Auditor Independence and Earnings' Quality: Evidence for ... - CiteSeerX

Auditor Independence and Earnings’ Quality: Evidence for Market Discipline vs. Proscriptive Regulation∗ James Brown Montana State University Dino Falaschetti Montana State University Michael Orlando Federal Reserve Bank of Kansas City September 1, 2006 Abstract Received research largely argues against auditor independence influencing the quality of earnings’ reports, but encounters several difficulties in doing so. Addressing these difficulties, we build additional confidence that auditor independence improves earnings’ quality, though any such effect appears to be small. Moreover, our research facilitates a more careful inference from audit fee data about the efficacy of Sarbanes-Oxley’s restriction on consulting for audit clients. Here, we develop more defensible evidence that moving past the Securities and Exchange Commission’s (SEC’s) fee disclosure mandates to proscribe non-audit services diminished financial market opportunities.

JEL: G14, G38, K22, M42 Keywords: Auditor Independence, Audit Fees, Non-Audit Services, Corporate Governance, Sarbanes-Oxley

∗

We thank Dan Covitz, Rob Fleck, Steven Hansen, Andy Hanssen, Mary Sullivan, Doug Young, participants at the 2006 meeting of the Washington Area Finance Association at George Washington University (WAFA) and the 81st Annual Conference of the Western Economic Association (WEA), and seminar audiences at Montana State University for helping us think about this research.

Auditor Independence and Earnings’ Quality: Evidence for Market Discipline vs. Proscriptive Regulation September 1, 2006 Abstract Does auditor “independence” improve the quality of financial disclosures? Popular characterizations of recent governance scandals strongly answer “yes.” Scholarly evidence appears less decisive, however, in part because it ignores several hypotheses about how independence can influence earnings’ quality. We examine proxy data on fees for audit and non-audit services (NAS) to address this issue as follows. 1.

We relax a priori linear restrictions in the literature, which assume that independence affects earnings’ quality in only one direction.

2.

We offer a finer evaluation of how markets value independence by better measuring unexpected disclosures and examining whether audit fee disclosures improve the efficiency of market valuations; and

3.

We look beyond internal effects of independence to consider the potential for one firm’s governance choice to influence other firms’ financial market opportunities.

In each case, we find evidence that auditor independence does not, by itself, materially degrade the quality of financial disclosures (either internally or externally). To the extent that significant relationships appear in our data, they are consistent with disclosure mandates exhausting opportunities to enhance financial market performance, and thus with proscriptive regulation (e.g., Sarbanes-Oxley’s restriction on NAS) having foreclosed such opportunities.

JEL: G14, G38, K22, M42 Keywords: Auditor Independence, Audit Fees, Non-Audit Services, Corporate Governance, Sarbanes-Oxley

1. Introduction High profile accounting scandals (e.g., Enron, WorldCom) squarely placed the topic of corporate governance in front of popular and business media. A widely held belief emerged that letting accountants consult for audit-clients compromises auditors’ independence and thus diminishes the quality of earnings’ reports (e.g., see Romano, 2004; Weil, 2004).1 Citing such conflicts of interest, US legislators built considerable support for the Sarbanes-Oxley Act of 2002 (SOX, hereafter), part of which restricts accountants from producing non-audit services (NAS).2 This support appears at odds, however, with scholarly examinations of disclosure mandates that preceded SOX – i.e., previous Securities and Exchange Commission (SEC) requirements that audit clients formally disclose fees paid for audit and nonaudit services.

These studies offer limited evidence that markets value the

information that such disclosures make available (e.g., see Glezen and Millar, 1985; Frankel et al., 2002; Ashbaugh et al., 2003), and employ such results to not only argue against disclosure mandates, but also against proscriptive regulations like SOX (e.g., see DeFond et al., 2002).3 We critically evaluate these results and find them wanting on several margins. Received research ignores, for example, several channels through which information about “fee dependence” can be influential. In addition, it leaves open the question of 1

Ezzamel, Gwilliam, and Holland (1996) document a similar perception for UK companies. See “Title II – Auditor Independence” of the Sarbanes-Oxley Act of 2002 (HR 3763), summarized in our Appendix A. The bill passed the House by a roll call vote of 423-3 and the Senate by a vote of 99-0 on July 25, 2002 (Source: Thomas (a service of the Library of Congress), accessed April 12, 2005 at http://thomas.loc.gov/home/thomas.html). 3 In addition, popular calls to loosen SOX and its regulatory constraint on producing NAS appear to be growing. The US Chamber of Commerce (2006, p. 16), for example, argued that prohibiting “Big Four firms” from “audit assignments when they have performed disqualifying services in prior years” unduly restricts competition. 2

1

how data from fee disclosures can inform regulations that would prescribe how market participants organize their governance services. Addressing these issues, we find more defensible evidence that mandating the disclosure of accounting fees can productively strengthen market discipline, though any such effect appears to be small. Moreover, to the extent that regulatory opportunities to enhance financial market efficiency existed, our evidence is consistent with disclosure mandates having exhausted them – i.e., any strengthening of market discipline that resulted from these mandates may have fully internalized the costs and benefits of commingling audit and non-audit services. Our research design thus offers more defensible evidence against the efficacy of SOX in having moved past disclosure mandates to restrict auditors from producing NAS. These contributions come from a more firmly grounded empirical investigation of how auditor independence relates to financial statement integrity. In the following section, we review the literature to identify channels through which auditor independence can plausibly influence earnings’ quality without being detected by received research designs. This review highlights several channels as being worthy of investigation. 1.

In addition to threatening the integrity of financial disclosures, jointly producing audit and non-audit services can leverage scope economies to improve disclosure-quality.4 But while such economies can offset associated agency costs, and thus give rise to a non-monotonic relationship between

4

Banks, for example, exhibit qualitatively similar economies when jointly producing lending and underwriting services (see, e.g., Drucker and Puri, forthcoming).

2

audit quality and auditor “independence,” contributions to the literature have a priori restricted this relationship to being linear. 2.

