THE SAMPLE AND VARIABLE MEASUREMENT Sample and Data

where the error term, v is assumed to be normally distributed with mean zero and variance σv

2 and E(u, v)=0. The latent variable (QUAL*) in Equation (2) represents the auditor’s judgment of the client’s financial condition. The remaining explanatory variables (Y) control for other determinants of audit opinions. The auditor applies threshold values, µ1, in determining the modified audit opinion. The auditor’s choice of opinions are clean (QUAL=0), shared (QUAL=1), and going concern (QUAL=2) when QUAL* falls in the ranges of (< 0), (0, µ1) and (>µ1), respectively.

The variable THREAT ^GC_Clea is used to test H1; accordingly, we expect the coefficient on THREAT ^GC_Clean is positive (γ1>0). As for H2, we predict the coefficient on THREAT ^GC_OA is negative (γ2>0).

IV. THE SAMPLE AND VARIABLE MEASUREMENT

Data

We apply this test to listed companies in Taiwan and establish the research period from 1999-2001. Since some variable measurements employ data from year t-2 to t+2, our data collection period covers 1997-2003. All the variables used to construct our empirical analysis are retrieved from the Taiwan Economic Journal (TEJ) database.

Sample Selection

Our sample is composed of publicly traded corporations listed on TSE and GSM, excluding financial institutions. To avoid the same company being classified into both switch and non-switch samples, companies in the non-switch sample are required to have kept (i.e., not switched) auditors for at least 3 consecutive years, the exact duration of the research period. In addition, we exclude insolvent companies that are judicially declared a special arrangement by TSE, since an auditor’s reporting discretion declines when a company has filed for bankruptcy (Carcello and Neal 2000). Furthermore, we do not include litigation qualifications because liability for lawsuits is not recorded in the financial statements (Krishnan and Krishnan 1996). Following this procedure, our sample is reduced to 1,926 companies-year combinations.

Auditors do not generally issue going-concern opinions for non-stressed companies that suddenly fail (McKeown et al. 1991). Therefore, from the preliminary sample we determine those companies that were potentially financially distressed. As in prior research, we define a company as stressed if it exhibits at least one of the following financial stress signals: (1) negative working capital in year t, (2) a bottom line loss in any of the 3 years prior to year t, and (3) negative operating cash flows in the consecutive 3 years prior to year t. After deleting non-stressed companies, we also exclude companies with insufficient data for estimating Equation (1). Therefore, our final sample for Model

(1) includes 791 companies-years. Owing to adding the lag variable of audit opinion in Equation (2), the final sample for Equation (2) is further reduced to 607 companies-years.

Table 1 presents the details of our sample selection procedure.

Table 1: Sample Selection Criteria

Sample selection for Equation (1)

Initial sample: industrial firms for 1999-2001 2,147 Less: Companies not retaining auditors at least 3 years in the non-switch

sample ( 145)

Insolvent ( 67)

Litigation qualification ( 9)

Preliminary sample 1,926

Less: Non-stressed firms (1,062) Insufficient data for Equation (1) ( 73) Final sample for Equation (1) 791 Breaking down as:

Non-switcher^a 760

Switcher^b 31

Sample selection for Equation (2)

Original sample from Equation (1) 791 Less: Insufficient data for Equation (2) ( 184) Final sample for Equation (2) 607 Breaking down as:

Clean opinions 344

Shared opinions 239

Going concern opinions 24

a Companies in the non-switch sample are required to have kept auditors for at least 3 consecutive years (financial institutions and service companies excluded).

b Companies in the switch sample are required to dismiss their auditors in year t+1 (financial institutions and service companies excluded).

Variable Measurement Auditor Switch (S)

A dummy switch (S) has been set to one if a company changes its auditor in the year following the issuance of the opinion, zero otherwise. By comparing both the audit firm and the individual auditors in the current year with that in the following year, we identify the auditor switch.⁴ Therefore, any one of following conditions shall not be coded as a switch: (1) same audit firm but different individual auditors, (2) audit firm merges, (3) same individual auditors who have joined a new audit firm, and (4) audit firm name changes.

4 The switching status of each company is verified by examining other sources such as the website information on Market Observation Post System in Taiwan.

Auditor Report (QUAL, OA, GC)

The auditor report (QUAL) is the dependent variable in Equation (2). QUAL is coded as zero for clean opinions, one for shared opinions, and two for going-concern opinions.⁵,⁶ Multiple qualifications arising in conjunction with the going-concern opinion are included in the going-concern opinions category. Two opinion dummies (i.e. OA and GC) are designated as independent variables in Equation (1) due to the three-level QUAL.

