The impact of office size on audit quality Abstract

(1)

Master Thesis Accountancy

The impact of office size on audit quality

Abstract

Multiple studies have found a relationship between office size and audit quality, where larger offices’ deliver higher audit quality. Almost all of these studies have been conducted in the United States. In this paper I will study the relation between office size and audit quality in Germany. I use a sample of 3981 observations of German public listed firms over the period of 1999 – 2009. In this study I differentiate between offices who are part of Big 3 audit firms and offices who are part of a non-Big3 audit firm. My findings are in contrast with results of previous studies as I cannot find evidence that office size has an influence on audit quality in Germany.

Student name: Jelmer Nutma Studentnumber: S2426065

Supervisor: dr. C.A. Huijgen Second assessor: dr. N. Hussain Word count: 8409

(2)

1

1. Introduction

In the last decade there were multiple financial and accounting scandals. These scandals leave us to guess whether the quality of the performed audits is of a sufficient level. The accounting scandals at Enron, WorldCom and Royal Ahold were the result of accounting errors, where also mistakes were made in the auditing process. The previously mentioned cases display the sufficiency of audit quality, or better lack thereof, at those firms. The lack of audit quality is however, a severe problem, as it is the role of the auditor to verify the information that the client is going to disclose in the annual report. By verifying the information the auditor plays a leading role in decreasing the information asymmetry between the client and its

stakeholders, making it important to deliver audits of sufficient quality. But what determines whether there is a high or a low level of audit quality? DeAngelo (1981) describes audit quality as the market assessed joint probability that a given auditor will a) detect a breach in the accounting system of the client and b) report this breach. The former statement is based on the competence of the auditor and the latter is based on the independence of the auditor. This definition however, is only a starting point, because to determine whether an audit is of sufficient quality, we have to look at the factors that have an influence on the quality of an audit.

In his review paper Francis (2004) has stated multiple influencing factors regarding audit quality. Francis (2004) argues that there are two discerning categories of research on the topic of audit quality, the first wave and the second wave of audit quality research. The first wave of audit quality is based on the big firm – small firm dichotomy as stated in the study of DeAngelo (1981). DeAngelo (1981) uses the size of an audit firm as a proxy for audit quality, since no single client is important enough for a big firm to risk losing its reputation and with it risking to lose its entire clientele. In contrast, in small firms where a single client might be important for the auditing firm’s existence, the auditor is more willing to go along with the client and misreport. In the first wave of research, Simunic and Stein (1987) and Francis and Wilson (1988) also came with results supporting this line of reasoning. Big N auditing firms have created a good and positive reputation and have therefore the incentive to deliver high quality audits to protect this reputation.

The second wave of research on audit quality moves beyond the big firm – small firm dichotomy. The second wave of research does not view the audit quality of the Big N firms homogeneous across the whole firm, but tries to differentiate. Within this differentiation there

(3)

2

are two dominant areas of research, whether the auditing firm is industry leader in the

industry of the client and whether there is an effect between different local offices. The notion that industry expertise has an impact on the audit quality is suggested by Solomon et al. (1999). They argue that the ‘industry expert’ has more experience within the industry and is therefore more capable to make sound audit judgments which have a positive effect on the audit quality. This coincides with the research of Francis et al (2005) who conclude that the industry expert receives a premium on its audit fee, implying a higher quality of the audit. The second aspect of the differentiation is based on the differences between offices of an auditing firm. The starting point here is the conclusion of DeAngelo (1981), that firm size can be used as a proxy for audit quality. However, when we scale the level of analysis down from the entire international big N firm to a local office of the big N firm, it is not so big anymore (Francis, 2004). Especially since the audit agreements are being handled by an engagement partner who often works at an office of the big N firm in the city where the client has its headquarters (Francis, 2004), gives reason that we might expect a difference in audit quality between the different offices.

Reynolds and Francis (2000) are the first who studied this level of differentiation. Their research shows that offices treat larger clients more conservative than smaller clients. This study also argues the importance of research regarding this topic since these relations are not found at firm level analysis, therefore making it important to study the relation between office size and audit quality. Choi et al (2010) continued on this topic and found a direct relation between office size and audit quality. This research was the first of its kind and based on a US setting.

This research will focus on the relationship between office size and audit quality in the

German setting. This will be conducted following the research question: What is the influence of office-size on audit quality?

This research paper contributes to the existing literature regarding the effect of the office size on audit quality as it is based upon the German setting. In my research I apply a sample of German public listed firms. Other studies such as those of Francis, Stokes, and

Anderson (1999), Reynolds and Francis (2000), Craswell et al. (2002), Francis and Yu (2009), Choi, et al. (2010) and Francis, et al. (2013) focus primarily on a US setting in which they use SEC registrants in their sample. The difference between the U.S. setting and the German setting is that the German audit market is structured in a different manner. The audit

(4)

3

market in the US is heavenly dominated by the Big 4 auditing firms, whereas the German market is dominated by a Big 3 (PwC, EY and KPMG) with Deloitte as a tag-a-long, followed by a series of small to medium sized firms, which combined represent a decent market share. My paper also contributes to the existing literature as I try to find evidence of a positive association between office-size and audit quality, implying that the audit quality is not

homogenous across all offices of a firm. Choi et al. (2010) studied this topic in the US setting and found direct evidence that this relation exists. Choi et al. (2010) also note that in future research, more focus should be laid on the local office level auditor behavior as the unit of analysis. This research will try to expand on the relationship found by Choi et al. (2010) and will try to find more evidence for this relationship to support generalization of the results. To my knowledge, this is the first research which tries to find this relation outside the US combining Big 3 firms and non-Big 3 firms together. This is a significant part of my contribution to the current literature as there are differences between the business’ models. Whereas the US setting is based on the Anglo-Saxon business model, the German setting is based on the Continental business mode. Joos and Lang (1994) concluded that substantial financial reporting differences exists between the two business models. Lastly, I contribute to the existing literature with this research because I differentiate of the current research by determining whether there are differences between Big 3 offices and non-Big 3 offices regarding the effect of office size on audit quality.

The remainder of this paper is structured as follows. The next chapter will elaborate on the literature and background. The third chapter discusses the research design, including the regression models and measurement of the variables. The fourth chapter shows the results of the research. In the fifth chapter I will discuss the results and the sixth chapter contains the conclusions.

