Printer Friendly

Long-Term Trends in Audit Fees.

There is little evidence on long-term trends in audit fees. While there have been many studies, their focus has generally been limited to a few years. This study examines audit fee trends over an 18-year period (1980-1997). The study involves Big 6-audited firms that voluntarily disclosed audit fees over this period. The design controls for changes in variables like firm size, complexity, and risk.

After controlling for audit wage rate and general price-level changes, we conclude that audit fees in the 1990s are not significantly different from 1980 levels. We find that there was a significant increase in audit fees in 1988, which we attribute to the Auditing Standards Board's introduction of the "expectation gap" standards. Many observers predicted that these standards would result in increased audit effort, and consequently increased audit fees. We also provide information on the question of whether mergers of large auditing firms have an anticompetitive effect in terms of pricing. Our study documents a short-lived increase in fees charged by the merged firms following the 1989 Big 8 mergers.

Finally, we provide evidence that audit fee drivers have changed. Specifically, we find that the influence of a client's accounts receivable and inventory on audit fees has declined apparently because auditors have become more efficient in the audit of these assets and have passed on the savings to their clients.

INTRODUCTION

The auditing profession experienced a wide variety of procedural, legal, and competitive changes through the 1980s and 1990s. Audit technology was altered by the increased use of computers, increased reliance upon analytical procedures, and the implementation of risk assessment models. This period was also characterized by high litigation against auditors. In the late 1980s, the profession made significant changes in the nature of audits with the "expectation gap standards." Finally, there have been changes in the competitive dynamics of the audit industry. Mergers have reduced the number of firms in the top tier of accounting firms from eight to five. Still, auditors claim that the auditing profession has experienced increased competition as the industry has shifted from a growth stage to a mature stage (Elliott and Pallais 1997).

These changes should affect the pricing of auditing services, or absent price adjustments, the profitability of audits. Audit production efficiencies should be reflected in lower fees, while increased audit procedures should trigger higher fees. Mergers may have an anticompetitive effect and drive up fees. On the other hand, if the audit marketplace is actually characterized by increasing competition, fees would be pushed down. Finally, the costs of auditor litigation should, other things being equal, be passed on to clients in the form of higher prices.

The purpose of this research is to examine long-term trends in audit fees. There is a large body of academic literature that uses cross-sectional analyses to explain the relationship between audit fees and client/auditor characteristics, but there is little evidence on how audit fees have behaved over the long term. The available evidence consists of anecdotal information, or surveys (e.g., Holman 1997) that provide limited controls for changes in client characteristics over time.

Our study examines audit fees over an 18-year period. We control for changes in auditee size, complexity and risk, as well as for audit wage rate changes and general price level changes. We study two factors related to audit production costs: the expansion of audit scope due to new audit standards and the reduction of audit costs due to production efficiencies. We also consider whether the 1989 mergers through which the Big 8 auditing firms became the Big 6 have had an anticompetitive effect on prices by reducing the number of available suppliers.(1)

The remainder of the paper is organized as follows. The next section discusses the major factors likely to have influenced audit prices over the period of the study. The following section develops the model we use in the empirical analysis. We then describe the method employed and present the results of our analyses. The final section provides some conclusions.

MAJOR INFLUENCES ON AUDIT FEE CHANGES

An important line of accounting research has sought to understand the market for audit services by studying audit fees. Audit fees are comprised of audit costs and profit. Following Simunic (1980), we characterize audit costs as having two components: production costs and expected future losses. Production costs are the costs incurred in carrying out the audit engagement, and are related to the amount of audit effort incurred, where audit effort is the time spent on the engagement. Production costs have typically been modeled in the audit fee literature as a function of auditee size, complexity, and risk. Expected future losses consist of costs that may be incurred by the auditor if the auditor's role in the engagement results in litigation.

The audit market has generally been viewed as being characterized by monopolistic competition. Some suppliers seek to differentiate their products by offering services of higher quality. The Big 6 auditors, for example, are generally perceived as offering a higher quality service than other auditing firms. Within each quality segment, however, the evidence is that the market is generally competitive.

Our analysis focuses on cost and competition changes within the Big 6 segment. Within this segment, any auditor-specific cost increase cannot easily be passed on to the client, since the client can obtain the same level of audit quality elsewhere. However, cost increases that are sustained by all suppliers will result in audit fees going up. Conversely, cost efficiencies are likely to be passed on to clients in the form of lower fees.

Increased Audit Scope Due to New Standards

New financial reporting rules and auditing requirements frequently have the effect of increasing the amount of work done on the audit and consequently increasing audit fees. The period under study saw a large number of financial accounting standards and auditing standards issued. While the effect of individual standards should be marginal in terms of impact on fees, in aggregate they are likely to have been material. In particular, it is likely that the "expectation gap" standards, which were expected to significantly increase the scope of audit work, had a corresponding significant effect on audit costs.

The "expectation gap" standards consist of ten standards issued by the AICPA's Auditing Standards Board in April 1988. The standards expanded the auditor's responsibility in such areas as evaluating the client's internal controls, providing early indication when the client is faced with financial distress, and detecting fraud. Many observers felt, at the time, that audit work would increase substantially following the issuance of these standards, which placed greater responsibility on the auditors in several areas (e.g., Berton 1987). These additional tasks were required to be conducted as a part of all audits, and thus were incurred by all suppliers. Given that audits are mandatory, the effect of an industry-wide cost increase should be to increase prices for all clients. While the standards were not mandated to take immediate effect, many accounting firms, including the Big 6, chose to adopt the standards early (Spires and Williams 1990).

In our empirical analysis, we assess whether audit fees rose following the implementation of the "expectation gap" standards.

Production Efficiencies

Since the early 1980s, there have been significant changes in the production technology in auditing. The increased integration of computers into the audit process, the greater use of analytical procedures, which has resulted in a decline in detailed testing, and the incorporation of risk-based auditing procedures are expected to have a significant increase in an auditor's production efficiency.(2) These changes have had the effect of replacing costly detailed tests of year-end balances with more cost-effective procedures. Many auditors have developed more structure in their audit approach (Cushing and Loebbecke 1986), which is also expected to yield production cost advantages (Gist 1994).

Improvements in audit production efficiency should cause the required audit effort (in terms of hours expended on a given audit task) to decline over time for all the big accounting firms. Elliott (1998) notes that audit technology has been making auditing less labor intensive. Given the nature of the audit market, production cost reductions should be reflected in decreasing audit fees. In our study, we attempt to determine if the changes in the audit fee structure over time can be attributable to production efficiencies.

Competition in the Audit Market

Many audit fee studies have been directed toward understanding the nature of competition in the audit market. Simunic (1980), in a seminal study, finds evidence of a Big 6 discount, attributable to scale economies. However, subsequent studies, such as Francis (1984), Craswell et al. (1995), and Palmrose (1986), detect a Big 6 price premium, plausibly attributable to a higher quality of assurance offered by these firms. Simon and Francis (1988) and Walker and Castarella (2000), among others, study the pricing of new engagements, to see if there is evidence of low-bailing, and show that auditors appear to discount prices in new engagements. Maher et al. (1992), study fee changes over 1977-1981, a period in which the profession dropped many of its restrictions against competition. They report a decline in audit fees over this period.

Our study examines the period 1980-1997. The competitive dynamics of the audit industry have changed over this period. What once were the Big 8 accounting firms were reduced through mergers to the Big 6. Because of the relatively small number of suppliers in this quality segment, there has been concern about the effects of mergers on audit fees. Iyer and Iyer (1996) summarize these arguments. They study large firm mergers in the U.K., but find no effect on the level of fees. Merging auditors may be able to raise prices if they benefit from increased market power. However, mergers may also convey cost advantages to the merging auditors, making them more efficient suppliers. In this event, the fees of the merging auditors should decline to reflect these lower costs.

Changes in the competitive environment may have the effect of increasing or decreasing prices for those who seek industry-specialist auditors. Industry specialists are auditors with a high share of the audit business in a particular industry. Hogan and Jeter (1999) report that large auditors have increased their market share in industries in which they are specialists, and indicate that a potential motive for audit firm mergers is to increase specialization in certain industries. Craswell et al. (1995), in a study of the Australian market, find that industry-specialist Big 6 auditors are able to charge a price premium over nonspecialist Big 6 auditors. Craswell et al. (1995) attribute their results to the investments made by industry specialists in order to acquire their industry-specific expertise. It is likely that greater market share also conveys the ability to reduce audit costs on an engagement both through greater expertise and through scale economies. However, Craswell et al. (1995) note that the premium charged to provide a return on the investment in industry expertise dominates any production economies achieved by the industry specialists.

