Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

ABNORMAL AUDIT FEES AND THE IMPROVEMENT OF UNFAVOURABLE AUDIT OUTCOMES 1 Charles Jie Ping Chen 2 , Xijia Su 3 , and Xi Wu 4 ABSTRACT When an audit opinion for the period t is unfavourable to the management of a listed company, the management often have a strong motivation to avoid it in the following period (t+1). This paper examines the association between the change in abnormal audit fees (AAF) and the improvement of unfavourable audit outcomes (UAO) from the period t to t+1, using a sample of Chinese listed companies with modified audit opinions during 2000-2002. We find that after controlling for the company’s fundamental financial characteristics and their change from period t to t+1, an increase in AAF without changing the auditor is significantly and positively associated with the improvement of UAO, while an increase in AAF concurrent with a change in the auditor is not significantly associated with improved UAO. The result suggests that economic interests provided by the management probably compromise auditor independence. Even for those companies with high risks, the management have the capacity to influence the audit outcomes to a great extent. Keywords: Unfavourable audit opinions, Abnormal audit fees, Auditor independence

I. INTRODUCTION Independent auditing is of great significance to the public as a mechanism to improve the credibility of financial information publicly disclosed by listed companies. However, the public’s confidence in the public accounting profession has been impaired by a series of high-profile financial scandals and audit failures since the beginning of the decade. Regulators are therefore promulgating rules and taking measures to reinforce the regulation of auditors’ behaviours. Recent academic research has focused on the potential impacts of various factors including economic interests on auditors’ independence. Extant literature has examined the impact of economic incentives on auditor independence 1

The authors thank Shimin Chen (the editor) and two anonymous reviewers for helpful comments and suggestions. Errors or omissions, however, remain our responsibility. Xi Wu claims that the opinions expressed in this paper are his personal views and do not represent any view or opinion of the organisation he is working for. 2 PhD, associate professor, the City University of Hong Kong, Kowloon, Hong Kong. 3 PhD, associate professor, the City University of Hong Kong, Kowloon, Hong Kong. 4 PhD, Professional Standards Department, the Chinese Institute of Certified Public Accountants, Beijing. 1

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

from different perspectives. Studies on how auditors’ reliance upon audit fees, or on auditors’ provision of non-audit services affect auditors’ independence, have produced inconsistent evidence. Using audit opinions as the dependent variable, Craswell et al. (2002) tested whether auditors’ propensity to issue unqualified audit opinions is associated with auditors’ reliance upon audit fees from a client, 5 and did not find a significant association. DeFond et al. (2002) found no significant association between non-audit service fees and impaired auditor independence, where auditor independence is proxied by auditors’ propensity to issue going concern audit opinions. They also found no association between going concern opinions and audit fees. Using earnings management as the dependent variable, Frankel et al. (2002) investigated the association between auditor fees and small earnings surprise (and discretionary accruals as well). They report a positive association between non-audit fees and measures of earnings management, and a negative association between audit fees and measures of earnings management. They conclude that the more the audit client purchases non-audit services from the auditor, the more auditor independence is impaired. Kinney and Libby (2002) point out in a discussion paper the weaknesses of Frankel et al. (2002) in research design and variable measurement and conclude that the results reported in Frankel et al. (2002) are difficult to interpret. Ashbaugh et al. (2003) further challenged the findings of Frankel et al. (2002) by re-performing a series of tests, and found that the results in Frankel et al. (2002) are sensitive to research design choices. The results from Chung and Kallapur (2003) also support the conclusion that there is no significant association between the variable of audit client importance and discretionary accounting accruals. Unlike prior studies, this paper focuses on the potential impact of audit fees, rather than non-audit fees, on auditor independence. In China’s auditing market, CPA firms are most likely to provide only audit services, with non-audit services accounting for a rather small proportion of total fees in a firm. As a result, it is less likely that studies on audit fees using Chinese observations will suffer from the joint determination effect that may possibly exist between audit fees and non-audit fees. 6 By the end of the 20th century, Chinese accounting firms had just undergone the delinking process. There is some gap between auditor independence in developing countries and that in developed countries. To exacerbate the situation, there is a general lack of demand for high-quality audit service and fierce competition in China’s domestic accounting market. All the above facts indicate that although audit fees paid by the management may not be very high in terms of absolute magnitude compared with those in the Western markets, they are 5

Craswell et al. (2002, 259) set the variable of fee dependence as the proportion of audit fees that a client contributes to the total audit and non-audit fees earned by the auditor. Audit opinions are classified into two types: qualified (coded as 1 and not including unqualified audit opinion with explanatory notes) and unqualified (coded as 0). 6 Whisenant et al. (2003) provide evidence that audit fees and non-audit fees are jointly determined. 2

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

significant enough to have a considerable effect on auditor independence in China. It is thus quite logical to assume that if the management provide some extra fees in addition to the normal level of audit fees, the impairment to auditor independence could be substantial. This paper attempts to investigate the association between the improvement of unfavourable audit outcomes (UAO) and the change in abnormal audit fees (AAF) based on China’s securities auditing market. 7 Unlike prior studies, we do not use audit opinions or discretionary accruals in a single period as the proxy for auditor independence. Instead, we analyse changes of audit outcomes during two consecutive periods. Those listed companies who had received unfavourable audit opinions in period t are firstly identified and then examined for the changes in period t+1 with respect to both the audit opinion and the reported earnings. Whether an improvement of both the audit opinion and the reported earnings occurs is our proxy for auditor independence (the dependent variable). Changes from period t to t+1 in AAF provided by those identified companies (the experimental variable) are estimated. 8 Then we examine the association between a change in AAF and the improvement of UAO. Our results show that an increase in AAF while retaining the incumbent auditor is significantly associated with an increase in the probability of the improvement of UAO, which suggests that the management of Chinese listed companies are able to avoid receiving UAO again in the second year by providing extra economic incentives to their incumbent auditors. The remainder of the paper is organised as follows: the next section introduces the institutional background and develops the research question; the third section presents the research design and methodology; the fourth section provides a description of the sample; the fifth section illustrates and analyses the empirical results; conclusions and limitations are presented in the last section.

II. BACKGROUND AND RESEARCH QUESTION 2.1 UAO and Listed Companies’ Incentives to Avoid UAO An audit outcome can be observed from either an audit opinion or audited financial numbers. If significant accounting and financial reporting issues occur in a listed company’s annual report,

7

Prior research in China’s auditing market focused on the association between auditor changes and the improvement of UAO (e.g. an internal research report in 2002 by the China Securities Regulatory Commission (CSRC); Li and Wu (2004b)). The potential contribution of this paper lies in an extension of the focus on the possible impact of AAF besides auditor changes on the improvement of UAO based on the previous framework of analysis. 8 We notice that Frankel et al. (2002) and DeFond et al. (2002) have used the idea of and approach to estimating the unexpected component from auditor fees. 3

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

two types of audit results may appear. One is that the management refuse to make adjustments suggested by the auditor and thus receive a modified opinion. The other is that the management agree to make adjustments suggested by the auditor and receive an unqualified opinion. However, as a result of the adjustment, the audited financial performance (most likely, net income) no longer meets the original intentions of the management. Both of the above two types of audit outcome are referred to as UAO in this study. The management of a listed company always have a strong incentive to avoid UAO either in the form of an unfavourable audit opinion or an audited accounting number that does not meet the intention of the management, since UAO would probably have a negative impact on a company’s financing costs, overall reputation, and ability to survive. The management, therefore, have a strong motivation to make all efforts to achieve both a favourable audit opinion and an audited accounting number that meets the expectation of the management. Users of the financial information usually expect the improvement of UAO to be supported by a real correction or improvement in financial reporting, either in the form of an adjustment to address the auditor’s concerns or a substantial improvement in the company’s financial conditions and performance. If the improvement of UAO lacks any economic substance and is only the result of the auditor’s compromise towards the management, such improvement will be very misleading. A study of the association between the improvement of UAO and potential impediments to auditor independence is thus warranted as it has important academic and policy implications. 9

2.2 A Possible Approach to Avoiding UAO: “Bribing” the Auditor Of the many approaches that the management can take to intervene in auditor independence and thus avoid UAO, providing extra economic incentives is especially cost effective for the following reasons. First of all, “bribing” the auditor involves a relatively low level of risk of being disciplined by the regulators. This is because, first, the regulators have given an increased attention to the type of audit opinion the company received in major regulatory decisions. Although the regulators may challenge the improvement of UAO, it is not likely that they will launch a serious investigation into the improvement due to the limited availability of expertise. 10 9

