Auditor Tenure and Quality of Financial Reporting

Abstract

Prior studies in general suggest a positive association between auditor tenure (the length of an auditor–firm relationship) and reporting quality (the informational content of reported earnings). In this study, we present evidence that the association is reversed when clients represent increased litigation risks to their auditors. Featuring downward biases in reported earnings as a measure of reporting quality that stem from auditors’ minimization of costs from potential audit errors, we argue that the magnitude of such downward bias decreases in auditors’ experiences with their clients (tenure improves reporting quality). Furthermore, we predict that longer auditor tenure is associated with larger downward bias for firms with increased audit risks (tenure impairs reporting quality). Using non-operating accruals as proxy for downward bias in reported earnings, we find robust empirical evidence in support of our prediction.

Keywords

auditor tenure bias in earnings litigation risks

Introduction

Financial reporting carries the ultimate objective of providing accurate financial information about a firm to its various stakeholders. As the firm’s auditor plays a critical role in the determination of reporting by attesting the reported accounting values, a natural question is whether auditor tenure—which helps the auditor acquire firm-specific knowledge and conduct effective audit examinations—affects positively or negatively the quality of financial reporting. Answers to the inquiry are of interest to regulators, as they bear directly on the recurring contentious debate of mandatory auditor rotations. More importantly, given the prevalent uses of Price to Earnings (P/E) ratios in financial analyses and investments, a clear understanding of the impact of auditor tenure on reporting quality is bound to improve economic efficiency. Motivated by this inquiry, literature on the nature of the tenure–quality relationship has been developed in the past decade. As suggested by Tepalagul and Lin (2015) in a comprehensive review, this literature in general documents a positive relationship between auditor tenure and reporting quality.¹

More recent studies have emphasized the multifacets of the tenure–quality relationship and explored factors susceptible to affecting the positive relationship established by prior research. For instance, Gul, Fung, and Jaggi (2009) find that the association between auditor tenure and earning quality is weaker for firms audited by industry specialists compared with non-specialists. Li (2010) finds that a positive relationship exists only for firms that represent great client importance to auditors, as a negative relationship emerges for firms that represent no such client importance. These additional but apparently conflicting evidences suggest that a closer examination of the relationship between auditor tenure and reporting quality is warranted. Continuing along this direction, we propose, in this study, a new analysis that highlights auditors’ litigation risks in examining the relationship, and provide new evidence on how auditor tenure affects reporting quality.

Our analyses are based on two generally accepted views: First, auditors with greater client-specific knowledge can estimate more accurately firms’ accounting values (e.g., Carcello & Nagy, 2004; Gul et al., 2009; Libby & Frederick, 1990; Myers, Myers, & Omer, 2003). Second, costs incurred to auditors from overstatement errors are higher than those from understatement errors (e.g., Kellogg, 1984; Kim, Chung, & Firth, 2003; Lys & Watts, 1994; St. Pierre & Anderson, 1984). Within this basic setup, we argue that after obtaining the estimate of an accounting value (e.g., earnings), an auditor would rather certify the reported value below her best estimate, creating thereby a downward bias in the audited reporting. Such bias obviously compromises the reporting quality, and we call it “auditor conservatism.”

Focusing on this downward bias, we first test whether the bias is smaller as an auditor gains more experience with her client. When the precision of the auditor’s estimate of an accounting value improves due to more client-specific experience, the risk of exposure to costs, of any misstatement errors, is reduced. Consequently, the auditor’s chosen downward bias in the reported accounting value is smaller. In other words, the client–auditor relation improves reporting quality, which is consistent with the widely documented positive association between auditor tenure and reporting quality. We then test whether the bias is larger with longer auditor tenure when a client-firm’s business suffers from deteriorations which represent increased litigation risks to its auditor. This is so because the same overstatement will entail higher expected costs to the auditor when a firm’s business deteriorates than when a firm’s business improves or remains unchanged. As a more experienced auditor is more able to detect such deterioration, the auditor anticipates greater expected costs of overstatement errors relative to understatement errors, and consequently incorporates larger downward bias in the reporting for the purpose of reducing her own costs. Thus, auditor tenure impairs reporting quality. This argument is novel and constitutes the focal point of our empirical investigations.

We use non-operating accruals, a construct of accruals proposed by Givoly and Hayn (2000), as our proxy for bias in reported earnings that is engendered by auditor conservatism. We use auditor tenure as the proxy for an auditor’s client-specific experience. As for variations in litigation risks, we use two alternative proxies. The first proxy is the change of a firm’s loss status during the most recent 2 adjacent years, which serves as a primary indicator of change in business conditions of a client and litigation exposure facing the auditor. The second proxy is derived from factor analysis using several variables related to a firm’s operating and financial conditions. This multidimensional measurement is intended to capture more aspects of a firm’s business risk. Once a factor has been identified, we use its change during the most recent 2 adjacent years as the proxy for the increased litigation exposure faced by the auditor.

Our sample period runs from 1988 to 2006. After controlling for firm age, cash flow, firm size, industry growth, firm capital structure, and firm performance, we find that, on average, auditor tenure is negatively associated with downward bias in non-operating accruals—confirming the widely documented positive relationship between audit tenure and reporting quality. But most interestingly, we find that when clients’ businesses deteriorate—which represents increased litigation risks to auditors—auditor tenure is actually positively associated with the downward bias in non-operating accruals. This empirical evidence is supportive of our argument on the relationship between auditor experience and downward bias in earnings attributable to auditor conservatism.

Our result of downward bias is reminiscent of accounting conservatism that has long been recognized as the prevailing feature of financial reporting.² While empirical evidence on accounting conservatism abounds, studies focusing on auditor conservatism are sparse. Moreover, there has been no attempt to characterize the tie between tenure-related auditor experience and auditor-induced reporting conservatism. Therefore, our study makes a contribution to the literature on conservatism in general by demonstrating the impacts of auditor tenure on the level of conservatism reflected in financial reporting.³

The remainder of our article is organized as follows: In Section “Hypothesis Development,” we discuss the relationship between auditor tenure and reporting quality, and formulate three testable hypotheses. In Section “Research Design,” we describe our research design. Our data and empirical results are presented in Section “Data and Empirical Findings.” In Section “Robustness Tests,” we outline robustness and sensitivity tests. Finally, in Section “Conclusion,” we summarize and conclude the article.

Hypothesis Development

Periodical financial reports strive to reflect the outcomes of firms’ continuous operations during fixed time intervals (accounting periods), but the transaction cycle of some operating activities often span more than one accounting period. Thus, the accounting representations for a particular period (e.g., earnings) involve estimations that may prove incorrect in hindsight—the accounting values will be either higher (overstatement errors) or lower (understatement errors) than the firms’ eventually realized values for the period. These misstatements can entail considerable costs to auditors who attest to the reporting. Costs of overstatement errors to auditors are usually potential legal liabilities and losses of reputation (e.g., Carcello & Palmrose, 1994), whereas costs of understatement errors to auditors are often strained relationships with clients and even terminations of audit engagements (Chow & Rice, 1982; Hubaid & Cooke, 2005; Krishnan, Krishnan, & Stephens, 1996). To minimize these costs, auditors have incentives to exercise their own influences on the reported earnings. Thus, the reported earnings of a firm are determined jointly by the firm and its auditor in equilibrium (Zhang, 1999).

When attesting the reported earnings, an auditor faces the prominent trade-off between the expected costs of overstatement and understatement errors, and this trade-off establishes a relationship between auditor tenure and reporting quality. Because the costs of an overstatement error of given size are, in general, larger than those of an understatement error of the same size (Kellogg, 1984; Kim et al., 2003; Lys & Watts, 1994; St. Pierre & Anderson, 1984), when an auditor seeks to minimize the expected costs associated with potential misstatements, this asymmetry in cost functions of the two misstatements will play a critical role in the auditor’s decision: It will induce the auditor to include a downward bias in the reported value, so that the endorsed earnings are smaller than the auditor’s best estimate of the earnings. If the reported earnings equaled the auditor’s best estimate, reporting would be unbiased.⁴

The downward bias is only relative to the auditor’s best possible estimate that is built, ex ante, into reported earnings, but this reported (and downward-biased) earnings may still turn out, ex post, to be either an overstatement or an understatement.⁵ The downward bias is driven by auditor’s lack of perfect knowledge of the true earnings and asymmetrically higher costs of audit overstatement error relative to understatement error. In other words, the auditor is conservative, and the conservatism has an impact on reporting quality.