(5)

4

2. Literature and background

In this chapter I will explain the role of the auditor in the society in section 2.1 and expand on the terms used and define these terms, as well as provide an overview in the current literature regarding audit quality in section 2.2. In section 2.3 I will formulate my hypotheses.

2.1 The role of the auditor in society

Before I can expand on what defines audit quality, we have to describe the role of the auditor within the society and what makes the role of the auditor so important.

A broadly accepted theory to describe the need for an auditor is the agency theory (Jensen and Meckling, 1976). This theory suggests the existence of two parties within a firm, being the principal (i.e. the shareholder of a firm) that engages another person, the agent, to perform services on the basis of a contract. For the agent to be able to fulfill these services, the principal has to delegate some decision making authority to the agent (Jensen and Meckling, 1976). The agent and the principal are often utility maximizers, which means that both parties try to receive maximum return for their input (Eisenhardt, 1989). Because of information asymmetry between both parties, which originates from the different roles that the two parties fulfil, agency costs arise (Jensen and Meckling, 1976). It is hard to reduce the agency costs because of the difference in roles. The agent is involved into the day-to-day business of the firm which he works for and thus gains a lot of insight and information. On the other side is the principal who is hardly involved with any of the day-to-day business of the firm and lacks the insight and information that the agent has acquired. Because of the existence of

information asymmetry, where the principal lacks information about the efforts of the agent, it is hard for the principal to evaluate the decisions made by the agent. This is where the

company’s financial report comes into play. In this way, the agent is able to inform the principal (shareholder) about the performance of the company. To increase the usefulness of the report an auditor is hired to review the information conveyed in the annual report. It is the auditor who decides whether the annual report gives a true and fair view of the company’s assets and liabilities. If the auditor certificates the financial report the principal receives a reassurance regarding the information the report contains which enables him to make deliberate decisions.

Apart from the agency theory there is also the stakeholder theory (Freeman, 1984) which follows the same path regarding information asymmetry and its problematic nature. It is however, not based on the same agent-principal role allocation. It views the shareholders as

(6)

5

one of the stakeholders of the company, next to a range of other stakeholders like employees, customers, suppliers, financiers, communities and governmental bodies (Freeman, 1984). In this framework there is also a key role for the auditor, as it not only approves the information within the financial report for the shareholders but for all of the other stakeholders as well. Therefore, a range of parties relies on the approval of the financial report by the auditor, giving the auditor an important role in society as information verifier.

2.2 Determinants of audit quality

Because of the important role that the auditor fulfills in society it is necessary that the auditor delivers a high level of audit quality. However, to determine what audit quality is and what defines it, we return to DeAngelo (1981). The definition provided by DeAngelo (1981) is the market-assessed joint probability that a given auditor will (a) discover a breach in the client’s accounting system, and (b) report the breach. Nevertheless, audit quality is hard to measure and also costly to evaluate for the users of the financial reports.

Audit quality can be conceptualized as a continuous spectrum with on the one end a very low audit quality and on the other end a very high quality (Francis, 2004). Quite clear is that audit failures find themselves at the lower end of the quality spectrum. Returning to the first part of DeAngelo’s (1981) definition suggests the failure that the auditor discovers a breach, for example the auditor fails to enforce the generally accepted accounting principles. And the second part of the definition suggests the failure that an auditor reports this breach, for example the auditor fails to modify the audit report. Both of these failures can be misleading for the intended user of the financial statements (Francis, 2004).

To determine the audit failure rate, one can consider the number of litigations against audit firms. However, Palmrose (2000) studied this phenomenon and concluded that the outright audit failure rate approaches zero, as this outright failure rate is so low that it is difficult to imagine what could be changed in either the profession or in the law to substantially lower the failure rate. The same matters for the research of Francis and Krishnan (2002), who studied outright business failures, as there is only a small number of outright business failures

annually. It is also not possible to argue that the cause of an outright business failure is caused by an audit failure. Research of Dechow et al. (1996) and the Government Accounting Office (GAO, 2003) regarding SEC sanctions and earnings restatements amount to a very low audit failure rate.

(7)

6

The difficulties of measuring audit failures has led to the creation of several proxies to be able to measure audit failures. DeAngelo (1981), Simunic and Stein (1987) and Francis and

Wilson (1988) use firm size as a proxy for the quality of the audit. Where they state that the larger the audit firm, the higher the audit quality. They do not argue that the larger audit firms do not make any mistakes or misreport, but they simply suggest that the larger audit firms deliver audits of an on average higher quality than the audits of smaller audit firms. This proxy receives however, criticism as the second wave of research regarding audit quality is trying to differentiate from the big firm – small firm dichotomy, rather than taking auditor firm size at the national or international level into account (Francis, 2004). Urges have been made to conduct future research focussed on the effect of the local office on audit quality (Reynolds and Francis, 2000 and Francis and Yu, 2009).

Reynolds and Francis (2000) commenced with research in this area. In their research,

Reynolds and Francis (2000) focus on the effect of client size on the auditor decision making. They signify the importance to focus on the office level of analysis as the decision and

judgement processes are being made on this level. This was reinforced by of Ferguson et al (2003) and Francis et al (2005), as these studies are more focussed on the effect of industry expertise on office level quality and make fair points why the office level of analysis is an important perspective. Most important is that the audit engagement is administered by a local office-based partner who is often located in the same city as the main office of the client. Francis and Yu (2009) conclude in their research that larger offices of Big 4 firms produce higher quality audits than the smaller offices of those Big 4 firms. They argue that local offices are handling most of the actions needed for the administration of the audit, contract the client, engage in the audit process and also issue the audit report. Consistent with the

previously suggested fact that the local audit office and the headquarters of the client are often located in the same city, provides the opportunity for the creation of office and client specific knowledge and expertise (Francis et al, 1999). Especially important is the decentralized office structure within Big 4 Firms, enabling the reduction of information asymmetry and providing the opportunity for the Big 4 auditors to develop better knowledge of the potential and

existing clients in the geographic local area. This also works the other way around where the client gains knowledge and confidence in the quality of the locally based auditors, who actually perform the audit (Carcello et al., 1992). This emphasizes the previously noted possibility that audit quality is not homogeneous across the audit firm and that offices might

(8)

7

deviate above or below the set standard of the auditing firm, meaning that part of the audit expertise is office specific (Francis et al. 2005, Vera-Munoz et al. 2006).