To the best of our knowledge, for example, in recent years, the CSRC has specifically analysed in their internal research report the phenomenon of the improvement of UAO and potential influencing factors. 10 As the CSRC (2002, 165-172) states, “since recent years, the CSRC has reinforced the regulation on unclean audit opinions. This year, top leaders of the CSRC have decided to implement a comprehensive and profound review on unclean audit opinions … at the early September 2001, a special task force on reviewing unclean audit opinions has been organized including professionals from the Accounting Department, IPO Department, and some regional branches of the CSRC, Shanghai Stock Exchange, and Shenzhen Securities Exchange, with an aim of investigating on listed companies who received unclean audit opinions on financial statements as for the year 2000…. The 4

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

Second, given the large number of regulatees, it is not practical for the regulator to identify and investigate all potential “briberies”. Third, in practice, regulators have to face the difficulty of systematically and efficiently defining, measuring, and identifying abnormal economic incentives provided by the management to the auditor. 11 As a result, regulators may have to be cautious in launching disciplinary investigations of CPA firms and companies. Secondly, the cost for a listed company to “bribe” the auditor is relatively low. According to descriptive statistics by Xia and Lin (2003, 43) and Li and Wu (2004a, 33), the fee for an annual audit paid by Chinese listed companies only averages RMB 3-400,000. Even if the management pay twice the price, such a cost remains negligible for a listed company, while the expected benefit for an improved audit outcome is significant. Thirdly, the current Chinese securities market is characterised by fierce competition and a lack of demand for high-quality audits (DeFond et al., 2000), which also contributes to an increased probability for the management to successfully “bribe” the auditor. Based on this understanding of the institutional background, we develop the following research question: For those listed companies who have strong incentives to avoid UAO, is the improvement of UAO significantly and positively associated with an increase in AAF?

III. RESEARCH DESIGN 3.1 Research Setting task force has interviewed 38 listed companies and 11 accounting firms with typical issues based on the analyses and discussions…. The CSRC has drafted The Notice on Unclean Audit Opinions and Related Issues for Listed Companies on the basis of issues identified in the specific review, with a plan to carry out more rigorous regulation on unclean audit opinions since 2001 annual audit.” As an update of regulatory practices mentioned in Li and Wu (2005, 14): “After 2002 annual audit, the CSRC has interviewed 32 listed companies and their auditors among the total 166 unclean audit opinions in 2002, and has issued inquiry letters to 19 listed companies and their auditors among the total 225 corrections of material accounting errors. During the disclosure of 2002 annual reports, Shanghai Stock Exchange and Shenzhen Securities Exchange suspended stock trading for 2 listed companies because of the nature of unclean audit opinions they received, and required the companies make adjustments to remove audit qualifications.” 11

Generally speaking, it is difficult and obscure to judge real reasons for a listed company to increase the payment of audit fees, especially to judge whether abnormal audit fees have been paid. Such difficulties and obscurities include the following: 1) Paying audit fees to the auditor by the management has been an established arrangement for a long time, which makes audit pricing behaviour be considered more as a normality than other market events such as auditor changes. 2) Information on payment of auditor fees has been undisclosed for a long time and has just been publicly disclosed for a very short time, which means the regulatory practice and public monitoring of audit pricing is less sensitive and experienced than other significant events such as auditor changes. 3) The determination of both the absolute and relative level of audit fees is complex and mixed with either an inherent demand for changes in audit workload and audit risks, or an incentive to “bribe” the auditor by the management for a more favourable audit result; thus one can hardly identify the decisive factor(s) or take responsive measures. 5

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

We need to control the research setting to make sure that the sample companies have received UAO in period t and thus have a strong incentive to avoid UAO in period t+1. In two types of UAO, unfavourable audit opinions are chosen as the starting point for the sample selection. 12 Companies that received unfavourable audit opinions usually have strong incentives to avoid them for the following reasons: 1) Once a listed company receives an unfavourable audit opinion, a substantial constraint will be imposed upon it, and it could be banned from refinancing or trading of its shares could be suspended. In addition, such companies will be subject to closer monitoring from the regulators (CSRC, 2001a, 2001b, 2002, 2003). 2) They may also suffer from negative publicity as criticism is often published in popular financial media. As a result, the public will pay more attention to these companies. 3) Prior literature (e.g. Chen, Su, and Zhao, 2000) documents a significantly negative market reaction towards unfavourable audit opinions. It is worth noting that we exclude unqualified audit opinions with explanatory notes from our sample, which means only three types of unclean audit opinions are considered (i.e. qualified opinions, adverse opinions, and disclaimers). Given its unqualified nature, the type of unqualified audit opinions with explanatory notes is often a compromised result by negotiation between the management and the auditor (Sun and Wang, 2000; Li and Wu, 2002b). Thus, unqualified audit opinions with explanatory notes are not deemed as UAO in the strict sense, and the management usually have a weaker incentive to avoid such type of opinion than non-unqualified audit opinions. By excluding unqualified audit opinions with explanatory notes, we can more explicitly and unambiguously attribute the improvements in audit outcomes to managers’ incentive to avoid UAO.

3.2 General Approach For those listed companies that received non-unqualified audit opinions in period t, we need to observe their audit results in period t+1, including whether the improvement of audit opinion occurs and the characteristics of the audited earnings number. The improvement of audit opinion is defined as a change in audit opinion type from a stricter one to a less severe one. The downward ranking of different types of audit opinions in terms of their severity is as follows: Adverse opinions / Disclaimers 13 →Qualifications→Unqualified 12

Given the difficulty of uniformly and clearly identifying and measuring the other type of UAO (i.e. the audited accounting number divergent from management intentions) and its improvement, we do not use it as the basis to collect the sample. In contrast, the criterion and measurement is more definitive for unfavourable audit opinions and their improvement. 13 Adverse opinions and disclaimers are theoretically not comparable in the extent of severity. But if an adverse 6

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

opinions. We also consider whether the audited earnings number has been significantly improved, where the earnings number is measured as the reported net income, which is most widely used as the performance measure for listed companies in China. The improvement of the audited earnings number is defined when a company reports, for example, a positive net income for period t+1 and a negative net income for period t, or a positive net income for period t+1 and a negative net income after deduction of non-recurring items for period t+1. For a detailed definition and description, see Section IV. Only when both the audit opinion and the earnings number are improved, is a company regarded as having an improvement of UAO, 14 which is then regressed on the change in AAF and other control variables.

3.3 Estimation of AAF According to Simunic’s (1980) audit pricing model, determinants of audit fees include the size of the audit client, audit complexity, client risks, and auditor characteristics. A simplified linear regression model of audit pricing is as follows: LnFEE= a0+∑aiXi+μ ,

(1)

where FEE is audit fees; X and i are specific explanatory variables and the number of such variables, respectively; and μ is the residual term, which can be interpreted as the abnormal component of audit fees that main economic factors cannot explain (i.e. AAF). A positive (negative) μ means an abnormally high (low) level of audit fees. To examine the impact of the change in AAF on the improvement of UAO, a variable is constructed as △ABNFEE (=μt+1-μt) to measure the change in AAF. We use audit fees and related financial and non-financial data for period t and t+1, respectively, to estimate μt and μ t+1 and thus to compute △ABNFEE. For details of the design and regression results of the audit fees

opinion occurs in one period while a disclaimer of opinion occurs in the other period, we will regard audit opinions as not improved. However, there is no case with adverse opinion in the sample. 14 According to the internal research report by the CSRC, the regulator believes that an audit opinion and business performance is interrelated. Since a risk-aversive accounting firm usually trades off between cleaner audit opinion and nice-looking reported net income, the improvement only in the audit opinion side is not deemed by the regulator as a success in “bribing” the auditor. For example, the management will not regard the situation as a satisfactory audit result from a loss with qualified audit opinion in 2001 to a loss with unqualified opinion in 2002, though such a situation may be theoretically deemed as an improvement of UAO. Thus, we use as the dependent variable the improvement of both audit opinion and the audited earnings number. In sensitivity tests not presented, the improvement either only in audit opinion or only in earnings number is defined as the dependent variable, and we find an insignificant or less significant correlation between our experimental variable and the dependent variable. 7

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

determination model, see the appendix.