To connect the downward bias to the generally accepted positive relationship between auditor tenure and reporting quality, we evaluate the influence of an auditor’s experience on the magnitude of the downward bias. When auditor tenure lengthens, an auditor acquires more client-specific experience, her accuracy of earnings estimations improves, which in turn reduces her expected costs associated with either type of misstatement and more so with overstatement.⁶ Consequently, her downward bias will be smaller. This outcome is in line with the generally claimed nature of relationship between auditor tenure and reporting quality; that is, longer auditor tenure improves reporting quality. In light of this result, we propose our first testable hypothesis as follows:

Hypothesis 1: Ceteris paribus, the size of the downward bias in financial reports is a decreasing function of an auditor’s experience with her client.

Hypothesis 1 suggests that the extent of auditor conservatism is reduced by the auditor’s experience. However, we argue that this relationship may not hold for all firms, in particular, for firms with increased audit risks. To corroborate our claim, we look into variations in operating and financial conditions of client-firms that represent changed audit risks, and to examine how the interaction of those changes and auditor tenure would affect the downward bias, and eventually, alter the nature of relationship between auditor tenure and reporting quality.

We posit that for the same earnings misstatement errors, associated costs to auditors should be higher when client-firms experience negative changes in business conditions than when firms are steady or experience positive changes. In particular, if a client-firm suffers a downturn in its underlying business but the auditor makes an overstating error, then it is more likely that the auditor will be sued and held liable by shareholders of the client-firm for the error, which increases the costs of overstatements. However, when a client-firm suffers a downturn and the auditor makes an understating error, it may also be more costly to the auditor, as managers of such client-firm simply penalize, more heavily, the auditor all the same.

Although costs of both types of misstatements may increase when client-firms undergo changes in their business conditions, it is most plausible that the increases in costs of overstatement errors are larger than the increases in costs of understatement errors when firms suffer downturns. Thus, the asymmetry in costs associated with the two types of misstatement errors becomes even more pronounced. With greater asymmetry, the net effect of changes in client business condition on auditor conservatism is easy to seize based on our characterization of the trade-offs facing auditors. That is, if downturns in client-firms’ business and financial situations increase asymmetry in costs of the two types of misstatement errors by auditors, we should expect auditors with longer tenure and greater client-specific experiences to better detect any downturns and respond with more severe downward biases for these clients.⁷ Thus, auditor tenure impairs reporting quality for firms hit by downturns in their business and financial conditions. In light of this new prediction, we propose our second testable hypothesis as follows:

Hypothesis 2: Ceteris paribus, the downward bias in financial report increases with an auditor’s experience with a client when the client’s business deteriorates.

In comparison with Hypothesis 1, Hypothesis 2 implies that the relationship between auditor tenure and reporting quality is reversed when clients become riskier to their auditors.

Hypothesis 1 may alternatively be supported by the argument that auditor tenure compromises auditor independence.⁸ For such an independence-based explanation to hold, two conditions are necessary: One, the client has an incentive to report earnings different from the ones assessed by the auditor; two, the auditor knowingly endorses client’s desired earnings. If auditors with longer tenure are more likely to accommodate clients’ reporting behaviors, then earnings manipulation would be positively associated with auditor tenure. This independence-based argument on the relationship between longer tenure and higher earnings may confound with our auditor experience-based explanation for Hypothesis 1 and audit risk-based explanation for Hypothesis 2.

To disentangle our explanations for the relationship between auditor tenure and reported earnings from the alternative independence-based argument, we propose the following hypothesis to test if auditors with longer tenure are more likely to accommodate firms’ manipulations to boost earnings.

Hypothesis 3: Ceteris paribus, the size of downward bias in financial reports decreases in auditor tenure for clients having explicit incentives to overreport.

Research Design

In this study, we use auditor tenure as proxy for auditors’ client-specific experience. Prior research suggests that experienced auditors have better developed knowledge structures.⁹Bonner and Lewis (1990) delineate their experience in more refined terms in positing that auditor expertise comprises three types of knowledge: general domain knowledge such as basic accounting and auditing knowledge; general business knowledge; and subspecialty knowledge such as knowledge on specific industry and business of the audit clients. While the first two types of knowledge are usually acquired through formal instruction and various personal experiences such as reading, subspecialty knowledge, however, is obtained only through specific experience with a client’s business, accounting system, and personnel. Auditors gain more subspecialty knowledge as the relationship with a specific client prolongs. In other words, auditors with longer tenure have greater knowledge of business and accounting practices of their clients. Choo (1996) suggests that, after analyzing auditor performance in a going-concern task, repeated exposure to a specific task may be a promising alternative proxy for auditors’ expertise.

In this article, we study how an auditor’s client-specific experience affects the bias in her client’s reported accounting value. Thus, auditor tenure serves as our most natural proxy for auditor’s client-specific experience due to its proximity to subspecialty knowledge of the auditor.

We use non-operating accruals to measure the impact that auditors exert on the reported earnings. As mentioned in Section “Hypothesis Development,” recognizing the joint determinations of earnings by both a client and its auditor, we can view total accruals as reflecting the combined influences on earnings by the two parties. We partition the total accruals into operating accruals and non-operating accruals, as in Givoly and Hayn (2000), and use the latter as proxy for the downward bias in earnings that is caused by auditor conservatism. The rationale for our choice is based on the following arguments: Auditors are likely to use accruals not related to operations to influence accounting values. To be specific, we note that auditors carry the responsibility to provide opinions on whether clients’ financial statements are presented in accordance with generally accepted accounting principles (GAAP), rather than be involved in clients’ routine or strategic operational management; thus, auditors can use GAAP—the main source of the auditors’ negotiation power—to incorporate their desired biases into the reported accounting values. For instance, auditors can encourage clients to select accounting principles that will lead to lower reported accounting values through conservative revenue recognitions, aggressive expense recognitions, lower asset valuation, and higher liability valuation. These choices and influences of auditors will be reflected in non-operating accruals which consist of items such as bad debt provisions; restructuring charges; changes in estimates; gains and losses on sales of assets; assets write-downs; accrual and capitalization of expenses; and deferral of revenues and their subsequent recognitions.

Accruals related to operations, however, include changes in current assets and liabilities that represent mainly the results of operating activities of a firm such as changes in growth, credit policies, inventory procurements, sales promotions, and payments of short-term liabilities. While the management of a client is able to manipulate accounting values through real operating activities, auditors have little influence to restrain or encourage such behaviors. Given this distinction between the two components of accruals, it is most relevant to use non-operating accruals, rather than total accruals, to measure the impact that auditors exert on the reported earnings.¹⁰

We calculate non-operating accruals as in Givoly and Hayn (2000):

N A C C_{i t} = T O T A L_A C C_{i t} - O P E R_A C C_{i t},

where NACC_it is non-operating accruals of firm i in year t; TOTAL_ACC is total accruals which is equal to income before extraordinary items minus cash flows from operation, scaled by total assets at the beginning of the year;¹¹OPER_ACC is operating accruals which is equal to changes in accounts receivable plus changes in inventory, plus changes in prepaid expense, minus changes in accounts payable, and minus changes in taxes payable, scaled by the beginning total assets.

In calculation of OPER_ACC, net balance of accounts receivable is used. As a result, bad debt provision related to accounts receivable is included in NACC. We use both current year’s non-operating accruals and 3-year average of non-operating accruals to measure bias in accounting value. We use 3-year average to mitigate measurement errors as in Givoly and Hayn (2000) and Ahmed and Duellman (2007). The average accruals over 3 years centered on the current year is denoted as AVERAGE_NACC.