In our analysis we consider the effect of the mergers on audit fees. We also consider whether industry-specialist auditors price audits differently from other auditors, and whether the mergers caused any changes in this behavior.

Litigation Environment

The litigiousness of the environment for accountants increased significantly over the period of this study (Palmrose 1994).(3) Expected litigation costs should be passed on either to all clients in the form of higher fees or to risky clients in the form of specific fee adjustments. If expected litigation costs are not reflected in higher fees, then they will result in lower profits for the auditing firms. The evidence on this issue is not clear, and the relationship between audit fees and variables representing liability exposure has been weak in most audit fee studies (Simunic and Stein 1996). However, on balance, the evidence seems to suggest that auditors make client-specific risk adjustments in fees. The adjustment appears to be made through increased audit effort rather than through a fee premium.(4)

We control for litigation risk in our study by incorporating the client's financial leverage (LEVERAGE) in our audit fee model. We use LEVERAGE to represent financial condition. Pratt and Stice (1994) show that auditors' perceived litigation risk is dependent upon the client's financial condition. While we believe it is important to control for litigation risk, however, we do not investigate the effect of changes in litigation risk on the pricing of audit fees for two reasons. First, our sample, which is small, includes firms that are relatively large with respect to the population of firms. These firms may be less risky, on average, than a randomly drawn sample, and thus be an inappropriate sample to test changes in the way in which auditors price litigation risk.(5) Second, though LEVERAGE has often been used in audit fee models, it may not be a sufficiently good proxy to capture litigation risk in our sample, which is not homogeneous with respect to industry.

AUDIT FEE MODEL

Empirical models of audit fees have generally estimated fees as a function of factors that drive the quantity of audit effort, where audit effort refers to the hours spent on an engagement. The common explanatory factors considered by prior studies have been auditee size, complexity, and risk (Simunic and Stein 1996). As DeFond et al. (2000) note, these models have demonstrated high explanatory power and have been robust across different samples, countries, and time periods.

Most audit fee models incorporate a large number of independent variables. Given the relatively small sample size in our study, we chose to use a parsimonious model. Our model is based on the model used by Simon and Francis (1988), which includes a subset of the independent variables used previously by Francis (1984) and Francis and Stokes (1986). The Simon and Francis (1988) model is as follows: LOGFEE = [b.sub.0] + [b.sub.1]LOGASSETS + [b.sub.2]SQSUBS + [b.sub.3]FOREIGN + [b.sub.4]INVREC + [b.sub.5]OPINION + [b.sub.6]AUDITOR + [b.sub.7]CNGAUD + e, where the dependent variable is the log of audit fees, LOGASSETS is the log of total assets, SQSUBS is the square root of subsidiaries, FOREIGN is the proportion of subsidiaries that are foreign operations, INVREC is the proportion of inventories and receivable to total assets, OPINION is a dummy variable that indicates whether the client received a qualified opinion, AUDITOR indicates the auditor type, and CNGAUD represents auditor tenure.

We modify the Simon and Francis (1988) model in the following way. We use the same dependent variable, renamed LFEE. We retain the variables LOGASSETS (renamed LASSETS), SQSUBS, and INVREC. LASSETS is our measure of auditee size, which has a well-established relationship with audit fees. Auditee size has been found to explain most of the cross-sectional variation in audit fees (Simunic 1980; Francis 1984; Craswell et al. 1995). SQSUBS is a measure of complexity. Audit fees are expected to increase as the complexity of the reporting entity increases because of the additional number of organizational units that require to be monitored. Many prior studies (e.g., Francis 1984; Francis and Stokes 1986; Maher et al. 1992) use SQSUBS to represent complexity and find it to be a significant explanatory variable. INVREC is the sum of accounts receivable and inventory, deflated by total assets. Simon and Francis (1986) argue that it is a measure of audit risk, though INVREC has been used also as a measure of client complexity (Johnson et al. 1995). It is related to the importance of specific audit procedures like the confirmation of receivables and verification of inventory.(6)

We omit the variable FOREIGN, which has had mixed results in the literature. The variable was found to be significant by Simon and Francis (1988) (as well as by Simunic [1980] and O'Keefe et al. [1994], among others) but it was found insignificant by Maher et al. (1992) and Craswell et al. (1995).(7)

We replace OPINION with LEVERAGE (the proportion of total debt to total assets). LEVERAGE is more suitable for our sample because of the low frequency of qualified opinions in our sample.(8) Going-concern audit reports were issued for only 4.4 percent of the observations (firm years) in our sample. LEVERAGE, a measure of the firm's financial condition, has been used often in audit fees studies to represent litigation risk (e.g., O'Keefe et al. 1994; Simunic and Stein 1996; Craswell et al. 1995).

Finally, we omit AUDITOR from our model because our study is restricted to clients audited by the Big 6, and we omit CNGAUD, which is included by Simon and Francis (1988) specifically to test the hypotheses of interest in their study. We replace AUDITOR with MSHARE, a measure of the auditor's specialization. MSHARE is calculated as the auditor's market share in the industry in which the client operates.(9) Greater market share may provide a firm the opportunity to charge premium prices (Craswell et al. 1995). At the same time, it is likely that greater market share also provides production economies.

In summary, we use the following cross-sectional model as the basis for our analysis:

[LFEE.sub.it] = [b.sub.0t] + [b.sub.1t] [LASSETS.sub.it]

+ [b.sub.2t] [SQSUBS.sub.it]

+ [b.sub.3t] [INVREC.sub.it]

+ [b.sub.4t] [MSHARE.sub.it]

+ [b.sub.5t] [LEVERAGE.sub.it] (Model 1)

where LFEE represents the log of the audit fee charged to the client, LASSETS is the log of the client's total assets, SQSUBS is the square root of the number of subsidiaries operated by the client, INVREC is the proportion of the client's assets represented in accounts receivable and inventory, MSHARE is the audit firm's market share, LEVERAGE is the proportion of the client's total debt to its total assets, and i and t are subscripts for firm and year, respectively. We modify this model for certain analyses.

Period of Study

Virtually all the published audit fee studies have been cross-sectional analyses that rely on fee data drawn from a short period of time, typically a single year. A notable exception is Maher et al. (1992), who model audit fee changes (rather than audit fees) from 1977 to 1981.(10) Maher et al. (1992) employ a sample of 78 firms for which data are publicly available. Maher et al. (1992) conclude that audit fees decreased over the 1977-1981 period after controlling for their covariates, since the intercept from their model is negative and significant. They attribute the result to increasing competition over the time period studied. However, the Maher et al. (1992) study is limited in that it uses only two years of data, and the period covered by the study is only five years. Our study covers the period 1980-1997, and we use data from each of the 18 years covered by the study.(11)

METHOD AND RESULTS

Sample

Companies are not required to make mandatory disclosures of the amounts they pay as audit fees. Data used in prior audit fee studies have generally either come from surveys (e.g., Simunic 1980), disclosures made in countries where audit fees are required to be disclosed, like Australia (e.g., Francis 1984), or from an auditing firm allowing access to its audit fee database (e.g., Davis et al. 1993). Our sample consists of firms that voluntarily disclosed audit fee data in SEC filings, chiefly the proxy statement, between 1980 and 1997 (both inclusive). Excluded from our sample were publicly traded ADRs, limited partnerships, and trusts. We also excluded firms that disclosed only the total compensation paid to their auditor for all services obtained and did not specify the amount paid for the audit. Finally, our sample was restricted to firms with the then Big 6 (8) auditors and with available market price data. The sample was restricted to Big 6 audit firms because there are known differences between Big 6 and non-Big 6 audit firms in pricing, presumably because the Big 6 firms offer a differentiated product (e.g., Palmrose 1986; Francis and Simon 1987). There were only a very small number of audit fee disclosures provided by clients that engaged non-Big 6 auditors.