3.4 Model Specification Given the binary and qualitative nature of the dependent variable, a logistic regression model is established as follows (presented as a linear form by logit transformation 15 ): logit p(OPBETTER*EMI=1)=a0+a1△ROA+a2△DEBT+a3GROWTH+a4PRELOSSES +a5PRELOSS+a6AUDCHG+a7SIZE+a8Y2002+a9Y2003 +a10△ABNFEE+a11△ABNFEE*AUDCHG+ε

(2)

3.4.1 Dependent Variable The dependent variable is defined as whether the audit opinion and the earnings number are improved simultaneously, multiplying two subdivided variables (i.e. OPBETTER and EMI). OPBETTER is the improvement of unfavourable audit opinion (= 1 when the unfavourable audit opinion received in period t is improved in period t+1; = 0 otherwise). EMI is the improvement of the earnings number (= 1 when an indication of earnings management in period t+1 is evident or otherwise a loss or consecutive losses would have taken place in period t+1; = 0 otherwise). When the dependent variable (OPBETTER*EMI) equals 1, the management receive a maximum of benefits by achieving a simultaneous improvement of a potentially unfavourable audit opinion and the earnings number.

3.4.2 Experimental Variables Our key experimental variable is the change in AAF for an annual audit, measured as △ABNFEE (=μt+1 － μt) (see Section 3.3). The other experimental variable is the interaction between △ABNFEE and AUDCHG (a dummy variable measuring the occurrence of auditor changes). Such an interaction variable is used to differentiate the impact of △ABNFEE on auditor independence between settings with and without a change in the auditor.

3.4.3 Control Variables Control variables include changes in financial conditions and performance during two consecutive periods, the consecutive losses condition, auditor change, company size, and fiscal year dummy variables. Since an improvement in financial conditions and performance from period t to t+1 may 15

Logit transformation (logit p=ln[p/(1-p)]) aims at transforming a logistic function into a linear form, where p=p(Y=1) indicating the probability that event Y will take place. 8

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

possibly contribute to the improvement of UAO, we introduce △ROA, △DEBT, and GROWTH to control for the effect of changes in corporate financial conditions and performance, where △ROA = [(operating incomet+1 – side business incomet+1) / total assetst+1] – [(operating incomet – side business incomet) / total assetst], 16 △DEBT = (total liabilitiest+1 / total assetst+1) – (total liabilitiest / total assetst), and GROWTH = (total assetst+1 – total assetst) / total assetst. Two dummy variables, PRELOSSES (1 for an observation reporting losses in both periods t and t-1, and 0 otherwise) and PRELOSS (1 for reporting a loss in period t only, and 0 otherwise), are used to capture potential pressure on and incentive for the management to avoid earlier UAO and especially to report a profit in period t+1 (otherwise the listed company would be subject to a suspension or limitation in stock trading, and suffer from a damage to its reputation as well). Previous literature has expressed great concern over whether the management can achieve a more favourable audit outcome by changing the auditor (Chow and Rice, 1982; Smith, 1986; Lennox, 2000), and the results of a few studies based on the Chinese market indicate that auditor changes are more likely to be associated with an improvement of UAO (Li and Wu, 2002a; CSRC, 2003; and an internal research report by the CSRC). Since an auditor change affects the auditor independence in the same direction as AAF does, it is necessary to control for the effect of such a variable. A dummy variable AUDCHG is thus introduced (1 for a change in the auditor taking place in period t+1, and 0 for retaining the incumbent auditor in period t+1). In addition, we control for the company size by setting SIZE as the natural log of total assets in period t+1, and set dummy variables Y2002 and Y2003 to control for potential difference in the economic and regulatory environment in various fiscal years.

IV. SAMPLE 4.1 Sample Selection As annual audit fees data are required for the study, we collect our sample starting from the fiscal year 2000, when the information on audit fees was publicly available for the first time. Meanwhile, we need to limit the sample to those who received non-unqualified audit opinions. Based on audit opinion statistics of the CSRC (2001, 2003) with necessary checks, we start our sample selection from 193 non-unqualified audit opinions issued during 2000-2002, further checking for the

16

Although inevitably subject to the management bias and manipulation in such items as main business revenues and costs, and general and administration expenses, the current setting of △ROA is free from earnings management through non-recurring earnings components including investment gains, extraordinary gains, subsidy gains, and minor business income, thus reflecting to a large extent the core and persistent profitability of a listed company. 9

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

availability of annual audit fees data for these companies during 2000-2003, and subsequent audit results correspondingly during 2001-2003. Li and Wu (2004a, Chapter 1) analysed problems found in the audit fees data disclosed by Chinese companies since 2001. They expressed concern over the reliability and comparability of the annual audit fees data, and introduced some procedures to safeguard data reliability and comparability (Li and Wu, 2004a, 27-28). We use their procedures to analyse annual audit fees data of 193 observations during two consecutive years, and finally select 97 observations to form the sample. 17 Table 1 presents the result of sample selection by audit opinion type and by fiscal year. Table 1 Composition of the Sample by Audit Opinion Type and by Fiscal Year Non-unqualified audit opinions

Qualified

Disclaimers

Adverse

Total

Population (per the year 2000)#

58

14

1

73

Number of the sample

10

2

0

12

Population (per the year 2001)##

44

20

0

64

Number of the sample

32

8

0

40

Population (per the year 2002)##

40

16

0

56

Number of the sample

34

11

0

45

142

50

1

193

76

21

0

97

Panel A: 2000 versus 2001

Panel B: 2001 versus 2002

Panel C: 2002 versus 2003

Panel D: Total during 2000-2002 Population (per the year t)### Number of the sample

# Population of non-unqualified audit opinions is based on statistics of the CSRC (2001) with necessary checks. ## Population of non-unqualified audit opinions is based on statistics of the CSRC (2003) with necessary checks. ### t is the year a listed company received a non-unqualified audit opinion, which is used as the starting point of our sample selection process.

4.2 Descriptive Statistics of Sample 4.2.1 Dependent Variable (1) The improvement of unfavourable audit opinion (OPBETTER) Table 2 depicts the changes of unfavourable audit opinions of the sample. Panel A presents a 17

In detail, we exclude observations: 1) lacking any disclosure of audit fees data for any fiscal year; 2) with difficulty in identifying the period audit fees are paid for; 3) with difficulty in separating annual audit fees from the company’s disclosure of mixed fees; 4) who are financial institutions; or 5) with mandatory rather than voluntary auditor changes (mostly in the year 2001). For observations in 2000 and 2001, we directly use the annual audit fees database of Li and Wu (2004a, Chapter 2) that has been identified through a strict filtering process. For observations in 2002, we analyse by reading annual reports the audit fees data case by case from 56 qualifications and disclaimers. Those observations that are excluded may meet one or more of the above-mentioned criteria. 10

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

movement matrix of non-unqualified audit opinions during two periods; panel B classifies the sample into two coding values of OPBETTER. According to panel B, 59 out of 97 observations (60.8%) received improved audit opinions. Table 2 Changes in Unfavourable Audit Opinions from Period t to t+1 of the Sample Panel A: Movement matrix of audit

Audit opinions in period t+1

opinions Unqualified opinions

Qualifications

Disclaimers

Total

Disclaimers

9

4

8

21

Qualifications

46

25

5

76

Total

55

29

13

97

Improved

Unimproved

unfavourable audit opinions

(OPBETTER=1)

(OPBETTER=0)

All sample companies

59

38

97

Disclaimers

13

8

21

Qualifications

46

30

76

Audit opinions in period t Panel

B:

Improvement

of

Total

Note: OPBETTER is a dummy variable, = 1 when an unfavourable audit opinion in period t has been improved in period t+1; = 0 when an unfavourable audit opinion in period t has not been improved in period t+1. The improvement of audit opinion is defined as a change in audit opinion type from a stricter one to a moderate one. The downward ranking of various types of audit opinions per their extent of severity is as follows: Disclaimers→Qualifications→Unqualified opinions.