Our basic regression model for testing Hypothesis 1 on the impact of an auditor’s experience on the bias in reported accounting value is as follows:

\begin{matrix} N A C C_{i t} = α + β_{1} T E N U R E_{i t} + β_{2} A G E + β_{3} Δ S A L E S_{i t} + β_{4} C F O + β_{5} S I Z E_{i t} + β_{6} S I Z {E_{i t}}^{2} \\ + β_{7} S I Z {E_{i t}}^{3} + β_{8} {R O A}_{i t} + β_{9} R O A_S D_{i t} + β_{10} L O S S_{i t} + β_{11} L E V + β_{12} C R_{i t} \\ + β_{13} I N D G R O W + β_{14} F I R M G R O W + β_{15} B I G + β_{16} S P E C I A L \\ + β_{17} T E N U R E_S P E C I A L + I n d u s t r y D u m m y + Y e a r D u m m y + ϵ_{i t}, \end{matrix}

where TENURE is the auditor tenure, measured by the number of consecutive years an auditor is engaged with the same client.

Hypothesis 1 implies that the coefficient of TENURE is positive. The control variables included in the regression model are similar to those used in Myers et al. (2003) and Gul et al. (2009). Firm age (AGE) is included to mitigate possible confounding effects of firm maturity on the effect of an auditor’s experience and to control for the differences in accruals because firms may have different life cycles (Anthony & Ramesh, 1992). Other variables documented as to be associated with accruals are also included. Specifically, we control for change in revenue (ΔSALES), as it is directly related to non-operating accruals; cash flow (CFO), because prior studies show an association between cash flow and accruals (e.g., Dechow, 1994); firm market value (SIZE, SIZE², and SIZE³) as larger firms have more stable accruals (Dechow & Dichev, 2002) and the size effect is non-linear (Gul et al., 2009). We also control for growth (INDGROW, FIRMGROW) as expanding or contracting industries and firms may have different accrual behavior (Gul et al., 2009); leverage (LEV, CR) as there are documented associations between leverage and accruals (e.g., Butler, Leone, & Willenborg, 2004); performance (LOSS, ROA, and ROA_SD) because it is documented that performance is associated with accruals non-linearly; and finally auditor size and specialization (BIG, SPECIAL, and TENURE_SPECIAL) as auditors’ specialization is found to be associated with the signed discretionary accruals (e.g., Gul et al., 2009). The measurements of these variables are as follows:

AGE is number of years the firm exists in the Compustat database;

ΔSALE is change of sales from year t − 1 to year t scaled by the beginning total assets;

CFO is cash flow from operations scaled by the beginning total assets;

SIZE is log transformation of the year-end market value of equity;

ROA is net income scaled by the beginning total assets;

ROA_SD is standard deviation of ROA for the last 3 years;

LOSS is an indicator variable equals 1 if net income is negative and 0 otherwise;

LEV is total liabilities to total assets ratio;

CR is ratio of current assets to current liabilities;

INDGROW is growth in sales in an industry;

FIRMGROW is growth in sales for a firm;

BIG is an indicator variable equals 1 when the auditor is one of the big 4 and 0 otherwise;

SPECIAL is an indicator variable equals 1 when the auditor audits the largest portion of total asset in an industry; and

TENURE_SPECIAL is an interactive variable equals the product of TENURE and SPECIAL.

Because accrual recognition varies across industries and time, we include two-digit standard industry code (SIC) to capture industry fixed effects and indicators for years to capture time fixed effects.

In testing Hypothesis 2, two alternative proxies for changes in a firm’s business conditions are used. First, we use the change of a firm’s loss status (ΔLOSS) in the last 2 years to capture the changed audit risks related to the firm’s business deterioration or improvement. In particular, the indicator variable ΔLOSS takes one of three values:

ΔLOSS = 1 if a firm’s net income is positive last year but turns to negative this year;

ΔLOSS = −1 if the firm’s net income is negative last year but turns to positive this year;

ΔLOSS = 0 if the loss status remain unchanged between the 2 years.

When a firm’s net income changes from positive to negative in the last 2 years, it implies that the firm’s business has deteriorated. Similarly, when a firm’s net income changes from negative to positive in the last 2 years, it implies that the firm’s business has improved. When the sign of the net income remains the same for the 2 years, it implies that the business situation is in a steady state.

To implement our identification mechanism that exploits variations in client-firms’ business conditions over time, we add an interactive term between auditor tenure (TENURE) and ΔLOSS as an independent variable to the basic regression model. This interactive term helps trace the root of the auditor conservatism (downward bias in earnings) to audit risks (increased costs of misstatement errors). We test Hypothesis 2 using this augmented regression model, where TENURE_ΔLOSS is the product of TENURE and ΔLOSS. Hypothesis 2 predicts that the coefficient of TENURE_ΔLOSS is negative.

Even though ΔLOSS captures an important aspect of change in a client’s business condition, it may not capture continuity of the changes and, more significantly, it may not capture changes in some other important aspects of the client’s operational and financial situations. To address this concern, we use factor analysis over a set of variables associated with clients’ operational and financial activities to generate an alternative proxy for change in business condition and audit risk. The factor analysis is over five variables: LOSS, −ROA, LEV, −CFO, and DSALE. The variables LOSS and LEV are the same as defined before. The variables −ROA and −CFO have the same absolute values of ROA and CFO but opposite signs. Finally, DSALE is a new variable that has a value of 1 if the change of sales (ΔSALE) is negative and 0 otherwise.

Based on the result of the factor analysis, we choose factors with eigenvalue higher than 1 to proxy for business condition and audit risk. The change of audit risk is the change in the value of these factors between the two most recent years, denoted as ΔRISK. The variable ΔRISK and the interactive term TENURE_ΔRISK are then added to the basic regression model in the same manner as ΔLOSS and TENURE_ΔLOSS, where TENURE_ΔRISK is the product of TENURE and ΔRISK. Hypothesis 2 predicts that the coefficient of TENURE_ΔRISK is negative.

To test Hypothesis 3, we identify firms that have strong incentives to boost earnings, such that the reported earnings deviate upward from the true earnings, and then test whether auditors with longer tenure are more likely to accommodate such deviations. Specifically, we use firms that make seasoned equity offerings in the period following the reporting in our tests as these firms have explicit incentives to manipulate and incorporate upward biases in earnings.¹² More upward bias with longer tenure would suggest that auditor tenure hinders auditors’ independence. Our design of using firms with seasoned equity offering can distinguish the effect of auditor experience from that of auditor independence on reported earnings quality.

When a firm issues seasoned stock, its auditor receives higher pressure exerted by firm, such that the reported value is less conservative; alternatively, the auditor faces greater litigation risks that can lead to more conservative report. The balance of these two opposite forces will lead to either a more or less conservative report. We cannot conclude that the auditor is independent if the reported value is less conservative. However, if the final report is more conservative or neutral, then we can conclude that the auditor does not lose independence under the pressure of the firm that issues seasoned stock.

To test for auditor independence, we add a variable FINANCE and an interactive variable TENURE_FINANCE to our regression models. The variable FINANCE is the measurement for the seasoned equity offering in the next period. The main measurement of FINANCE is the amount of seasoned equity offering in the next period scaled by total assets at the end of the current period. TENURE_FINANCE is the interactive term that equals the product of TENURE and FINANCE. Based on the findings in Rangan (1998); Teoh, Welch, and Wong (1998); and Shivakumar (2000), the coefficient of FINANCE should be positive. If auditors with longer tenure are more likely to succumb to the pressure from clients prior to the seasoned equity offerings through endorsing overstated earnings, the coefficient of TENURE_FINANCE should be positive. We also run the regressions with changes of FINANCE because it is more directly related to the changes of incentives. In this case, we use variable ΔFINANCE and interactive term TENURE_ΔFINANCE in the regression model, where ΔFINANCE is the difference in FINANCE between year t+ 1 and year t, and TENURE_ΔFINANCE is the product of TENURE and ΔFINANCE. If auditors with longer tenure are more likely to succumb to the pressure from clients prior to a seasoned equity offering, the coefficients of both TENURE_FINANCE and TENURE_Δ FINANCE should be positive.

Data and Empirical Findings

Sample Composition and Descriptive Statistics

The initial sample consists of all firms for the years 1988 to 2006 in the Compustat database. We exclude firms in the financial services industries (SIC between 6000-6999). The firm’s age and the auditor’s experience variables are calculated from all available data in the Compustat. A firm’s age is the number of years from the time it first appeared in the Compustat database to the current year, and an auditor’s experience is the number of years that the firm has retained the same auditor till the current year. The change of auditor code due to audit firm mergers is treated as a continuation of the predecessor auditor.

