krishnan et al - 1994 - auditor switching and conservatism

It takes the value 1 for unqualified opinions U, 2 for asset realization including litigation opinions AR, and 3 for going concern opinions GC.4 Consistency exceptions for voluntary and

Auditor Switching and Conservatism

Author(s): Jagan Krishnan

Source: The Accounting Review, Vol 69, No 1 (Jan., 1994), pp 200-215

Published by: American Accounting Association

Stable URL: http://www.jstor.org/stable/248267

Accessed: 09/05/2014 00:48

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at

http://www.jstor.org/page/info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive We use information technology and tools to increase productivity and facilitate new forms

of scholarship For more information about JSTOR, please contact support@jstor.org.

American Accounting Association is collaborating with JSTOR to digitize, preserve and extend access to The Accounting Review.

http://www.jstor.org

Vol 69, No.1

January 1994

pp 200-215

Jagan Krishnan

Temple University

SYNOPSIS AND INTRODUCTION: The relation between audit opinions and auditor switching has received considerable attention in recent years Chow and Rice (1982) report a positive association between a firm's propensity to switch auditors and the receipt of a qualified opinion in the year prior to the switch.1 However, firms that switch auditors do not seem to receive "improved" opinions in the year following the switch (Chow and Rice 1982; Smith 1986) Despite this evidence, the Securities and Exchange Commission (e.g., Release No 33-6594 [1985] and Financial Reporting Release No 34 [1989]) and the popular financial press (e.g., Power 1984) continue

to express concerns about opinion shopping

This study focuses on the auditor's opinion formulation process for switching and non-switching clients in the year prior to the switch In particular, it examines the possibility that auditor switches are triggered not by the receipt of qualified opinions, but by auditors' use of conservative judgments for some clients While conservatism refers to a numberof accounting and auditing issues, the overall conservatism of the auditor is assumed to be reflected in a tendency to issue qualified opinions An ordered probit model of the qualification decision is estimated with different threshold values for prospective switchers and non-switchers measuring different judgments applied to the two groups of clients The results support the hypothesis that threshold values for switchers are significantly lower (more conservative) than those for non- switchers Further analysis of switching patterns suggests that when qualified opinions are based on conservative standards, the switching rate is higher than when average standards are applied

The observed conservatism could be the auditor's reaction to negative private information gathered during the audit that makes continued association with the

I Under Statement on Auditing Standards No 58 (AICPA 1988), auditors no longer issue a qualified opinion for a material uncertainty, which is now reported as an unqualified opinion with an explanatory paragraph Hence, the term

"qualification" for the purpose of this study will also refer to an unqualified opinion issued for a material uncertainty

This paper is based on my dissertation at the Ohio State University I am indebted to my dissertation committee, Ray Stephens (Chairman), Stephen Cosslett, Douglas Schroeder, and Eric Spires for their guidance and suggestions I have also benefited from the comments on earlier versions of this article by Celal Aksu, Steven Balsam, Ann Gabriel, Penelope Greenberg, Ralph Greenberg, Jayanthi Krishnan, Eric Press, SridharRamamoorti, A Rashad Abdel-khalik (editor), two anonymous reviewers, and seminar participants at the Ohio State University, Temple University, and the 1992 American Accounting Association Annual Meeting

Submitted December 1991 Accepted May 1993

200

Krishnan-Auditor Switching and Conservatism 201

client uneconomical or undesirable However, the evidence indicating that opinions

do not improve after switches suggests that opinion shopping is generally futile This condition may exist because private information is obtained during the audit process, and a client usually will not be able to shopfora less conservative auditorwith a prior commitment of a favorable treatment Recent concerns about opinion shopping and its effect on auditor independence may therefore not be justified

Key Words: Conservatism, Auditor switching, Audit opinion, Opinion shopping

Data Availability: The author has spent considerable time and effort in the

collection of the data used for this study Data will be made available upon completion of related projects

T n HE remainder of this article is organized as follows Section I provides the motivation and develops the hypothesis Section II outlines an empirical model to test for auditor con- servatism Sections III and IV describe the data and the results Sensitivity analyses are presented in section V Concluding comments appear in section VIL