(2) The improvement of earnings number (EMI) As noted before, EMI is a dummy variable reflecting the improvement of the earnings number. EMI is coded 1 when an indication of earnings management in period t+1 is evident or otherwise a loss or consecutive losses would have taken place in period t+1. A detailed analysis and criteria to judge whether EMI for an observation is coded 1 are illustrated in this section. Panel A of Table 3 presents the movement matrix of the earnings number during two consecutive periods as our starting point for analysis, and panel B presents the EMI coding criteria and results for observations in each cell of the matrix. As indicated in panel B, for 44 observations with a reported loss in period t+1, regardless of whether they report positive or negative net income in period t, the audited earnings number is deemed as unimproved and thus EMI is coded 0. For 33 observations with a reported profit in period t+1 compared with a loss in period t, we do not set EMI as 1 for granted. Rather, a sequential set of criteria is established as follows: (a) Has an observation suffered consecutive losses on and before period t (i.e. loss in both period t and t-1)? If yes, EMI is coded 1 (19 observations meet criterion (a)); if not, criterion (b) should be 11

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

considered. (b) Has an observation publicly reported a negative net income after deducting non-recurring earnings components in period t+1? If yes, EMI is coded 1 (3 observations meet criterion (b)); if not, criterion (c) is then resorted to. (c) Will the net income after deducting non-recurring earnings components turn to negative if the necessary adjustment is made? 18 If yes, EMI is coded 1 (6 observations meet criterion (c)); if not, EMI is coded 0. To sum up, 28 out of 33 observations have their EMI coded 1 and 5 observations have their EMI coded 0. Table 3 Changes in the Audited Earnings Number from Period t to t+1 of the Sample Panel A: Movement matrix of earnings number

Net income in period t

Net income in period t+1 Profit

Loss

Total

Profit

20

9

29

Loss

33

35

68

Total

53

44

97

Benchmark setting

Alternative setting

of EMI

of EMI

Panel B: EMI coding criteria and results for observations in each cell of the matrix Cells of the matrix (Changes in net income from period t to t+1)

EMI=1

EMI=0

EMI=1

EMI=0

Profit→Loss

9

9

Loss→Loss

35

35

Loss→Profit (a) Consecutive losses both in period t and t-1

19

19

(b) A reported negative net income after deducting non-recurring

3

3

6

6

earnings components in period t+1 (c) An adjusted# negative net income after deducting non-recurring earnings components in period t+1 Observations other than those with the feature of (a), (b), or (c)

5

5

Profit→Profit (d) A reported negative net income after deducting non-recurring

2

2

earnings components in period t+1 (e) An adjusted# negative net income after deducting non-recurring

6

6

earnings components in period t+1 Observations other than those with the feature of (d) or (e) Total

12 30

67

12 36

61

Note: EMI is a dummy variable, = 1 when an indication of earnings management in period t+1 is evident or otherwise a loss or consecutive losses would have taken place in period t+1; = 0 otherwise. # We adjust the reported net income after deducting non-recurring earnings components by further excluding those non-recurring items such as minor business income, investment gains, and retrospective corrections to the previous earnings number, which have not been computed by the management as non-recurring components to deduct from net income.

18

We adjust the reported net income after deducting non-recurring earnings components by further excluding those non-recurring items such as minor business income, investment gains, and retrospective corrections to the previous earnings number, which have not been computed by the management as non-recurring components to deduct from net income. 12

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

For 20 observations with reported profits in both periods t+1 and t, we do not set EMI as 0 for granted. Rather, we are concerned with whether there is a clear indication for such observations to avoid reporting a loss in period t+1. Similar criteria are set as follows: (d) Has an observation publicly reported a negative net income after deducting non-recurring earnings components in period t+1? If yes, EMI is coded 1 (2 observations meet criterion (d)); if not, criterion (e) is then resorted to. (e) Will the net income after deducting non-recurring earnings components turn to negative if the necessary adjustment is made? 19 If yes, EMI is coded 1 (6 observations meet criterion (e)); if not, EMI is coded 0. The criteria for coding EMI for 20 observations with reported profits both in period t+1 and t are slightly different, because they generally have weaker incentives to manage earnings than 33 observations with a reported profit in period t+1 but with a loss in period t. The benchmark setting is only to code EMI 1 for two observations meeting criterion (d), while the alternative setting is to code EMI 1 for eight observations meeting either criterion (d) or (e). In the benchmark setting, there are a total of 30 out of 97 observations with EMI coded 1 and 67 observations with EMI coded 0. In the alternative setting, there are a total of 36 out of 97 observations with EMI coded 1 and 61 observations with EMI coded 0. (3) Subdivided Items Combined: OPBETTER*EMI Table 4 describes the interactive characteristics between OPBETTER and EMI. Table 4 Interactions between OPBETTER and EMI Panel A: Benchmark setting of EMI

EMI=1

EMI=0

Total

OPBETTER=1

27

32

59

OPBETTER=0

3

35

38

Total

30

67

97

Pearson Chi-square between OPBETTER and EMI=15.515, p-value=0.000 Panel B: Alternative setting of EMI

EMI=1

EMI=0

Total

OPBETTER=1

30

29

59

OPBETTER=0

6

32

38

Total

36

61

97

Pearson Chi-square between OPBETTER and EMI=12.172, p-value=0.000 Note: OPBETTER is a dummy variable, = 1 when an unfavourable audit opinion in period t has been improved in period t+1; = 0 when an unfavourable audit opinion in period t has not been improved in period t+1. The improvement of audit opinion is defined as a change in audit opinion type from a stricter one to a moderate one. The downward ranking of various types of audit opinions per their extent of severity is as follows: Disclaimers→Qualifications→Unqualified opinions. EMI is a dummy variable, = 1 when an indication of earnings management in period t+1 is evident or otherwise a 19

See footnote 18. 13

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54. loss or consecutive losses would have taken place in period t+1; = 0 otherwise. For details of benchmark and alternative settings of EMI, see Table 3.

As panels A and B indicate, no matter how EMI is set, OPBETTER and EMI are positively correlated (significant at the .01 level). For the benchmark (alternative) setting of EMI, there are 27 (30) out of 97 observations with both OPBETTER=1 and EMI=1, and only 3 (6) out of 30 (36) observations with EMI=1 that fail to achieve the improvement of unfavourable audit opinions in period t+1. In general, the interaction table indicates a strong motive by observations with EMI=1 to avoid unfavourable audit opinions.

4.2.2 Experimental and Control Variables Table 5 presents descriptive statistics for experimental and control variables. Panel A reports the mean level of △ROA, △DEBT, and GROWTH as -0.023, 0.197, and -0.097, respectively, indicating a slight decrease in financial performance and conditions for the sample from period t to t+1. Thirty-four (33) per cent of sample companies have suffered losses (a loss) for two consecutive years by period t. Auditor changes have taken place in 24 out of 97 observations. The mean (median) of △ABNFEE is 0.000 (0.024). (See the Appendix for detailed descriptive statistics and implications of △ABNFEE.) The mean of △ABNFEE with AUDCHG=1 is -0.021, slightly lower than that with AUDCHG=0. Partitioned by OPBETTER*EMI, as indicated in panels B and C, the mean of △ABNFEE with OPBETTER*EMI=1 is lower (but not significant) than that with OPBETTER*EMI=0. A multivariate analysis is thus warranted to examine the potential relationship between △ABNFEE and the dependent variable.

Table 5 Descriptive Statistics for Experimental and Control Variables Mean

Std. Dev.

Minimum

Median

Maximum

△ROA

-0.023

0.869

-5.199

0.001

4.189

△DEBT

0.197

1.065

-3.620

0.032

5.301

GROWTH

-0.097

0.328

-0.866

-0.074

1.055

SIZE

11.299

1.126

7.674

11.427

13.757

△ABNFEE

0.000

0.276

-0.924

0.024

0.696

-0.924

0.000

0.655

Panel A-1: Continuous Variable

-0.021

0.200

Number

Proportion

PRELOSSES (=1)

33

34.0%

PRELOSS (=1)

32

33.0%

AUDCHG (=1)

24

25.0%

Y2002 (=1)

40

41.0%

Y2003 (=1)

45

46.0%

△ABNFEE*AUDCHG Panel A-2: Categorical Variable

14

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54. Panel B: Benchmark setting of EMI

OPBETTER*EMI=1

OPBETTER*EMI=0

Independent-sample

(n=27)

(n=70)

T test

Mean

Std. Dev.

Mean

Std. Dev.

t-statistic

p-value

△ABNFEE

0.025

0.305

-0.010

0.266

0.553

0.581

△ABNFEE*AUDCHG

-0.032

0.209

-0.016

0.197

-0.353

0.725

Panel C: Alternative setting of EMI

OPBETTER*EMI=1

OPBETTER*EMI=0

Independent-sample

(n=30)

(n=67)

T test

Mean

Std. Dev.

Mean

Std. Dev.

t-statistic

p-value

△ABNFEE

0.012

0.310

-0.005

0.262

0.289

0.773

△ABNFEE*AUDCHG

-0.048

0.221

-0.009

0.190

-0.895

0.373

Definition of variables: OPBETTER*EMI is the interaction of OPBETTER and EMI. OPBETTER

A dummy variable, = 1 when an unfavourable audit opinion in period t has been improved in period t+1; = 0 when an unfavourable audit opinion in period t has not been improved in period t+1. The improvement of audit opinion is defined as a change in audit opinion type from a stricter one to a moderate one. The downward ranking of various types of audit opinions per their extent of severity is as follows: Disclaimers→Qualifications→Unqualified opinions.

EMI

A dummy variable, = 1 when an indication of earnings management in period t+1 is evident or otherwise a loss or consecutive losses would have taken place in period t+1; = 0 otherwise. For details of benchmark and alternative settings of EMI, see Table 3.