We delete firms with negative total assets, sales, debts, and market values of equity because such observations bring noises into the regressions. To mitigate problems caused by extreme observations in cash flows and accruals, we exclude observations in the top and bottom 0.5 percentile of the data. To avoid confounding effect of start-up firms on the effect of short auditor experiences with those particular firms, we drop observations of all firms in their first 5 years of existence. Equally, auditors whose tenure did not last longer than 5 years are also left out from the sample, because firms that switch auditors early in the relationship may be systematically different from other firms (Myers et al., 2003). Following Myers et al. (2003) and Gul et al. (2009), we eliminate firms that underwent merger and acquisitions, as Collins and Hribar (2002) show that estimating accruals for those firms is problematic. After imposing all the necessary requirements to calculate the average non-operating accruals and other control variables, we obtain a final sample of 22,572 firm-year observations.

Figure 1 illustrates the relationship between non-operating accruals and auditor tenure. The first graph (Panel A) is a scatter illustration of the relationship between the two variables. Two salient features emerge in the graph: One, the range of NACC across firms is reduced as auditor tenure increases; two, the range of NACC is reduced mainly due to a smaller negative NACC. When auditor tenure is longer than 17 years, even the largest negative NACC in the earnings is very modest.

Figure 1.

The graphic relationship between non-operating accruals and auditor tenure.

The second graph (Panel B) illustrates the relationship between auditor tenure and several statistics of NACC. Both the mean and the median of NACC are clearly increasing in auditor tenure. The 10th percentile of NACC has a very strong positive relationship with auditor tenure, while the 90th percentile does not show a systematic relationship with the tenure. The third graph (Panel C) illustrates statistics of NACC in time series, which shows no clear trend along the time for any of these statistics.

Panel B:

Graphic illustration of several NACC statistics and auditor tenure.

Panel C:

Graphic illustration of several NACC statistics in time series.

Panel A in Table 1 reports the descriptive statistics for non-operating accruals, auditor experience, and the control variables for the sample. For our two key variables, the average of non-operating accruals (proxy for auditor conservatism) is −8.3% of the total assets, and the average of auditor tenure (proxy for client-specific experience) is 10.34 years. The average age of the firms is slightly over 27 years. Panel B in Table 1 reports several statistics of non-operating accruals by auditor tenure. In general, the mean and median of non-operating accruals are negative, and the absolute size of non-operating accruals decreases when auditor tenure lengthens. Furthermore, the standard deviation of non-operating accruals across all firms decreases as the length of auditor tenure increases. These two univariate results indicate that the auditors with greater experience make more accurate estimates.

Table 1.

Descriptive Statistics.

Panel A: Sample Distribution of Variables.
Variable	M	SD	10th percentile	25th percentile	Median	75th percentile	90th percentile
NACC	−0.083	0.113	−0.179	−0.109	−0.065	−0.033	−0.001
AVERAGE_NACC	−0.088	0.090	−0.176	−0.113	−0.070	−0.040	−0.015
SD_NACC	0.094	0.121	0.013	0.026	0.054	0.113	0.217
TENURE	10.338	7.199	3	5	8	15	21
AGE	27.177	12.814	11	16	26	35	46
ΔSALE	0.107	0.452	−0.154	−0.022	0.058	0.195	0.412
CFO	0.053	0.194	−0.116	0.011	0.080	0.142	0.214
SIZE	4.996	2.270	1.994	3.316	4.917	6.568	8.042
ROA	−0.014	0.229	−0.219	−0.034	0.035	0.083	0.145
LEV	0.536	0.456	0.175	0.310	0.504	0.668	0.847
CURR	2.871	6.623	0.781	1.222	1.906	3.095	5.262
INDGROW	1.091	0.189	0.946	1.014	1.074	1.143	1.232
FIRMGROW	1.276	8.149	0.846	0.973	1.066	1.190	1.411
Categorical variable (%)
BIG	84.5
LOSS	32
SPECIAL	23
Observations	22,572

Panel B: Descriptive Statistics for Non-Operating Accruals by Auditor Tenure.
Auditor tenure	Observations	M	SD	10th percentile	Median	90th percentile
1	657	−0.097	0.151	−0.218	−0.065	0.002
2	774	−0.091	0.131	−0.199	−0.069	−0.003
3	1,839	−0.097	0.152	−0.219	−0.068	0.007
4	2,107	−0.091	0.125	−0.204	−0.068	−0.001
5	1,949	−0.091	0.131	−0.202	−0.071	0.005
6	1,751	−0.094	0.126	−0.222	−0.070	0.004
7	1,493	−0.093	0.124	−0.202	−0.069	−0.002
8	1,269	−0.086	0.114	−0.190	−0.069	0.001
9	1,131	−0.080	0.102	−0.181	−0.063	−0.004
10	981	−0.085	0.117	−0.176	−0.066	−0.005
11	810	−0.083	0.105	−0.173	−0.067	0.000
12	691	−0.075	0.089	−0.172	−0.063	0.004
13	582	−0.075	0.101	−0.159	−0.066	0.003
14	551	−0.083	0.102	−0.166	−0.067	−0.012
15	429	−0.074	0.125	−0.155	−0.061	0.008
16	449	−0.069	0.082	−0.146	−0.061	0.000
17	707	−0.068	0.071	−0.136	−0.063	−0.004
18	670	−0.068	0.067	−0.131	−0.061	−0.012
19	592	−0.064	0.068	−0.124	−0.061	−0.004
20	505	−0.062	0.065	−0.121	−0.057	−0.006
21	448	−0.067	0.059	−0.135	−0.065	−0.011
22	387	−0.064	0.065	−0.128	−0.056	−0.007
23	343	−0.062	0.054	−0.124	−0.056	−0.004
24	299	−0.060	0.076	−0.123	−0.056	0.008
25	257	−0.064	0.068	−0.136	−0.056	−0.011
26	211	−0.056	0.051	−0.121	−0.053	−0.006
27	176	−0.057	0.064	−0.120	−0.050	−0.002
28	158	−0.055	0.071	−0.130	−0.050	0.007
29	114	−0.067	0.065	−0.118	−0.057	−0.008
30	101	−0.065	0.047	−0.121	−0.058	−0.018
31	83	−0.060	0.060	−0.122	−0.054	0.009
32	58	−0.044	0.046	−0.088	−0.042	−0.010
Total	22,572	−0.083	0.113	−0.179	−0.065	−0.001

Note. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 22,572 firm-year observations.

Table 2 presents pairwise correlations between the main variables. Note that auditor tenure and firm age are highly correlated, with a correlation coefficient of .583. We will address this issue in the section of sensitivity analyses.

Table 2.

Pairwise Correlations Among Variables.

Variable	NACC	NACC_SD	TENURE	AGE	ΔSALE	CFO	SIZE	ROA	LEV	CURR	INDGROW	FIRMGROW	LOSS	BIG	SPECIAL
NACC	1.000
NACC_SD	−.309**	1.000
TENURE	.103**	−.159**	1.000
AGE	.139**	−.182**	.583**	1.000
ΔSALE	−.114**	.040**	−.063**	−.088**	1.000
CFO	.066**	−.287**	.147**	.154**	.055**	1.000
SIZE	.102**	−.245**	.284**	.304**	−.002	.280**	1.000
ROA	.458**	−.376**	.143**	.160**	.162**	.801**	.264**	1.000
LEV	−.147**	.102**	.009	.073**	−.054**	−.060**	.069**	−.158**	1.000
CURR	.044**	.012	−.039**	−.073**	−.005	−.074**	−.115**	−.027**	−.174**	1.000
INDGROW	−.030**	.013**	−.041**	−.032**	.036**	.015**	−.049**	.012	.009	−.006	1.000
FIRMGROW	−.055**	.051**	−.016**	−.017**	.032**	−.043**	−.019**	−.056**	−.016**	.011	.000	1.000
LOSS	−.224**	.262**	−.149**	−.185**	−.180**	−.481**	−.268**	−.618**	.135**	.046**	−.012	.020**	1.000
BIG	.040**	−.106**	.130**	.061**	.001	.090**	.413**	.079**	−.005	−.027**	−.012	.003	−.077**	1.000
SPECIAL	.019**	−.050**	.079**	.047**	−.011	.050**	.208**	.044**	.019**	−.022**	−.001	−.002	−.038**	.232**	1.000

Indicates significance at the .05 level. See appendix for variable descriptions.

