much lower, because the sample is less than a sixth of the original size. 11 This ratio has been used in prior literature, e.g. Maydew and Shackelford [2005]. We also conduct our tests with ATR defined as fees for tax-services divided by fees for audit services plus fees for tax services. This leads to similar results in terms of significance. The coefficients are slightly larger. 12 We are aware that internationalization of companies is usually proxied by control variables such as foreign income or foreign sales. Unfortunately, we cannot get reliable data for this approach from the relevant databases. Therefore, we construct an own indicator for internationalization by the given ratio. Data are obtained from the Amadeus database by Bureau van Dijk. 11

12 (1) to identify the audit firm. Industry-fixed effects and year-fixed effects are included in the model. The coefficient of interest is β 1. When including the proxy for internationality in the basic regressions as an additional control variable, the number of observations decreases a lot due to data limitations compared to other control variables. We show the effect of including this variable for hypothesis 1 while not including the proxy in hypotheses two to four. However, untabulated results show that including the proxy in the other hypotheses does not change the inferences of this paper. In hypothesis 2, we analyze the relation between APTS and the sustainability of tax strategies a company executes. Following the emerging literature on tax sustainability, e.g. Neuman et al. [2013] or McGuire et al. [2013], we assume that a tax strategy is sustainable if there is a low variation in the effective tax rates over the years. Specifically, the coefficient of variation of the ETR is used as a proxy: CV_ETR i,t = N ETR i,t AvgETR i 2 t=1 abs 1 N N t=1 ETR i,t, (2) where ETR is either CASHETR or GAAPETR. The coefficient of variation is determined by scaling the standard deviation of the respective ETR over three years13 (N=3) by the absolute value of the mean of the respective ETR over the same three-year period. As pointed out in Neuman et al. [2013], the coefficient of variation reveals how consistently a company maintains a certain tax outcome over the chosen period. In this respect, it is a measure of volatility without unit. Lower values of CV_CASHETR and CV_GAAPETR indicate higher sustainability. In order to investigate the hypothesis we run the OLS regression in equation (3) where the coefficient of variation of both versions of ETR is the dependent variable. Our focus lies on the coefficient on ATR, which is the same fee ratio as above. To control for the volatility of earnings we also use the coefficient of variation of pretax book income (CV_PTBI) which is calculated analogously to equation (2) (Newman et al. [2013]). 13 Neuman et al. [2013] and McGuire et al. [2013] use a window of five years to calculate the coefficient of variation. However, we use only three years instead because our sample would be diminished too much otherwise. The three year period ranges from t-2 to t. 12

13 CV_CASHETR i,t or CV_GAAPETR i,t = β 0 + β 1 ATR i,t + β 2 CV_PTBI i,t + β 3 SIZE i,t + β 4 LEV i,t + β 5 PPE i,t + β 6 ROA i,t + β 7 INTANG i,t + Industry Fixed Effects + Year Fixed Effects + ε i,t (3) Again, we use indicator variables for the big-4 companies to control for a possible influence. All continuous variables are winsorized to the 1 and 99 percentiles. 5.2 Descriptive Statistics and Empirical Results Descriptive Statistics Our sample comprises the four major indices of the DAX-family (number of companies in brackets): DAX (30), MDAX (50), SDAX (50), and TecDAX (30), see Panel A of Table 1. This makes 160 firms which are all listed in the Prime Standard of the Frankfurt Stock Exchange. Data on auditor fees are hand collected from the respective financial statements for years 2009 through We use the indices composition as of January Companies are excluded from the sample if they do not provide information about fees and if their headquarter is not located in Germany because these firms do not necessarily disclose auditor fee information according to German law.14 Panel B of Table 1 shows the distribution of audit firms.15 We obtain financial statement data from Compustat Global. All data are publicly available. <Insert Table 1 here> The sample selection is shown in Table 2. We collect fee data for 829 firms. In 495 firm-years (60% of the sample) clients buy APTS. <Insert Table 2 here> Descriptive statistics are provided in Table 3. <Insert Table 3 here> Fees paid to the auditor for tax services make on average about 15% of the fees paid for audit services. However, compared to the median firm s ratio of 3 to 4%, it seems that there are some companies with very high ratios in ATR. Within the subsample of those firms with non-zero values for APTS the mean of ATR is 26% while the median amounts to 14.5% (not reported). 14 The number of firms varies between 131 and 142 per year (instead of 160). 15 It is interesting to see that the trend in our sample is towards concentration at the big-4 audit firms. There is a drastic decline at the smaller companies in the item Other, while PwC and EY experience strong increases in their audit contracts and KPMG and Deloitte remain relatively steady. 13