If markets are efficient and auditor independence matters, then equity prices should

respond

to

disclosures

about

unexpected

independence.

Contributions to the literature, however, examined how markets responded to “gross” disclosures – i.e., disclosures that confound expected and unexpected independence.5

Even if auditor independence influences

earnings’ quality, the errors-in-variable problem that this treatment creates can attenuate coefficient estimates of interest and thus hide evidence of an effect. 3.

Received research designs attempt to measure the own-firm effect of auditor independence on the quality of earnings’ reports. They do not consider, however, the potential for disclosures about auditor independence at one firm to inform markets about others. Understanding the extent of such informational externalities is important for evaluating the efficiencyconsequences of mandated disclosures and more proscriptive governance regulations.

Each of these difficulties can bias inference toward rejecting the hypothesis that separating the production of audit and non-audit services expands financial market opportunities. After carefully addressing these issues in Section 3, however, we find 5

To be sure, the literature does not completely ignore this issue. DeFond et al. (2002) and Frankel et al. (2002), for example, evaluated how proxies for earnings’ quality relate to measures of “unexpected” nonaudit fees. These measures ignore, however, the potential for organizational features (e.g., audit committee independence) to substitute for auditor independence in producing corporate governance services (Falaschetti and Orlando, 2004). Moreover, while information sets must be available before they can facilitate expectations, DeFond et al. (2002) and Frankel et al. (2002) estimate “expectations” from contemporaneous information. Even received measures of unexpected non-audit fees thus appear prone to the errors-in-variables problem that we attempt to address more carefully in the present paper.

3

even stronger evidence against the efficacy of regulatory proscriptions on how financial market participants produce governance services (e.g., the SOX restriction). Indeed, to the extent that significant relationships appear in our data, they support the hypothesis that mandated fee disclosures exhausted what was possible in expanding financial market opportunities. Corporate governance in general, and accounting systems in particular, play an important role in defining an economy’s potential. For example, market discipline can expand the set of feasible organizational opportunities, such as the ability to separate ownership from control, and may itself benefit from informative financial disclosures. Likewise, holding organizational opportunities constant, financial capital can more easily find productive employment in rich informational environments. However, our evidence that any responses to news about auditor independence are “local” (i.e., they do not spillover to other firms) suggests that regulating how that independence is established can compromise, rather than bolster, the integrity of financial disclosures. We thus conclude in Section 4 by considering how political forces may have pushed US governance regulations in a direction that works against the public’s interest, and how future research might improve our understanding of this important political dimension of economic performance.

2. Potential Difficulties with Received Research Designs Even before legislators responded to recent governance scandals, the issue of accountants producing NAS for audit clients received considerable regulatory attention.

The Securities and Exchange Commission’s (SEC) Accounting Series

Release (ASR) No. 250: Disclosure of Relationships with Independent Public 4

Accountants, for example, required subject companies to disclose fees paid to auditors for NAS (via proxy statements filed after September 30, 1978). Glezen and Millar (1985) found, however, that shareholder voting on auditor-retention negligibly responded to these disclosures.

At least on its face, this evidence supports the

hypothesis that producing NAS for audit clients does not materially compromise an accountant’s integrity (Glezen and Millar 1985, p. 859-60). It also appears to bolster the SEC’s rationale for withdrawing ASR 250 in February of 1982 – i.e., shareholders lack interest in fee disclosures. 2.1 Linear-restrictions can bias inference But drawing such strong inference from Glezen and Millar’s (1985) evidence can be problematic. For example, a negligible relationship between shareholder voting and fee disclosures also supports the normatively opposing, but observationally equivalent, hypothesis that auditor independence influences earnings’ quality in a non-monotonic manner.

The following figure illustrates one such possible

relationship.

5

Figure 1 Possible Observed vs. Actual Relationship An auditor’s dependence on non-audit fees can, in principle, improve or degrade the quality of reported earnings. If, for example, informational inputs for producing audit services intersect those for producing NAS, then jointly producing audit and non-audit services can improve earnings quality by facilitating scope economies. Joint production can, in addition, increase the cost of certifying misstated financial statements – e.g., the reputational costs of any such certification might include foregone profits from audit and non-audit services. On the other hand, by endowing

6

managers with the capacity to threaten auditors with the loss of non-audit business, jointly producing audit and non-audit services increases the pressure that managers can place on auditors to endorse compromised financial statements.6 In this light, earnings’ quality appears capable of sharing a non-monotonic relationship with fee dependence. Quality may, for example, first decrease with fee dependence if marginal forces associated with managerial influence overwhelm those associated with scope economies or reputational incentives. If these forces’ relative magnitudes ultimately reverse, then earnings’ quality can also share a positive relationship with fee dependence over domains where this dependency is more considerable. To the extent that such non-monotonicities characterize the relationship between earnings quality and dependence on non-audit fees, research designs that a priori restrict that relationship to being linear can spuriously produce evidence against the hypothesis that NAS matters. For example, simple correlations between proxies for quality and fee dependence, as well as corresponding coefficient estimates from linear regressions, can appear negligible even if quality and fee dependence share an important non-linear relationship. Glezen and Millar (1985) appear to have evaluated simple hypotheses about how fee dependence relates to earnings’ quality, however, instead of considering more flexible methods for evaluating joint hypotheses about this relationship and its functional form.7 Their reported results thus cannot dismiss 6 Arruñada (1999) offers a comprehensive evaluation of how the joint production of audit and non-audit services enhances the quality of earnings reports. Frankel et al. (2002) do the same for how joint production degrades earnings quality. Bratton (2003, pp. 12-13) reviews the conventional wisdom that “nonaudit consulting rents, employment opportunities at clients, and audit industry concentration” compromise the “professional relationship” between auditors and management. 7 Subsequent authors appear to have followed Glezen and Millar (1985) in this regard. We review some of these contributions below.