The Effect of Auditors’ Experience on the Biases in Earnings

Table 3 presents the results of the basic regression model where the dependent variable is, respectively, current non-operating accruals, average non-operating accruals for 3 years centered on the current year, and standard deviation of non-operating accruals calculated over the 3 years. From the table, coefficient of TENURE loads positively for NACC and AVERAGE_NACC models (t = 2.52 and 3.57, respectively), consistent with the univariate results presented in Panel B of Table 1. These results suggest that auditors, on average, endorse financial statements with higher non-operating accruals when the auditors have gained more experience with the clients. As NACC represents auditor’s downward bias, the positive coefficient on TENURE implies that the reported earnings are less downwardly biased when auditors are more familiar with their clients’ business, operations, and accounting policies. The results are consistent with Hypothesis 1.

Table 3.

The Relationship Between Auditor Tenure and Non-Operating Accruals.

Variables	Dependent variable
	NACC		AVERAGE_NACC		NACC_SD
	Coefficient	t statistics	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0617***	−3.49	−0.1245***	−5.22	0.1574***	4.41
TENURE	0.0003**	2.52	0.0004***	3.57	−0.0008***	−4.51
AGE	0.0004***	6.05	0.0005***	7.47	−0.0003***	−3.23
ΔSALE	−0.0602***	−5.26	−0.0379***	−5.38	0.0228***	4.48
CFO	−0.5047***	−32.29	−0.2631***	−24.47	0.0485***	3.94
SIZE	−0.0009	−0.23	0.0095**	2.21	−0.0130**	−2.21
SIZE ²	0.0001	0.11	−0.0018**	−2.27	0.0012	1.17
SIZE ³	0.0000	0.16	0.0001**	2.50	−0.0001	−0.91
ROA	0.6085***	33.76	0.3404***	23.23	−0.1741***	−10.43
ROA_SD	−0.0117	−1.64	−0.0279*	−1.92	0.0650**	2.09
LOSS	0.0258***	9.34	0.0166***	7.04	0.0004	0.13
LEV	−0.0078	−1.30	−0.0086	−1.11	0.0090	1.04
CR	0.0001	1.52	0.0003**	2.12	−0.0002	−1.41
INDGROW	−0.0025	−0.75	−0.0020	−0.69	−0.0024	−0.58
FIRMGROW	−0.0002*	−1.70	−0.0001*	−1.83	0.0003***	2.76
BIG	0.0051**	2.09	−0.0038	−1.39	−0.0032	−0.82
SPECIAL	0.0007	0.31	0.0013	0.53	−0.0045	−1.17
TENURE_SPECIAL	−0.0001	−0.38	0.0000	0.20	0.0002	0.70
Year Fixed Effect	Yes		Yes		Yes
Industry Fixed Effect	Yes		Yes		Yes
Observations	22,572		22,572		22,572
Overall R²	.55		.40		.251

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals, 3-year average non-operating accruals, and standard deviation of non-operating accruals. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 22,572 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

Although the coefficient of TENURE seems small with a value of 0.0003, the economic impact of auditor tenure on reported earnings is still significant. To intuit, notice that the average tenure in the sample is 10 years. If we compare a new auditor of 0-year tenure with an auditor of 10-year tenure (all other things equal), the difference in NACC between these two auditors is about 0.0003 × 10 = 0.3% of total assets. Considering that the average ROA is (−1.4%) in the sample, the value of 0.3% is quite significant.

Non-operating accruals are also significantly affected by LOSS and all the control variables AGE, ΔSALE, CFO, ROA, except for auditor industry specialization, which may suggest that firm-specific knowledge includes industry knowledge, such that the impact of industry knowledge is captured by variable TENURE.

Note that SIZE effect is not significant. This can be explained by the fact that our dependent variable, NACC, is measured in actual (signed) value. The means of NACCs are actually similar for large and small firms. Thus, the NACC is not much affected by firm size, although the standard deviation of NACC is a lot higher for smaller firms.

We also check whether the standard deviation of non-operating accruals decreases in auditor experience by regressing NACC_SD as a dependent variable. The result shows that the coefficient of TENURE is significantly negative (t = −4.51), indicating that the dispersion of auditors’ estimates of earnings is reduced by auditors’ client-specific experience, which is also consistent with the univariate results presented in Panel B of Table 1. As auditors gain more experience, they make more accurate estimates of earnings, and their desired downward biases in earnings become smaller. As we feature downward bias—a signed measure—to characterize reporting quality, we will focus on NACC rather than the unsigned measure of NACC_SD in the following tests.

The Effect of Auditors’ Experience on the Biases in Earnings for Firms With Changed Business Conditions

In Tables 4 and 6, we report results on the impacts of changes in audit risks induced by transitions in firms’ business conditions on the downward biases in earnings. We use two respective risk proxies: The first proxy is the change in a firm’s loss status, and the second proxy is the change in a risk factor derived from a factor analysis using several variables relating to a firm’s operational and financial activities.

Table 4.

The Relationship Between Auditor Tenure and Non-Operating Accruals Controlling for Change in Client Loss Status.

Variables	Dependent variable
	NACC		AVERAGE_NACC
	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0646***	−3.69	−0.1194***	−5.12
TENURE	0.0003***	2.60	0.0004***	3.35
TENURE_ΔLOSS	−0.0005***	−3.22	−0.0005***	−4.12
ΔLOSS	0.0008	0.28	0.0177***	7.40
AGE	0.0004***	6.16	0.0005***	7.41
ΔSALE	−0.0602***	−5.25	−0.0377***	−5.41
CFO	−0.5037***	−32.35	−0.2664***	−24.87
SIZE	−0.0008	−0.20	0.0089**	2.08
SIZE ²	0.0001	0.11	−0.0018**	−2.26
SIZE ³	0.0000	0.14	0.0001**	2.54
ROA	0.6098***	33.80	0.3389***	23.32
ROA_SD	−0.0119	−1.64	−0.0273*	−1.92
LOSS	0.0289***	9.21	0.0086***	2.96
LEV	−0.0081	−1.32	−0.0076	−1.02
CR	0.0001	1.47	0.0003**	2.19
INDGROW	−0.0027	−0.81	−0.0017	−0.61
FIRMGROW	−0.0002*	−1.71	−0.0001*	−1.76
BIG	0.0048**	1.99	−0.0031	−1.13
SPECIAL	0.0008	0.34	0.0011	0.46
TENURE_SPECIAL	−0.0001	−0.42	0.0000	0.24
Year Fixed Effect	Yes		Yes
Industry Fixed Effect	Yes		Yes
Observations	22,572		22,572
Overall R²	.55		.40

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals or 3-year average non-operating accruals, using change in loss status to proxy change in client’s risk. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 22,572 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

Table 4 presents the results of the augmented regression models after adding the variable, ΔLOSS, and its interactive term with the variable TENURE, TENURE_ΔLOSS. In Table 4, the coefficient of TENURE is positive for both models (t = 2.60 and 3.35, respectively), which is consistent with our previous results. Coefficient of the interactive term, TENURE_ΔLOSS, is negative in both models (t = −3.22 and −4.12, respectively). Such results of negative coefficient suggest that more experienced auditors incorporate larger downward biases into earnings than less experienced auditors do when firms are suffering from business deteriorations, while more experienced auditors reduce the downward biases more than less experienced auditors do when firms are experiencing business improvements. These results are supportive of our Hypothesis 2.

To mitigate the concern that a single indicator variable, ΔLOSS, may not adequately measure the changes in clients’ underlying business conditions, we employ factor analysis over a set of variables relating to clients’ operating and financial aspects to generate an index variable to proxy business conditions. Table 5 presents the results of the factor analysis. From the table, the eigenvalues of all factors except for Factor1 are well below 1. All five variables are mainly loaded onto Factor1, particularly those variables associated with firm performance. Therefore, we only need to use Factor1 to measure a firm’s risk, which we denote as RISK.