Larger offices have more employees, more engagement hours and most likely, more and/or bigger clients, leading to a build-up of more expertise and experience within larger offices. Becker (1993) argues that this is very important, as experience is a crucial part of human capital, giving a larger auditing office more human capital than offices of a smaller size. Danos et al (1989) show that auditors are more likely to consult colleagues within the same office rather than consulting another office or the organisation on the national level, causing larger offices to be better equipped for inter office consulting and exchanging experience as well as having more resources available. In addition to the experience and expertise that the larger auditing firms have access to, it also has the possibility to provide extra man hours to work for the client because there is a more staff available. This supports the idea that a larger office is better able to deliver a higher quality audit than a smaller office.

2.3 Hypotheses

From the previous arguments we can argue that the proxy for audit quality used by DeAngelo (1981), the size of the auditing firm, is not valid when it comes to determining audit quality on the office level of analysis. In this study I will differentiate from the proxy set by

DeAngelo (1981) and study the relation between office size and audit quality to receive a proxy that fits better with audit quality. To obtain this proxy I will utilize the research question stated in the introduction: What is the influence of office-size on audit quality? To answer the research question I turn to the hypothesis H1: Larger offices have a positive effect on audit quality. This hypothesis lays the basis for the answer of the research question. I will however, differentiate by adding two more hypotheses namely: H2: Larger Big 3 offices have a positive effect on audit quality, and H3: Larger non-big 3 firms have a positive effect on audit quality. By adding these hypotheses I focus on two different spectrums within the auditing market, the Big 3 firms and the non-Big 3 firms. H2 focusses on whether office size has an effect on audit quality within Big 3 firms. H3 in turn focusses on whether office size has an effect on audit quality applying a sample containing only non-Big 3 firms.

(9)

8

3. Research Design

In this chapter I will elaborate on the measurement of audit quality in section 3.1. In section 3.2 I describe the measurement of the independent and control variables, followed by section 3.3 in which I expand on the sample that I base my research on.

3.1 Measurement of audit quality

As a measureable proxy for audit quality I turn to the earnings quality of the client. This is a much used and accepted proxy in this particular research area. Becker et al (1998) studied the relationship between earnings quality and audit quality and found that a high quality auditor is more likely to detect accounting practices that are questionable. Moreover, the higher quality auditor is more likely, in comparison to a low quality auditor, to object the application of these practices or modify the audit report. To determine the level of the earnings quality I will use the behaviour of the client regarding abnormal accruals in practising of earnings

management (Frankel, Johnson and Nelson 2002).

To measure earnings quality I use the modified Jones model (Dechow et al. 1995) which is a generally used and accepted model to detect earnings management. However, some concerns exist regarding the use of the modified Jones model as it is prone to estimation errors. To deal with this problem I will also use the model of Ball and Shivakumar (2006). This model controls for the asymmetric timeliness of accruals in recognizing economic gains and losses. I add the second measure of earnings quality to increase the robustness of the results via a robustness check.

The following equation illustrates how I obtain the discretionary accruals following the modified Jones model (Dechow et al, 1995):

First, I determine the total accruals (1):

TAt/At-1 = a1[1/Aτ-1] + a2[(∆REVτ - ∆RECτ)/At-1] + a3[PPEτ/At-1]+ εt (1)

After that I determine the total accruals of an organisation for year t, I estimate the nondiscretionary accruals with the following equation (2):

NDAt/At-1 = α1[1/At-1] + α2[(∆REVt - ∆RECt)/At-1] + α3[PPEt/At-1] (2)

Now that I have estimated the nondiscretionary part of the total accruals, I calculate the discretionary accruals according to the following formula (3):

(10)

9

DAt/At-1 = TAt/At-1 – NDAt/At-1 (3)

Where:

TAt = Total accruals scaled by total lagged assets at t-1;

NDAt = estimated nondiscretionary accruals scaled by total assets at t-1;

t = a year subscript indicating a year in the event period;

∆REVt = revenues in year t less revenues in year t -1 scaled by total assets at t-1;

∆RECt = net receivables in year τ less net receivables in year t -1 scaled by total assets at t-1;

PPEt = gross property plant and equipment in year t scaled by total assets at year t -1

At-1 = total assets at t -1;

ε = error term;

a1, a2, a3 = denotation of OLS estimates of α1, α2, α3;

α1, α2, α3 = industry specific parameters; and

DAt = Discretionary accruals scaled by total assets at t-1.

For the robustness check I use the model of Ball and Shivakumar (2006). The following equations illustrate how I obtain the discretionary accruals using the model of Ball and Shivakumar (2006):

First, I calculate the total accruals utilizing the following equation (4): TAt / At-1 = b1[1/At-1] + b2[(∆REVt - ∆REC)/At-1] + b3[PPEt / At-1]

+ b4[CFOt / At-1] + b5DCFOt + b6 [(CFO / At-1) * DCFOt] + εt (4)

Now that I have determined what the total accruals consist of, I calculate the nondiscretionary accruals with the following equation (5):

NDAt / At-1 = β1[1/At-1] + β2[(∆REVt - ∆REC)/At-1] + β3[PPEt / At-1]

+ β4[CFOt / At-1] + β5DCFOt + β6 [(CFO / At-1) * DCFOt] (5)

Finally, I determine the discretionary accruals according to the following formula (6):

DAt/At-1 = TAt/At- - NDAt /At- (6)

Where:

TAt = Total accruals scaled by total lagged assets at t-1;

NDAt = estimated nondiscretionary accruals scaled by total assets at year t -1;

t = a year subscript indicating a year in the event period;

(11)

10 ∆RECt = net receivables in year τ less net receivables in year t -1 scaled by total assets at t-1;

PPEt = gross property plant and equipment in year t scaled by total assets at year t -1;

CFO = cash flow from operation scaled by total assets at year t -1;

DCFO = a dummy variable that equals 1 if the CFO is negative and 0 otherwise;