7

the normatively opposing hypothesis that dependence on non-audit fees significantly influences earnings’ quality, but in a non-monotonic manner.8 In addition to being subject to the above criticisms, Glezen and Millar’s (1985) evidence appears consistent with fee disclosures being important, but voting costs discouraging even rational owners from collectively acting against compromised auditors. More recent authors (e.g., Frankel et al., 2002; Ashbaugh et al., 2003) have thus tended to look not at how approval voting responds to fee disclosures, but rather at how market valuations respond.

In doing so, however, they too ignored the

potentially confounding issue of functional form, and have thus done little to distinguish the observational equivalencies that Glezen and Millar (1985) left open.9 Exploiting a more recent SEC reporting requirement, for example, Ashbaugh et al., (2003) estimated that firm-level market valuations negligibly responded to disclosures about the proportion of fees paid to auditors for NAS (i.e., “fee ratios”).10 On their face, these results largely support the hypothesis that having accountants produce NAS for audit clients does not degrade earnings’ quality.11 But while the 8

See, e.g., Glezen and Millar (1985, Tables 6 and 7). Other widely cited contributions that we reviewed also encounter difficulty in making valid inference available. Francis and Ke (2003), for example, examined whether the market valuation of earnings surprises depends on auditor independence. In doing so, however, they not only omitted the reporting of potentially important sensitivity analyses, they drew inference from an indicator of whether surprises occurred after the SEC implemented its fee disclosure mandate (e.g., equation (5) formally characterizes each firm as maintaining the same filing date). This methodology thus treats earnings surprises as having occurred when audit-fee information was available, even for firms that file proxies in late quarters – i.e., firms for whom such information could not have been available. Such difficulties do not appear confined to studies that report evidence that fee disclosures matter. DeFond et al. (2002), for example, reported that the propensity for auditors to issue going concern opinions is unrelated to auditor independence. While their evaluation restricts consideration to only distressed firms, however, it ignores the potential for bias to emerge from non-random selection. 10 Ashbaugh et al. (2003) follow Frankel et al. (2002) in exploiting the SEC’s “Final Rule S7-13-00, Revision of the Commission’s Auditor Independence Requirements,” which demands that companies disclose, via proxy statements filed after February 5, 2001, information regarding fees that the auditor billed to it during the previous year (Frankel et al., 2002, p. 4). 11 To be sure, Frankel et al. (2002) find evidence from accrual data that jointly producing audit and nonaudit services degrades earnings’ quality, but little in the way of an economically meaningful market 9

8

opposing forces highlighted above suggest that the fee dependence/earnings’ quality relationship may be non-monotonic, the regression specifications from which received “non-results” develop a priori restrict this relationship to being linear. In this light, published results that might have empirically distinguished the weight of these forces appear inconclusive, especially those that draw on capital market responses to audit/non-audit fee disclosures.12 2.2 Ignoring market efficiency implications can bias inference In addition to having offered evidence on how a proxy for earnings’ quality (i.e., approval voting) linearly relates to a proxy for auditor independence (i.e., the proportion of total fees that are attributable to NAS), Glezen and Millar (1985) conducted an event study of how retention voting responded to fee disclosures. In doing so, they found that voting negligibly responded to disclosures, and characterized this evidence as supporting the hypothesis that fee dependence does not influence earnings’ quality. But this evidence also supports the joint hypothesis that fee dependence affects earnings’ quality and shareholders rationally form expectations. Fee disclosures can be informative without systematically changing the direction in which shareholders vote. Indeed, evidence that the number of companies with higher approval ratings after the fee disclosure equals the number with lower ratings (see Glezen and Millar, 1985, Tables 3 – 5) supports the hypothesis that jointly producing audit and non-audit response to fee disclosures. Ashbaugh et al. (2003) argue that Frankel et al.’s (2002) accrual evidence is spurious, and question whether Frankel et al.’s (2002) event study results are even statistically significant. Kinney and Libby (2002) and Falaschetti and Orlando (2004) offer related critiques. 12 We do not address evidence that draws on non-market data, such as those on accruals or audit opinions (e.g., see Craswell et al. (2002) and Ashbaugh et al. (2003)). Evidence that rests on such measures tends to support the hypothesis that producing non-audit services does not materially diminish the quality of reported earnings (e.g., see Romano, 2004, Table 3).

9

services compromises earnings quality, but shareholders correctly anticipate (on average) this joint production. More recent contributions also exhibit this difficulty. These studies look at how firm valuations react to the disclosure of audit and non-audit fees. In other words, they estimate parameters from the following equation:

AR = α1 Fee Dependence +

∑

n

i=2

α i ⋅ Controlsi + εi

(1)

where AR measures “abnormal returns” and Fee Dependence equals the ratio of fees paid for NAS to fees paid for all services (i.e., audit and non-audit).13 If markets are efficient, however, abnormal returns will not depend on Fee Dependence per se, but rather on the fee ratio’s unexpected portion. To the extent

that the set of “control variables” is incomplete, the above specification can thus create an important errors-in-variables problem that hides the influence of auditor independence. To be sure, notice that the above equation can be rewritten as follows.

AR = α1 (Expected Dep. + Unexpected Dep.) +

∑

n

i=2

α i ⋅ Controlsi + εi

(2)

We see here that the variable Fee Dependence is a relatively noisy proxy for the variable of interest if markets are efficient – i.e., the unexpected fee dependence.

13

See Frankel et al. (2002) and Ashbaugh et al. (2003).

10

Such errors-in-variables problems give rise to an “attenuation bias” that can drive estimates of α1 toward zero (see our Appendix B for proof). Received results that fee disclosures don’t matter may be an artifact of this bias rather than evidence that auditor independence, by itself, does not compromise the integrity of financial disclosures. Frankel et al. (2002) attempted to address this issue by constructing measures of unexpected fee dependencies.14 In doing so, however, they ignored the potential for organizational features (e.g., board and ownership structures) to act as substitutes for auditor independence in producing corporate governance services (Falaschetti and Orlando, 2004). In addition, while information sets must be available before they can influence expectations, Frankel et al. (2002) estimate “expectations” from contemporaneous information (i.e., the timing of their expectation equals that of the information from which it is estimated). By ignoring such factors, this research design leaves open the potential for a considerable errorsin-variable problem. Ashbaugh et al. (2003) take an alternative approach by comparing cumulative abnormal returns between the first year in which information about fee dependencies was available and the last year in which such information was unavailable (i.e., fiscal years 2000 and 1999, respectively). Rather than evidencing a negligible market reaction to the disclosure of fee dependencies, however, Ashbaugh et al.’s (2003) results may instead reflect an efficient market where abnormal returns average to zero over time (i.e., new information affects market prices, but not systematically in one direction or the other). 14

DeFond et al. (2002) employ a similar method to investigate the sensitivity of going concern opinions to NAS.