Table 5.

Generating Variable for Firm Risks Using Factor Analysis.

Panel A: Factor Loading.
Variable	Factor1	Factor2	Factor3	Factor4
LOSS	0.68	0.31	−0.04	−0.03
−ROA	0.95	−0.09	0.11	−0.01
LEV	0.15	0.14	0.24	0.02
−CFO	0.83	−0.29	−0.10	−0.02
DSALE	0.26	0.35	−0.12	−0.03

Panel B: Factor Analysis.
Factor	Eigenvalue	Proportion	Cumulative
Factor1	2.15	.83	0.83
Factor2	0.33	.13	0.96
Factor3	0.10	.04	1.00
Factor4	0.00	.00	1.00
Factor5	−0.00	.00	1.00

Table 6 presents the results of the augmented regression models after adding the variable, ΔRISK, and its interactive term with TENURE, TENURE_ΔRISK, where ΔRISK is the change in RISK between the previous year and the current year, and TENURE_ΔRISK is the product of TENURE and ΔRISK. The results in Table 6 show positive coefficient for TENURE and negative coefficient for TENURE_ΔRISK, consistent with the results in Table 4. Thus, longer auditor tenure increases the downward biases in earnings for firms whose businesses deteriorate which represent greater audit risks, and again, Hypothesis 2 is supported.

Table 6.

The Relationship Between Auditor Tenure and Non-Operating Accruals Controlling for Changes in Client Risks.

Variables	Dependent variable
	NACC		AVERAGE_NACC
	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0561***	−3.19	−0.1215***	−5.69
TENURE	0.0003***	2.60	0.0004***	3.23
TENURE_ΔRISK	−0.0005***	−3.11	−0.0007***	−5.52
ΔRISK	0.0020	0.87	0.0188***	9.20
AGE	0.0004***	6.12	0.0005***	7.54
ΔSALE	−0.0606***	−5.15	−0.0344***	−5.33
CFO	−0.5049***	−32.21	−0.2615***	−24.87
SIZE	−0.0008	−0.19	0.0076*	1.81
SIZE ²	0.0001	0.11	−0.0016**	−2.13
SIZE ³	0.0000	0.15	0.0001**	2.49
ROA	0.6084***	34.49	0.3526***	24.64
ROA_SD	−0.0120	−1.63	−0.0238*	−1.88
LOSS	0.0273***	9.95	0.0138***	6.06
LEV	−0.0080	−1.32	−0.0077	−1.11
CR	0.0001	1.50	0.0003**	2.21
INDGROW	−0.0029	−0.88	−0.0010	−0.36
FIRMGROW	−0.0002*	−1.76	0.0000	0.27
BIG	0.0049**	2.00	−0.0023	−0.86
SPECIAL	0.0007	0.31	0.0012	0.50
TENURE_SPECIAL	−0.0001	−0.39	0.0000	0.17
Year Fixed Effect	Yes		Yes
Industry Fixed Effect	Yes		Yes
Observations	22,572		22,572
Overall R²	.55		.42

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals or 3-year average non-operating accruals, using change in a risk factor derived from factor analysis to proxy for change in client’s risk. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 22,572 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

The Effect of Auditor Tenure on Auditor Independence

Table 7 presents the results of regressions with variables for seasoned financing. In the table, both coefficients of TENURE_FINANCE and TENURE_ΔFINANCE are insignificant (t = 1.28 and −1.38). These results do not support Hypothesis 3, and suggest that the positive relation between lengthened auditor tenure and a smaller downward bias in earnings is not caused by reduced auditor independence, implying that longer auditor tenure does not lead to more accommodating behaviors from auditors, and our results for Hypotheses 1 and 2 are not attributable to auditor independence. We also measure FINANCE in an indicator variable which equals 1 if the seasoned equity offering in the next period is 5% or more of the current total assets at year-end, and 0 otherwise, the results remain unchanged.

Table 7.

The Relationship Between Auditor Tenure and Non-Operating Accruals Controlling for Seasoned Equity Offerings.

Variables	Dependent variable: NACC
	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0603***	−3.43	−0.0627***	−3.56
TENURE	0.0002**	2.23	0.0003***	2.64
ENURE_FINANCE	0.0007	1.28
FINANCE	−0.0063	−1.10
TENURE_ΔFINANCE			−0.0011	−1.38
ΔFINANCE			0.0093*	1.81
AGE	0.0004***	6.07	0.0004***	6.06
ΔSALE	−0.0601***	−5.25	−0.0602***	−5.26
CFO	−0.5052***	−32.73	−0.5049***	−32.30
SIZE	−0.0012	−0.30	−0.0006	−0.16
SIZE ²	0.0001	0.17	0.0000	0.05
SIZE ³	0.0000	0.10	0.0000	0.21
ROA	0.6081***	33.46	0.6086***	33.76
ROA_SD	−0.0116	−1.62	−0.0117	−1.63
LOSS	0.0257***	9.32	0.0258***	9.34
LEV	−0.0079	−1.32	−0.0078	−1.30
CR	0.0001	1.48	0.0001	1.59
INDGROW	−0.0025	−0.75	−0.0025	−0.75
FIRMGROW	−0.0002*	−1.70	−0.0002*	−1.70
BIG	0.0051**	2.11	0.0051**	2.09
SPEICAL	0.0007	0.29	0.0008	0.34
TENURE_SPECIAL	−0.0001	−0.34	−0.0001	−0.43
Year Fixed Effect	Yes		Yes
Industry Fixed Effect	Yes		Yes
Observations	22,572		22,572
Overall R²	.55		.55

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals, controlling for firms issuing equity financing. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 22,572 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

Robustness Tests

In this section, we discuss and conduct additional tests to see whether the main results presented in previous section are robust. First, we calculate the non-operating accruals with the following alterations. We exclude depreciation expenses when calculating total accruals and non-operating accruals. We use net income rather than income before extraordinary items to calculate total accruals and non-operating accruals. We further use abnormal non-operating accruals to proxy for bias in earnings. These alternative measurements do not alter our inferences as presented in Tables 3 through 6 (not tabulated).

We then use different measures for auditor tenure. We use an indicator variable that equals 1 if the term of an auditor–client relation is at least k years (in our tests, the values assigned to k include 4, 5, 6, 7, 8, and 9) and 0 otherwise. We also use logarithm-transformed value of auditor tenure. Again, all main inferences are unaffected when these alternative measures are used (not tabulated).

The dependent variable of downward bias and the independent variable of auditor tenure in our model can be endogenously determined. On one hand, auditors may want to continue their engagements with clients that have high-quality earnings and drop those engaging in earnings management or adopting poor accounting practices. On the other hand, clients reporting high-quality earnings may have incentives to retain their auditors, while clients with low-quality earnings may shop for new auditors. To address such endogeneity, we use a two-stage least-squares regression to obtain consistent estimators. At the first stage, we obtain the prediction value for auditor tenure using an auditor tenure estimation model similar to that in Gul et al. (2009), and then, at the second stage, we substitute the original tenure with the predicted value from the model for audit tenure. The regressions in the second stage produce similar inferences to those shown in Tables 3 through 6 (not tabulated).¹³

We rerun all the regressions using several alternative samples. The alternative samples are generated by combinations of the following variations: Observations of all firms in their first 5 years of existence and of auditors who did not last longer than 5 years are both included. The observations of firms in their first 10 years of existence are excluded. All observations of non-big 4 auditors are excluded. The purpose of using these alternative samples is to determine whether our results are affected by new firms and to avoid confounding effects of systematic differential client characteristics on the effect of auditors’ experiences, as small auditors may have different clientele (Lawrence, Minutti-Meza, & Zhang, 2011). The inferences documented in Tables 3 to 6 continue to hold for all these alternative samples.

