Does office size matter in client acceptance decisions? Evidence from big 4 accounting firms

According to Bray (2002), premiums for professional liability insurance have soared in wake of the Enron-Andersen affairs and large firms may face rate increases of more than 100 percent.

SOX 404 requires management and the external auditor to report on the adequacy of the company’s internal control over financial reporting, and thus caused an exogenous shock to the demand for Big 4 audit services. SOX 404 is effective for audits of all accelerated filer clients with fiscal years ending on or after November 15, 2004. The AS5, An Audit of Internal Control over Financial Reporting that is Integrated with an Audit of Financial Statements, was adopted by the PCAOB with the goal of improving the efficiency of integrated audits. The effective date for compliance with AS5 is for fiscal years ending on or after November 15, 2007. The financial recession began in December 2007 and ended June 2009 based on the National Bureau of Economic Research.

As Schroeder and Hogan (2013) indicate, it is difficult to disentangle the impacts of AS5 from those of the economic recession, as both alter the auditor’s relationships with existing clients and influence client acceptance and retention decisions.

Audit offices that experience high levels of recent growth measured by increase in audit fees and offices that gain major industry clients will encounter temporary capacity constraint and thus impair office-level audit quality (Bills et al. 2016; Francis et al. 2017).

Except for using private data as Johnstone and Bedard (2003, 2004), it is difficult to control for the demand side effect. Prior studies of client portfolio management using public data suffer from similar limitations (Hogan and Martin 2009; Hsieh and Lin 2016; Schroeder and Hogan 2013).

Shipman et al. (2017) suggest that imposing a caliper distance, which restricts the distance between propensity scores for a successful match, is a best way to decrease the likelihood of poor matches. We also re-perform the match using a caliper distance of 0.03 which is a commonly used caliper distance in accounting research, and the main results remain unchanged.

We use the information on auditor changes (dismissal or resignation) provided by Audit Analytics database.

The factor analysis indicates that there are two factors with an eigenvalue greater than 1. We use the factor with the greatest eigenvalue for each set of variables. The factor loadings of the risk measure ROA, LOSS, LEV, and CASH are − 0.0937, 0.7017, 0.0819, and 0.7015, respectively. As expected, FINR decreases with ROA and increases with LOSS and LEV. Despite an unexpected positive correlation between CASH and FINR, the findings for the other variables suggest that the increases in FINR are consistent with greater financial risk.

\(\frac{{TAC{C_t}}}{{T{A_t}}} = {\beta _1}\frac{1}{{T{A_t}}} + {\beta _2}\frac{{\Delta RE{V_t} - \Delta RE{C_t}}}{{T{A_t}}} + {\beta _3}\frac{{PP{E_t}}}{{T{A_t}}} + {\beta _4}ROA + {\varepsilon _t}\)

The factor analysis indicates that there are three factors with an eigenvalue greater than 1. We use the factor with the greatest eigenvalue for each set of variables. The factor loadings of the risk measure GROWTH, ABSDACC, INVREC, GC, MODOP, and TENURE are 0.3372, 0.6769, 0.0803, 0.4161, -0.1416, and -0.4779, respectively.

Following Francis et al. (1994), we classify the following industries as litigious: bio-technology (SIC codes 2833–2836 and 8731–8734), computer hardware (SIC codes 3570–3577), electronics (SIC codes 3600–3674), retailing (SIC codes 5200–5961), and computer software (SIC codes 7370–7374).

According to extant studies (Francis et al. 2005; Cassell, Giroux, Myers, and Omer 2012), we specify the following audit fee model for estimation by year: \(FEE = {\beta _0} + {\beta _1}SIZE + {\beta _2}INVREC + {\beta _3}LEV + {\beta _4}ROA + {\beta _5}SEG + {\beta _6}MAO + {\beta _7}INDUSTRY + \varepsilon\) where FEE is the natural logarithm of total audit fees, SIZE is the natural logarithm of total assets, INVREC is inventory and receivables divided by total assets, LEV is total liabilities divided by total assets, ROA is return on assets, SEG is the natural logarithm of the number of business segments, MAO is one if auditor issue modified opinions for anything, and zero for unqualified opinions, and INDUSTRY is 2-digit SIC code industry dummy variables. Then, abnormal audit fees are estimated as the residual from this model.

We thank for the reviewer’s suggestion.

We extend our sample period to 2015 to reexamine H1 which is not restricted to the specific period. As we described in footnote 18, the main results remained qualitatively unchanged.

According to U.S. Census Bureau, a MSA has one or more counties or county equivalents that have at least one urban core area of at least 50,000 populations which adjacent territory has a high degree of social and economic integration with the core as measured by commuting ties. We exclude observations without MSA codes because it is problematic to calculate city-level auditor industry expertise for those firms.

To provide a full view of the audit market, Tables 2 and 3 include a sample of 16,533 firm-year observations before the propensity matching procedure.

Another possible explanation for the largest reduction of Big 4 clients in 2005 is that many companies go private after the passage of SOX (Engel et al. 2007).

Before the propensity matching procedure, clients audited by large Big 4 audit offices have significantly higher financial risk than clients audited by small Big 4 offices (mean value of FINR for large offices = − 0.1646, mean value of FINR for small office = − 0.2961, p-value from a test of difference in means across large offices and small offices = 0.000). Large Big 4 offices have clients with significantly lower audit risk compared to small offices (mean value of AUDR for large offices = − 0.1998, mean value of AUDR for small office = − 0.1630, p-value from a test of difference in means across large offices and small offices = 0.000). Therefore, the propensity score model seems effective in forming a matched sample, so there are no significant differences in risk attributes between clients in large offices and small offices after the matching procedure.

For new client observations, there are 140 observations switching between Big 4 audit firms and 38 observations switching from non-Big 4 audit firms to Big 4 audit firms.

Since we use auditor resignations to proxy discontinued clients in client retention decisions, there are only 16 observations resigned by Big 4 auditors. Although we intend to examine the same scenarios regarding client retention decisions, we encounter some difficulties in analyzing subsamples. Specifically, our discontinued clients are much fewer in the subsample and can be perfectly predicted by several control variables.

We thank for the reviewer’s suggestion.

We thank for the reviewer’s suggestion.

We use an alternative measure for LOCAL, where LOCAL is coded one if clients and auditors are located in the same MSA, and zero otherwise. The findings suggest that the coefficient on LOCAL × AUDR is significantly positive (coefficient = 0.5595, p = 0.024), which is robust.