At-1 = total assets at t -1;

ε = error term;

b1, b2, b3 = denotation of OLS estimates of β 1, β 2, β 3;

β1, β2 β3,… =industry specific parameters; and

DAt = Discretionary accruals scaled by total assets at year t -1;

To estimate the industry specific parameters I use the Ordinary Least Squared (OLS)

regression. I organize the categorization of the different categories according to the grouping put forth by Barth et al (1998). This is a grouping which denominates each industry with a value between 1 and 14 where each value represents a different industry. The resulting variables of these equations are the dependent variables DA1, from the modified Jones model (1995) and DA2, from the Ball and Shivakumar (2006) model. To actually apply the variables in a regression model I use the absolute value of DA. This is to counteract the possibility that the values of DA balance out around the value of 0. As DA1 and DA2 can have positive or negative values, it does not have an impact whether the value of DA is positive or negative but the distance between the value and zero is important. Therefore, I use absolute values to deal with this problem as it keeps the distance between zero and the value of DA intact.

3.2 Measurement of variables

To measure office-size I apply the same methods that are used in other studies regarding this topic, i.e. I sum all the revenues of the clients of the office and use the result as a proxy for office size. To account for the possibility that extremely large and extremely small clients influence the results, I scale total revenues with the natural logarithm.

In addition to the total revenues of the client I also use another variable to measure office size. This additional variable is measured by summing the number of clients of each office. This additional variable is utilized to add robustness to the results of the research.

The variable OFSIZE1 is calculated with the natural logarithm of the total revenues of the client of an office. OFSIZE2 is composed of the number of clients per office, these are the two independent variables in this research.

(12)

11

To adjust for known effects on discretionary accruals I add several control variables to the model. I add LN_Assets (natural logarithm of the total assets) to control for the client size effect on accruals quality (Dechow and Dichev 2002) as larger clients often have a higher earnings quality. Loss_reporting is a dummy variable (with a value of 1 if the firm reported a loss and a value of 0 if the firm reported a profit) to control for potential differences in the accrual quality for firms that have reported a loss and firms that have made a profit (Choi et al. 2007). Firms that report a loss often have a higher incentive to manage earnings than firms that have reported a profit. LEV (total liabilities divided by the total assets) is a control variable to control for the fact that highly leveraged firms often have a greater incentive to manage earnings (Becker et al. 1998). To control for growth of the firm I add the control variables B_TO_M (book to market ratio) and Change_Rev (changes in revenues deflated by the lagged total assets) to control for growth of the firm. As high change in revenues and a low book to market value are often an indicator for more earnings management. To control for effects regarding the tenure of the auditor at the client (Johnson et al. 2002) I include a control dummy TENURE (with a value of 1 if the tenure is three years or less and a value of 0 otherwise) as the audit quality is on average lower in the first three years of an audit tenure. I add the dummy variably BIG 3 (1 if the firm is being audited by a Big 3 firm and 0 otherwise) to control for the effect that larger auditing firms deliver on average a higher level of audit quality (DeAngelo 1981).

Lastly I add control dummies for the years in the sample to control for macro-economic events like the financial crisis in 2007-2008.

3.3 Sample

I base my research on a database of German public listed firms for the time period of 1999 until 2009. The dataset also contains data for the years 1997, 1998 and 2010 to account for lag and lead effects. The database contains data of 5012 observations for the years 1999 until 2009. For this research I excluded 1032 observations as they represent audit firms that have only a single office. As I try to measure the effect of the office size, firms with a single office cannot measure the effect and are therefore eliminated in the dataset. The remaining dataset contains 3981 observations.

To test the hypothesis 1: Larger offices have a positive effect on the audit quality. I include all 3981 observations. To test the 2nd and 3rd hypothesis:

H2: Larger Big 3 offices have a positive effect on audit quality. H3: Larger non-big 3 firms have a positive effect on audit quality.

(13)

12

I exclude all the non-Big 3 offices to test H2, leaving 2487 observations in the sample. To test H3 I exclude all Big 3 offices leaving a sample of 1494 observation. For both hypotheses I exclude the dummy variably BIG 3 as it is redundant to include this variable since either all the offices in the sample are Big 3 or all the offices are non-Big 3.

(14)

13

4. Results

In this chapter I show the descriptive statistics in section 4.1 and the correlation matrix in section 4.2. The results of the regression analyses are presented in section 4.3 and lastly I will address the robustness check in section 4.4.

4.1 Descriptive statistics

Table 1 shows the descriptive statistics of the variables that act as a proxy for audit quality, |DA1| and |DA2|, the independent variables OFSIZE1 and OFSIZE2 as well as the control variables. I exclude the year dummies in the descriptive statistics since they do not show relevant information regarding the sample.

TABLE 1 Descriptive Statistics

N Minimum Maximum Mean

Std. Deviation |DA1| 3981 ,00 1,10 ,107 ,12 |DA2| 3981 ,00 1,10 ,107 ,12 OFSIZE1 3981 ,46 13,01 8,16 2,67 OFSIZE2 3981 1 22 7,20 5,61 LNAssets 3981 -,74 12,48 5,34 2,14 B_TO_M 3981 -17,32 10,32 ,73 ,938 Change_Rev 3981 -5,20 144,95 ,10 2,44 LEV 3981 -,73 6,35 ,59 ,29 Big3 3981 0 1 ,62 ,48 Loss_reporting 3981 0 1 ,29 ,46 TENURE 3981 0 1 ,66 ,47

Based on the descriptive statistics it is noteworthy that the means of |DA1| and |DA2| are close to each other. Regarding office size we see that, measured by the total revenue of the clients, the average (mean) size for OFSIZE1 is 8.157 which is about €348 million. The maximum number of clients measured with OFSIZE2 is 22. The average size of the firms used in the sample is 5.345 which translates to about €210 million. The table shows that about 62% of the clients in the sample are audited by a Big 3 office and 66% of the sample consists of an audit in the first three years of a client-auditor relationship.

(15)

14 4.2 Correlation matrix

Table 2 shows the Pearson correlation matrix among the variables (n = 3981). The correlation matrix shows that the measures of earnings quality, |DA1| and |DA2| are significantly

correlated with a coefficient of 0.922 (p<0.01).