14 Looking at the different tax planning measures, we observe that the one-year measures CASH- ETR and GAAPETR are slightly below the statutory tax rate, while both long-run measures appear to be quite close to the statutory tax rate. This is a sign that the average firm in our sample does not tend to engage in aggressive tax planning. With regard to hypothesis 2, for the sake of brevity we include descriptive statistics for control variables only for the regression using CV_CASHETR and on CV_GAAPETR. The variation in the coefficients of variation is quite large even though the data are winsorized. Panels A and B of Table 4 present the relevant correlation coefficients. Empirical Results for Tax Planning <Insert Table 4 here> Table 5 shows the regression results of equation (1). The first and second (third and fourth) columns of Panel A are designed to explain LRCASHETR (LRGAAPETR) as the dependent variable. Since using INTLTY limits the number of firm-year observations more than any other control variable, we show the regression results in column one (three) without INTLTY but include it in column two (four). Anyway, the measure is not statistically significant in both cases. The coefficient of interest is β1 explaining the effect of ATR on the ETRs. The coefficients for estimating LRCASHETR are statistically significant on a level of 1%. Column one and two report a coefficient of about The coefficient can be interpreted as the influence of ATR on ETR. If the audit-tax ratio rises by 0.1 the LRCASHETR rises by 0.9 percentage points. Since ATR is a ratio, it is difficult to interpret a change by 0.1 units. For reasons of simplicity, we assume a mean company characterized by a mean APTS fee ( million), a mean audit fee ( million), a mean LRCASHETR (30.38%) and a mean pretax income ( million). This firm would have an ATR of A rise in ATR by 0.1 units leads to a new ATR of Assuming the audit fee not to change, the new ATR equals a tax service fee to the audit firm of million (a rise of about 90%).16 Given the above coefficient the mean firm s LRCASHETR therefore rises to 31.28%. If the long-run cash effective tax rate rises by 0.9 percentage points we assume this increase to be due to a rise in income taxes paid in time t 16 Assuming the audit fee not to change is a restrictive assumption but we apply the audit fee as a scaling variable to generate the tax fee-to-audit fee ratio. The assumption is weak and limits the interpretation of the coefficient derived in the regression but is the only possibility to deal with a variable containing information about tax fees and audit fees. The ceteris paribus assumption is crucial for our interpretation. 14

15 only. If the rise of a three-year cash ETR is due to a single-year rise of income taxes paid, the effect on an theoretical single year ETR (which is easier to interpret) is three times as big. Assuming the mean pretax income in all three years and the mean amount of taxes paid in years t-2 and t-1 not to change the calculated amount of income taxes paid in t rises from million to million, an increase of in taxes. We are aware that this interpretation only holds true for the mean firm while having constant values for pretax income. Nevertheless, it gives an impression of the economic significance. Most control variables are not statistically significant. Firms SIZE and ROA show significant results meaning that bigger and less profitable firms tend to have higher CASHETRs. In the first column it appears that being audited by KPMG leads to a lower ETR, but the significance is weak. Running the regression without control variables gives consistent results. In the LRGAAPETR case we do not find a significant association between ATR and tax planning. However, in Panel B using annual GAAPETR and not including the internationality measure shows significant results for the coefficient of interest on a 5%-level, including the internationality measure yields no statistical significance anymore. However, one might wonder if this effect is due to INTLTY or due to the change in the sample size. Even in the absence of INTLTY, the economic significance of the ATR coefficient is half compared to the results in column one and two. SIZE is negatively associated with GAAPETR (on a significance level of 10%) indicating that bigger firms tend to be characterized by lower GAAPETRs, while in the CASHETR case we have an opposite finding. Regarding the association to ATR using annual CASHETR yields similar inferences to the long-run case. <Insert Table 5 here> Taken together, these findings indicate that knowledge spillover from the audit to the tax department in favor of more tax planning are not an important issue in the German setting. Moreover, economic bonding does not appear to drive the association between APTS and tax planning. Instead reputational and liability concerns may be the key factors. Empirical Results for Tax Sustainability As Table 6 shows, we obtain negative and significant coefficients on ATR regardless of using CV_CASHETR or CV_GAAPETR. The coefficient is slightly more significant in the first case. This finding indicates that the volatility of the respective effective tax rates decreases with increasing fees for APTS relative to audit fees. In our research setting, this means that there is a 15