As seen in Table 2, auditor tenure and firm age are highly correlated. This correlation is inherent as long auditor tenure is conditional on the long existence of a client, that is, the client must have a long life for the auditor tenure to be long. The high correlation between auditor tenure and firm age not only creates a multicollinearity problem, but more significantly it also makes the effects of firm age and auditor tenure on all dependent variables similar. Thus, it is unclear whether our estimated effects of auditor tenure on earnings reporting are truly due to auditor tenure rather than due to firm age. To mitigate this concern, we repeat all the tests using observations from all firms after their first 15 years of existence, so as to better isolate age effect from tenure effect on NACC. The rationale is that while earnings quality is greatly influenced by age in a firm’s growing period, it is much less so for established and mature firms. Thus, focusing on more mature stage of firms will mitigate the confounding effect of age on the impact of tenure on NACC.

Table 8 contains the regression results using the subsample excluding firms in their first 15 years of existence. In Panel A for our basic regression model, coefficients of TENURE continue to be significantly positive. In Panel B for our augmented regression model, coefficients of TENURE_ΔLOSS and TENURE_ΔRISK remain significantly negative. These two results are again consistent with Hypotheses 1 and 2, respectively. Thus, the inferences in Tables 3 and 6 are not altered by such restriction on firm age.

Table 8.

Panel A: The Relationship Between Auditor Tenure and Non-Operating Accruals With Firm Age Longer Than 15 Years.
Variables	Dependent variable
	NACC		AVERAGE_NACC		NACC_SD
	Coefficient	t statistics	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0507**	−2.51	−0.1234***	−9.35	0.1566***	4.32
TENURE	0.0003**	2.45	0.0004***	3.45	−0.0008***	−4.46
AGE	0.0004***	6.05	0.0006***	7.16	−0.0002*	−1.85
ΔSALE	−0.0508***	−4.07	−0.0335***	−4.19	0.0185***	3.40
CFO	−0.5206***	−27.92	−0.2719***	−21.16	0.0478***	3.33
SIZE	0.0017	0.42	0.0118***	2.65	−0.0162***	−2.71
SIZE ²	−0.0003	0.39	−0.0022***	−2.72	0.0017	1.64
SIZE ³	0.0000	0.55	0.0001***	2.92	−0.0001	−1.36
ROA	0.6157***	27.47	0.3558***	20.04	−0.1769***	−9.75
ROA_SD	−0.0083	−1.40	−0.0206*	−1.65	0.0512*	1.84
LOSS	0.0248***	7.77	0.0175***	6.39	0.0019	0.63
LEV	−0.0068	−1.12	−0.0079	−1.02	0.0090	1.08
CR	0.0005**	2.29	0.0006*	1.89	−0.0003	−0.64
INDGROW	−0.0022	−0.61	−0.0026	−0.79	0.0036	0.83
FIRMGROW	−0.0005***	−3.18	−0.0001	−1.44	0.0005***	2.85
BIG	0.0027	1.01	−0.0038	−1.26	−0.0030	−0.70
SPECIAL	0.0022	0.92	0.0051*	1.85	−0.0107	−2.53
TENURE_SPECIAL	−0.0001	−0.59	−0.0001	−0.86	0.0005	1.97
Year Fixed Effect	Yes		Yes		Yes
Industry Fixed Effect	Yes		Yes		Yes
Observations	17,887		17,887		17,887
Overall R²	.555		.402		.243

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals, excluding firms with age below 15 years. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 17,887 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

Panel B: The Relationship Between Auditor Tenure and Non-Operating Accruals Controlling for Change in Risks With Firm Age Longer Than 15 Years.
Variables	Dependent variable: NACC
	Coefficient	t statistics	Coefficient	t statistics
INTERCEPT	−0.0538***	−2.66	−0.1157***	−8.93
TENURE	0.0003**	2.50	0.0003***	2.92
TENURE_ΔLOSS	−0.0005***	−2.87
ΔLOSS	0.0026	0.83
TENURE_ΔRISK			−0.0007***	−4.59
ΔRISK			0.0181***	7.57
AGE	0.0004***	6.15	0.0006***	7.04
ΔSALE	−0.0508***	−4.07	−0.0312***	−4.20
CFO	−0.5198***	−28.03	−0.2700***	−21.43
SIZE	0.0017	0.43	0.0104**	2.40
SIZE ²	−0.0003	−0.39	−0.0021***	−2.70
SIZE ³	0.0000	0.54	0.0001***	3.01
ROA	0.6168***	27.50	0.3649***	21.17
ROA_SD	−0.0084	−1.40	−0.0182	−1.63
LOSS	0.0270***	7.40	0.0124***	4.53
LEV	−0.0070	−1.13	−0.0069	−0.97
CR	0.0005**	2.26	0.0006*	1.96
INDGROW	−0.0024	−0.65	−0.0017	−0.52
FIRMGROW	−0.0005***	−3.17	−0.0000	−1.00
BIG	0.0026	0.95	−0.0023	−0.77
SPEICAL	0.0022	0.94	0.0046*	1.67
TENURE_SPECIAL	−0.0001	−0.62	−0.0001	−0.73
Year Fixed Effect	Yes		Yes
Industry Fixed Effect	Yes		Yes
Observations	17,887		17,887
Overall R²	.558		.415

Note. This table presents the results of relationship between auditor tenure and current non-operating accruals, controlling for client’s change in risk using a sample with firm age above 15 years. The t statistics are calculated using clustered standard errors by firm for the multivariate analyses. The sample consists of all firm-years from 1988 to 2006 in Compustat. We delete firms in the financial services industries (SIC between 6000-6999). We also delete firms with negative total assets, sales, and market values of equity. We exclude 0.5% of the cash flow and non-operating accruals at each extreme. Finally, we exclude observations of auditors whose total tenure with a client is shorter than 5 years and firms whose existence is less than 5 years. The final sample consists of 17,887 firm-year observations. See appendix for variable descriptions.

*, **, ***Indicate significance at the .10, .05, and .01 levels, respectively, using two-tailed tests.

Conclusion

We study how an auditor’s tenure affects financial reporting quality. Based on an analysis that captures the trade-off of risks facing an auditor, we show first that auditor tenure is generally associated with better reporting quality as measured by less bias in audited earnings. In contrast, we argue that when conditioned on a deterioration in the client-firm’s business conditions (which leads to increased litigation risks to their auditors), the positive relationship between auditor tenure and reporting quality is reversed—a novel hypothesis that was previously unavailable in the literature. The validity of this new hypothesis is tested empirically, and the results provide confirmative evidence that is robust with respect to a range of measurements and specification issues, including auditor independence.

Our study extends the literature on the relationship between quality of reported earnings, measured through the size of bias in earnings, and auditor tenure by examining the effects of increased (auditor) litigation risks. Our signed indicator of quality has the advantage of providing directional corrections, and thus provides an effective metric of earnings quality. Litigation risk is a major feature in the trade-offs facing an auditor, and we show that it can counteract the general trend toward lower reporting bias that comes with a longer auditor–client relationship.

Combining the findings from the existing literature and this article, we have a more complete picture of the ways by which auditor experience affects quality of earnings information, and thus raise the level of our understanding on this topic of great importance in financial reporting.

Footnotes

Appendix

Definition of Variables

AGE Number of years a firm exists on the Compustat data base;

$A V E R A G E_N A C C_{i t} = \frac{N A C C_{i t - 1} + N A C C_{i t} + N A C C_{i t + 1}}{3}$ ;

BIG An indicator variable equals 1 if the auditor is one of the big 4 firms, 0 otherwise;

CFO Cash flows from operations scaled by the beginning total assets;

CR Ratio between current assets to current liabilities;

DSALE An indicator variable equals 1 if ΔSALE is negative and 0 otherwise;

FINANCE Value of seasoned equity offering in the next period scaled by the total assets at the end of current period;

FIRMGROW Growth in sales for a firm;

INDGROW Growth in sales in an industry;

LEV Total liabilities to total assets ratio;

LOSS An indicator variable equals 1 if net income is negative and 0 otherwise;

MB Market to book value ratio of equity;

NACC Non-operating accruals, is equal to TOTAL_ACC - OPER_ACC;

$N A C C_S D_{i t} = \sqrt[2]{{(N A C C_{t - 1} - A V E R A G E_N A C C_{i t})}^{2} + {(N A C C_{t} - A V E R A G E_N A C C_{i t})}^{2} + {(N A C C_{t + 1} - A V E R A G E_N A C C_{i t})}^{2};}$

OPER_ACC Operating accruals which are equal to change in accounts receivable plus changes in inventory plus changes in prepaid expense minus changes in accounts payable and minus changes in taxes payable, scaled by the beginning total assets;

PPE Net property, plant, and equipment in year t, scaled by the beginning total assets;

QUICK Ratio of current assets net of inventory to current liabilities;

RISK A factor measuring firm risk generated from a set of variables using factor analysis;

ROA Net income scaled by the beginning total assets;

ROA_SD The standard deviation of ROA for the last three years;

SIZE Log transformation of the year-end market value of equity;

SPECIAL An indicator variable equals 1 if an auditor audits highest portion of assets in an industry;

TENURE The number of consecutive engagements with a client;

TOTAL_ACC Total accruals which are equal to the income before extra-ordinary items minus cash flow from operation, scaled by the beginning total assets;

ΔFINANCE Difference between FINANCE in year t+1 and FINANCE in year t;

ΔLOSS Difference between LOSS in year t and LOSS in year t-1;

ΔREC Change in accounts receivable from year t-1 to year t, scaled by beginning total assets;

ΔRISK Difference between RISK in year t and RISK in year t-1;

ΔSALE Change of sales from year t−1 to year t, normalized by the beginning total assets. An interaction variable is the product of two variables and is denoted by the two variables linked with an underscore in between.

Acknowledgements

The authors thank Barbara M. Grein, Hai Lu, Robert Mathieu, Miguel Minutti, Wally Smieliauskas, Minlei Ye, an anonymous referee, Bharat Sarath (the Editor), and participants at the 2010 American Accounting Association (AAA) and 2011 Administrative Sciences Association of Canada (ASAC) annual conferences for their valuable comments.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support from Social Sciences and Humanities Research Council of Canada is greatly appreciated.

Notes

References

Ahmed

Duellman

(2007). Accounting conservatism and board of director characteristics: An empirical analysis. Journal of Accounting & Economics, 43, 411-437.

Anthony

J. H.

Ramesh

(1992). Association between accounting performance measures and stock prices: A test of the life cycle hypothesis. Journal of Accounting & Economics, 15, 203-227.

Ball

Shivakumar

(2005). Earnings quality in U.K. private firms: Comparative loss recognition. Journal of Accounting & Economics, 38, 83-128.

Basu

(1997). The conservatism principle and the asymmetric timeliness of earnings. Journal of Accounting & Economics, 24, 3-37.

Bonner

S. E.

Lewis

B. L.

(1990). Determinants of auditor expertise. Journal of Accounting Research, 28(Suppl.), 21-28.

Butler

Leone

Willenborg

(2004). An empirical analysis of auditor reporting and its association with abnormal accruals. Journal of Accounting & Economics, 37, 139-165.

Carcello

J. V.

Nagy

A. L.

(2004). Audit firm tenure and fraudulent financial reporting. Auditing: A Journal of Practice & Theory, 23, 55-69.

Carcello

J. V.

Palmrose

Z.-V.

(1994). Auditor litigation and modified reporting on bankrupt clients. Journal of Accounting Research, 32(Suppl.), 1-30.

Choo

(1996). Auditors’ knowledge content and judgement performance: A cognitive script approach. Accounting, Organizations and Society, 21, 339-359.

10.

Chow

Rice

(1982). Qualified audit opinions and auditor switching. The Accounting Review, 57, 326-335.

11.

Collins

D. W.

Hribar

(2002). Errors in estimating accruals: Implications for empirical research. Journal of Accounting Research, 40, 105-134.

12.

Dechow

P. M.

(1994). Accounting earnings and cash flows as measures of firm performance: The role of accounting accruals. Journal of Accounting & Economics, 18, 3-42.

13.

Dechow

P. M.

Dichev

(2002). The quality of accruals and earnings: The role of accrual estimation errors. The Accounting Review, 77, 35-60.

14.

Francis

J. R.

Lennox

C. S.

Wang

(2012). Selection models in accounting research. The Accounting Review, 87, 589-616.

15.

Geiger

Raghunandan

(2002). Auditor tenure and auditor reporting failures. Auditing: A Journal of Practice & Theory, 21, 67-78.

16.

Givoly

Hayn

(2000). The changing time-series properties of earnings, cash flows and accruals: Has financial reporting become more conservative? Journal of Accounting & Economics, 29, 287-320.

17.

Gul

F. A.

Fung

S. Y. K.

Jaggi

(2009). Earnings quality: Some evidence on the role of auditor tenure and auditors’ industry expertise. Journal of Accounting & Economics, 47, 265-287.

18.

Hubaid

Cooke

T. E.

(2005). The impact of managing director changes and financial distress on audit qualifications and auditor switching. Journal of Business Finance & Accounting, 32, 1703-1739.

19.

Johnstone

K. M.

Bedard

J. C.

Biggs

(2002). Generating financial reporting alternatives: Enhancing and inhibiting roles of task environment and task knowledge. Auditing: A Journal of Practice & Theory, 2, 47-65.

20.

Jones

(1991). Earnings management during import relief investigations. Journal of Accounting Research, 29, 193-228.

21.

Kellogg

R. L.

(1984). Accounting activities, security prices, and class action lawsuits. Journal of Accounting & Economics, 6, 185-204.

22.

Kim

J.-B.

Chung

Firth

(2003). Audit conservatism, asymmetric monitoring, and earnings management. Contemporary Accounting Research, 20, 323-359.

23.

Krishnan

Stephens

(1996). The simultaneous relation between auditor switching and audit opinion: An empirical analysis. Accounting and Business Research, 26, 224-236.

24.

Lawrence

A. M.

Minutti-Meza

Zhang

(2011). Can big 4 versus non-big 4 differences in audit-quality proxies be attributed to client characteristics? The Accounting Review, 86, 259-286.

25.

(2010). Does auditor tenure affect accounting conservatism? Further evidence. Journal of Accounting and Public Policy, 29, 226-241.

26.

Libby

Frederick

D. M.

(1990). Experience and the ability to explain audit findings. Journal of Accounting Research, 28, 348-367.

27.

Lys

Watts

R. L.

(1994). Lawsuits against auditors. Journal of Accounting Research, 32, 65-93.

28.

Myers

J. N.

Myers

L. A.

Omer

T. C.

(2003). Exploring the term of the auditor-client relationship and the quality of earnings: A case for mandatory auditor rotation?The Accounting Review, 78, 779-799.

29.

Rangan

(1998). Earnings management and the performance of seasoned equity offerings. Journal of Financial Economics, 50, 101-121.

30.

Securities and Exchange Commission. (1994). Staff report on auditor independence. Washington, DC: Government Printing Office.

31.

Shelton

S. W.

(1999). The effect of experience on the use of irrelevant evidence in auditor judgment. The Accounting Review, 74, 217-224.

32.

Shivakumar

(2000). Do firms mislead investors by overstating earnings before seasoned equity offerings?Journal of Accounting & Economics, 29, 339-371.

33.

Pierre

St. Anderson

(1984). An analysis of the factors associated with lawsuits against public accountants. The Accounting Review, 59, 242-263.

34.

Teoh

Welch

Wong

(1998). Earnings management and the under-performance of seasoned equity offerings. Journal of Financial Economics, 50, 63-100.

35.

Tepalagul

Lin

(2015). Auditor independence and audit quality: A literature review. Journal of Accounting, Auditing & Finance, 30, 101-121.

36.

Tubbs

R. M.

(1992). The effect of experience on the auditor’s organization and amount of knowledge. The Accounting Review, 67, 783-801.

37.

U.S. Senate. (1976). The accounting establishment (Prepared by the Subcommittee on Reports, Accounting and Management of the Committee on Governmental Affairs). Washington, DC: Government Printing Office.

38.

Watts

R. L.

(2003). Conservatism in accounting part 1: Explanations and implications. Accounting Horizons, 17, 207-221.

39.

Zhang

(1999). A bargaining model of auditor reporting. Contemporary Accounting Research, 16, 167-184.