The two measures OFSIZE1 and OFSIZE2 which measure the office size are significantly correlated with a coefficient of 0.709 (p<0.01).They are however, not correlated with the proxies for audit quality |DA1| and |DA2|. |DA1| and |DA2| are significantly correlated with LNAssets with a coefficient of -0.158 (p<0.01), Big 3 with a coefficient of -0.050 (p<0.01) and TENURE with a coefficient of 0.056 (p<0.01). These correlations indicate that these variables have a significant influence on the dependent variables |DA1| and |DA2|. The relation of the correlation is in line with the expected outcome, LNAssets and Big 3 are

expected to be negatively correlated and TENURE is expected to be positively correlated with the variables |DA1| and |DA2|.

The correlation between the control variables is very low which suggest that there will be no problems regarding multicollinearity. The highest correlation is between CFO and

CHANGE_SALES with a correlation coefficient of 0.387. To validate the suggestion of no multicollinearity concerns, I have calculated all the variance influence factors (VIF) values in the regression models. None of the variables show a VIF value greater than 10 reinforcing the suggestion that multicollinearity is of no concern in this research.

(16)

15 TABLE 2

Correlation Matrix

|DA1| |DA2| OFSIZE1 OFSIZE2 LNAssets B_TO_M

Chang

e_Sales LEV BIg3 Lossreporting TENURE CFO

|DA2| ,922*** ₁ ,000 OFSIZE1 ,011 ,018 1 ,472 ,267 OFSIZE2 ,007 ,016 ,709*** ₁ ,660 ,321 ,000 LNAssets -,158*** _-,161*** _,050*** _,015 ₁ ,000 ,000 ,002 ,355 B_TO_M -,025 -,030 ,019 ,007 -,014 1 ,115 ,056 ,221 ,639 ,369 Change_Sales ,002 ,005 -,006 -,002 -,026 -,001 1 ,909 ,746 ,710 ,903 ,099 ,972 LEV -,013 -,016 ,099*** _,048*** _,026 _,033** _-,008 ₁ ,401 ,303 ,000 ,002 ,100 ,036 ,630 Big3 -,050*** _-,032* _,001 _,008 _,246*** _-,051*** _-,028 _-,008 ₁ ,002 ,042 ,930 ,620 ,000 ,001 ,082 ,608 Loss_reporting ,139*** _,147** _,028 _,028 _-,318** _,005 _-,007 _,010 _-,095*** ₁ ,000 ,000 ,078 ,075 ,000 ,734 ,639 ,533 ,000 TENURE ,056*** _,061** _,021 _,044*** _-,132** _,003 _,013 _,020 _-,109*** _,095*** ₁ ,000 ,000 ,188 ,005 ,000 ,848 ,407 ,213 ,000 ,000 CFO -,016 -,008 ,075*** _,028 _,003 _,027 _,387***_-,079*** _,001 _-,011 _,003 ₁ ,321 ,614 ,000 ,074 ,842 ,086 ,000 ,000 ,947 ,502 ,857

(17)

16 4.3 Regression results

Table 3 presents the results of the regression analyses for Hypothesis 1: Larger offices have a positive effect on the audit quality. Table 3 shows the independent variable |DA1| regressed on OFSIZE1 and OFSIZE2 and the control variables.

TABLE 3

Results of regression of office size on audit quality |DA1| and OFSIZE1 and OFSIZE2

OFSIZE1 OFSIZE2

Predicted

effect B Sig. B Sig.

OFSIZE - ,001 ,292 0.0009 ,790 LNAssets - -,007 ,000*** -,007 ,000*** Change_Rev + 0.0005 ,940 0.0006 ,937 B_TO_M - -,004 ,087* -,004 ,090* Loss_reporting + ,026 ,000*** ,026 ,000*** Big 3 - -,002 ,583 -,002 ,577 TENURE + ,008 ,056* ,008 ,054* LEV + -,005 ,457 -,004 ,510 Intercept ,138 ,000*** ,143 ,000***

Year dummies Included Included

N 3981 3981

R2 _0.038 _0.038

Dependent Variable: |DA1|

*.**.*** Denote p < 10%, < 5% and < 1 %, respectively, with two tailed tested

Table 3 shows the results of regressing the variable |DA1| on OFSIZE1 in the second, third column. The fourth and fifth column in the table shows the results of |DA1| regressed on OFSIZE2. The results in table 3 show that OFSIZE1 is insignificantly positive even at the ten percent level with a coefficient magnitude of 0.001 and a p value of 0.292. Furthermore, the results of the second regression are consistent with the outcome of the first regression. Table 3 shows that OFSIZE2 is insignificantly positive even at the ten percent level with a coefficient magnitude of 0.0009 and a p value of 0.790. It is noteworthy to take into account that the expected negative effect between |DA1| and OFSIZE1 and OFSIZE2 is actually the opposite and in fact shows a positive effect. These results indicate that office size has no influence on earnings quality.

(18)

17

When interpreting the results of the control variables we see that the predicted effect is consistent with the effect found in the results, except for the control variable LEV which shows a negative effect instead of the expected positive effect. The control variables

LNAssets and Loss_reporting are significantly negative and positive respectively at the one percent level. The book to market ratio (B_TO_M) and TENURE significant at the ten percent level. These results indicate that LNAssets, Loss_reporting, B_TO_M and TENURE have an influence on earnings quality.

The results of the regression analyses for Hypothesis 2: Larger Big 3 offices have a positive effect on audit quality are shown in table 4. Table 4 shows the regression results of office size on audit quality based on the observations of Big 3 offices only. The table shows |DA1| regressed on OFSIZE1 and OFSIZE2 and the control variables.