11

2.3 Focusing on firm-level effects can hide information- and governance spillovers

Finally, an important issue to which neither early nor more recent evidence speaks is the potential for firms to shy away from disclosures that are privately costly but publicly beneficial. Suppose, for example, that learning about one firm’s fee ratio informs market participants about fee ratios at related firms. Then, to the extent that disclosing information is costly, firms will reveal less information about their auditors’ independence than is socially optimal (e.g., see Admati and Pfleiderer, 2000). This correlation can emerge from at least two sources. First, firms that confront similar market forces (e.g., those that operate in related sectors) may choose similar governance structures. Here, one firm’s disclosure can inform markets about forces that are common to several firms. In this case, the reporting firm does not internalize the costs and benefits of disclosing governance information. Consequently, absent disclosure mandates, firms will release less information than is optimal from the market’s perspective. This correlation can also emerge from an arguably more obstinate source – i.e., from one firm’s governance structure affecting others’ governance choices (even those that operate in unrelated sectors). Here, firms continue to disclose too little information (because information is still correlated), but govern themselves in a less than optimal manner because their governance choices affect others. Moreover, disclosure mandates may not mitigate this inefficiency – i.e., the external consequences of governance choices are independent of disclosure choices. In this case, pushing past disclosure mandates to proscribe certain governance practices may

12

be necessary to internalize the consequences of governance choices (rather than disclosure decisions) and thus expand financial and economic opportunities. Received research on how auditor independence relates to earnings’ quality appears to ignore these channels, and may have thus reached too far in (i) dismissing the efficacy of disclosure mandates and (ii) extending that inference to the efficacy of proscriptive regulations. Indeed, each of the prominent contributions that we focus on above (i.e., Glezen and Millar, 1985; DeFond et al., 2002; Frankel et al., 2002; and Ashbaugh et al., 2003) examines how disclosures affect own-firm performance. They would thus fail to find evidence of regulatory efficacy if those regulations’ main benefit comes from internalizing the consequences of disclosure- or organizationalchoices.

When such spillovers are salient, governance regulations can expand

economy-wide production opportunities without revealing their effectiveness to research that only looks for internal effects.

3. Empirical Results By (i) a priori restricting the relationship between fee dependence and earnings’ quality to being linear, (ii) employing noisy proxies for unexpected fee dependencies, and (iii) ignoring the potential for one firm’s disclosure to inform markets about others, received research designs may be unable to discover evidence that fee dependence matters, even if it does. In addition, they cannot distinguish what any

such evidence means for the efficacy of disclosure mandates that might enhance market discipline and proscriptive regulations that might address organizational externalities.

13

We address these difficulties by exploring more completely the manner in which data from fee disclosures might evidence a relationship between auditor independence and earnings’ quality. Our departure from the literature begins with an examination of whether this relationship (observed at the audit-client level) is non-monotonic. We then implement a research design that can distinguish what have heretofore appeared as observationally equivalent hypotheses – i.e., (i) jointly producing audit and nonaudit services does not materially influence earnings’ quality and (ii) jointly producing these services influences earnings’ quality, but markets are efficient. Finally, we consider whether disclosures about auditor independence produce informational externalities – i.e., market benefits that competitive forces might not create on their own and would hide themselves from received research designs. Our results offer stronger evidence that fee dependencies compromise the information that financial statements make available, though this effect appears to be small. They also argue, however, against the ability of proscriptive regulations to address this relationship in a productive manner. We thus conclude that disclosure mandates may have facilitated a more complete pricing of how firms choose their auditors’ independence, but find no evidence that fee disclosures informed markets about others’ governance decisions or that the manner in which firms organize their governance services created external effects. In this light, the SOX restriction on producing NAS appears unlikely to improve upon the market discipline that disclosure mandates may have already strengthened.

14

Formatted: Bullets and Numbering

Data

Our investigation begins where Frankel et al.’s (2002) and Ashbaugh et al.’s (2003) ended – i.e., with an event study of how valuation levels responded to “initial” disclosures of fees paid to auditors.

Here, we’re interested in recreating the

literature’s results to gain confidence that our evidence is not a data-artifact. To pursue this interest, we measured auditors’ “independence” for each firm in the Audit Analytics database that filed a definitive proxy statement between February 5, 2001 (the first day that firms were required to disclose such fees) and February 4, 2002, leaving us with 3,313 sampled firms.15 Following Frankel et al. (2002) and Ashbaugh et al. (2003), we defined this measure (i.e., Fee Dependence) as the ratio of non-audit fees to total fees. This ratio’s average equals 0.47 for our sampled firms (i.e., 47% of total fees are, on average, attributable to NAS), and its standard deviation equals 0.26. Treating the filing date as our “event date” (i.e., day 0), we then used stock price data from the Center for Research in Securities Prices (CRSP) to compute each sampled firm’s abnormal return on the event date. Here, our variable Abnormal Return (AR) equals the difference between the firm’s raw return and the CRSP

equally weighted market return on the disclosure date.16 Evaluated at their mean (i.e., 0.08%), our Abnormal Returns exhibit the same magnitude as do Ashbaugh et al.’s (2003) (i.e., 0.04%).

15

Data from Audit Analytics is available with subscription at www.auditanalytics.com. Our results are invariant to a number of alternative measures of excess returns. For example, we obtained similar results from comparing raw returns around the announcement to the return of a beta matched portfolio over the same interval. In addition, while our reported results are based on the disclosure date (t = 0), we found similar results when employing both two and three day windows around the announcement (e.g., -1 to 0, 0 to +1, and -1 to +1).