I Motivation The relation between audit opinions and auditor switching has been examined around the themes of opinion shopping and reasons for switching The relation between switching and the opinion issued in the preceding year has received particular attention Chow andRice (1982) show

a significant positive coefficient for the opinion variable in estimating the decision to switch DeAngelo (1982), Haskins and Williams (1990), Schwartz and Menon (1985) and Smith (1986) examine the issue (although only Smith [19861 makes it the main focus of analysis), but do not find a significant relation between switching and the audit opinion Given that different meth- odologies are used, no consensus has emerged concerning the degree of association between switching and issuing qualified audit opinions

A smaller number of studies concentrate specifically on opinion shopping by examining the relation between switching and the subsequent audit opinion These results are more consistent Chow and Rice (1982, 334) report that firms that switched after a qualified opinion were no more likely to receive a clean opinion in the year after switch than others Using cases, Smith (1986) compared pre- and post-switch opinions, looking for conflicts between the predecessor and successor auditors He found that, in a sample of 139 firms that changed auditors after a qualified report, only five cases suggested the possibility of opinion shopping

In the absence of clear evidence of successful opinion shopping in the post-switch period, thA continuing concerns about such shopping expressed by the Securities and Exchange Commission suggest that a further examination of the prior-year opinion and switching would be useful to policy makers, practitioners, and academicians Previous studies that examine the opinion formulation process have not considered differences between switchers and non-switchers This study addresses the possibility that auditors apply relatively stricter interpretations of accounting and auditing issues for certain clients, in response to which these clients switch auditors in the hopes of securing a less conservative auditor The hypothesis tested is that, given a set of observed client financial characteristics, switchers receive more conservative treatment in the audit opinion decision than non-switchers While relatively more conservative treatment will cause switchers

to receive qualified opinions more often, it is the differential treatment rather than the issuance

of the qualified opinion that triggers the switch

202 The Accounting Review, January 1994

II Empirical Approach Previous work on audit qualifications (e.g., Bell and Tabor 1991; Dopuch et al 1987; Mutchler 1985) has used a binary dependent variable to represent qualified and clean opinions

In this article, opinions are classified into three categories of increasing seriousness: unqualified, asset realization (including litigation), and going concern The audit opinion formulation process

is modeled using ordered probit This approach has the advantage of allowing measurement of movements between qualification levels that would be obscured in a qualified/clean dependent variable It also allows for a measure of auditor conservatism in terms of the threshold values, after controlling for observed differences in the financial and market conditions of client furns Let ye denote the auditor's judgment of the client's (observable) condition, specified as a function of a vector of financial and market variables X and a random error Ec Thus for client i,

The components of Xi are discussed in detail in the next section The auditor applies threshold values in determining the qualification Let ye take values between y and y Denoting threshold values by g 1 and g2' the auditor's choice of opinions are unqualified, asset realization, and going concern when ye falls in the ranges (y, Q), (Ro, g) and (g2, y) respectively

Let (4yS' I2SW) and (11NSW, INsw) denote the thresholds of switchers and non-switchers respectively, and let SWITCH be a dummy that takes the value of one if an observation is a prospective switcher, and zero otherwise The possibility of different thresholds for the two groups is taken into account by specifying the threshold values g, and i2 as:

M1 = 91NSW + (4u91 - g1NSW) * SWITCH, and J2 = t2NSW + (2SW- U2NSW) * SWITCH (2) For the purpose of identification the first threshold for non-switchers, I1NSW, equals zero The model is an extended form of the standard ordered probit model.2 (See Dietrich and Kaplan 1982 and McKelvey and Zavoina 1975 for detailed descriptions of the standard ordered model.) The parameters to be estimated are ,1SW 92SW' g2Nsw and P/ The test of the hypothesis that switchers receive more conservative treatment than non-switchers is that [tysw - gk1NSW < 0 and