TABLE 4

Results of regression of office size on audit quality within the big3 |DA1| and OFSIZE1 and OFSIZE2

Predicted

OFSIZE - ,001 ,095* ,000 ,481 LNAssets - -,006 ,000*** -,006 ,000*** Change_Rev + ,002 ,764 ,002 ,751 Loss_reporting + ,025 ,000*** ,025 ,000*** B_TO_M - -,003 ,166 -,003 ,179 TENURE + ,002 ,690 ,002 ,687 LEV + ,003 ,650 ,005 ,549 Intercept ,128 ,000*** ,136 ,000***

Year dummies Included Included Included Included

N 2487 2487

R2 _0.037 _0.036

In the second and third column of table 4 the results of the regression of |DA1| on OFSIZE1 are presented. The fourth and fifth column show the results of |DA1| regressed on OFSIZE2. In table 4 we can see that OFSIZE1 is positively significant at the ten percent level with a coefficient magnitude of 0.001 and a p value of 0.095. The direction of the effect between

(19)

18

OFSIZE1 and earnings quality is however, not aligned. This means larger offices of a Big 3 firm are less effective at detecting and constraining opportunistic earnings management than smaller offices of a Big 3 firm. The results of the second regression of |DA1| on OFSIZE2 shows opposing results. Table 4 shows that the results are insignificant positive even at the ten percent level with a coefficient magnitude of 0.000 and a p value of 0.481 indicating that office size does not have an influence on earnings quality.

The control variables used in the regressions shown in table 4 behave mostly in the same manner as the control variables used for the regressions in table 3. The control variables LNAssets and Loss_reporting are both significant at the 1 percent level which is similar to the results in table 3. TENURE and B_TO_M are however, insignificant at the ten percent level in contradiction with the results in table 3 where these variables are significant at the ten percent level. All control variables do show the same effect as the predicted effect in table 4 which is consistent with the results in table 3. The only effect that shows a different direction between the predicted and actual effect is LEV. Table 3 presents the outcome which is opposite of the expected effect. In table 4, the variable shows a direction that is aligned with the predicted effect.

(20)

19

In table 5 the results for the third hypothesis, Larger non-big 3 firms have a positive effect on audit quality, are shown. The content of table 5 is composed of the observations of all the offices which are not part of a Big 3 firm. The table shows the results of the regression of |DA1| on OFSIZE1, OFSIZE2 and the control variables.

TABLE 5

Results of regression of office size on audit quality within the non-big 3 |DA1| and OFSIZE1 and OFSIZE2

Predicted

OFSIZE - ,001 ,369 -,001 ,684 LNAssets - -,007 ,000*** -,007 ,000*** Change_Rev + ,000 ,905 ,001 ,874 Loss_reporting + ,000 ,940 ,000 ,919 B_TO_M - ,028 ,000*** ,028 ,000*** TENURE + -,002 ,825 -,001 ,915 LEV + ,002 ,730 ,002 ,733 Intercept ,115 ,000 ,125 ,000

Year dummies Included Included Included Included

N 1494 1494

R2 _0.045 _0.039

Dependent Variable: |DA1

In table 5 we see that OFSIZE1 is insignificant positive at the ten percent level with a

coefficient magnitude of 0.000, and a p value of 0.369. The regression of |DA1| on OFSIZE2 show that OFSIZE2 is insignificant negative with a coefficient magnitude of -0.001 and a p value of 0.684. These results indicate that that a larger non-big 3 office is not more effective at detecting and constraining opportunistic earnings management than a smaller non-Big 3 office. This also implies that office size, for non-Big 3 organizations, does not have an influence on audit quality. It is noteworthy to highlight the coefficient of OFSIZE2, which is the only regression with an effect that matches the predicted effect, whereas the effect of OFSIZE1 does not match with the predicted effect.

(21)

20

Regarding the control variables we can see that LNAssets behaves the same as with the previous models in table 3 and 4, being significant at the 1 percent level. Loss_reporting is however, insignificant is contrast to the results in table 3 and 4 in which it shows significant results. Table 5 shows B_TO_M to be significant at the one percent level whereas in table 3 the variable is only significant at the ten percent level and table 4 where B_TO_M is

insignificant. In addition the results of the control variable B_TO_M show that the expected effect and the actual effect do not match. The effect of TENURE is also not matched with the predicted effect. All other control variables follow an effect that is matched by the predicted effect

4.4 Robustness check

To add robustness to the results of the regressions shown in the previous section I applied another measure of earnings quality which is |DA2| calculated with the model of Ball and Shivakumar (2006). I present the results of the robustness checks in tables in the appendix. In the following section I will briefly summarize the important results.

In the robustness check for hypothesis 1: Larger offices have a positive effect on the audit quality, in which I regress |DA2| on OFSIZE1 and OFSIZE2 and the control variables, I find that OFSIZE1 is insignificantly positive even at the ten percent level with a coefficient

magnitude of 0.001 and a p value of 0.145. OFSIZE2 is insignificantly positive even at the ten percent level with a coefficient magnitude of 0.00 and a p value of 0.386. These results are similar to the results as shown in table 3, reinforcing the indication that office size does not have an influence on earnings quality. The control variables behave mostly the same as in table 3 with LNAssets, Loss_reporting, B_TO_M and TENURE being significant. The effects of the control variables are aligned with the predicted effects, except for the variable Big 3 which shows a positive effect in the robustness checks, contrasting with negative effect shown in table 3.

The results of the robustness check for hypothesis 2: Larger Big 3 offices have a positive effect on audit quality, show that OFSIZE1 is significantly positive at the ten percent level with a coefficient magnitude of 0.00 and a p value of 0.062. This result corresponds with the results of table 4 which also shows a significant positive relation at the ten percent level. The results of the regression of |DA2| on OFSIZE 2 presents an insignificant positive relation with a coefficient magnitude of 0.000 and a p value of 0.270. These results match the outcomes of

(22)

21

the regression analyses shown in table 4. These outcomes provide robustness to the observation that offices of a Big 3 firm are less effective at detecting and constraining opportunistic earnings management.

The control variables in the robustness check behave in a same manner as the control variables of the regression analyses in table 4. LNAssets and Loss_reporting have a

significant relation with |DA2|. The variable B_TO_M shows also a significant relation at the five percent level in contrast to the results in table 4 where B_TO_M does not display such a relation. Lastly the control variable Change_Rev does not match with the predicted effect. The robustness check for hypothesis 3: Larger non-big 3 firms have a positive effect on audit quality results in an insignificant relation between |DA2| and OFSIZE1 and between |DA2| and OFSIZE2. OFSIZE1 is insignificantly positive even at the ten percent level with a coefficient magnitude of 0.002 and a p value of 0.272 and OFSIZE2 is insignificantly negative at the ten percent level with a coefficient magnitude of -0.00007 and a p value of 0.953. The results correspond to the results presented in table 5 which show the same effects. This robustness check reinforces the suggestion that office size does not have an influence on audit quality at non-Big 3 offices.