16

15

Finally, we employed a specification similar to that of Frankel et al. (2002) to establish a baseline relationship between abnormal returns and non-audit fees. In particular, we estimated parameters from equation (1) and report results from this baseline specification in the first two columns of our Table 1.

Table 1 Dependent Variable = Abnormal Returns Estimation Method = OLS (standard errors in parentheses)

Variable

(1)

(2)

(3)

(4)

Constant

Fee Dependence 2

0.005 (0.002)*** -0.008 (0.003)** -

0.012 (0.005)** -0.006 (0.004)* -

Fee Dependence 3

-

-

0.011 (0.006)* -0.004 (0.012) -0.003 (0.013) -

Log(Market Value)

-

-0.001 (0.000)

-0.001 (0.000)

0.013 (0.006)** -0.028 (0.026) 0.068 (0.069) -0.053 (0.051) -0.001 (0.000)

Adj. R2 Obs.

0.002 3,313

0.002 3,313

0.002 3,313

0.002 3,313

Fee Dependence

The magnitude of our coefficient estimate on Fee Dependence (i.e., -0.01) mimics that of Frankel et al.’s (2002). In particular, like others before us, we find that abnormal market returns share a statistically significant and negative linear relationship with (gross) fee dependencies. While statistically significant, however, this relationship appears to be economically small – e.g., a standard deviation increase in Fee Dependence is associated with a decrease in market capitalization of about $7,600 (where the average market capitalization is almost $3 million).

16

Formatted: Bullets and Numbering

Does ‘Fee Dependence’ share a non-monotonic relationship with AR?

Recall, however, that the relationship between earnings quality and an auditor’s dependence on non-audit fees may be non-monotonic – i.e., theory does not support an a priori linear restriction.17 To evaluate whether inference from received results is sensitive to this restriction, we added both squared and cubed fee ratios to our baseline specification (i.e., Fee Dependence2 and Fee Dependence3). These additions let the data speak for themselves about what is an appropriate specification, but do not change the inference that linear specifications make available – i.e., the relationship between fee ratios and auditor independence appears negligible for higher ordered independence-terms (see specifications (3) and (4) in Table 1). Here, received evidence against the influence of fee dependencies begins to exhibit an interesting robustness.

Formatted: Bullets and Numbering

Do unexpected fee dependencies affect market valuations?

Our literature review also suggests, however, that if markets are efficient, then the coefficient estimates of interest in Table 1 can be biased toward zero. Our concern is that fee dependence may matter, while evidence of such an effect hides behind the attenuation bias from estimating coefficients with noisy proxies. We address this issue by more finely partitioning Fee Dependence into its expected and unexpected components, and measuring how markets respond to the unexpected component. In addition, we evaluate the resulting evidence’s robustness by considering whether market valuations became more efficient after being informed about fee dependencies. Having carefully addressed this important potential for bias, we find 17

See our Figure 1 above, and discussion therein.

17

more defensible evidence that relaxing auditor independence compromises financial disclosures, but that this effect is a small one. Formatted: Bullets and Numbering

Abnormal returns decrease with unexpected fee dependencies DeFond et al. (2002) and Frankel et al. (2002) attempted to estimate the unexpected portion of fee disclosures from proxy statement information about the financial performance and operating characteristics of sampled audit clients. On at least two dimensions, however, this methodology offers only a coarse partition of Fee Dependence into its expected and unexpected components, and thus leaves open a

potentially considerable errors-in-variable problem. First, this methodology ignores the capacity for organizational attributes (e.g., board structure) to substitute for auditor independence in producing governance services (Falaschetti and Orlando, 2004). Second, it estimates “expectations” from information that was unavailable to market participants – i.e., the method examines data on fee dependencies, financial performance, and operating characteristics from the same proxy statement. Expectations about fee dependencies, however, must be formed before proxy statements are disclosed. Indeed, if markets are efficient, then information disclosed with that about fee dependencies cannot systematically relate to prior expectations about fee dependencies. We address the potential for such treatments to have biased available inference by restricting attention to fee dependencies that are more likely to have been unexpected. In particular, we develop a new variable, Unexpected Fee Dependence, from the following model’s residuals.

18

Fee DependenceProxy 2001 = f (Firm CharacteristicsProxy 2000) + u

(3)

Note that our set of “firm characteristics” pre-dates our data on fee dependencies, and thus satisfies a necessary condition for information to act as a basis for expectations. In addition, this set includes data on organizational features that might act as substitute factors in producing governance services, including those that Falaschetti and Orlando (2004) find to be plausibly exogenous. Data requirements to estimate equation (3) cause a reduction in our sample size to 927 audit clients, and we disclose the results of that estimation in Appendix C. Using these results to partition Fee Dependence into its expected and unexpected components, we re-estimate equation (1) with a less noisy proxy for unexpected dependencies and report the results in the following Table 2.

Table 2 Dependent Variable = Abnormal Returns Estimation Method = OLS Variable

(1)

(2)

Constant

0.006 (0.009) -0.011 (0.005)** 0.010 (0.019) -

0.005 (0.009) -

Unexpected Fee Dep. Expected Fee Dep.

-

Log(Market Value)

0.000 (0.001)

-0.013 (0.005)** 0.000 (0.001)

Adj. R2 Obs.

0.003 927

0.006 927

Fee Ratio

19

If markets are efficient and fee dependence matters, then abnormal returns should vary only with Unexpected Fee Dependence (and not its expected counterpart). Such a relationship makes itself evident in column (1) – i.e., the coefficient estimate on Unexpected Fee Dependence is statistically significant, while that on Expected Fee Dependence is not. Even more, the negative and significant coefficient estimate on Unexpected Fee Dependence suggests that the results reported in Table 1 are not an

artifact of what, in principle, can be an important source of endogeneity bias. This consistency also appears in comparison to our “restricted sample” estimate on the relatively noisy proxy for unexpected dependencies, Fee Ratio – i.e., the coefficient estimate reported in the second column of Table 2 differs insignificantly from its “full sample” counterpart reported in Table 1. Formatted: Bullets and Numbering

Auditor independence negligibly affects the efficiency of market valuations To further evaluate this inference’s sensitivity to how we measure unexpected fee dependencies, we examine a deeper implication of the joint hypothesis that markets are efficient and fee dependency is informative.

In particular, we evaluate the

implication that, if news about fee dependence matters, then errors in forecasting a firm’s financial performance should decrease. This method appears attractive for our present purposes because it relies less heavily on correctly estimating the unexpected portion of Fee Disclosure. To see this attractiveness, suppose that v ∈ R+ represents an asset’s true valuation, and let the relationship v ∈ V ⊂ R+ represent information about that valuation. Under this representation, market efficiency implies that E [v | v ∈ V ] − v = 0 – i.e., forecast errors equal zero on average. 20

In the present context, E [v | v ∈ V ] represents the market’s valuation of a firm for a given set of information. Now suppose that disclosures about audit and non-audit fees contain additional information about earnings’ quality so that v ∈ V ′ where V ′ ⊂ V .18 Forecast errors may continue to equal zero in this richer informational

environment, while the variability of errors decreases.

This implication follows

immediately from our definition of information and that for calculating variances –

[

] [

]

i.e., E (E [v |v ∈ V ′,V ′ ⊂V ]−v ) < E (E [v |v ∈ V ]−v ) . 2

2

Looking at how valuation levels and variation respond to revelations about fee dependence can thus facilitate a more thorough evaluation of how auditor “independence” might relate to earnings’ quality.

To the extent that we have

continued to mis-specify the “expectation regression” (i.e., equation (3)), the

economically negligible magnitude of our estimate on Unexpected Fee Ratio may also be an artifact of attenuation bias. If information enters markets in the manner that we’ve modeled here, however, then examining changes in the variance of firmvaluations offers an important robustness check. We thus extend our event study to include an evaluation of how the variance in market valuations responded to fee disclosures. Here, forecast errors decrease under the hypothesis that jointly producing audit and non-audit services influences earnings’ quality, but remain constant under the hypothesis that joint production does not affect earnings’ quality. To the extent that we also observe an immaterial response from 18

To fix ideas, suppose that taken on its face, an earnings report implies that a firm’s fundamental valuation equals $25/share, but uncertainty about that report’s quality implies that [$15, $35] is a reasonable confidence interval. Now suppose that new information reduces the uncertainty about this report’s quality. Here, the firm’s fundamental valuation can continue to be $25/share (i.e., the disclosure does not adjust reported earnings per se), while the associated confidence interval shrinks (i.e., the disclosure can increase the precision with which “true” performance is estimable from “reported” performance).

21

volatility, we can thus gain additional confidence that (over observed domains) any effect from jointly producing audit and non-audit services is small.19 To measure volatility we focus on one- and two-month intervals immediately preceding and immediately following the proxy filing date. Our first measure of volatility (Volatility 1) equals the standard deviation of the stock price over the interval, divided by the average stock price during the interval. To evaluate whether our results are sensitive to changes in overall market volatility, we compute a second measure of volatility (Volatility 2) that scales firm level volatility over the interval with the volatility of the S&P 500 during the same period.

Table 3 Fee Disclosure and Variance in Market Valuations Days Relative to Announcement:

-62 to -2

2 to 62

t, z stat

-32 to -2

2 to 32

t, z stat

Volatility 1 mean median

0.401 0.110

0.392 0.099

-0.04 -2.91***

0.199 0.072

0.360 0.072

1.30 0.58

Volatility 2 mean median

8.258 2.414

13.573 3.226

0.78 3.71***

7.275 2.178

14.943 2.956

1.29 7.92***

Variance Measure

Table 3 offers evidence about whether volatility declined after audit fees were disclosed. Looking first at the two-month window immediately before (-62 to -2) and

19

Note that the manner in which we modeled information in Section 2 has the information in fee disclosures being independent of the disclosure’s content (i.e., the disclosure per se reduces uncertainty). If, instead, information about earnings’ quality varies with the level of fee dependence, then our volatility measures should exhibit a positive relationship with our variable Fee Dependence. Evaluating this implication in several unreported regressions, we did not find evidence for such a relationship.

22

immediately following (2 to 62) the proxy date, Volatility 1 decreases at both the mean and the median, though only at the median is the decline statistically significant. After scaling by the volatility of the S&P 500, volatility in the two months following the proxy filing actually increased slightly at the mean, and significantly so at the median. Examining volatility over a narrower one-month window also suggests that, if anything, stock price volatility increased following the proxy filing.20 These results corroborate those reported in Table 2 that, while auditor independence may share a statistically significant relationship with earnings’ quality, that relationship appears to be economically negligible.

Formatted: Bullets and Numbering

Do fee disclosures create governance spillovers?

Results that we report above argue more strongly than do those in the literature that any negative relationship between fee dependencies and earnings’ quality is economically small (e.g., see Frankel et al., 2002; Ashbaugh et al., 2003). They do not support claims in the literature (at least not by themselves), however, that relevant

disclosure mandates and proscriptive regulations exerted little or deleterious effects (e.g., see Glezen and Millar, 1985; DeFond et al., 2002). Indeed, to the extent that such regulations encouraged firms to internalize the consequences of disclosure or organizational

decisions,

such

regulations

could

have

expanded

financial

opportunities in ways that received research designs are unable to see.21

20

We examined the change in volatility over many other windows (both shorter and longer) and obtained similar results. 21 Anat Admati and Paul Pfleiderer (2000) offer a model where mandated disclosures can improve welfare when informational spillovers are salient. They also caution, however, that the optimal method of disclosure can be difficult to find.

23

We address this issue by recreating our event study for sampled firms that filed proxies after April 5, 2001 (approximately half of our sampled firms filed after this date). If information from private disclosures or governance choices “spills over” to others, then firms that filed late in the proxy year should have exhibited a smaller response to disclosures than did firms that filed early in the year (i.e., markets would have already had information about late disclosers).22 Coefficient estimates reported in the following table, however, argue against such effects being salient.