III Data and Variables The primary data source is the Disclosure Inc database An initial sample of companies was drawn for the years 1986, 1987 and 1988.3 The process of moving from the initial collection of data to the final sample involved several steps First, only companies with data on the CRSP tapes (New York, American and NASDAQ daily files) were retained Market variables were con- structed using data from CRSP, but companies with price data for less than 70 trading days were deleted Second, financial service companies (SIC Codes 6000-6999) were excluded because 2The log likelihood function 11 for a sample of size N is given by:

L*= EZil In[O(ulsw * SWITCHi - Xip)] + Zi2 ln[(L)(P2NSW + (M2SW -

SWITCHi - Xip)] + Zi3 11 - D(P2NSW + (P2SW - P2NSW) * SWITCHi - Xip)]}

where ZQ = 1, if OPINION, = j, zero otherwise, and j=1,2,3

3The initial list of companies was drawn from March 1988 to March 1990 data discs

Krishnan-Auditor Switching and Conservatism 203

variables used in studies such as this one may be inappropriate for these firms (Dopuch et al 1987, 434) Third, companies that filed for bankruptcy in the year before switch were dropped Finally, some observations were deleted because of incomplete data The main analysis is based on 2,989 observations for 1986 and 1987, while the 1988 data, comprising 808 companies, were retained

as a time series holdout sample to test the predictive ability of the model

The dependent variable, OPINION, represents the auditor's opinion in the year before the switch It takes the value 1 for unqualified opinions (U), 2 for asset realization (including litigation) opinions (AR), and 3 for going concern opinions (GC).4 Consistency exceptions for voluntary and mandatory accounting changes are included in the unqualified category Litigation qualifications include tax uncertainties The asset realization category includes future financing opinions in which the possible violation of the going concern assumption is not stated, and multiple qualifications involving litigation and asset realization Multiple qualifications arising

in conjunction with the going concern opinion are included in the going concern category The switching variable, SWITCH, denoting the prospective switching status of the firm, is coded one if the firm changed auditors in the year following the issuance of the opinion, zero otherwise Prospective switchers were identified by comparing the current year auditor with that

in the next year.5 The switching rate in the sample is 6.6 percent

Since this study focuses on the different treatment of switchers and non-switchers, the choice

of independent variables is based on previous work, especially that of Dopuch et al (1987) The variables consist of a set of financial and market variables, and controls representing auditor type, time listed, and industry affiliation Ratios of receivables and inventory to assets are used to capture high risk accounts Firm size (measured by the logarithm of assets) and a dummy for operating losses are included as proxies for possible financial distress The liabilities/assets ratio measures leverage Beta and residual standard deviation of returns from the standard market model are used as measures of systematic and fin-specific risks, respectively The firm's performance relative to the market is measured by its returns minus market return

Descriptive statistics for the full sample and for qualification and switching groups are presented in table 1 The frequency of switching increases significantly as one moves from the unqualified opinion to the going concern opinion Among the explanatory variables, the incidence

of loss and the residual standard deviation of returns increase, on average, with the seriousness

of the qualification The liabilities/assets ratio and firm size do not differ across the unqualified and asset realization qualification groups However, firms in the going concern category are significantly smaller and have higher liabilities/assets ratio than those in the asset realization category The receivables/assets and inventory/assets ratios are both lower, on average, for the asset realization category than for the unqualified group While the latter ratio is higher for the going concern group than the asset realization group, the former is similar over the two qualification groups One surprising finding is that clients' systematic risk, as measured by beta, decreases as one moves from the unqualified opinion to the asset realization opinion However, the going concern clients have a higher mean beta than that of the asset realization category Returns minus market return is not significantly different across the unqualified and asset realization groups, but is distinctly lower for the going concern category Firms with going

4The audit opinions were verified through comparisons with other sources, such as 10-K filings, annual reports, and COMPUSTAT Companies for which audit opinions could not be verified from a second source were dropped IThe switching status of each company was verified by examining other sources such as COMPUSTAT, WhoAudits America, 10-K and 8-K filings, and annual reports