The control variables show similar behavior as LNAssets and Loss-reporting are both significant at the one percent level. The effects resulting from the regression analyses match with the predicted effects for all the control variables except for LEV which shows a negative effect in contrast with a predicted positive effect.

(23)

22 5.

Discussion

In this chapter I will discuss the results of the regression analyses. The previous chapter shows that the results of the regression used to test hypothesis 1: Larger offices have a positive effect on the audit quality are not significant. This means that office size does not have a significant influence on audit quality with the sample of Big 3 and non-Big 3 offices. This contradicts with the results of Choi et al. (2010) who found evidence supporting the notion that larger offices provide higher audit quality. The difference with this research however, is that Choi et al. (2010) used only Big N firms in their sample.

Based on the results found in table 3 I have to reject hypothesis 1: Larger offices have a positive effect on the audit quality. The differentiation in the sample could be cause for the insignificant results of the analyses since table 5, which shows the results of the analysis of non-Big 3 offices, indicates insignificant results with a p value of 0.369. This could be the catalyst for the insignificant results for the whole sample.

Interestingly, the results of table 4 shows a significant positive relation between office size and audit quality of offices that are part of a Big 3 firm. However, the effect that it describes is the opposite of what was expected as it shows that larger offices audits result in more discretionary accruals. More discretionary accruals lower the level of earnings quality, which means that the level of audit quality is also lower. This relation is rather weak with a

coefficient of 0.001. The results of the robustness checks, which I have added in the appendix, only show a significant relation when OFSIZE1 is used as a measure of office size. This leads to ambiguous results regarding the outcome of the regressions that analyses hypothesis 2: Larger Big 3 offices have a positive effect on audit quality. The positive effect that the results show is not the result that was expected. The results indicate that Big 3 offices are less effective at detecting and constraining opportunistic earnings management. This contradicts the predicted relation that larger offices of Big 3 firms are more effective at detecting and constraining earnings management. This means that I have to reject hypothesis 2.

Hypothesis 3: Larger non-Big 3 firms have a positive effect on audit quality has to be rejected because the results of the regression analyses are insignificant, with a p value of 0.369. The robustness checks reinforce this finding as all other combinations of dependent and

independent variables show insignificant results between office size and audit quality for non-Big 3 firms.

(24)

23

Based on the results of the three hypotheses designed to answer the research question: What is the influence of office size on audit quality? I conclude that office size does not have a significant influence on audit quality. This is similar for offices that are part of a Big 3 firm and offices that are part of a non-Big 3 firm.

These results are prone to some limitations as the research is based on a German setting operating in the Continental business model and utilizing a sample consisting of observations of Big 3 and non-Big 3 firms. This can have an effect on the results compared to previous research, like Choi et al. (2010) and Francis and Yu (2009) as they focus only on offices that are part of a Big N firm and operate within the Anglo-Saxon business model. The descriptive statistics show that 62% of the audits have been conducted by a Big 3 office and 38% of the audits have been conducted by a non-Big 3 office. Comparing my sample with Choi et al (2010) who have a sample in which 80% of the audits have been conducted by a Big 4 firm and Francis and Yu (2009) who use a sample consisting of only offices of Big 4 firms, the sample composition is different.

The results of table 5 might give some more insight in why my results are insignificant. Table 5 shows the results of the regression of office size on earnings quality. The outcome suggests that office size does not have a significant influence on audit quality for offices of non-Big 3 auditing firms. This could mean that non-Big 3 offices in Germany are better organized to audit bigger clients, suggesting that smaller offices can deliver audits at the same level of quality as a larger office. This is reinforced by the way the German auditing market is structured, where the Big 3 firms are less dominant in their market share as compared to the Big N firms market share in, for example, the US.

Another factor that possibly influences the results is the control variable Big 3 in table 3 as it shows a weak negative relation between being part of a Big 3 firm and audit quality,

suggesting that being part of a Big 3 firm does not warrant a higher audit quality. Lastly it is also possible that the client characteristics have influenced the results even after controlling for a number of control variables.

(25)

24

6. Conclusion

Previous researchers have studied the topic of the influence of office size on audit quality. These studies found a relation between the aforementioned relation. In this research I studied the relation between office size of Big 3 and non-Big3 firms and its effect on audit quality. The sample that I utilized for this research consists of observations of German public listed firms.

After analyzing the results I have found that there is no relationship between office size and audit quality in Germany. This can possibly be explained by the organization of the German auditing market. The results of the research do not fit to be generalized as they are based on a German setting which has its own characteristics (legislation, business culture, audit market structure) and thus limits the generalizability of the results.

For future research I would recommend to try to find this relationship in an Anglo-Saxon business model setting, preferably in the US setting. A lot of previous research has been conducted in this setting and it would be interesting to see if the results of this research would be generalizable to the US.

I would also recommend a cross-country study focusing on the effect of office size on audit quality, with a sample covering multiple countries to try to proof country specific effects (i.e. legislation or business culture). This research could also be expanded on by adding

(26)

25

References

Ball, R., & Shivakumar, L. (2006). The role of accruals in asymmetrically timely gain and loss recognition. Journal of accounting research, 44(2), 207-242.

Barth, M. E., Beaver, W. H., & Landsman, W. R. (1998). Relative valuation roles of equity book value and net income as a function of financial health. Journal of Accounting and Economics, 25(1), 1-34.

Becker, C, M. DeFond, J. Jiambalvo, and K. R. Subramanyam. 1998. The effect of audit quality on earnings management. Contemporary Accounting Research 15

Becker, G. S. 1993. Human Capital. 3rd edition. Chicago, IL: University of Chicago

Carcello, J., R. Hermanson, and N. McGrath. 1992. Audit quality attributes: The perceptions of audit partners, preparers, and financial statement users. Auditing: A Journal of Practice & Theory 11: 1-15

Choi, J-H., and R. Doogar. 2005. Auditor tenure and audit quality: Evidence from going-concern qualifications issued during 1996–2001. Working paper, Seoul National University Choi, J. H., C. F. Kim, J. B. Kim, and Y. Zang. 2010. Audit office size, audit quality, and audit pricing. Auditing: A Journal of Practice and Theory 29 (1): 73–97.

