Disclosure of fees paid to auditors and the market valuation of earnings surprises

Frankel et al. (2002) report evidence that clients are more likely to manage earnings to meet benchmark targets if they also pay their auditors high levels of nonaudit fees, implying that auditors may have compromised their objectivity and allowed clients greater discretion to manage earnings. However, Ashbaugh et al. (2003), Chung and Kallapur (2003), Francis and Ke (2003), Larcker and Richardson (2004), and Reynolds, Deis, and Francis (2004), report evidence that this is not the case, or that the results in Frankel et al. (2002) are driven by subsets of firms and do not generalize to a large set of companies. In related research, DeFond, Raghunandan, and Subramanyam (2002) and Craswell, Stokes, and Laughton (2002) document that nonaudit services have no effect on the auditor’s likelihood of issuing a going concern opinion. Finally, Reynolds and Francis (2000) document that auditors treat “larger” clients more conservatively: that is, large clients are more likely to have smaller abnormal accruals and auditors are more likely to issue going concern reports, both of which are opposite to what would be expected if fee dependence from large clients reduces auditor independence.

Despite the SEC’s concern that nonaudit service fees have the potential to impair auditor independence, we know of no litigation or SEC actions that explicitly relate the provision of nonaudit services to poor audit quality or a deliberate violation of auditor independence. While auditor independence cannot be observed directly ex ante, the impairment of auditor independence is determinable ex post through litigation or SEC investigations. However, the deliberate violation of independence constitutes fraud and there are historically very few proven cases of auditor fraud.

Our study assumes the disclosure of nonaudit fees allows investors to re-evaluate auditor independence. However, it is possible that fee disclosures do not change investors’ perception of auditor independence but are instead correlated with other information that affects investors’ valuation of accounting earnings such as restructuring activity. This limitation also applies to Frankel et al. (2002) and Ashbaugh et al. (2003). We have attempted to control for such alternative explanations by including control variables, and by examining the market valuation of earnings surprises for the period before and the period after the fee disclosures separately. Still, caution should be exercised when interpreting this study’s empirical evidence.

The initial proxy statement fee disclosures required the disclosure of three types of fees paid to auditors for the most recent fiscal year: (1) audit fees related to form 10-K and 10-Q filings; (2) financial information systems design and implementation fees; and (3) all other fees. See SEC (2000c).

Transparency through disclosure is a feature of longstanding SEC policy as evidenced by timely reporting in forms 10-Q and 8-K, as well as the annual 10-K and registration statements, and prospectuses required under the 1933 Act. See Kripke (1979) and Seligman (1982).

In addition to influencing investors’ beliefs on the quality of reported earnings (i.e., \(\phi^2_{j}\), the second moment), fee disclosures may also reveal information that alters investors’ expectation on the earnings persistence, firm risk, and future growth opportunities (i.e., the first moment). We do not consider these effects in the paper because they are not our main interest and are difficult to isolate from the common ERC determinants using our research design.

To control more explicitly for stale forecasts, we also measure FERR using only those analysts’ forecasts issued within 60 days of earnings announcements, and the results using this alternative definition are qualitatively the same as those reported in the paper.

We did not include X and NAS as main effects in order to be consistent with the ERC literature. However, including the control variables and NAS as main effects in the model, as well as the interaction terms, does not alter our inference. Following Collins and Kothari (1989), we also defined our control variables X using dummies to avoid potential measurement errors in X and obtained similar results.

Elliott and Hanna (1996) also use  −5% as the cutoff. Results are similar if we use a cutoff of 0%,  −1%, or  −2%.

Results are similar if the abnormal return is measured over 5 days centered on the earnings announcement date.

Frankel et al. (2002, p. 82) note that a large value in the ratio of nonaudit fees to total fees is unlikely to capture auditor independence problems unless the absolute dollar amount of the fees is also large. Ashbaugh et al. (2003) demonstrate this with two firms in their sample having an identical FEERATIO of 73%. Yet in one case total fees are only $71,000 and in the other case total fees are $5.7 million. It is difficult to imagine any scenario in which total fees of $71,000 would create an economic bond that threatens auditor independence even if virtually all fees were for nonaudit services.

We do not differentiate among types of nonaudit services in defining NAS because the SEC’s initial disclosure regulations did not mandate a detailed classification of nonaudit fees. In addition, existing auditor dependence theories focus on the magnitude of nonaudit fees rather than the type. There is no strong reason to believe that one type of nonaudit service might cause more auditor independence problems than another type.

As a further sensitivity check, we also allowed the coefficient on FERR to vary with the variance of analysts’ earnings forecasts. The coefficient on PERIOD*FERR*NAS remains significantly negative (−193, p < .001) and the coefficient on FERR*NAS is significantly positive (115, p = .022). However, because the variance of analysts’ earnings forecasts is missing for 25% of our sample firms, we do not include this control variable in regression model specified in Eq. 5.

Our sample includes only quarterly earnings announcements within 1 year of the fee disclosure date because we wish to study the initial effect of the fee disclosures on investors’ perception of auditor independence. The fee disclosures began with proxy statements filed on or after February 5, 2001.

This test assumes that the magnitudes of audit and nonaudit fees are stable from year to year. Francis and Ke (2003) report that yearly correlations are in excess of .85 for audit fees and nonaudit fees based on a recent Australian data where fees are publicly reported. In addition, we also estimate the regression model (5) using only the two quarterly earnings surprises immediately before and immediately after the initial fee disclosure date and obtain similar results (see Section 3).

To eliminate potential outliers, we deleted influential observations in each regression model in Tables 3, 4 and 5 based on Cook’s (1977) distance statistic. In addition, the reported standard errors allow heteroskedasticity and any type of correlation for observations of the same firm but assume independence for observations across different firms (Rogers, 1993).

Untabulated regression results indicate that the results in Table 4 hold for both positive and negative accruals.

The test variables of interest are PERIOD*FERR*NAS for the models in columns (2) through (7) of Table 3, and PERIOD*NAS for the model in column (8) of Table 3.

We also use a cutoff of 67th ($738,000) and 50th ($369,000) percentiles in defining NAS2, the absolute magnitude of nonaudit fees. The coefficient on PERIOD*FERR*NAS2 remains significantly negative for the 67th percentile cutoff, but not for the 50th percentile cutoff. This sensitivity analysis suggests that nonaudit fees start to cause perceived auditor dependence once they reach a material dollar threshold somewhere between $369,000 and $738,000 for our sample firms.

As an additional sensitivity check, we also define high levels of nonaudit fees using the natural log of one plus the dollar amount of nonaudit fees, and high levels of audit fees using the natural log of one plus the dollar amount of audit fees. The coefficients on PERIOD*FERR interacted with these continuous measures of nonaudit fees and audit fees are never significant.

The bankruptcy probability used in Whisenant et al. (2003) is from a model in Zmijewski (1984) and is based on bankruptcy data from the 1970s which may not apply to our sample period. However, the two key variables in Zmijewski’s model, firm performance and financial leverage, are directly included as controls in our model.