Table 1, Panel A presents the final sample distribution across years. The largest sample is obtained in 1982 (90 firms) while the smallest sample is obtained in 1994 (66 firms). 1,330 firm year disclosures are available for the 18-year period, generated by 249 firms. The number of disclosures per firm ranged from 1 (22 firms) to 18 (8 firms), with the average number of disclosures per firm being 5.3. On average, 53.7 percent of the sample firm-years are drawn from the manufacturing industry, 14.8 percent are transportation or utility firms, 10.9 percent are retailers, 14.3 percent are financial services firms, and 6.7 percent are other firms.(12)
TABLE 1
Sample Description(a)

Panel A: Sample Composition by Year

          No. of        Mean Audit Fee    Mean Assets
Year    Observations    (in thousands)    (in millions)

1980        85             $1,280            $8,106
1981        87              1,347             8,302
1982        90              1,500            10,304
1983        80              1,584             9,179
1984        71              1,811             9,013
1985        71              1,915            10,155
1986        74              2,019            12,466
1987        76              2,355            13,838
1988        76              2,633            14,810
1989        73              2,680            15,941
1990        70              3,080            17,219
1991        68              3,065            15,421
1992        70              2,969            15,701
1993        68              3,004            15,747
1994        66              2,976            16,177
1995        68              2,835            15,371
1996        69              2,762            16,722
1997        68              2,752            19,475

Panel B: Descriptive Data

                                 Sample Companies
                                     (n = 68)

                                Mean    Std. Dev.

Assets (billions of $d)        $19.5     $51.9
Return on Assets (%)             1.8      12.6
Debt-equity ratio                3.1       5.9
Current ratio                    2.1       3.4
Price-earnings ratio            23.1      12.9
Book-market ratio (%)           37.2      30.4

                                Nondisclosing Firms
                                    (n = 5,684)

                              Mean   Std. Dev.   t-value

Assets (billions of $d)        $2.3     $13.9     2.73(*)
Return on Assets (%)          -0.7      16.7      1.68
Debt-equity ratio              2.6       5.3      0.86
Current ratio                  2.8       3.0     -1.70
Price-earnings ratio          23.2      13.4     -0.02
Book-market ratio (%)         44.1      58.4     -1.83

(*), (**) = p < 0.01, p < 0.05, respectively.
(a) Panel A breaks the audit fee-disclosing sample down by year. Panel
B compares some characteristics of the audit fee-disclosing sample with
nondisclosing finns for the year 1997.


Given that only a small number of firms elect to provide such voluntary disclosures, it is important to compare the characteristics of these firms with nondisclosure firms to examine for any systematic bias in the sample. Table 1, Panel B presents descriptive information for the 68 sample firms which made audit fee disclosures in 1997, and provides, as a comparison, data for 5,684 nondisclosing firms. This comparison sample consists of firms included in Compact Disclosure that meet two restrictive criteria applied to the audit fee-disclosing sample: (1) they have a Big 6 auditor and (2) they have available market price data. Common descriptive statistics were collected to measure auditee size (assets), profitability (return on assets), leverage (debt/equity ratio), liquidity (current ratio), and market valuation (price/earnings ratio and book value of equity/market value of equity). Univariate statistics show that firms voluntarily disclosing audit fees tend to be larger than nondisclosing firms. No statistically significant differences were found on any other dimensions.(13)

Our sample is biased in that it contains only firms that voluntarily disclose audit fees, and the factors that lead to voluntary fee disclosure are unknown.(14) However, the sample is interesting in its own right. Although the sample has a relatively small number of firms, comprising only 1.2 percent of the firms that meet the auditor-type and market value-availability criteria in 1997, the asset value of these firms is material, accounting for 9.3 percent of the asset value of all firms meeting these criteria.

Are Audit Fees Rising or Falling?

In examining the audit fee trend over the 18-year period of the study, it is important to control for changes in the audit fee drivers, auditee size, complexity, and risk. We use the following procedure. First, we estimate the audit fee model (Model 1) for 1980. Using this 1980 model as a fee expectations model, we predict the audit fee (PFEE) for each firm for each year from 1981-1997. We label the prediction error as FEEDIFF, such that [FEEDIFF.sub.it] = ([LFEE.sub.it] - [PFEE.sub.it])/[PFEE.sub.it], where LFEE is the natural logarithm of the actual fee paid, PFEE is the predicted log fee using the 1980 model, and i and t are subscripts for firm and year, respectively. If the fee changes only in proportion to changes in levels of the independent variables in the audit fee model, then FEEDIFF, the audit fee prediction error, should not be significantly different from zero. Nonzero values of FEEDIFF capture the cumulative effects of changes in pricing.

Note that in this analysis we are interested only in whether there is a prediction error using the 1980 model as a benchmark. Positive (negative) prediction errors in a given year indicate that audit fees in that year are higher (lower) than predicted by the 1980 model. That is, positive (negative) numbers for FEEDIFF suggest that audit fees have increased (declined) when compared with 1980, after controlling for changes in the audit fee model independent variables. We do not attempt at this stage to assess whether the source of the prediction error is a change in the audit fee model over time, though this is an analysis we conduct later in the paper.

Model 1 is first estimated for the 85 firms included in the sample for 1980. Table 2 presents the results. It is evident that the model has good fit, with the adjusted [R.sub.2] at 0.854. Only three of the variables in the model (LASSETS, SQSUBS, and INVREC, representing auditee size, complexity, and proportion of risky assets, respectively) are significant in 1980.
TABLE 2
1980 Audit Fee Model Regression Results(a)
                   (n = 85)

Variable       Mean      Std. Error     t-value

INTERCEPT     -1.003       0.432       - 2.3(**)
LASSETS        0.442       0.032        13.8(*)
SQSUBS         0.140       0.025         5.6(*)
INVREC         0.020       0.003         5.9(*)
MSHARE        -0.001       0.004        -0.2
LEVERAGE      -0.018       0.130        -0.1

[R.sup.2]      0.854
F-value       99.5(*)

(*), (**) p < 0.01, p = < 0.05, respectively, two-tailed tests.
(a) This table provides statistics from the regression of the audit fee
model using 1980 data: LFEE = [b.sub.0] + [b.sub.1] LASSETS + [b.sub.2]
SQSUBS + [b.sub.3] INVREC + [b.sub.4] MSHARE + [b.sub.5] LEVERAGE.


The estimated intercept and coefficients are used to calculate PFEE, using realized values for the independent variables for each firm year. FEEDIFF is then calculated for each firm for each year. Table 3 shows the calculated annual mean values for FEEDIFF. Table 3 shows a positive and significant value for FEEDIFF first in 1983. Significant positive values in FEEDIFF persist through 1997, with the exception of 1995-1996, suggesting that starting in 1983, audit fee levels are higher than the 1980 audit fees after controlling for changes in auditee size, complexity, and risk.
TABLE 3
Fee Prediction Errors (FEEDIFF) for Full Sample(a)

  Year      FEEDIFF     t-value

  1980       0.001        0.1
  1981       0.001        0.1
  1982       0.006        0.8
  1983       0.018        2.0(**)
  1984       0.034        4.1(*)
  1985       0.031        2.9(*)
  1986       0.042        4.1(*)
  1987       0.041        4.2(*)
  1988       0.043        4.2(*)
  1989       0.046        4.4(*)
  1990       0.044        4.2(*)
  1991       0.040        3.3(*)
  1992       0.039        3.4(*)
  1993       0.038        3.4(*)
  1994       0.034        2.9(*)
  1995       0.020        1.4
  1996       0.019        1.3
  1997       0.028        2.4(**)

(*), (**) p < 0.01, p < 0.05, respectively, two-tailed tests.
(a) The table provides mean values of FEEDIFF, the prediction error for
audit fees, for all companies reporting audit fees for the specified
year. The t-values are for differences from zero. Predicted audit fees
are calculated using the 1980 audit fee model in Table 2.


Price-Level Adjustment

One possible reason for the observed increase in audit fees is that the price of the principal input into the audit process, the wage rate for audit labor, has gone up. If audit effort remains the same but audit wages have increased, then the actual fee should increase relative to the predicted fee, and we should observe an increase in FEEDIFF over time. Similarly, FEEDIFF values may be due to changes in general price levels. For example, if all other factors stay the same, but a firm's total assets increase simply due to inflation, then the predicted audit fee would increase and the prediction error become more negative.

If FEEDIFF is observed to increase even after adjusting for audit wage-rate changes and general price-level changes, the number of possible explanations for the change is reduced. Two remaining plausible explanations are that audit effort increased or that margins on audit engagements increased. In the next analysis, we adjust for audit wage-rate changes and general price-level changes. The adjustment effectively reduces audit fees to 1980 wage rates and client assets to 1980 price levels.(15)

We adjust only one of the independent variables in the audit fee model (Model 1), LASSETS. SQSUBS is not measured in dollars and thus does not require adjustment. We do not adjust INVREC, LEVERAGE, or MSHARE, all ratios or percentages, assuming that the numerator and denominator in each case would require similar adjustments.