16 positive relation between APTS and the sustainability of tax strategies. This finding contradicts the one by Francis et al. [2015] in an US-American setting. Nevertheless, our result is in line with the view that audit/advisory firms market sustainable tax strategies as part of their services and effectively implement these strategies in their clients firms. Furthermore, because of their increased knowledge of the companies due to the audit cycles the audit firms may find it easier to suggest suitable strategies, which lead to sustainable tax outcomes. <Insert Table 6 here> The control variables have various significance levels. It is worth mentioning that in contrast to e.g. Francis et al. [2015] and Neuman et al. [2013] the volatility of earnings or pretax income respectively appears to have no significant effect on the volatility of the tax outcomes (although it is close in the case of CV_CASHETR). Moreover, in both cases we find a highly significant negative coefficient on ROA indicating that more profitable firms have more sustainable tax strategies, which seems to be plausible. Running the regressions without control variables produces consistent results. 5.3 Additional Tests Controlling for Potential Endogeneity between Tax Planning and APTS Several studies, e.g. Omer et al. [2006], Lassila et al. [2010], McGuire et al. [2012], or Krishnan et al. [2013], argue that companies do not randomly decide on whether to buy tax services from their auditor. Instead several factors which may be both observable and unobservable appear to influence this decision. McGuire et al. [2012] explicitly point out that the level of tax avoidance which is achieved by buying tax services from outside the audit firm may have an influence. Since our German setting implies quite a regulated environment this reasoning is especially important. Due to the regulations the auditor may not be in a good position to advise very aggressive strategies. Therefore, it may be more useful for the firm to rely on the work of third party tax advisory. Hence, the degree of tax planning a firm wants to achieve may also affect the decision on how much APTS are to be bought. This would induce a serious endogeneity problem in our model of hypothesis 1. To control for this issue we employ a Heckman [1979] two-stage model and a simultaneous equations approach. In the first stage of the Heckman procedure we estimate a probit model that predicts whether a company will resort to its auditor for tax services. The model is similar to the ones used by McGuire et al. [2012] and Krishnan et al. [2013], although some variables of these models 16

17 cannot be used due to data restrictions. APTS is an indicator variable set to 1 if the firm obtains any tax services from its auditor and 0 otherwise. For the definitions of the other variables we refer to the appendix. A peculiarity of our model is that we explicitly control for tax planning activity by including LRCASHETR. APTS i,t = β 0 + β 1 AUDIND i,t + β 2 LAF i,t + β 3 SIZE i,t + β 4 LEV i,t + β 5 PPE i,t + β 6 ROA i,t + β 7 CASH i,t + β 8 DEP i,t +β 9 LRCASHETR i,t + β 10 Deloitte i,t + β 11 EY i,t + β 12 KPMG i,t + β 13 PwC i,t + Industry Fixed Effects + Year Fixed Effects + ε i,t (4) Results from this regression are reported in column 1 of Table 7. It is worth noticing that the long-run cash ETR has a slightly significant positive impact on the probability to buy APTS. The estimates from equation (4) are used to construct an inverse Mills Ratio which will be included as an additional control variable in the second step. Apart from that, the second step is basically identical to equation (1). Results are shown in columns 2 and 3 of Table 7. Column 2 follows Krishnan et al. [2013] and includes all remaining observations while column 3 omits those observations with no APTS according to McGuire et al. [2012]. In both columns we still find a highly significant positive association between ATR and LRCASHETR which is consistent with our previous finding of less tax planning in the presence of APTS. Furthermore, the inverse Mills Ratio being significant in both column 2 and 3 indicates that there may be potential endogeneity associated with the decision to buy APTS (Krishnan et al. [2013]). However, this does not change our inferences. Additionally, it is striking that in column 3 the explanatory value more than doubles compared to column 2 while all big-4 companies are now significantly positive associated with tax planning. <Insert Table 7 here> In a second approach to deal with the endogeneity problem we use a simultaneous equations estimation, specifically a three-stage least squares model. The detailed system of equations is shown in Table 8. The first equation is identical to equation (1) and the second equation which is estimated simultaneously has ATR as the dependent variable and the same independent variables as equation (4), because these are supposed to influence the decision on whether and to what extent APTS are obtained. The results in column 1 of Table 8 are consistent with APTS negatively influencing tax planning even after controlling for endogeneity by estimating the 17