Table 4 Dependent Variable = Abnormal Returns Estimation Method = OLS Variable Constant Unexpected Fee Dep Fee Ratio Log(Market Value) Adj. R2 Obs.

“early” filers (1) (2) 0.021 0.021 (0.011)* (0.011)* -0.008 (0.007) -0.011 (0.006)* -0.001 -0.001 (0.001)* (0.001) 0.005 460

0.009 460

“late” filers (1) (2) -0.006 -0.007 (0.014) (0.014) -0.015 (0.007)** -0.015 (0.007)** 0.000 0.001 (0.001) (0.001) 0.004 467

0.005 467

Rather, market responses to disclosures about fee-dependence appear to be, if anything, stronger for firms filing statements late in the proxy year. This evidence opposes the hypothesis that informational spillovers motivate either disclosure

mandates or more proscriptive regulations in the present setting.

22

Foster (1980) anticipated this type of method.

24

Coupled with the results of relaxing linearity restrictions and examining how price-volatility responds to disclosures, this additional evidence creates considerable difficulty for “public interest” rationalizations of regulations like the SEC’s mandated disclosure, and even more so for the SOX restriction on NAS. Here, our results offer stronger support for informal rationalizations of why the SEC withdrew previous disclosure requirements (i.e., because shareholders lacked interest in such disclosures (Glezen and Millar, 1985 (pp. 859-60)) than they do for public interest rationalizations of the SOX restriction.

4. Conclusion Prominent studies have argued that financial regulations like those considered here (i.e., disclosure mandates and organizational proscriptions) exhibit little potential to expand financial (and, ultimately, economic) opportunities (e.g., see Glezen and Millar, 1985; Kroszner and Rajan, 1997; DeFond et al., 2002). In making this case, however, they left open to various degrees the possibilities that (i) governance attributes share a non-monotonic relationship with economic performance, (ii) forces associated with market efficiency mask evidence of regulatory efficacy, and (iii) informational and organizational spillovers retard market efficiency. By explicitly addressing these issues, we develop stronger evidence about the efficacy of disclosure mandates and related organizational proscriptions as they apply to the issue of auditor independence. To the extent that our data exhibit significant relationships, they suggest that disclosure mandates expand financial market opportunities, but that any such expansion is unlikely to be economically important. Our methodology also produces relatively strong evidence against the hypothesis that 25

disclosures about auditor independence create “information externalities” from firms being subject to related forces or organizational choices creating external effects. This evidence argues much more forcefully than does that in the literature against the efficacy of proscriptive regulations like the SOX restriction on producing NAS. While we consider previously overlooked channels through which fee dependencies might matter, we cannot evaluate every such channels. Our results are, nevertheless, provocative in suggesting that an absence of evidence that auditor independence matters cannot be easily dismissed as a methodological artifact. In this light, the SEC’s re-introduction of fee disclosure mandates and SOX restrictions on producing NAS appear puzzling – i.e., if such rules advance the public’s interest, then why is evidence that markets value related information so difficult to find? Pushing the present research in this political economy direction may be an important objective for future research. Romano (2004, p. 8) begins to do so by investigating why “Congress would enact legislation that in all likelihood would not fulfill its objectives.” This rationalization appears to rest, however, on the implicit assumption that legislators take as their objective the maximization of public utility, arguing that election year and media pressures encouraged legislators to ignore “a literature at odds with their policy recommendations” (Romano 2004, p. 9). Moreover, the SEC’s mandated disclosure predates the phenomena at which Romano’s (2004) criticism aims. Future work might thus do well to consider why self-interested regulators, whether they reside in an agency like the SEC or a national legislature like the US Congress, might implement mandated disclosures or related governance restrictions

26

that maintain little known capacity to advance the public’s interest. This type of understanding might encourage organizational reforms that enhance the quality of collective decision-making ex ante, and thus improve upon “remedies” that rely on ex post reactions to particularly unproductive policies.

27

Appendix A Section 201: Services Outside The Scope Of Practice Of Auditors; Prohibited Activities23 It shall be “unlawful” for a registered public accounting firm to provide any non-audit service to an issuer contemporaneously with the audit, including: (1) bookkeeping or other services related to the accounting records or financial statements of the audit client; (2) financial information systems design and implementation; (3) appraisal or valuation services, fairness opinions, or contribution-in-kind reports; (4) actuarial services; (5) internal audit outsourcing services; (6) management functions or human resources; (7) broker or dealer, investment adviser, or investment banking services; (8) legal services and expert services unrelated to the audit; (9) any other service that the Board determines, by regulation, is impermissible. The Board may, on a case-by-case basis, exempt from these prohibitions any person, issuer, public accounting firm, or transaction, subject to review by the Commission. It will not be unlawful to provide other non-audit services if they are pre-approved by the audit committee in the following manner. The bill allows an accounting firm to “engage in any non-audit service, including tax services,” that is not listed above, only if the activity is pre-approved by the audit committee of the issuer. The audit committee will disclose to investors in periodic reports its decision to pre-approve non-audit services. Statutory insurance company regulatory audits are treated as an audit service, and thus do not require pre-approval.

23

Source: The American Institute of Certified Public Accountants (AICPA). “Summary of Sarbanes-Oxley Act of 2002.” Accessed at http://www.aicpa.org/info/sarbanes_oxley_summary.htm on April 26, 2005.

28

The pre-approval requirement is waived with respect to the provision of non-audit services for an issuer if the aggregate amount of all such non-audit services provided to the issuer constitutes less than 5% of the total amount of revenues paid by the issuer to its auditor (calculated on the basis of revenues paid by the issuer during the fiscal year when the non-audit services are performed), such services were not recognized by the issuer at the time of the engagement to be non-audit services, and such services are promptly brought to the attention of the audit committee and approved prior to completion of the audit. The authority to pre-approve services can be delegated to one or more members of the audit committee, but any decision by the delegate must be presented to the full audit committee.