To adjust audit fees, we construct an accounting wage index using auditor salary data. The salary data are obtained from annual salary surveys conducted by Robert Half & Associates. We use the surveys for 1980-1997. The selected data pertain to large public accounting firms (firms with over 1,000 employees). The survey provides mean salaries for "staff," " advanced staff," "senior," " manager-supervisor," and "manager" levels. We calculate the percentage change in salary for these five positions each year. The accounting wage index is then constructed using the annual mean of the percentage change in salary for the five positions and setting the base year (1980) to 100.(16)

The accounting wage index (labeled ACINDEX) is shown in Table 4 for the years 1980-1997. Table 4 also shows the CPI, adjusted to the base year 1980, for the corresponding years. It is evident that the CPI has outpaced the accounting index over the period of the study, suggesting that the input costs for accounting firms have not risen as quickly as prices for other goods and services, on average. Most of the cumulative difference in the CPI and ACINDEX, however, is attributable to the first few years of the period studied. The difference between the two indices through 1997 is 21.2, while the difference through 1984 is 11.5.
TABLE 4
Price Level Adjustment(a)

  YEAR        ACINDEX       CPI

  1980         100.0       100.0
  1981         104.2       110.3
  1982         108.6       117.1
  1983         110.6       120.9
  1984         114.6       126.1
  1985         118.5       130.6
  1986         122.2       133.0
  1987         127.3       137.9
  1988         131.1       143.6
  1989         138.0       150.5
  1990         141.7       158.6
  1991         148.8       165.3
  1992         153.9       170.3
  1993         156.9       175.4
  1994         160.7       179.9
  1995         164.0       185.0
  1996         168.7       190.4
  1997         173.6       194.8

(a) This table provides an accounting wage index and the consumer price
index for the years 1980-1997, both using 1980 as the base year.


To remove these price-level effects from our consideration of audit fee trends, the price-level-adjusted audit fee prediction error (FEEDIFF2) is calculated as LFEE2 - PFEE2. Here, LFEE2 is log of the deflated audit fee, where the deflator is the accounting wage index (ACINDEX). To calculate PFEE2, Model 1 is used to predict audit fees for each client for each year as before, but LASSETS is replaced by LASSETS2, where LASSETS2 is the log of the firm's assets deflated by the CPI for the corresponding year. Table 5 shows FEEDIFF2 values using the price-level-adjusted calculations.
TABLE 5
Deflated Fee Prediction Errors (FEEDIFF2) for Full Sample(a)

   Year       FEEDIFF2     t-value

   1980         0.001        0.1
   1981         0.002        0.2
   1982         0.004        0.5
   1983         0.015        1.7
   1984         0.030        3.5(*)
   1985         0.024        2.1(**)
   1986         0.031        2.6(*)
   1987         0.026        2.6(*)
   1988         0.026        2.6(*)
   1989         0.025        2.3(**)
   1990         0.023        2.1(**)
   1991         0.014        1.1
   1992         0.010        0.7
   1993         0.007        0.6
   1994         0.001        0.1
   1995        -0.018       -1.1
   1996        -0.021       -1.4
   1997        -0.014       -1.1

(*), (**) p < 0.01, p < 0.05, respectively, two-tailed tests.
(a) The table provides mean values of FEEDIFF2, the deflated prediction
error for audit fees, for all companies reposing audit fees for the
specified year. Predicted audit fees are calculated using the 1980 audit
fee model in Table 2. FEEDIFF2 is adjusted for changes in accounting
wage rates and the general price level index.


It is interesting to note from Table 5 that making the price-level adjustments results in removing the significant cumulative price changes in the 1990s. Starting from 1991, the sample does not show a significant mean prediction error in audit fees. In other words, when adjusted for accounting wage-rate changes and general price-level changes, audit fees in the 1990s appear to return to the 1980 price levels.

Sample Composition Problem

One problem with the analysis reported in Table 5 is the changing sample composition. Audit fee disclosing firms enter and exit the sample each year. For reasons of sample size, we could not restrict the sample to firms disclosing audit fees for all 18 years covered by the study--only eight firms would meet this criterion. While we seek to explain Model 1 prediction errors as being due to underlying economic trends, it may be that these prediction errors exist simply due to changes in sample composition. In this analysis, we test the result shown in Table 5 for sensitivity to sample composition changes.

From the original 18-year sample we create four subsamples: Sample 1: 1980-1988, Sample 2: 1983-1991, Sample 3: 1986-1994, and Sample 4: 1989-1997. To be included in a subsample, a firm needed to make continuous audit fee disclosures for every year covered by the subsample. Using this criterion ensured that an analysis of cross-sectional regressions for a subsample over time would not be confounded by changes in the composition of the subsample. There were 30 firms in Sample 1, 40 firms in Sample 2, 43 firms in Sample 3, and 31 firms in Sample 4.

Mean FEEDIFF2 values are calculated for each subsample for each year and reported in Table 6. Results generally are consistent with the finding from Table 5. Significant positive values observed for Sample 1 and Sample 2 through the 1980s, but none of the samples shows positive significant values for FEEDIFF2 in the 1990s. Samples 1 and 2 show inconsistent results for some years (1983, 1985, 1986, and 1987), but suggest a trend of positive significant values for FEEDIFF2 in the 1983-89 period, in contrast to the 1984-89 period for which the full sample showed positive significant values for FEEDIFF2.(17)
TABLE 6
Deflated Prediction Errors (FEEDIFF2) for Time-Based Subsamples(a)

          Sample 1     Sample 2    Sample 3   Sample 4

1980       -0.008
1981        0.007
1982        0.021
1983        0.029(**)   0.017
1984        0.029(**)   0.022(**)
1985        0.037(*)    0.013
1986        0.033(**)   0.010       0.002
1987        0.037(*)    0.013       0.005
1988        0.050(*)    0.024(**)   0.013
1989                    0.024(**)   0.015       0.007
1990                    0.020       0.014       0.015
1991                    0.019       0.014       0.018
1992                                0.021       0.019
1993                                0.017       0.020
1994                                0.016       0.022
1995                                            0.024
1996                                            0.015
1997                                            0.018

(*), (**) p < 0.01, p < 0.05, respectively, two-tailed tests.
(a) The table provides mean values of FEEDIFF2, the deflated prediction
error for audit fees, for companies in each of the four time-based
subsamples. Predicted audit fees are calculated using the 1980 audit fee
model in Table 2. FEEDIFF2 is adjusted for changes in accounting wage
rates and the general price-level index.


Did the "Expectation Gap" Standards Increase Audit Fees?

To detect the effect on audit fees of significantly increasing audit scope, it is necessary to determine whether fee changes in individual years are significant. We break down the prediction error FEEDIFF2 into two components: [FEEDIFF2.sub.it] = [FEEDIFF2.sub.it]-1 + [CHGDIFF2.sub.it]. The first component, the cu-mulative audit fee prediction error for firm i through year t-1, impounds the effect of events from 1980 through year t-1. The second component (CHGDIFF2) consists of the error uniquely attributable to factors in year t. CHGDIFF2 allows us to identify whether significant fee shocks took place in a particular year. In contrast, FEEDIFF2 identifies the cumulative effect of annual shocks in audit fees. FEEDIFF2 can be significant in a particular year without CHGDIFF2 being significant in the corresponding year because the aggregation of insignificant (or significant) changes in prior years created a significant cumulative difference.