18 opposite direction, i.e. the amount of APTS depending on the level of tax planning, simultaneously. Moreover, the opposite direction in column 2 does not show a significant coefficient on LRCASHETR suggesting that this direction may not be very important.17 <Insert Table 8 here> Taken together, the above additional tests corroborate the findings in the main analysis that APTS do not have an increasing influence on tax planning due to knowledge spillover to the tax department of the audit firm. Although the decision on how much tax services are purchased from the auditor and the pursued level of tax planning may be endogenous, our finding still holds. Splitting the Sample Depending on SIZE In Panel B of Table 5, the statistically significant coefficients of SIZE for both GAAPETR and CASHETR show different signs. This finding indicates that the firm size is somehow relevant and the effect on the respective ETR measure is not quite obvious. To further analyze the effect of ATR on the ETR while taking into account the special role of firm size, we split up the sample into three groups based on the natural logarithm of total assets. Within the terciles we repeat the baseline regression in equation (1). Untabulated results show that the effect found for the whole sample using CASHETR seems to be driven by bigger firms since the coefficient for ATR does not have statistical significance for the lowest tercile. Firms within the second and highest tercile seem to provide evidence for an economically more significant influence of ATR on CASHETR although being less statistically significant. We explain the lower statistical significance by the small sample of less than 230 firm-year observations per group. However, the adjusted R 2 rises to over 33% in the second and third tercile. In contrast, the effect of ATR on GAAPETR appears to be driven by those firm-year observations ranked in the lowest tercile. The economic as well as the statistical significance increase for the first tercile while not yielding any statistical significance for the second and third tercile. This finding is interesting because we interpret the altering statistical significance of ATR coefficients as evidence for the different influence of APTS on ETR measures. It seems that for smaller18 firms rising levels of APTS are associated with higher levels of GAAPETR while not affecting the CASHETR. For 17 However, the explanatory power in column 2 of Table 8 is poor. 18 Although labeling this group of firms smaller firms it is worth noting that all firms are listed in the German prime standard and that these firms are not necessarily small firms but the smallest within the biggest listed firms in Germany. 18

19 bigger and medium firms we provide evidence that rising levels of APTS are associated with higher levels of CASHETR while not affecting the GAAPETR. 6 The Accounting Perspective 6.1 Research Design We evaluate accrual quality (our proxy for audit quality and earnings quality in general) according to the Francis et al. [2005] accrual quality model. In a first step, the following linear time series regression is conducted in order to estimate total current accruals, while the focus is on the remaining residuals. TCA i,t = α 0 + α 1 CFO i,t-1 + α 2 CFO i,t + α 3 CFO i,t+1 + α 4 Rev i,t + α 5 PPE i,t + ε i,t (5) TCA i,t is a measure of total current accruals of firm i in year t calculated as TCA i,t = CA i,t - CL i,t - CASH i,t - DCL i,t where CA i,t is the change in current assets between year t-1 and t, CL j,t denotes the change in current liabilities between year t-1 and t, CASH i,t is the change in cash between year t-1 and t and DCL i,t denotes the change in debt in current liabilities between year t-1 and t. The total current accruals are regressed on CFO i,t which is firm i s cash flow from operations in year t, its lag and lead values, on Rev i,t indicating the change in sales revenue between year t-1 and t and on PPE i,t denoting the value of property, plant and equipment in year t. As established by Francis et al. [2005] we use current accruals because there are no long time lags between current accruals and cash flow realizations, which could be the case if total accruals were used. We conduct a cross-sectional regression where we pool all firms in the sample, which is different compared to Francis et al. [2005] where firms are pooled on an industry level with each industry group having at least 20 observations. Due to the relatively small size of our sample, this approach is not possible. We expect the same signs to appear as in previous studies, which should be in line with the reasoning of Dechow and Dichev [2002]. The standard deviation of the residuals of equation (5) is the accrual quality proxy. However, instead of a rolling window of five years we take a three-year window from t-2 through t. The shorter window is due to the fact that we do not have time series in our sample that are sufficiently long. To sum up, our measure of accruals quality is as follows: AQ i,t = σ(ε i ) t, calculated over t-2 through t. Higher values of AQ i,t indicate a lower accrual quality because the standard deviation is higher. Following Francis et. al [2005] in our further analysis next to the raw data we also employ 19