Craswell, A. T., D. J. Stokes, and J. Laughton. 2002. Auditor independence and fee dependence. Journal of Accounting and Economics 33 (2): 253–75.

Danos, P., J. W. Eichenseher, and D. L. Holt. 1989. Specialized knowledge and its communication in auditing. Contemporary Accounting Research 6: 91-109.

Dechow, P. M., and I. D. Dichev. 2002. The quality of accruals and earnings: The role of accrual estimation errors. The Accounting Review 77 Supplement: 35–59.

Dechow, P., R. Sloan en A. Sweeney. 1995. Detecting earnings management. The Accounting Review 70: 193-225.

Dechow, P., Sloan, R., Sweeney, A., 1996. Causes and consequences of earnings

manipulation: an analysis of firms subject to enforcement actions by the SEC Contemporary Accounting Research (spring), pp. 1–36

(27)

26

DeAngelo, L., 1981, Auditor size and auditor quality, Journal of Accounting and Economics, 3(3): pp. 183–199.

Eisenhardt, K. M. (1989). Agency theory: An assessment and review. Academy of management review, 14(1), 57-74.

Ferguson, A., Francis, J., Stokes, D., 2003. The effects of firm-wide and office-level industry expertise on audit pricing. The Accounting Review (April), 429–448.

Francis, J., D. J. Stokes, and D. J. Anderson. 1999. City markets as a unit of analysis in audit research and the re-examination of Big 6 market shares. Abacus 35 (2): 185–206. Francis, J.R., 2004, What do we know about audit quality?, The British Accounting Review

36(4): pp. 345–368.

Francis, J., Krishnan, J., 2002. Evidence on auditor risk-management strategies before and after the private securities litigation reform act of 1995 Asia-Pacific Journal of Accounting and Economics (December) pp. 135–157

Francis, J., Reichelt, K., Wang, D., 2005. The pricing of national and city-specific reputations for industry expertise in the US audit market. The Accounting Review (January), Francis, J., & Yu, M. (2009). Big 4 Office Size and Audit Quality. The Accounting Review,

84(5), 1521-1552.

Francis, J. R., Michas, P. N., & Yu, M. D. 2013. Office Size of Big 4 Auditors and Client Restatements. Contemporary Accounting Research, 30(4), 1626-1661.

Frankel, R., M. Johnson, and K. Nelson. 2002. The relation between auditors’ fees for nonaudit services and earnings quality. The Accounting Review 77 Supplement: 71– 105.

Freeman, R. E. 1984. Strategic management: A stakeholder approach. Boston: Pitman. Jensen, M. C., & Meckling, W. H. (1976). Theory of the firm: Managerial behavior, agency

costs and ownership structure. Journal of financial economics, 3(4), 305-360. Johnson, V, I. Khurana, and J. K. Reynolds. 2002. Audit firm tenure and the quality of

accounting earnings. Contemporary Accounting Research 19: 637-660

Jones, J.J. 1991. Earnings Management During Import Relief Investigations. Journal of Accounting Research, 29, (2), 193-228.

(28)

27

Joos, P., & Lang, M. 1994. The effects of accounting diversity: Evidence from the European Union. Journal of Accounting Research, 141-168.

Palmrose, Z., 2000. Empirical research in auditor litigation: considerations and data, studies in accounting research No. 33. American Accounting Association, Sarasota, FL.

Reynolds, J. K., and J. R. Francis. 2000. Does size matter? The influence of large clients on office-level auditor reporting decisions. Journal of Accounting and Economics 30 (3): 375–400.

Simunic, D., Stein, M., 1987. Product Differentiation in Auditing: Auditor Choice in the Market for Unseasoned New Issues (The Canadian Certified General Auditors Research Foundation)

Solomon, I., Shields, M., Whittington, O.R., 1999. What do industry-specialist auditors now? Journal of Accounting Research (Spring), 191–208.

Vera-Munoz, S., J. Ho, and C. Chow. 2006. Enhancing knowledge sharing in public accounting firms. Accounting Horizons 20: 133-155.

Wallman, S. 1996. The future of accounting, Part III: Reliability and auditor independence. Accounting Horizons 10: 76-97

(29)

28

Appendix

TABLE 3A

Results of regression of office size on audit quality |DA2| and OFSIZE1 and OFSIZE2

Predicted

OFSIZE - ,001 ,145 0.0009 ,790 LNAssets - -,007 ,000*** -,007 ,000*** Change_Rev + ,000 ,883 0.0006 ,937 B_TO_M - -,004 ,046** -,004 ,090* Loss_reporting + ,027 ,000*** ,026 ,000*** Big 3 - ,003 ,486 -,002 ,577 TENURE + ,009 ,024** ,008 ,054* LEV + -,006 ,325 -,004 ,510 Intercept ,131 ,000*** ,143 ,000***

N 3981 3981

R2 _0.038 _0.038

(30)

29 TABLE 4A

Predicted

OFSIZE - ,002 ,062* ,000 ,270 LNAssets - -,007 ,000*** -,007 ,000*** Change_Rev + -,001 ,859 -,001 ,877 Loss_reporting + ,026 ,000*** ,027 ,000*** B_TO_M - -,005 ,045** -,005 ,050** TENURE + ,002 ,612 ,002 ,616 LEV + ,001 ,886 ,002 ,771 Intercept ,131 ,000*** ,140 ,000***

Year dummies Included Included Included

N 2487 2487

R2 _0.046 _0.045

TABLE 5A

Predicted

OFSIZE - ,002 ,272 -0.000 ,953 LNAssets - -,005 ,000*** -,005 ,000*** Change_Rev + -,000 ,918 -,000 ,947 Loss_reporting + ,001 ,548 ,001 ,564 B_TO_M - ,027 ,000*** ,027 ,000*** TENURE + -,004 ,686 -,002 ,781 LEV + ,008 ,227 ,007 ,234 Intercept ,102 ,000*** ,112 ,000***

N 1494 1494

R2 _0.033 _0.032

(31)