Table 7 presents mean calculated values of CHGDIFF2 for each of the subsamples. Table 7 shows that there are only three years in which statistically significant values of CHGDIFF2 are obtained. A statistically significant increase is reported in 1981 and 1982 for Sample 1, the only sample covering these years. The only year showing a statistically significant price change across multiple samples is 1988. All three of the samples covering the year 1988 show significant values for CHGDIFF2. Since changes in audit wage rates are ruled out in this analysis, the evidence is strong that there was an expansion in audit fee effort in 1988. The evidence from Table 7 provides support for the assertion that the "expectation gap" standards, which were introduced in 1988, increased audit fees.(18)
TABLE 7
Change in Deflated Prediction Errors (CHGDIFF2) for
Time-Based Subsamples(a)

         Sample 1    Sample 2    Sample 3     Sample 4

1980     --
1981     0.015(**)
1982     0.015(*)
1983     0.007           --
1984     0.001           0.005
1985     0.007          -0.009
1986    -0.003          -0.003        --
1987     0.004           0.003        0.003
1988     0.013(*)        0.011(*)     0.008(**)
1989                    -0.001        0.001      --
1990                     0.003       -0.001      0.007
1991                    -0.001       -0.001      0.003
1992                                  0.007      0.001
1993                                 -0.005      0.001
1994                                  0.000      0.002
1995                                             0.002
1996                                            -0.009
1997                                             0.003

(*), (**) p < 0.01, p < 0.05, respectively, two-tailed tests.
(a) The table provides mean values of CHGDIFF2, the change in the
deflated prediction error for audit fees, for companies in each of the
four time-based subsamples. Predicted audit fees are calculated using
the 1980 audit fee model in Table 2. CHGDIFF2 is adjusted for changes
in accounting wage rates and the general price-level index.


Changes in the Audit Fee Model

The previous analyses provide evidence that, using the 1980 model as a benchmark, audit fees increased over the period 1980-1997, though when predicted fees are adjusted for changes in wage rates and general price levels, the fee levels through the 1990s appears to return to the 1980 level. It is possible that this change is due to a change in the audit fee model. For example, if the audit fee prediction error is due to increased audit effort (in terms of hours spent on audit tasks) and this increase is proportional with the size of the client, then we should see the audit fee model change to reflect a larger coefficient for size. In this section we investigate whether the fee increases are due to a change in the audit fee model. We estimate Model 1 for each of the 18 years in the sample. The data for these regressions are not adjusted for price-level changes. The number of observations for each year's regression is shown in Table 1 Panel A. Results for the regressions are shown in Table 8.
TABLE 8
Audit Fee Model Regression Coefficients(a)

Year       Intercept      LASSETS        SQSUBS        INVREC

1980      -1.003(**)    0.442(*)       0.140(*)      0.020(*)
1981      -1.442(*)     0.509(*)       0.102(*)      0.018(*)
1982      -1.411(*)     0.528(*)       0.133(*)      0.014(*)
1983      -1.411(*)     0.526(*)       0.113(*)      0.017(*)
1984      -0.995        0.493(*)       0.128(*)      0.016(*)
1985      -1.700(*)     0.535(*)       0.097(*)      0.018(*)
1986      -1.226(**)    0.519(*)       0.075(*)      0.014(*)
1987      -1.835(*)     0.559(*)       0.084(*)      0.014(*)
1988      -2.029(*)     0.569(*)       0.093(*)      0.015(*)
1989      -1.667(*)     0.534(*)       0.088(*)      0.011(*)
1990      -2.265(*)     0.576(*)       0.081(*)      0.012(*)
1991      -2.350(*)     0.631(*)       0.095(*)      0.015(*)
1992      -1.987(*)     0.599(*)       0.089(*)      0.011(*)
1993      -1.756(*)     0.547(*)       0.103(*)      0.012(*)
1994      -1.000(*)     0.512(*)       0.146(*)      0.010(*)
1995      -2.001(*)     0.573(*)       0.107(*)      0.014(*)
1996      -1.352(*)     0.511(*)       0.146(*)      0.011(*)
1997      -0.672        0.457(*)       0.169(*)      0.011(*)

Year        MSHARE       LEVERAGE    [R.sup.2] (adj)

1980        -0.001        -0.018          0.854
1981        -0.001        -0.420          0.856
1982        -0.005        -0.819(**)      0.858
1983        -0.004        -0.674          0.840
1984        -0.002        -0.514          0.821
1985        -0.010        -0.063          0.801
1986        -0.003        -0.246          0.818
1987        -0.002        -0.251          0.857
1988        -0.001        -0.254          0.870
1989         0.001         0.131          0.873
1990        -0.004         0.221          0.885
1991        -0.006        -1.095(*)       0.879
1992        -0.004        -0.775(**)      0.882
1993         0.007        -0.361          0.894
1994         0.008        -1.080(**)      0.870
1995        -0.007        -0.457          0.849
1996        -0.001        -0.468          0.859
1997        -0.006        -0.240          0.884

(*), (**) p < 0.01, p < 0.05, respectively, two-tailed tests.
(a) This table provides coefficients from the 18 annual regressions
of the audit fee model using 1980 data: LFEE = [b.sub.0] + [b.sub.1]
LASSETS + [b.sub.2] SQSUBS + [b.sub.3] INVREC + [b.sub.4] MSHARE +
[b.sub.5] LEVERAGE.


Model 1 performs well for each of the 18 years under study. The [R.sup.2] values are high, averaging 0.858 across the 18 years. The variables that were significant in the 1980 regression in Table 2 (LASSETS, SQSUBS, and INVREC) are significant also in each of the following 17 years. This is consistent with prior research. MSHARE is not significant in any of the regressions. We comment on this point later. LEVERAGE is significant in four of the 18 years, but the sign on the coefficient is negative in those years rather than positive. We would expect a positive coefficient if LEVERAGE is a good proxy for litigation risk and if litigation risk is included in audit fees.(19)

If the relationship between audit fees and the audit fee drivers has remained constant over time, then the coefficients derived from annual cross-sectional regressions should be stable over the 18-year period. On the other hand, if there are systematic effects on audit fees from changes in the audit environment, then these effects should be revealed in changes in the coefficients. One apparent trend is a downward drift in the coefficient for INVREC. In a later section we test to see if this trend is statistically significant.

Competition in the Audit Market

We examine two different aspects of audit market competition in this study. First, we consider the auditor's expertise in the client's industry. Next, we evaluate the effect of the Big 6 mergers in 1989.(20)

The auditor's expertise in the client's industry, as measured by the market share, can have two potential effects on audit pricing. One possible effect is that the auditor becomes the preferred choice in the industry segment, and consequently is able to charge a fee premium. The second potential effect is that the auditor's industry expertise allows the auditor to conduct audit engagements in the industry segment more economically than competitors. If either explanation holds true for our sample, then the variable MSHARE should be significant in Model 1 regressions. A positive coefficient would provide support for a pricing premium by industry specialists. A negative coefficient is consistent with industry specialists having cost advantages.

However, Tables 2 and 8 show that MSHARE is not significant in any of the years of the study, suggesting the absence, in our sample, of either a fee premium or a fee discount for auditors who are industry specialists. This result is inconsistent with Craswell et al. (1995). Additionally, MSHARE tends to have a negative coefficient for this sample (15 out of 18 regressions), as against the positive relationship obtained by Craswell et al. (1995). The 1989 Big 6 mergers do not result in MSHARE becoming significant.

To consider the effect of the Deloitte & Touche merger and the Ernst & Young merger on fees, we modify Model 1 as follows:

LFEES = [b.sub.0t] + [b.sub.1t] LASSETS + [b.sub.2t] SQSUBS + [b.sub.3t] INVREC + [b.sub.4t] MSHARE + [b.sub.5t] LEVERAGE + [b.sub.6t] DTEY (Model 2)

where DTEY is defined as 1 if the auditor is Deloitte & Touche or Ernst & Young or one of the predecessor auditors of these two firms, and the other variables are as previously defined. If the merger resulted in the two merged firms charging higher prices than competitors, then the DTEY term should be positive and significant in the post-merger years. If the merger resulted in cost advantages for the merged firms that are passed on to clients, then DTEY should be negative and significant.

Note that if the 1989 mergers result in increasing fees for all firms because of the reduced number of suppliers, then we should not expect to see significant positive coefficients for DTEY after 1989. Instead, we should observe significant positive values of CHGDIFF in the post-1989 period in Table 7. However, Table 7 does not show significant values for CHGDIFF in this post-merger period.