OA, shared opinion, is 1 when QUAL=1, zero otherwise. GC, a going-concern opinion, is 1 when QUAL=2, zero otherwise.

Control Variables (X) Included in the Auditor Switch Model

In addition to auditor reports of interest, we control for the effects of other factors likely to affect a client’s decision to dismiss its auditor: (1) changes in client characteristics, (2) characteristics of the incumbent auditor, (3) financial distress, and (4) miscellaneous, which are designated as year dummies (YEARj).

(1) Client Changes

We expect a positive relation between auditor switch and each of the following variables that reflect changing auditee characteristics. Johnson and Lys (1990) argue that audit firms achieve competitive advantages through specialization, and that clients purchase audit services from the least cost supplier. Client-auditor realignments thus represent efficient responses to changes in client operations and activities over time.

Following the model used by Johnson and Lys, four variables are used as proxies for expansion, profitability, financing and audit risk: changes in asset growth (|GROWCH|), changes in cash flow (|CFOCH|), changes in financing (|FINCH|), and changes in times-interest-earned (|TIECH|), respectively. |GROWCH| is constructed by the absolute value of the difference obtained from subtracting the pre-switch two-year average assets growth rate from the post-switch average growth rate. The absolute value is used since the primary focus is on whether the client changes its auditor, not the direction of auditor changes. Similarly, the variable, |CFOCH|, is the absolute value of change in two-year average operating cash flows (deflated by total assets) before and after an auditor switch.

|FINCH| is measured by the absolute difference of two-year mean proceeds from external financing before and after the switch, where external financing is measured by the proceeds from newly-issued equity and debt (public or private), divided by total assets.

We construct |TIECH| from the absolute difference between the two year mean times-interests-earned (TIE) before and after the switch. TIE is defined as earnings before interests and taxes divided by interest expenditure. We winsorize both the upper and lower 5% of TIE because TIE is inflated by minor interest expenditures.

A change in top management is often associated with a change in auditors.⁷ A new

5 In prior auditor report research, an audit opinion was classified as ‘modified’ for material uncertainties and going-concern problems depending on the severity of qualifications. In Taiwan, as most material uncertainties involve litigation, which cannot be predicted by financial variables, we eliminate the litigation category as has been done by prior research (e.g. Krishnan and Krishnan 1996).

6 In line with previous research (e.g., Krishnan 1994; Jeter and Shaw 1995), consistency exceptions for voluntary and mandatory accounting changes are one cause for a modified auditor report, but they are included in the clean opinion category because the auditor has little discretion in such matters.

7 See Chow and Rice (1982), Williams (1988), and Carcello and Neal (2003).

manager may change auditors in order to obtain a fresh perspective on the company’s financial results, or because he or she had positive experience with another audit firm (Carcello and Neal 2003). In Taiwan, both the chairman of the board and the chief executive officer (CEO) are charged with the execution of the company’s decisions.

Therefore, we set dummy variable MGTCH to 1 if both chairman and CEO changed in the year the auditor was dismissed or in the previous year and 0 otherwise.

(2) Incumbent Auditor Characteristics

Following Krishnan et al. (1996), we use IMS and BIG5 to represent auditor-related factors and predict a negative relation between these auditor-related factors and auditor switch. The auditor’s industry market share (IMS) is measured as the percentage of the log of total assets that the auditor audits for all companies in the client’s industry⁸. The auditor’s industry market share can reflect audit expertise and can also proxy for reputation effect. The greater the auditor’s market-share in the client’s industry, the less likely the client is to dismiss its auditor (Krishnan et al. 1996). Previous studies have used the Big 5 auditing firms to proxy for both auditor quality and reputation effects. A client is less likely to switch from a Big 5 (Krishnan et al. 1996). Therefore, variable BIG5 takes the value of one for the member firms of Big 5 in Taiwan, zero otherwise.

(3) Financial Distress

Previous studies (e.g., Schwartz and Menon 1985; Krishnan and Stephens 1995) have suggested that financially distressed companies may be more likely than healthy companies to change auditors. The motivation for such a change could be a need for different services, an inability to pay audit fees or disagreements with the incumbent auditor over accounting policies or disclosures. We use a 2-year consecutive net loss (LOSS2) to represent financial distress and predict a positive sign for this variable.

Control Variables (Y) Included in the Auditor Report Model

Besides the threat variables of interest, the choice of independent variables in the auditor report model is classified into four categories: (1) contrary factors, (2) mitigating factors, (3) auditor characteristics (quality and tenure), and (4) miscellaneous.

(1) Contrary factors

Prior studies have found that the greater the client’s financial distress, the greater the probability of receiving modified auditor reports (Carcello and Neal 2000; Geiger and Raghunandan 2002). We use dummy variable DISTRESS to identify whether the entity has declared insolvency in the subsequent year.⁹ We expect a positive relationship between DISTRESS and the receipt of a going concern opinion.

Mutchler et al. (1997) find that debt covenant violations are positively associated with the probability of receiving a going-concern opinion. We include leverage variable LEV to capture proximity to covenant violations because firms close to violation are

8 Each company’s industry comes from TEJ’s classification of companies into industries, which is based on the TSE version SIC and adjusted by primary products.

9 Dopuch et al. (1987) consider whether a loss was reported (LOSS) in measuring a client’s financial health.

In contrast, we use LOSS as one of several financial stress signals mentioned earlier in section 4.1.2. To avoid multicollinearity, we use the DISTRESS variable, which reflects the ex post event of being classified as insolvent, as a proxy for the auditor’s ex ante perception that the client’s financial condition is deteriorating. The variable’s validity depends on whether one can reasonably assume that the auditor is aware of the client’s financial condition at the time of issuing the report.

likely to have high leverage (Beneish and Press 1993), and predict a positive sign. LEV is measured by total liabilities over total assets at the end of the year.

Following Dopuch et al. (1987), we also include the ratio of receivables and inventories to assets (RIA) to capture high-risk accounts, which call for greater caution and exercise of independent auditor judgment. RIA is expected to have a positive association with a going concern opinion.

(2) Mitigating factors

We include several factors that are likely to mitigate the probability of receiving a going concern opinion and expect the sign on the coefficients for each mitigating factor to be negative. Client size is generally positively associated with its financial health, which in turn decreases with the likelihood of its receiving qualified opinion (Dopuch et al.

1987; Francis and Krishnan 1999). Furthermore, large companies have more negotiating power in the event of financial difficulties and hence are more likely to avoid bankruptcy (Reynolds and Francis 2001, DeFond et al. 2002). Client size (SIZE) is measured as a log of total assets.

Two other mitigating factors in our model include FASTSALE and FFINANCE because SAS No.59 specifies managerial actions that mitigate the effect of contrary factors, including plans to sell assets, issue new financing or refinance existing debt, and increase ownership equity. Using methodology similar to that of Reynolds and Francis (2001), we examine the subsequent fiscal year financial statements to identify sales of assets or the issuance of new debt or equity. FASTSALE is the sum of the proceeds from selling investment and fixed asset in year t+1, scaled by total assets in year t. FFINANCE is measured by the issuance of new debt or equity (public or private) in year t+1, divided by total assets in year t.

(3) Auditor characteristics

We include an industry specialist dummy variable (SPEC) to control for the impact auditor quality could have on the exercise of independent judgment. High-quality auditors have a greater investment in technology to detect errors and irregularities and are therefore more likely to issue a qualified opinion (Krishnan and Krishnan 1996; Craswell et al. 2002). We expect the sign on the coefficient for SPEC to be positive. According to Craswell et al. (1995), we identify the industry specialist if the auditor’s market share is greater than 20% in the client’s industry with at least 30 companies.¹⁰

We measure auditor tenure (TENURE) as the number of consecutive years that the client has retained the auditor.¹¹ The longer the auditor tenure, the more complacent the auditor becomes and the less independent the auditor’s judgment is (Jeter and Shaw 1995;

Geiger and Raghunandan 2002; Craswell et al. 2002). It is also the case, however, that over the years an auditor develops in-depth knowledge of the client’s business, which is crucial in performing an effective audit, and thus is more likely to be vigilant in exercising independent auditor judgment (Geiger and Raghunandan 2002; Craswell et al.

2002). Therefore, the sign on the coefficient for TENURE could be either positive or

10 Craswell et al. (1995) used the thresholds of 10% and 20%, respectively, to identify an industry specialist.

We chose 20% because both mean and median of IMS are greater than 10% (see Table 2).

11 We truncate auditor tenure at 12 years because of data limitation. Moreover, truncation can reduce the effect of extreme values for clients that have retained their auditors for many years (Carcello and Neal 2003).

negative.

(4) Miscellaneous

The remaining control variables are prior year audit opinion (PRIOROP), time listed (AGE), and indicator variables for year j (YEARj). PRIOROP captures the effect of persistence in audit reporting and is expected to have a positive sign. AGE controls firm maturity and is measured as log of the number of years the company has been publicly traded. Finally, YEARj control for any year-specific effects. As with Francis and Krishnan (1999), no directional signs are predicted for AGE and YEARj.

V. RESULTS