204 The Accounting Review, January 1994

;~~~~~~~x CD Ct C CD o O CD ^t o CD oN It m CD r- CD oF

U~~~~~~~~~~ ~~~~r)W s k t t > -4 - m W) I -4 -4 m t r- M - CD O -4 t M 0 CD CD N CD N C\ C\ t- t 00 m It o)

=~~~~~~~~- -S 1, _s _st > t kn ? N oo CD ? t- CD t N N t m r- r

C) b = m X m ^

C '' oo OC - CD C CD OoC CD 00 C oo CD ND

~~~ o6N D 6546 6i6 6 6o6 e c v 66 66c o.6o6 66 (>

C

;~~~~~~ CD O M CD \ CD M CD oo o, t CD N CD CD o " o A o

~~~~~oO N^ 'o 00 ~o OO ~o 'o O\ '-o o oo?oU

o

to v~~~~~~~I 0 C\ CD "O o ?? N "t N 't 't N N t- ?? ?- V)V m cr

oZ r1C NCIt 00 W ) flO CD CD O) N CD N CD ? CI CD

-

t~~~~~~~~~~ S\ -O CD, O0 O) CD O 1, O \O N m 0, t- O C\ CD Ot O

~~~~~~~~~~C O oo 'I In In -4 eO 0 O CD W m - CD N C\ -4 kn C

o6 6 56666666c c c 4c 6 6 6 6 6p 5

~~ 0C~ 'n 'n'n c~~~ 0cm ON- D 0 oeq NtO \C 0 Om

c5C54 66 6 656 c 66 -4c 6 65c 5 5c6 66 o o66 dq

U 0

K-rishnan-Auditor Switching and Conservatism 205

0 ~~~~q

o

Po

1.0

P4

u~~~~~

> ~ ~ ~ ~ ~ *~

S

>lb

0