Annual regressions are run for Model 2. Because of the entry and exit of firms from the sample, the number of Deloitte & Touche and Ernst & Young clients changes from year to year. Table 9 reports the number of sample firms that are audited by the two merged accounting firms or their predecessors each year. The coefficients for the nonauditor terms are virtually unchanged from those reported in Tables 2 and 8, and are not shown here. Table 9 shows the annual coefficients for the auditor term DTEY only. We provide the coefficients for all 18 years to see if the post-merger pricing is a continuation of pre-merger practices. Interestingly, the coefficients are positive for 16 of the 18 years for which regressions are run. However, they are significant only for the three years following the mergers (1991-1993), after a one-year lag. This is consistent with a short-lived premium that is obtained in the three years following the merger but dissipates thereafter. It should be noted, however, that each firm only has a small number of clients in the sample, so the generalizability of the results may be limited.
TABLE 9
Effect of Big 8 Mergers on Audit Fees Model(a)

         No. of Observations

Year    DTEY = 0    DETY = 1    DTEY Coefficient   t-value

1980       55          30            0.188           1.5
1981       57          30            0.194           1.5
1982       60          30            0.204           1.8
1983       53          27            0.175           1.4
1984       46          25            0.143           1.1
1985       51          20            0.159           1.0
1986       52          22            0.231           1.6
1987       54          22            0.128           0.9
1988       51          25            0.092           0.7
1989       52          21            0.134           1.0
1990       50          20            0.228           1.8
1991       47          21            0.326           2.3(**)
1992       48          22            0.333           2.2(**)
1993       48          20            0.326           2.2(**)
1994       45          21            0.017           0.1
1995       47          21           -0.032          -0.1
1996       50          19            0.288           1.3
1997       51          17           -0.013          -0.1

(*), (**) p < 0.01, p < 0.05, respectively.
(a) This table provides partial results from the regression model
LFEE = [b.sub.0] + [b.sub.1] LASSETS + [b.sub.2] SQSUBS + [b.sub.3]
INVREC + [b.sub.4] MSHARE + [b.sub.5] LEVERAGE + [b.sub.6] DTEY. Only
the coefficient for DTEY for each of the years 1980 through 1997 is
shown. DTEY = 1 if the auditor is Deloitte & Touche or Ernst & Young or
one of their predecessor firms.


Production Efficiencies

Recall Model 1, in which audit fees are estimated as a function of the client's size, complexity and risk.

[LFEE.sub.it] = [b.sub.0] + [b.sub.1] [LASSETS.sub.it] + [b.sub.2] [SQSUBS.sub.it] + [b.sub.3] [INCVREC.sub.it] + [b.sub.4] [MSHARE.sub.it] + [b.sub.5] [LEVERAGE.sub.it] (Model 1)

As the levels of the independent variables in this model increase (decrease) over time, reflecting greater (less) size, complexity or risk, the amount of audit effort increases (decreases), resulting in higher (lower) audit fees. However, the coefficients can change as well, and for several reasons, such as: (1) change in audit wages, (2) change in audit scope, and (3) change in production methods. A change in audit wages will change audit fees without a corresponding change in audit effort, while changes in audit scope and production methods will change audit effort without any underlying changes in the client's size, complexity or risk.

Table 5 shows that the audit wage rate has increased over the period of the study. An increasing audit wage rate in itself increases the coefficients for LASSETS, INVREC, SQSUBS, and LEVERAGE. Since MSHARE may represent both the premium earned by the auditor from specialization and the specialized auditor's cost advantages, the effect on MSHARE is unclear. The effect on the coefficient for LASSETS should be partially offset by changes in general price levels reflected in LASSETS.

Increases in the assurance level, or expansions of audit scope, have taken place in the period of our study, most notably with the "expectation gap standards," but also through other auditing and financial accounting standards, as well as through new business practices such as the widespread use of derivatives. These increases in scope should also be reflected in greater coefficients for LASSETS and SQSUBS, since expanded scope should require proportionately more effort in larger and more complex firms. The effect of the expanded scope on INVREC is less certain, and depends on whether the change in scope disproportionately affects firms with a greater intensity of accounts receivable and inventory. Likewise, the effect of expanded scope on LEVERAGE depends on whether the increase in scope varies with the riskiness of the firm.

We argued earlier that since the early 1980s, there have been significant advances in the production technology being employed in audits, such as the increased use of analytical procedures and of computer-aided audit methods. These improvements should lead to declining coefficients in the audit fee model. In general, we should observe declining coefficients for LASSETS, SQSUBS, INVREC, and LEVERAGE, since audit effort should decline across the board for all firms, holding all else constant. In particular, we expect an effect on the coefficient for INVREC ([b.sub.3]). Louwers (1998) argues that the majority of audit work is devoted to inventory verification and receivables confirmation audit procedures. With increasing computerization and other audit technology advances, the tasks required for verifying inventory and confirming accounts receivable should consume less effort over time, which should result in a decrease in [b.sub.3].(21)

Considering the number of factors that affect the audit fee model coefficients and the offsetting effects of these factors, the final expectation for a trend in each coefficient is unclear. If positive coefficients are observed for LASSETS, INVREC, SQSUBS, or LEVERAGE, then it would appear that wage rate or audit scope effects dominate production efficiencies. However, if negative coefficients are observed, then a reasonable conclusion is that production efficiencies dominate wage rate and audit scope effects.

An inspection of Table 8 shows that [b.sub.3] appears to decline over the study period. The other coefficients do not show any obvious trend, perhaps because of the counteracting effects of assurance levels, wage rates, and production efficiencies. We formally test to see if the coefficients have changed over time by using the following model:

[b.sub.x] = [d.sub.0x] + [d.sub.1x] YEAR (Model 3)

Six different regressions are estimated for Model 3, one for each set of coefficients estimated from Model 1 (including the intercept). Each regression has 18 observations, corresponding to the 18 years covered by the sample. The dependent variable for these regressions takes the form [b.sub.0] through [b.sub.5], where [b.sub.0] is the intercept and [b.sub.1] through [b.sub.5] are the coefficients for the independent variables estimated in Model 1. YEAR takes the value 1 (if the coefficient was obtained from the 1980 regression) through 18 (for the 1997 regression).

Table 10 presents the results from the six Model 3 regressions. Only one of the [b.sub.x] coefficients appears to change with time.(22) The INVREC regression has a high adjusted [R.sup.2] (0.63). The independent variable YEAR is negative and significant in this regression, suggesting that INVREC has decreased over time. The argument that auditors have introduced production efficiencies in the past two decades is a plausible explanation for this finding. In auditing current assets it appears that the fee-reducing effect of production efficiencies dominates the fee-increasing effect of higher accounting wages and increased audit scope. Given that several factors affect the audit fee model coefficients, the only inference we can draw from the lack of significance in the YEAR coefficients for the other variables in the audit fee model is that there is no dominating trend in those instances that is evident in our sample.
TABLE 10
Longitudinal Trends in Audit Fee Model Coefficients(a)

                                                   Intercept
Variable from Audit Fees Model                    ([d.sub.0])

INTERCEPT
Model: [b.sub.0] = [d.sub.0] + [d.sub.1] YEAR      -1.447(*)

LASSETS
Model: [b.sub.1] = [d.sub.0] + [d.sub.1] YEAR       0.513(*)

SQSUBS
Model: [b.sub.2] = [d.sub.0] + [d.sub.1] YEAR       0.101(*)

INVREC
Model: [b.sub.3] = [d.sub.0] + [d.sub.1] YEAR       0.018(*)

MSHARE
Model: [b.sub.4] = [d.sub.0] + [d.sub.1] YEAR      -0.003

LEVERAGE
Model: [b.sub.5] = [d.sub.0] + [d.sub.1] YEAR      -0.310

                                                     YEAR
                                                  Coefficient
Variable from Audit Fees Model                    (d.sub.1])

INTERCEPT
Model: [b.sub.0] = [d.sub.0] + [d.sub.1] YEAR       -0.012

LASSETS
Model: [b.sub.1] = [d.sub.0] + [d.sub.1] YEAR        0.002

SQSUBS
Model: [b.sub.2] = [d.sub.0] + [d.sub.1] YEAR        0.001

INVREC
Model: [b.sub.3] = [d.sub.0] + [d.sub.1] YEAR       -0.001(*)

MSHARE
Model: [b.sub.4] = [d.sub.0] + [d.sub.1] YEAR        0.000

LEVERAGE
Model: [b.sub.5] = [d.sub.0] + [d.sub.1] YEAR       -0.010(b)

Variable from Audit Fees Model                  Model [R.sup.2]

INTERCEPT
Model: [b.sub.0] = [d.sub.0] + [d.sub.1] YEAR       -0.043

LASSETS
Model: [b.sub.1] = [d.sub.0] + [d.sub.1] YEAR        0.010

SQSUBS
Model: [b.sub.2] = [d.sub.0] + [d.sub.1] YEAR       -0.018

INVREC
Model: [b.sub.3] = [d.sub.0] + [d.sub.1] YEAR        0.630

MSHARE
Model: [b.sub.4] = [d.sub.0] + [d.sub.1] YEAR       -0.038

LEVERAGE
Model: [b.sub.5] = [d.sub.0] + [d.sub.1] YEAR       -0.039

(*), (**) p < 0.01, p < 0.05, respectively.
(a) This table provides results from six different regression models.
The dependent variables ([b.sub.x]) are coefficients obtained from the
annual audit fee model regressions described in Table 8
(LFEE = [b.sub.0] + [b.sub.1] LASSETS + [b.sub.2] SQSUBS + [b.sub.3]
INVREC + [b.sub.4] MSHARE + [b.sub.5 LEVERAGE]). The independent
variable in each regression is the year of audit fee disclosure (YEAR).
The reported number is the coefficient for YEAR.