29

Appendix B Hypothesis: If markets are efficient, then the OLS estimate of α1 in equation (1) is biased toward zero. Proof: To ease exposition, let y ≡ AR and x ≡ Fee Dependence in equation (1) so that the OLS estimate of α1 equals cov(x, y ) var (x ) . In addition, let αˆ 1 denote the OLS estimate of α1, and α˜ 1 denote the unbiased estimate. Finally, let x E and xU denote the expected and unexpected components of Fee Dependence, respectively (i.e., let x = x E + x u ). Then

αˆ 1 = cov(x E + xU , y ) var (x E + xU ) = (cov(x E , y ) + cov(xU , y )) (var (x E ) + 2cov(x E , xU ) + var (xU )). Without loss of generality, let cov(xU , y ) < 0 , and notice that market efficiency implies that cov(x E , y ) = cov(x E , xU ) = 0. The following relationship thus emerges. 0 > αˆ 1 = (cov(xU , y )) (var (x E ) + var (xU )) > (cov(xU , y )) var (xU ) = α˜ 1. In words, unexpected fee dependence diminishes earnings quality (i.e., cov(xU , y ) < 0 ), but αˆ 1 maintains an upward bias toward zero (letting cov(xU , y ) > 0 creates a downward bias toward zero).

30

Appendix C Dependent Variable = Fee Dependence Estimation Method = OLS Variable Constant 0.467 (0.153)*** Fraction of shares held by CEO -0.024 (0.059) Fraction of shares held by Institutions -0.007 (0.025) (Log)Boardsize 0.015 (0.026) Fraction of Audit Committee Independent -0.058 (0.033)* Fraction of Nominating Committee Independent 0.064 (0.029)** Dummy=1 if Chairman separate from CEO 0.022 (0.014) (Log)Sales -0.020 (0.010)** Sales Growth 0.015 (0.030) Return on Assets -0.011 (0.153) Dummy=1 if Net Loss during year 0.024 (0.028) Cash Flow-to-Assets 0.097 (0.112) Leverage 0.043 (0.034) (Inventories + Receivables)-to-Assets -0.010 (0.037) Dummy=1 if New Stock or Debt Issue during year 0.040 (0.026) (Log)Market Value 0.062 (0.009)*** Market Value-to-Book Value -0.004 (0.002)* Annualized Stock Return -0.011 (0.016) Adj. R2

0.162

Note: Though not reported, we also include industry dummy variables in the above regression. 31

References Admati, Anat R. and Paul Pfleiderer. 2000. “Forcing Firms to Talk: Financial Disclosure Regulation and Externalities.” Review of Financial Studies 13(3):479519. Arruñada, Benito. 1999. The Economics of Audit Quality: Private Incentives and the Regulation of Audit and Non-Audit Services. Norwell, MA: Kluwer Academic Publishers. Ashbaugh, H., R. LaFond, and B. Mayhew. 2003. “Do Nonaudit Services Compromise Auditor Independence? Further Evidence.” Accounting Review 78(July):611-639. Bratton, William W. 2003. “Rules, Principles, and the Accounting Crisis in the United States.” Villanova Law Review 48(4). Bushee, Brian J. and Christian Leuz. 2005. “Economic Consequences of SEC Disclosure Regulation: Evidence from the OTC Bulletin Board.” Journal of Accounting & Economics 39(2):233-264. Craswell, Allen, Donald J. Stokes, and Janet Laughton. 2002. “Auditor Independence and Fee Dependence.” Journal of Accounting & Economics 33:253-275. DeFond, Mark L., K. Raghunandan, and K. R. Subramanyam. 2002. “Do Non-Audit Service Fees Impair Auditor Independence? Evidence from Going Concern Audit Opinions.” Journal of Accounting Research 40(4):1247-1274. Drucker, Steven and Manju Puri. Forthcoming. “On the Benefits of Concurrent Lending and Underwriting.” Journal of Finance. Ezzamel, M., D. R. Gwilliam, and K. M. Holland. 1996. “Some Empricial Evidence from Publicly Quoted UK Companies on the Relationship Between the Pricing of Audit and Non-Audit Services.” Accounting and Business Research 27(1):3-16. Falaschetti, Dino and Michael J. Orlando. 2004. “Is Auditor Independence Endogenous? Evidence and Implications for Public Policy.” Federal Reserve Bank of Kansas City, Research Division Working Paper No. 03-13. Foster, George. 1980. “Externalities and Financial Reporting.” Journal of Finance (May):521-533. Francis, Jere R. and Bin Ke. 2003. “Disclosure of Fees Paid to Auditors and the Market Valuations of Earnings Surprises.” Working Paper. 32

Frankel, Richard M., Marilyn F. Johnson, and Karen K. Nelson. 2002. “The Relation Between Auditors’ Fees for Non-Audit Services and Earnings Management.” Accounting Review (Special Issue on Quality of Earnings):71-105. Glezen, G. William and James A. Millar. 1985. “An Empirical Investigation of Stockholder Reaction to Disclosures Required by ASR No. 250. Journal of Accounting Research 23(2):859-870. Kinney, William R. Jr. and Robert Libby. 2002. “Discussion of The Relation Between Auditors’ Fees for Nonaudit Services and Earnings Management. Accounting Review (Special Issue on Quality of Earnings):107-114. Kroszner, Randall S. and Raghuram G. Rajan. 1994. “Is the Glass-Steagall Act Justified? A Study of the US Experience with Universal Banking Before 1933.” American Economic Review 84(September):810-832. Romano, Roberta. 2004. “The Sarbanes-Oxley Act and the Making of Quack Corporate Governance.” New York University Law and Economics Working Papers, No. 3, September 26. Scheiner, J. H. 1984. “The Impact of SEC Nonaudit Service Disclosure Requirements on Independent Auditors and Their Clients: An Empirical Investigation.” Journal of Accounting Research (Autumn):789-797. Weil, Jonathan. 2004. “Auditing Firms Get Back to What They Do Best.” Wall Street Journal (March 31):C3.

33