~~~ -~~~~ u ~~c - 0 CL

0

~~~~~~~~~~o ~ ~ ~ ~ ~ ~~r

0~~~~~~

-~~~~~~ ~ ~ ~ :a - - 4

44 O 0 0 O

0~~~~

C d *.H.c5E

206 The Accounting Review, January 1994

Table 2 Extended Ordered Probit Estimates

Threshold/ Expected Parameter

Variable Sign Estimate t-statistic

Threshold:

I1SW - g1NSW - 0.498a -4.604

2SW - 92NSW - 0.559a -4.234 Variable:

Industry Dummies C

Cli-squared 577.780

a One-tailed test (two-tailed otherwise)

b 1 if the observation belongs to year 1987

cOne digit industry dummyvariables were used Agriculture, forestry, andfishing and mining were combined because there were few observations in each group The following industry variables were significant: Manufacturing (-0.183,

t = -1.839), Transportation (0.636, t = 5.039), and Retail (-0.915, t = -3.677)

OPINION (Dependent Variable) = 1 if unqualified, 2 if asset realization or litigation, and 3 if going concern J1lNSW,

j2Nsw9 pls~w and p2sw are the first and second thresholds for non-switchers and switchers, respectively The first threshold for non-switchers is normalized to zero Other variables are defined in table 1

Krishnan-Auditor Switching and Conservatism 207

concern opinions have the lowest mean for time listed on an exchange However, time listed is higher on average for the asset realization category than for the unqualified group

The incidence of qualified opinions is significantly higher for switchers as compared to non- switchers Relative to non-switchers, switching clients tend to be smaller, and incur losses more often The residual standard deviation of returns is higher for switchers, while beta and time listed are significantly lower The three financial ratios and returns minus market return are not significantly different for the two groups

IV Empirical Results

Extended Ordered Probit Model Estimates

Table 2 presents the estimates of the extended ordered probit.6 The model chi-squared statistic has a p-value less than 0.001 The R2 (as defined in McKelvey and Zavoina 1975) for the model is 0.371 A Hausman specification test (Hausman 1978) was also conducted with the binary probit model defined as the alternative specification The results support the null hypothesis of

no misspecification (p = 0.247)

Among the explanatory variables, the liabilities/assets ratio, losses, residual standard deviation of returns, returns minus market return, and firm size have significant coefficients in the expected direction The coefficients of the receivables/assets ratio and time listed have unexpected signs.7 Beta and the inventory/assets ratio have statistically insignificant coeffi-

cients.8

The first threshold for non-switchers is normalized to zero The second threshold for non- switchers is significantly positive, suggesting the appropriateness of the ordered probit model The differences between the first thresholds of switchers andnon-switchers, ,t sw - ,t NSW and that between the second thresholds, 12SW - 2Nsw are both negative and statistically significant at below the one percent level.9

The finding that switchers have lower thresholds than non-switchers suggests that switching may be motivated by auditor conservatism, rather than by the issuance of a qualified opinion per

se Consider an alternative naive model in which clients are expected to switch in response to a qualified opinion, regardless of treatment In such a model, a client would switch auditors even when a qualified opinion is deserved By contrast, the model presented here suggests that clients would switch auditors only when they perceive the qualification issued to be "undeserved"; i.e., the result of a conservative treatment The client's deserved opinion can be approximated by using a simple ordered probitmodel with uniform thresholds for all clients According to the naive model, the switching rate for clients that deserve an asset realization or going concern opinion will

be higher than that for clients that deserve an unqualified opinion However, the conservatism

6Pooled data for 1986 and 1987 is used in this study because the numberof observations in each qualification category

is small in each year, particularly when categorized by switching status A likelihood ratio test did not reject the null hypothesis of no difference in slope coefficients across years (p = 0.431)

7Dopuch et al (1987) also report finding an opposite sign on their change in receivables/total assets ratio variable

8The correlation matrix (not reported) showed that the correlations between residual standard deviation of returns and (log) assets (-0.535) and between time listed and (log) assets (0.621) may result in multicollinearity Reestimation of the model after dropping residual standard deviation of returns and time listed (separately), however, showed no change in the qualitative results None of the other correlations exceeded 0.45

9 To check the robustness of the results, the model was reestimated on 100 randomly generated samples of 2,989 observations, each drawn with replacement from the estimation sample The differences in both thresholds for switchers and non-switchers were negative and statistically significant in each of these 100 samples

208 The Accounting Review, January 1994

Table 3 Impact of Conservatism on Switching Rates

Predicted

Opinion from Predicted Opinion

Ordered Probit from Extended Ordered Switching

Model a Probit Modelb Rate (%)

Asset Realization 11.36 Going Concern 22.22

Asset Realization 0.00 Going Concern 20.55

aDerived from a simple ordered probit (estimates not reported) using average thresholds for all clients

bDerived from the extended ordered probit reported in table 2, using conservative thresholds for all clients

model has the following implications: (1) among firms that deserve unqualified opinions, the switching rate will be higher for those that had actually received asset realization qualifications

or going concern opinions than for those that received unqualified opinions; (2) among clients that deserve the asset realization qualification, the switching rate will be higher for clients that had actually received the going concern opinion than for those that received the asset realization qualification.10 No predictions can be made for the going concern opinion, since the application

of conservative standards cannot yield a more serious outcome

Predicted opinions from the ordered probit (in which all clients face the same thresholds) and the extended ordered probit (using the conservative thresholds) were cross-classified and switching rates were computedfor the subgroups mentioned above The results are shown in table

3 Looking first at the naive model, the switching rates for firms that deserve unqualified, asset realization, and going concern opinions are 6.22 percent, 20.55 percent, and 8.33 percent respectively These rates support the hypothesis of the naive model: clients deserving qualified opinions switch more often However, the application of conservative thresholds reveals a different picture Among clients that deserve unqualified opinions but receive qualified opinions due to apparent conservative treatment, the switching rate is significantly higher (11.36 percent for asset realization and 22.22 percent for going concern) than for clients that both deserve and receive unqualified opinions (5.75 percent) Similarly, while firms that deserve an asset realization qualification have a switching rate of 20.55 percent, all of these clients receive going concern opinions because of the apparent conservative application of standards The pattern of

'Othe model also predicts that clients who receive unqualified opinions by the application of conservative standards have a higher probability of switching than those who receive unqualified opinions by the application of average standards However, such behavior cannot be tested, since the actual thresholds are not observed