CONCLUSION

The audit fees literature has focused on cross-sectional studies and there is little evidence of long-term trends. Using a sample of Big 6 (8) audited clients, this study examines audit fees over the 18-year period beginning in 1980. The study provides some insights into audit fee trends, and into the influence on audit fees of major changes in the audit environment, such as mergers among the big accounting firms, significant new standards, and improvements in audit technology.

The sample size is small, due to the voluntary nature of audit fee disclosures. Restricting the sample to Big 6 auditees that voluntarily disclose audit fees likely results in selection biases, which limits the generalizability of the study's findings. For example, sample firms on average are larger than nondisclosing Big 6 auditees, and are likely to be larger also than non-Big 6 audited clients. It may be that larger clients have more bargaining power in relation to the auditor than smaller clients and are thus able to better manage the growth of audit fees. Also, auditors may introduce innovations in production methods first to large clients, which would result in fee reductions due to cost savings from production efficiencies being realized later for small clients.(23) Further, non-Big 6 auditors may have been slower, on average, in introducing production efficiencies than the Big 6. At the same time, some of our results, such as the effect on audit fees of the expectation gap standards, should apply to all segments of the audit market.

Despite these limitations, we believe this study makes several contributions to the accounting literature. First, there is no other academic study on long-term audit fee pricing that controls for variables like size, complexity, and risk, as well as for wage rate and general price-level changes. The absence of these studies is unsurprising, given the voluntary nature of audit fee disclosures. Our study is able to document the trend in audit fees over an extended period. Using 1980 as the base year, and applying price deflators to adjust audit fees to 1980 accounting wage levels and firm assets to 1980 general price levels, we conclude that the net effect of the dynamics in the audit environment is that audit fees in the period 1983-89 were significantly higher than the 1980 level. Audit fee levels in the 1990s are not significantly different from 1980 levels.

Second, we show that audit fees increased significantly in 1988. We attribute the increase to the Auditing Standards Board's introduction of the "expectation gap" standards, which many observers predicted would result in audit effort, and consequently audit fees, going up. This evidence of significant costs on auditees imposed by audit standards, provides important information for regulators and standard-setters, including the Auditing Standards Board.

Third, we document a short-lived increase in fees charged by the merged firms following the 1989 Big 8 mergers in the U.S. This provides some additional information on the question of whether mergers of large auditing firms have an anti-competitive effect in terms of pricing. Concerns about the level of competition in the audit market were expressed in the 1977 Cohen Commission report, and have been reiterated since then. Several empirical studies have provided evidence of price competition in the audit marketplace. Our study suggests that the 1989 mergers did not have a long-term effect on audit fees.

Finally, we provide evidence that audit fee drivers have changed over time. Specifically, we observe a decline in the coefficient for INVREC, the variable representing the proportion of accounts receivable and inventory in the balance sheet. This decline suggests that production efficiencies dominate increases in audit wage rates and audit scope effect in the audit of these assets. It appears that auditors have become more efficient in the audit of these assets over time and have passed on the cost savings to their clients.

Also of interest to the academic research in this area, we find no evidence to support Craswell et al.'s (1995) Australian findings that auditors charge a premium for a reputation as industry specialists. Auditors specializing in a particular industry do not appear to set audit fees differently from nonspecialists in our sample.

We would like to thank Peter Easton, WanYing Lin, Eric Spires, and Gary Taylor, the workshop participants at The Ohio State University, Jane Mutchler, and two anonymous referees. We are also appreciative of the data provided by Robert Half & Associates. Financial support for this project was provided by the Fisher College of Business.

(1) The Big 8 accounting firms became the Big 6 during the period covered in this study, and have since become the Big 5. We generally refer to these firms as the Big 6 in this paper.

(2) See Blocher and Loebbecke (1993) for a review of studies on the effectiveness of audit procedures. Ameen and Strawser (1994) and Hirst and Koonce (1996) provide evidence that the use of analytical procedures has increased in recent years.

(3) See Cloyd et al. (1998) for a review of recent developments in auditor litigation.

(4) See Simunic and Stein (1996) for a review of studies relevant to the issue of whether auditors price litigation risk.

(5) Table 1 shows that sample firms are larger than the average for the population as a whole. However, these firms are not leveraged to a lesser extent than the population as a whole.

(6) Some prior studies have used inventory and accounts receivable as individual variables (e.g., Simunic 1980), deflated by total assets, rather than the sum of accounts receivable and inventory. Several studies also use total current assets, deflated by total assets.

(7) When FOREIGN (measured as foreign assets divided by total assets, as in Simunic [1980]) is included in our regressions, it is not significant. Also, including FOREIGN does not change the other coefficients significantly or increase the [R.sup.2] by a material amount. Using 1997 data for example, the adjusted [R.sup.2] is 0.885 with FOREIGN and 0.884 without.

(8) Using OPINION rather than LEVERAGE did not change the reported results in any significant way.

(9) MSHARE is calculated as follows: [MSHARE.sub.nt] = [SALES.sub.ntj]/[SALES.sub.jt], where [SALES.sub.ntj] is the sum of the sales of auditor n's clients in industry j (measured by two-digit SIC code) for period t, and [SALES.sub.jt] is the sum of the sales of all firms in industry j in period t.

(10) Iyer and Iyer (1996) also use Maher et al. (1992) model and approach.

(11) The model used by Maher et al. (1992) is as follows: FEE = [b.sub.0] + [b.sub.1]REV + [b.sub.2]DIVERS + [b.sub.3]SUBS + [b.sub.4]FORGN + [b.sub.5]INV + [b.sub.6] RECV + e, where FEE is the change in audit fees from 1977 to 1981, REV is the change in the auditee's revenue, DIVERS is the change in the number of two-digit SIC codes in which the client is engaged, SUBS is the difference in the square root of subsidiaries, FORGN is the change in the ratio of foreign assets to total assets, INV is the change in the ratio of inventories to total assets, and RECV is the change in the ratio of receivables to total assets. We use assets as the size measure in our study. We do not include DIVERS or FORGN, but these measures are not significant if included in our regression model.

(12) The industry composition of the sample varies from year to year. Manufacturing firms range from a low of 46.4 percent of the sample in 1996 to 60.6 percent in 1984. Ranges for the other industries are as follows: transportation and utilities 11.8-17.1 percent; retail 6.8-14.1 percent; financial services 9.9-18.8 percent.

(13) A review of the descriptive statistics for the other 17 years (1980-1996) showed similar results.

(14) Eight of the firms in the sample have made disclosures for every year since 1976, when the SEC issued ASR No. 276. This ASR required that clients provide disclosures of all material (over 5 percent) nonaudit services provided by their auditor, as a percentage of the audit fee. ASR No. 276 was repealed in 1979, but these firms continued to voluntarily provide audit fee disclosures.

(15) Recall that FEEDIFF = (LFEE - PFEE)/PFEE, where LFEE and PFEE are the logs of the actual and predicted audit fees, respectively. Actual (predicted) audit fee = actual (predicted) audit effort in hours x actual (predicted) wage rate. If actual wage rates increase due to inflation, then LFEE and FEEDIFF should both increase, since predicted wage rates are based on 1980 numbers. If the CPI increases, then nominal values of LASSETS will increase, without corresponding increases in audit effort. This results in PFEE increasing and FEEDIFF decreasing.

(16) If we were to construct the index using the mean salary for the five positions each year, then we would likely overweight the managerial positions, which have higher salaries but are fewer in number in the accounting firms than staff positions.

(17) In contrast, FEEDIFF values (calculated using numbers unadjusted for price and wage level changes) are significant for the subsamples in all years starting with 1982. This is inconsistent with Table 3 for 1982, 1995 and 1996.