△ROA

A change in main business returns, = [(operating incomet+1 – minor business incomet+1) / total assetst+1] – [(operating incomet – minor business incomet) / total assetst].

△DEBT

A change in leverage, = (total liabilitiest+1 / total assetst+1) – (total liabilitiest / total assetst).

GROWTH

Ratio of an increase in assets, = (total assetst+1 – total assetst) / total assetst.

PRELOSSES

A dummy variable, = 1 for an observation with consecutive losses both in period t and t-1; = 0 otherwise.

PRELOSS AUDCHG

A dummy variable, = 1 for an observation with a loss only in period t; = 0 otherwise. A dummy variable, = 1 for a change in the auditor taking place in period t+1; = 0 for retaining the incumbent auditor in period t+1.

SIZE

Natural log of total assets in period t+1.

Y2002

A dummy variable, = 1 for an observation in the fiscal year 2002; = 0 otherwise.

Y2003

A dummy variable, = 1 for an observation in the fiscal year 2003; = 0 otherwise.

△ABNFEE

The change in abnormal audit fees, = μt+1-μt , where μ is the residual term of the audit pricing determinant model (see Equation (A1) of Table A1), and t is the year a listed company received a non-unqualified audit opinion, which is used as the starting point of our sample selection process.

△ABNFEE*AUDCHG is the interaction of △ABNFEE and AUDCHG.

As the correlations reported in Table 6 indicate, △ROA is negatively correlated with △DEBT, and positively correlated with GROWTH; both significantly. The correlation between △ABNFEE and △ABNFEE*AUDCHG is positive and significant at the 0.01 level, while the correlation between △ABNFEE and AUDCHG is negative and significant at the 0.1 level,

15

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

indicating that a change in auditor is less likely to take place when the AAF increases, but if a change in auditor takes place, it is more likely for the AAF to increase. There is a positive and significant (at the 0.1 level) correlation between AUDCHG and PRELOSSES, suggesting a change in auditor is more likely to take place in an observation with consecutive losses both in period t and t-1.

Table 6 Pearson Correlation Matrix for Control and Experimental Variables △ROA △ROA

1.000

△DEBT

-0.208** ***

0.436

GROWTH

0.167

PRELOSSES

-0.166

PRELOSS

-0.157

AUDCHG

0.009

SIZE

△DEBT GROWTH PRELOSSES PRELOSS AUDCHG SIZE

Y2002

Y2003 △ABNFEE

-0.605***

1.000

0.028

0.033

0.086 0.067 ***

-0.365

1.000

**

-0.227

*

-0.174

***

0.276

-0.540***

1.000

*

-0.071

1.000

***

-0.056

-0.052

1.000

***

**

0.219

0.054

0.001

0.180 -0.306

0.008

0.060

-0.133

Y2003

0.009

-0.001

0.030

0.226**

-0.120

-0.102

0.074 -0.779*** 1.000

△ABNFEE

0.141

-0.092

0.055

-0.048

0.058

-0.177*

0.056

0.000

0.000

1.000

0.083

*

0.046

-0.106

0.088

0.732***

0.160

AUDCHG

1.000

Y2002

△ABNFEE* AUDCHG

△ABNFEE*

-0.088

0.118

-0.264

-0.060

-0.184

1.000

1.000

***, **, and * indicate significance at the 0.01, 0.05, and 0.1 levels, respectively. Definition of variables: △ROA

A change in main business returns, = [(operating incomet+1 – minor business incomet+1) / total assetst+1] – [(operating incomet – minor business incomet) / total assetst].

△DEBT

A change in leverage, = (total liabilitiest+1 / total assetst+1) – (total liabilitiest / total assetst).

GROWTH

Ratio of an increase in assets, = (total assetst+1 – total assetst) / total assetst.

PRELOSSES

A dummy variable, = 1 for an observation with consecutive losses both in period t and t-1; = 0 otherwise.

PRELOSS

A dummy variable, = 1 for an observation with a loss only in period t; = 0 otherwise.

AUDCHG

A dummy variable, = 1 for a change in the auditor taking place in period t+1; = 0 for retaining the incumbent auditor in period t+1.

SIZE

Natural log of total assets in period t+1.

Y2002

A dummy variable, = 1 for an observation in the fiscal year 2002; = 0 otherwise.

Y2003

A dummy variable, = 1 for an observation in the fiscal year 2003; = 0 otherwise.

△ABNFEE

The change in abnormal audit fees, = μt+1-μt , where μ is the residual term of the audit pricing determinant model (see Equation (A1) of Table A1), and t is the year a listed company received a non-unqualified audit opinion, which is used as the starting point of our sample selection process.

△ABNFEE*AUDCHG is the interaction of △ABNFEE and AUDCHG.

V. MULTIVARIATE ANALYSES Results pertaining to our multivariate tests are shown in Columns A and B of Table 7, where the 16

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

benchmark and alternative setting of EMI apply, respectively. Model χ2 of both logistic regressions are statistically significant (p<0.01), and pseudo-R2 is 68.6 per cent and 59.2 per cent, respectively, suggesting that the model fits the data well. There is no serious multicollinearity for the model. 20 The results shown in Columns A and B are similar in general. Table 7

Results of Logistic Regression Model for Explaining the Improvement of UAR

Dependent variable:

Column A: Benchmark setting of EMI 2

Column B: Alternative setting of EMI

Coefficient

Wald χ

p-value

Coefficient

Wald χ2

p-value

△ABNFEE

3.819*

3.112

0.078

3.435*

3.187

0.074

△ABNFEE*AUDCHG

-5.106*

3.014

0.083

-6.153**

4.811

0.028

△ROA

3.396**

5.422

0.020

3.452**

5.859

0.015

△DEBT

-2.005**

4.050

0.044

-2.400**

5.910

0.015

3.153

2.508

0.113

1.922

1.513

0.219

PRELOSSES

5.229***

13.470

0.000

2.996***

9.956

0.002

PRELOSS

3.410***

6.662

0.010

1.366

2.482

0.115

AUDCHG

0.220

0.067

0.796

0.492

0.401

0.527

SIZE

1.293***

7.677

0.006

1.025**

6.319

0.012

Y2002

-2.117

2.437

0.119

-1.109

1.160

0.281

OPBETTER*EMI Experimental variables

Control variables

GROWTH

Y2003 Intercept Model χ2

-2.856**

4.312

0.038

-1.298

1.517

0.218

-17.367***

9.308

0.002

-13.477***

7.662

0.006

62.695***

52.903***

Pseudo-R2

68.6%

59.2%

Overall Classification Accuracy

92.8%

86.6%

97

97

Number of Observations

***, **, and * indicate significance at the 0.01, 0.05, and 0.1 levels, respectively. Definition of variables: OPBETTER*EMI is the interaction of OPBETTER and EMI. OPBETTER

A dummy variable, = 1 when an unfavourable audit opinion in period t has been improved in period t+1; = 0 when an unfavourable audit opinion in period t has not been improved in period t+1. The improvement of audit opinion is defined as a change in audit opinion type from a stricter one to a moderate one. The downward ranking of various types of audit opinions per their extent of severity is as follows: Disclaimers→Qualifications→Unqualified opinions.

EMI

A dummy variable, = 1 when an indication of earnings management in period t+1 is evident or otherwise a loss or consecutive losses would have taken place in period t+1; = 0 otherwise. For details of benchmark and alternative settings of EMI, see Table 3.

△ABNFEE

The change in abnormal audit fees, = μt+1-μt , where μ is the residual term of the audit pricing determinant model (see Equation (A1) of Table A1), and t is the year a listed company received a non-unqualified audit opinion, which is used as the starting point of our sample selection process.

20

If a linear regression is run to Equation (2), the result of a diagnosis for multicollinearity indicates that the values of Variance Inflation Factor (VIF) for all explanatory variables are less than the critical value 10 (the maximum value is less than 3). 17

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54. △ABNFEE*AUDCHG is the interaction of △ABNFEE and AUDCHG. △ROA

A change in main business returns, = [(operating incomet+1 – minor business incomet+1) / total assetst+1] – [(operating incomet – minor business incomet) / total assetst].

△DEBT

A change in leverage, = (total liabilitiest+1 / total assetst+1) – (total liabilitiest / total assetst).

GROWTH

Ratio of an increase in assets, = (total assetst+1 – total assetst) / total assetst.

PRELOSSES

A dummy variable, = 1 for an observation with consecutive losses both in period t and t-1; = 0 otherwise.

PRELOSS

A dummy variable, = 1 for an observation with a loss only in period t; = 0 otherwise.

AUDCHG

A dummy variable, = 1 for a change in the auditor taking place in period t+1; = 0 for retaining the incumbent auditor in period t+1.

SIZE

Natural log of total assets in period t+1.

Y2002

A dummy variable, = 1 for an observation in the fiscal year 2002; = 0 otherwise.

Y2003

A dummy variable, = 1 for an observation in the fiscal year 2003; = 0 otherwise.

For the experimental variables, the results in Columns A and B indicate that the coefficient of △ABNFEE is significantly positive at the 0.1 level, in contrast to a negative coefficient of △ABNFEE*AUDCHG significant at the 0.1 or 0.05 level. A Chi-square test shows the combined coefficient of (△ABNFEE+△ABNFEE*AUDCHG) is not significantly different from zero. 21 This implies that when the auditor is unchanged (AUDCHG=0), the higher △ABNFEE is, the more likely UAO is improved; however, when a change in the auditor takes place (AUDCHG=1), an increase in abnormal audit fees is not significantly associated with the improvement of UAO. The result generally supports our conjecture that the management have the capacity to avoid UAO by increasing AAF without changing the incumbent auditor. Further interpretations of the above result are as follows: 1) A change in the auditor is more likely to be subject to public attention and regulation because of its extraordinary nature and the required governance procedures (including public disclosure) for listed companies; 2) The CSRC and the CICPA (Chinese professional accounting body) have been putting increasing emphasis and efforts on the regulation of auditor changes for recent years; 3) Companies who change their auditors in our sample are inherently risky, as evidenced by a significant positive correlation between PRELOSSES and AUDCHG shown in Table 6, and by the further multivariate testing result of the auditor changes determinant model as well, 22 thus being treated by successor auditors 21

For the regression model in Column A, Wald χ2=0.484, p-value=0.487. For the regression model in Column B, Wald χ2=2.213, p-value=0.137. 22 Using a sample of 97 observations in our paper, we run a simplified regression model for determinants of auditor changes in period t+1 as follows: logit p(AUDCHG=1)=b0+b1PRELOSSES+b2PRELOSS+b3DEBT+b4GROWTH+δ, (3) where four variables including AUDCHG, PRELOSSES, PRELOSS, and GROWTH have the same definition as those defined in Table 5, DEBT is the leverage in period t (= total liabilitiest / total assetst). The model fits the data well with Pseudo-R2 as 22.0 per cent. The regression result shows a significant positive association between PRELOSSES and AUDCHG (at the .05 level). If a new explanatory variable PREDISCL (1 for a disclaimer of opinion issued by the auditor in period t, and 0 for a qualified opinion in period t) added in Equation (3), a significant positive association between PRELOSSES and AUDCHG is still there (at the 0.1 level), and a positive association (nearly significant at the 0.1 level) between PREDISCL and AUDCHG is observed. 18

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

with considerable scruples and conservatism. In general, an auditor-changing setting would weaken the management capacity to avoid UAO by providing more economic impetus, while maintaining the incumbent auditor and then “bribing” them would be more covert. As to the control variables, there is a significant positive (negative) association between △ROA (△DEBT) and the dependent variable at the 0.05 level, suggesting that the improvement of corporate financial performance and conditions contributes to the improvement of UAO. SIZE is positively correlated with the dependent variable at a significance level of 0.01 or 0.05, indicating that large companies have a stronger influence on the audit outcomes. In addition, the coefficient of GROWTH in Column A is positive and nearly significant at the 0.1 level. The above results are generally consistent with the economic rationale of the improvement of UAO; that is, it is the substantive improvement of the corporate operating performance and the financial condition that helps to make UAO improve correspondingly. We also find a significant positive association between PRELOSSES and the dependent variable (at the 0.01 level), suggesting that observations reporting consecutive losses both in period t and t-1 are more likely to realise the improvement of UAO. Given that no significant or high correlations were found between PRELOSSES and △ROA, △DEBT, or GROWTH in the Correlation Matrix of Table 6, we are concerned about the potential illegitimacy and management opportunism of the improvement of UAO for those observations with PRELOSSES = 1. The association between PRELOSS and the dependent variable is similar to but weaker than that between PRELOSSES and the dependent variable, and we have a similar concern about the improvement of UAO for observations with PRELOSS=1. 23 The results in both Columns A and B show that the coefficient of AUDCHG, though positive, is not significantly different from zero, suggesting that, in our sample, a change in auditor generally is not accompanied with an improvement of UAO.

VI. CONCLUSION AND DISCUSSION The management of a listed company usually have a strong incentive to avoid an audit outcome that is unfavourable to the company. This paper investigates the effect of a change in abnormal audit fees on the improvement of an unfavourable audit outcome. Starting from identifying a sample of listed companies who received non-unqualified audit opinions during 2000-2002, we observe their subsequent audit outcomes and the association with the change in 23

The Correlation Matrix in Table 6 shows a low and insignificant correlation between PRELOSS and △ROA, or △DEBT, and even shows a significant negative correlation between PRELOSS and GROWTH at the 0.05 level, suggesting no improvement in financial performance and conditions for observations with PRELOSS=1. 19

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

abnormal audit fees. We find that after controlling for the company’s fundamental financial characteristics and their change from period t to t+1, an increase in AAF without changing the auditor is significantly and positively associated with the improvement of UAO, while an increase in AAF concurrent with a change in the auditor is not significantly associated with the improvement of UAO. The result suggests that the economic impetus provided by the management probably compromise auditor independence. As long as the management make substantive decisions on the selection of an auditor and related fees, the management have the capacity to manipulate the audit outcomes to a considerable extent even for those companies with high risks. Considering the difficulty for information users to detect the abnormal component of audit fees and its change, an approach that applies the result of our study is suggested for information users by only observing the change in actual audit fees to roughly assess its implications for auditor independence, which is warranted by a significant, positive, and high correlation between the change in AAF and the change in actually observed audit fees. 24 This paper has preliminarily explored the potential impact of economic interests provided to the auditor on the audit outcome. Several issues remain that may impose limitations upon the generalisability of the conclusions of this study. Firstly, in order to assure a typical research setting, we have limited our sample to a small number of observations with an outstanding conflict between the management and the auditor. The short history of and weaknesses in audit fees disclosure make the sample size even smaller. Secondly, limitations are inevitable in estimating the abnormal component of audit fees, which are similar to those in estimating the discretionary component of accounting accruals. 25 For example, if some client risk factors are unintentionally omitted in the audit pricing model, an error will appear in the estimation of AAF and its change during periods. Thirdly, besides the limitations of the research design, the quality of audit fees disclosure, including non-disclosure by problem companies 26 and potential incompleteness for audit fees disclosure, 27 has a great impact on the measurement of AAF. 24

We define △FEE as the change in actually observed audit fees, computed as (FEEact/t+1 – FEEact/t) / FEEact/t, where FEEact is the amount of annual audit fees actually observed, and t is the year a listed company received a non-unqualified audit opinion. Pearson correlation between △ABNFEE and △FEE is 0.510 (p-value=0.000). 25 For example, it is not clear whether cross-sectional or time-series data should be used, nor is it clear whether the data should be estimated in a pool or by each accounting firm. These issues warrant further study on the formation and estimation of abnormal audit fees. Actually, limitations are inevitable in any estimation of an unexpected component of certain economic value, no matter how comprehensive the estimation model seems. Nevertheless, it is rational to conclude that the audit pricing model for estimating abnormal components of audit fees, because of its relative maturity, is more powerful than the accounting accruals model for estimating discretionary accruals. For a detailed discussion of research design issues in estimating discretionary accruals, see McNichols (2000). 26 In the process of data collection, we noticed that a considerable number of listed companies who successively avoid UAR do not disclose any information about annual audit fees, which limited our sample size to a large extent. 27 Even if a problem company discloses annual audit fees as required, there is a possibility that extra economic interests provided by the management to the auditor are not disclosed, or are disclosed in alternative forms, such as: 1) disclosure of fees for a service that would have been unnecessary; 2) mix of extra payments into fees for services other than an annual audit, where the conventional audit pricing model would not apply; and 3) disclosure 20

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

APPENDIX Estimation of Abnormal Audit Fees and △ABNFEE On the basis of prior studies on audit fees, a regression model is constructed as shown in Table A1, incorporating such factors as client size, audit complexity, audit risk, and auditor characteristics. In Equation (A1), the variable SIZE is used to control for the auditee’s size, measured by the natural log of total assets, SQSUBS, controlling for audit complexity measured by the square root of the number of consolidated subsidiaries. The variables DEBT, CATA, LIQUID, and ROA are used to control for the auditee’s business and financial risk, measured by the leverage, the proportion of current assets to total assets, the ratio of current assets to current liabilities, and main business returns, respectively. To reflect auditor characteristics, we use the variables AUDCHG, TENURE, and AUDITORi (i represents the audit firm) 28 to control for the auditor change, audit firm tenure, and specific auditor identity. The dummy variables YEARj (j denotes the specific time-period) are used to control for potential systematic difference among periods. Table A1 OLS Regression Model for Annual Audit Fees LnFEE

=

b0+b1SIZE+b2SQSUBS+b3DEBT+b4CATA+b5LIQUID+b6ROA+b7AUDCHG +b8TENURE+∑biAUDITORi+∑bjYEARj+μ

(A1)

Definition of variables: FEE

= Annual audit fees.

LnFEE

= Natural log of annual audit fees.

SIZE

= Natural log of total assets.

SQSUBS

= Square root of number of consolidated subsidiaries.

DEBT

= Total liabilities to total assets.

CATA

= Ratio of current assets to total assets.

LIQUID

= Ratio of current assets to current liabilities.

ROA

= Main business returns (operating income – minor business income) / total assets

AUDCHG

= Dummy variable, 1 for a change in auditor; 0 for no change in auditor.

TENURE

= Audit firm tenure of continuously conducting the annual audit for the client.

of an increased budget for covering accommodation and traffic expenditures in an audit. The substantial freedom owned by the management for how to decide on forms of auditor-related fees and how to disclose such information does create great difficulties for information users including researchers in judging the completeness of annual audit fees and identifying the abnormal component of audit fees. 28 Besides the variables BIG4 and SMAUD (for their definitions, see Table A1), every audit firm with five or more listed clients in our sample is designated with a corresponding dummy variable AUDITORi to enhance the power and predictability of the OLS model for audit pricing. We include those dummy variables AUDITORi with a significant coefficient in the final model to predict abnormal audit fees. Such procedures improve the adjusted R2 of the OLS model for observations in period t (t+1) from 0.45 (0.42) to 0.55 (0.50). 21

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54. AUDITORi

= Dummy variables. The variable BIG4 is set to identify Big 4 auditors (1 for a Big 4 firm of PWC, KPMG, DDT, or EY). The variable SMAUD is used to identify relatively small firms with a criterion of less than 10 listed companies (1 for a small firm with less than 10 listed companies). In addition, every audit firm with five or more listed clients in our sample is designated with a corresponding dummy variable AUDITORi.

YEARj

= Dummy variables. For observations in period t, fiscal years cover 2000-2002, with two dummy variables Y2001 and Y2002. For observations in period t+1, fiscal years cover 2001-2003, with two dummy variables Y2002 and Y2003.

μ

= Residual term, which can be interpreted as the abnormal component of audit fees that main economic factors cannot explain (i.e. AAF).

For observations that received non-unqualified audit opinions in period t, we need to compute the change in abnormal audit fees in period t+1. Audit fees and other necessary data for period t and t+1 will be used to estimate residual terms μt and μt+1, respectively, and thus to compute △ABNFEE (=μt+1-μt). Tables A2 and A3 present descriptive statistics and regression results using necessary data in period t and t+1 for 97 observations that received non-unqualified audit opinions during 2000-2002 (statistics for specific AUDITORi variables other than BIG4 and SMAUD are not reported). Table A2 Descriptive Statistics for Variables in the Annual Audit Fees Regression Model Observations in period t

Observations in period t+1

Panel A: Continuous variables

Mean

Std. Dev.

Mean

Std. Dev.

FEE

48.208

48.930

49.789

55.607

LnFEE

3.665

0.562

3.677

0.572

SIZE

11.480

1.019

11.299

1.126

SQSUBS

1.845

1.282

1.883

1.295

DEBT

0.930

1.439

1.211

2.533

CATA

0.571

0.192

0.574

0.240

LIQUID

1.216

1.258

1.080

0.876

ROA

-0.224

0.687

-0.247

0.670

TENURE

3.880

2.790

3.990

3.010

Number

Proportion

Number

Proportion

AUDCHG (=1)

26

27.0%

24

25.0%

BIG4 (=1)

4

4.1%

5

5.2%

SMAUD (=1)

23

24.0%

27

28.0%

Panel B: Categorical variables

other AUDITORi variables

Unreported

Unreported

Unreported

Unreported

Y2001 (=1)

40

41.0%

-

-

Y2002 (=1)

45

46.0%

40

41.0%

Y2003 (=1)

-

-

45

46.0%

Number of observations

97

Note: see Table A1 for definitions of variables. 22

97

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

Table A3 shows a significant F-statistic of 7.93 (6.95) (p-value=0.000) and adjusted R2 of 0.54 (0.50) of the audit-pricing model for observations in period t (t+1), suggesting the model fits the data well. 29 Regression results indicate that both client size (SIZE) and audit complexity (SQSUBS) are significantly and positively correlated with annual audit fees, consistent with audit pricing theory and prior studies. There is a significant positive association (at the 0.1 level) between the current ratio variable (CATA) and audit fees for observations in period t. Big 4 firms (BIG4=1) earn significantly higher audit fees. Audit firms with few listed clients (SMAUD=1) earn a lower level of audit fees (especially evident for observations in period t+1). Table A3 Regression Results of OLS Model for Annual Audit Fees Observations in period t Intercept

Observations in period t+1

Coefficient

t-statistic

p-value

Coefficient

t-statistic

p-value

0.692

1.157

0.251

1.997***

2.964

0.004

SIZE

0.196***

3.704

0.000

0.134**

2.459

0.016

SQSUBS

0.087**

2.331

0.022

0.158***

3.556

0.001

DEBT

0.052

1.096

0.276

0.017

1.479

0.143

CATA

0.349*

1.682

0.096

0.050

0.316

0.753

LIQUID

0.025

1.183

0.241

-0.044

-1.050

0.297

ROA

-0.012

-0.141

0.888

-0.019

-0.260

0.795

AUDCHG

0.297**

2.263

0.026

-0.104

-0.712

0.479

TENURE

-0.001

-0.059

0.953

-0.054**

-2.546

0.013

0.831**

2.394

0.019

0.818*

1.800

0.076

BIG4

-0.049

-0.575

0.567

-0.178**

-2.064

0.042

Unreported

Unreported

Unreported

Unreported

Unreported

Unreported

Y2001

0.205

1.607

0.112

-

-

-

Y2002

0.301**

2.333

0.022

0.119

0.940

0.350

Y2003

-

-

-

0.150

1.141

0.257

SMAUD other AUDITORi variables

F-statistic

7.93***

6.95***

2

0.54

0.50

Number of observations

97

97

Adjusted R

***, **, and * indicate significance at the 0.01, 0.05, and 0.1 levels, respectively. See Table A1 for definitions of variables.

Table A4 reports descriptive statistics of μt, μt+1, and △ABNFEE (=μt+1-μt). In order to understand their real economic meanings, we present descriptive statistics of the proportion of abnormal audit fees to actually observed total audit fees using the value of μ. 30

29

The regression results for observations in periods t and t+1 are adjusted for white heteroskedasticity. The Kolmogorov-Smirnov test indicates a normal distribution of μt and μt+1 (p-values are 0.845 and 0.534, respectively). Multicollinearity is not a concern since the values of Variance Inflation Factor (VIF) for all explanatory variables are less than the critical value 10 (the maximum value is around 3).

30

See table note # in Table A4 for the deduction. 23

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

As Table A4 indicates, all the mean values of μt, μt+1, and △ABNFEE are zero. For period t (t+1), the minimum and maximum values of μt (μt+1) are -0.855 (-1.306) and 0.889 (0.834), respectively, suggesting a range of the proportion of abnormal audit fees to actual audit fees between -135.0 (-269.3) per cent and 58.9 (56.6) per cent. The upper quartile statistics show that a quarter of the sample companies in period t (t+1) are estimated to provide an unexpected 21.8 (24.9) per cent of actually observed audit fees, suggesting our statistical results of μt and μt+1 are consistent with the general understanding of the association between abnormal audit fees and the economic benefit to the clients. Table A4 Descriptive Statistics of μt, μt+1, △ABNFEE, and Proportion of Abnormal Audit Fees to Actually Observed Total Audit Fees Observations in period t μt

FEEabn t/FEEact t #

Observations in period t+1 FEEabn t+1/FEEact t+1 #

μt+1

△ABNFEE ##

Mean

0.000

0.0%

0.000

0.0%

0.000

Std. Dev.

0.350

29.5%

0.370

30.9%

0.276

Minimum

-0.855

-135.0%

-1.306

-269.3%

-0.924

Maximum

0.889

58.9%

0.834

56.6%

0.696

Lower quartile (25%)

-0.264

-30.2%

-0.246

-27.9%

-0.132

Median

-0.016

-1.6%

0.041

4.0%

0.024

Upper quartile (75%)

0.246

21.8%

0.286

24.9%

0.174

Number of observations

97

97

97

Note: See Table A1 for the definition of μ. # The deduction from the value of μ to the proportion of abnormal audit fees to actually observed total audit fees is as follows. Based on Equation (1), μ= LnFEE – (a0 +∑aiXi), then eμ = e actually observed annual audit fees, presented as FEEact, while e

a0 +

LnFEE − ( a0 +

∑ ai X i ) . Since eLnFEE is the

∑ ai X i is the expected component of audit fees

that can be explained by explanatory variables in the audit pricing model, presented as FEEest, we come to eμ = FEEact / FEEest, then FEEest / FEEact = 1/eμ, and finally FEEabn / FEEact = (FEEact – FEEest) / FEEact = 1 – 1 / eμ, where FEEabn indicates the unexpected (abnormal) component of audit fees. ## △ABNFEE=μt+1-μt .

The correlation between μt and μt+1 is also tested. For 97 sample companies (52 observations with μt<0), the Pearson correlation between μt and μt+1 is 0.707 (0.564), significant at the 0.01 level, suggesting that an increase in △ABNFEE during two consecutive periods is generally not the effect of low-balling in audit pricing. 31 It is worth noting that we do not estimate the abnormal audit fees in all available listed companies for the following reasons: 1) Focusing on these 97 observations allows us to avoid the 31

If there is a significant negative correlation between μt and μt+1 (especially for observations with μt<0), an alternative interpretation of the increase in △ABNFEE (i.e. lowballing pricing effect) may not be excluded, since such an effect features a contrast between a relatively low level of audit fees in a prior period and a higher level of audit fees in a subsequent period (DeAngelo, 1981). 24

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

problem of structural difference in the audit fee model. The 97 observations under investigation are a sub-group of poor-performing firms. Their ability to pay the auditors and their incentives to improve both the reported financial results and audit opinion are different from those of firms that are not financially distressed. The auditor may have to use a different audit fee structure in order to retain these financially distressed clients. Extending the estimation sample to all listed firms would force the observations from two distinctively different client groups to fit into one audit fee model despite the fact that one group of observations has much less resources to pay the auditors but much stronger incentives to improve audit opinion and financial results than the other. 32 2) Since the requirement for reporting audit fees in China is not unambiguous and enforcement of it is weak. For example, some companies use cash basis and others use accrual basis in reporting audit fees. To further complicate the situation, auditors may charge a lower audit fee but have their audit-related expenses reimbursed from their clients in some years but charge a higher audit fee without getting any expenses reimbursed. This actually is another reason why focusing on a smaller sample is better than using all observations: the more observations we use for estimating the abnormal audit fees, the higher the probability that measurement errors may contaminate our results. 33 To sum up, abnormal audit fees estimated by focusing on the 97 observations could be superior to that generated by using all listed firms.

REFERENCES 李爽，吴溪.2002a.《审计师变更研究：中国证券市场的初步证据》.中国财政经济出版社。 李爽，吴溪.2002b.审计意见变通及其监管：经验证据.《中国会计与财务研究》第 4 卷第 4 期，1-28。 李爽，吴溪.2004a.《审计定价研究：中国证券市场的初步证据》.中国财政经济出版社。 李爽，吴溪.2004b.不利审计意见的改善与自愿性审计师变更：1997~2003 年间的趋势描述及 其含义.《审计研究》第 5 期，13-19。 李爽，吴溪.2005.后中天勤时代的中国证券审计市场.《会计研究》第 6 期，10-15。 32

Research students at City University of Hong Kong have collected annual audit fees data for all available listed companies during 2001-2003, and we have run the audit pricing model using all available audit fees data. F-test results show clearly that the two sets of regression coefficients are significantly different most of the time. For example, the coefficient of the variable SIZE when using all available observations (0.271) is significantly larger than those when only using 97 observations (0.196 and 0.134, for periods t and t+1, respectively, with an F-statistic of 64.91 and 215.00 both significant at the 0.0001 level). Also, the coefficient of the variable DEBT when using all available observations (0.097) is also significantly larger than those when only using 97 observations (0.052 and 0.017, for periods t and t+1, with an F-statistic of 3.45 and 10.87, significant at the 0.1 and 0.001 levels, respectively). Such evidence suggests that for our special sample with 97 observations, auditors seems to charge less for the same size of client and are less sensitive to the financial risk of the client. 33 Compared with 97 observations that have been verified by reading the annual report carefully, it would be a formidable task to read over annual reports for all recent years to verify all available audit fees data. Even if one could do that, the probability of having measurement errors in the dataset would go up as the sample size increases with insufficient and incomparable disclosure practices. 25

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

孙铮，王跃堂.2000.说明段与变更审计意见的实证分析.见：孙铮，李增泉主编.《中国证券市 场财务与会计透视》.上海财经大学出版社，33-43。 夏冬林，林震昃.2003.我国审计市场的竞争状况分析.《会计研究》第 3 期，40-46。 中国证监会.2001a.《谁审计中国证券市场：审计市场分析（1997-1999）》.中国财政经济出版 社。 中国证监会.2001b.《谁审计中国证券市场：审计市场分析（2000）》.中国财政经济出版社。 中国证监会.2002.《注册会计师说“不”：中国上市公司审计意见分析（1992-2000）》.中国 财政经济出版社。 中国证监会.2003.《谁审计中国证券市场：审计市场分析（2001-2002）》.中国财政经济出版 社。 Ashbaugh, H., LaFond, R., and Mayhew, B. W. (2003), ‘Do Nonaudit Services Compromise Auditor Independence? Further Evidence’ The Accounting Review 78: 611-639. Chen, C. J. P., Su, X, and Zhao, R. (2000), ‘An Emerging Market’s Reaction to Initial Modified Audit Opinions: Evidence from the Shanghai Stock Exchange’ Contemporary Accounting Research 17: 429-455. Chow, C. W., and Rice, S. J. (1982), ‘Qualified Audit Opinions and Auditor Switching’ The Accounting Review 57: 326-335. Chung, H., and Kallapur, S. (2003), ‘Client Importance, Nonaudit Services, and Abnormal Accruals’ The Accounting Review 78: 931-955. Craswell, A., Stokes, D. J., and Laughton, J. (2002), ‘Auditor Independence and Fee Dependence’ Journal of Accounting and Economics 33: 253-275. DeAngelo, L. (1981), ‘Auditor Independence, Low Balling, and Disclosure Regulation’ Journal of Accounting and Economics 3: 113-127. DeFond, M. L., Raghunandan, K., and Subramanyam, K. R. (2002), ‘Do Non-Audit Service Fees Impair Auditor Independence? Evidence from Going Concern Audit Opinions’ Journal of Accounting Research 40: 1247-1274. DeFond, M. L., Wong, T. J., and Li, S. (2000), ‘The Impact of Improved Auditor Independence on Audit Market Concentration in China’ Journal of Accounting and Economics 28: 269-305. Frankel, R. M., Johnson, M. F., and Nelson, K. K. (2002), ‘The Relation between Auditors’ Fees for Nonaudit Services and Earnings Management’ The Accounting Review 77 (Supplement): 71-105. Kinney, W. R. Jr., and Libby, R. (2002), ‘Discussion of The Relation between Auditors’ Fees for Nonaudit Services and Earnings Management’ The Accounting Review 77 (Supplement): 107-114. Lennox, C. (2000), ‘Do Companies Successfully Engage in Opinion Shopping? Evidence from the UK’ Journal of Accounting and Economics 29: 321-337.

26

Chen, Charles J.P., Xijia Su, and Xi Wu. Abnormal Audit Fees and the Improvement of Unfavourable Audit Outcomes. China Accounting and Finance Review Vol.7 No.4 (2005 December): 29-54.

McNichols, M.F. (2000), ‘Research Design Issues in Earnings Management Studies’ Journal of Accounting and Public Policy 19: 313-345. Simunic, D.A. (1980), ‘The Pricing of Audit Services: Theory and Evidence’ Journal of Accounting Research 18: 161-190. Smith, D. B. (1986), ‘Auditor “Subject to” Opinions, Disclaimers, and Auditor Changes’ Auditing: A Journal of Practice & Theory 6: 95-108. Whisenant, S., Sankaraguruswamy, S., and Raghunandan, K. (2003), ‘Evidence on the Joint Determination of Audit and Non-Audit Fees’ Journal of Accounting Research 41: 721-744.

27