(18) We conduct the same analysis with FEEDIFF, decomposing [FEEDIFFit] as FEEDIFFit-1 + CHGDIFFit. Results were similar to those reported in Table 7 with only one difference. For Sample 1, CHGDIFF was not significant in 1981. This analysis also shows a significant shock in 1988 for all three samples again. The CHGDIFF2 analysis is more useful for our purpose since it rules out audit wage changes.

(19) It is plausible that the results obtained for LEVERAGE are due to sample characteristics. To consider the robustness of the model with respect to the proxy used for litigation risk, we used two alternate measures as well: MAXLOSS, the maximum loss sustained by the firm's equity holders over the year for which the audit fee was paid, and INDUSTRY, coded as 1 if the firm is in one of the four industries identified by Francis et al. (1994) (biotechnology, computers, electronics, and retailing) as having a high incidence of accounting-related litigation. The two alternate measures of litigation are used as replacements for LEVERAGE in Model 1. Neither INDUSTRY nor MAXLOSS are significant in any year. While significant inconsistent coefficients are not obtained in these regressions, there is no strong evidence that they serve as better proxies for litigation risk, assuming litigation risk is priced in audit fees. Using these alternate measures in the regression does not disturb the results obtained for the other variables. To obtain MAXLOSS we calculate all values of ([P.sub.d] - [P.sub.d+i]) /[P.sub.d], where [P.sub.d] and [P.sub.d+i] are the price on days d and d + i respectively, such that d goes from 1 to n - 1, n being the number of trading days in the 12-month period, and i goes from 1 to n-d. MAXLOSS is the lowest calculated value of ([P.sub.d] - [P.sub.d+i])/[P.sub.d].

(20) The mergers were announced in May 1989, but completed in late 1989.

(21) Auditors' use of computers has often involved inventory and accounts receivable applications (AICPA 1986). Inventory and accounts receivable also have commonly been suggested as areas where analytical procedures can be productively applied.

(22) The precision of the estimates of the slope coefficients [b.sub.x] may vary across the firms in our sample. This could lead to a heteroskedasticity problem in the Model 3 regressions (Saxonhouse 1977). To address this problem, we run additional analyses deflating [b.sub.x] and the independent variable in the cross-sectional regressions by the standard error of the slope coefficient estimates. Results are consistent with those reported in Table 10, except that the cross-sectional regression for INVREC ([b.sub.3]) produces a higher [R.sup.2] (=0.78).

(23) Hirst and Koonce (1996) provide some evidence that auditors are likely to use analytical procedures more extensively in larger clients.

REFERENCES

Ameen, E. C., and J. R. Strawser. 1994. Investigating the use of analytical procedures: An update and extension. AUDITING: A Journal of Practice & Theory (Fall): 69-76.

American Institute of Certified Public Accountants (AICPA). 1986. Auditors' Use of Microcomputers. New York, NY: AICPA.

Berton, L. 1987. Panel to recommend stiffer penalties, more audit safeguards to fight fraud. Wall Street Journal (April 27): A3.

Blocher, E., and J. K. Loebbecke. 1993. Research in analytical procedures: Implications for establishing and implementing auditing standards. In The Expectation Gap Standards: Progress, Implementation Issues, Research Opportunities. New York, NY: AICPA.

Cloyd, C., J. Frederickson, and J. Hall. 1998. Independent auditor litigation: Recent events and related research. Journal of Accounting and Public Policy (Summer): 121-142.

Craswell, A. T., J. R. Francis, and S. L. Taylor. 1995. Auditor brand name reputations and industry specializations. Journal of Accounting and Economics 20: 297-322.

Cushing, B., and J. Loebbecke. 1986. Comparison of Audit Methodologies of Large Accounting Firms. Sarasota, FL: American Accounting Association.

Davis, L. R., D. N. Ricchiute, and G. Trompeter. 1993. Audit effort, audit fees, and the provision of nonaudit services to audit clients. The Accounting Review 68 (January): 135-150.

DeFond, M. L., J. R. Francis, and T. J. Wong. 2000. Auditor industry specialization and market segmentation: Evidence from Hong Kong. AUDITING: A Journal of Practice & Theory (Spring): 49-66.

Elliott, R. K., and D. Pallais. 1997. Are you ready for new assurance services. Journal of Accountancy (June): 47-51.

--. 1998. Assurance services and the audit heritage. AUDITING: A Journal of Practice & Theory 17 (Supplement): 1-7.

Francis, J. R. 1984. The effect of audit firm size on audit prices. Journal of Accounting and Economics 6: 133-151.

--, and D. Stokes. 1986. Audit prices, audit differentiation, and scale economies: Further evidence from the Australian audit market. Journal of Accounting Research 24 (Autumn): 383-393.

--, and D. Simon. 1987. A test of audit firm pricing in the small client segment of the U.S. audit market. The Accounting Review (January): 145-167.

--, D. Philbrick, and K. Schipper. 1994. Shareholder litigation and corporate disclosures. Journal of Accounting Research 32 (Spring): 137-164.

Gist, W. E. 1994. Empirical evidence on the effect of audit structure on audit pricing. AUDITING: A Journal of Practice & Theory (Fall): 25-40.

Hirst, D. E., and L. Koonce. 1996. Audit analytical procedures: A field investigation. Contemporary Accounting Research (Fall): 457-486.

Hogan, C. E., and D. C. Jeter. 1999. Industry specialization by auditors. AUDITING: A Journal of Practice & Theory (Spring): 1-17.

Holman, F. W. 1997. Manufacturers Alliance Survey on Outside Audit Fees, 1997. Arlington, VA: Manufacturers Alliance.

Iyer, V. M., and G. Iyer. 1996. Effect of Big 8 mergers on audit fees: Evidence from the United Kingdom. AUDITING: A Journal of Practice & Theory (Fall): 123-137.

Johnson, E. N., K. B. Walker, and E. Westergaard. 1995. Supplier concentration and pricing of audit services in New Zealand. AUDITING: A Journal of Practice & Theory (Fall): 74-89.

Louwers, T. J. 1998. The relation between going-concern opinions and the auditor's loss function. Journal of Accounting Research 36 (Spring): 143-156.

Maher, M. W., P. Tiessen, R. Colson, and A. J. Broman. 1992. Competition and audit fees. The Accounting Review 67 (January): 199-211.

O'Keefe, T., D. Simunic, and M. Stein. 1994. The production of audit services: Evidence from a major public accounting firm. Journal of Accounting Research 32 (Autumn): 241-261.

Palmrose, Z-V. 1986. Audit fees and auditor size: Further evidence. Journal of Accounting Research 24: 97-110.

--. 1994. The joint and several vs. proportionate liability debate. Stanford Journal of Law, Business & Finance (Fall): 53-72.

Pratt, J., and J. D. Stice. 1994. The effects of client characteristics on auditor litigation risk judgments, required audit evidence, and recommended audit fees. The Accounting Review 69 (October): 639-656.

Saxonhouse, G. R. 1977. Regressions from samples having different characteristics. Review of Economics and Statistics (May): 234-237.

Simon, D. T., and J. R. Francis. 1988. The effects of auditor changes on audit fees: Tests of price cutting and price recovery. The Accounting Review (April): 255-269.

Simunic, D. A. 1980. The pricing of audit services: Theory and evidence. Journal of Accounting Research 18: 161-190.

--, and M. T. Stein. 1996. The impact of litigation risk on audit pricing: A review of the economics and the evidence. AUDITING: A Journal of Practice & Theory 15 (Supplement): 119-134.

Spires, E., and D. D. Williams. 1990. Auditors' adoption of SAS 58 audit reports. Accounting Horizons (September): 76-82.

Walker, P. L., and J. R. Castarella. 2000. The role of auditor profitability in pricing new engagements. AUDITING: A Journal of Practice & Theory 19 (Spring): 157-167.

Submitted July 1999

Accepted August 2000

Krishnagopal Menon is a Professor at Boston University and David D. Williams is an Associate Professor at The Ohio State University.
COPYRIGHT 2001 American Accounting Association
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2001 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Menon, Krishnagopal; Williams, David D.
Publication:Auditing: A Journal of Practice & Theory
Geographic Code:1USA
Date:Mar 1, 2001
Words:12202
Previous Article:Board Independence and Audit-Firm Type.
Next Article:Additional Evidence on Audit Report Lag.
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters