2.1 New Zealand accounting and auditing environment

The N.Z. accounting and auditing environment has experienced significant changes as regulators have undergone a comprehensive review and reform of its financial reporting landscape. At the turn of the century, New Zealand had similar economic systems and auditing procedures to countries like the United States and Australia but lagged in reforms of the audit framework as regulators argued New Zealand does not experience financial scandals. In fact, New Zealand was one of the last countries with a self-regulating audit profession with reforms accelerating after major financial scandals rocked the nation (Davis & Hay, 2012). However, the institutional landscape in New Zealand differs from the United States in important ways, such as being much less litigious than the United States and adopting voluntary rather than mandatory rules-based governance systems (Davis & Hay, 2012; Sharma et al., 2011). With specific regard to legal and market penalties against auditors, New Zealand tends to assess them with much lower severity. Sharma et al. (2011) examine 18 years of corporate litigation cases in New Zealand and find only 14 of 417 total cases resulted in a judgement against the auditor.

Leaders of the accounting profession in New Zealand, including CEOs of the Big 4 audit firms, began to recognize the need for independent regulation and lobbied for reform just before the global financial crisis (Diplock, Jack, McLeod, Dawson, & Hunt, 2007). The catalyst for urgent reform was the series of unprecedented major finance company collapses beginning in 2006. These collapses were largely due to poor corporate governance, risky lending, and audit failures (Harris, 2009). With regard to financial reporting, New Zealand listed companies were mandated to adopt the N.Z. equivalents of International Financial Reporting Standards (IFRS) beginning in 2007. In addition to the aforementioned factors, this decision was motivated by the Australian and European Union adoptions of IFRS (Bradbury & Van Zijl, 2005). With regard to audit regulations, New Zealand implemented an External Reporting Board (XRB) and through the FMA replaced the self-regulated audit system with an independent and more stringent, prescriptive approach similar to that of the PCAOB in the United States (Davis & Hay, 2012).

With specific regard to executive compensation, shareholders of N.Z. public companies have become increasingly vocal in their skepticism of, and opposition to, large salaries, bonuses, and share options granted to top-level management, as the compensation structure is perceived to be unrelated to actual firm performance (Morrall, 2011; Taylor, 2018). Recent publicized scandals such as that of Fuji Xerox New Zealand in 2017 exemplify this skepticism. The company launched legal action against its former executives, charging them with inappropriate accounting practices (e.g., sales at any cost) and the pursuit of incentive-based remuneration through these aggressive practices. Accordingly, there has been growing concern that executives can use their positions of power to reward themselves independently of the firm's best interest.³ Gunasekaragea and Wilkinson (2002) posit that CEO compensation practices in New Zealand have actually incentivized financial misreporting. There has also been particular concern that the pay-setting influence of the CEO may allow excess extraction of funds to the ultimate detriment of shareholders (Roberts, 2005). And most recently, initiatives such as “Closing the Gap” continue to attempt to rein in exorbitant executive salaries, stating it as good for N.Z. society and companies alike (Closing the Gap, 2018).

Shareholders' concerns over excessive and unjustified CEO compensation is also supported by the information asymmetry surrounding CEO compensation (Taylor, 2018). The nature of disclosure of CEO compensation generally, and stock options specifically, varies widely in New Zealand, making auditor assurance over this compensation potentially more valuable. It is no surprise that given the state of affairs over CEO compensation, the IFAC urged New Zealand to reform its governance over executive compensation as it relates to firm risk management (IFAC, 2010). Specifically, they note among various measures to help “prevent future financial crises” to include that “governance mechanisms need to be strengthened at banks and other companies to ensure better oversight of risk management and executive compensation” (IFAC, 2010, p. 1). Also of note is the lack of executive compensation clawback policy in New Zealand relative to the United States. Following several major financial reporting scandals in the United States, the Sarbanes-Oxley Act of 2002 (SOX) mandated publicly traded firms to establish executive compensation clawback policies, which were reinforced by the Dodd-Frank Act of 2010. Such policies help firms recover incentive-based payments made to executives on the basis of financial results that turn out to be false and require restatement. Studies of clawback policy in the United States have shown improved financial reporting quality and lower audit fees (Chan, Chen, Chen, & Yu, 2012, 2014; Dehaan, Hodge, & Shevlin, 2013). Of course, the United States has much greater magnitude and complexity of incentive compensation, but the total lack of clawback policy and voluntary corporate governance practices in New Zealand may provide opportunities for management to misreport financial results that maximize incentive-based executive compensation. The auditor's recognition of this possibility (i.e., increased financial reporting risk) could manifest in higher audit fees.

2.2 Extant literature

While keeping in mind the less-established and voluntary N.Z. governance practices over executive compensation, our review of U.S. literature informs this study. As one of the first such studies, Vafeas and Waegelein (2007) examine the association between compensation incentives related to bonus plans, stock options, and long-term performance, and audit fees. They posit that the financial reporting risks inherent in incentive-based compensation may put an upward pressure on audit fees. Conversely, they also posit that incentive-based compensation may be a sound agency risk-monitoring mechanism that may be associated with less financial misreporting risks, manifesting in lower audit fees. They label this a “substitution explanation” and report some results consistent with this notion.

Later U.S. studies extend Vafeas and Waegelein (2007) and focus on analyzing the sensitivity of executive equity compensation to changes in share price (delta) and changes in return volatility (vega). Arguing the potential fraud risk associated with equity incentives, Kannan et al. (2014) examine audit fees in relation to these incentives for CEOs and CFOs. They find evidence that auditors consider the risk implicit in executive compensation arrangements and price this risk accordingly. Specifically, they find a positive association between audit fees and vega for both the CEO and CFO, but no significant results for delta. In contrast, Kim et al. (2015) find that vega is related to audit fees, but only for the CEO and not the CFO. They further document that auditors charge higher fees when firms increase the use of stock options in executive compensation.

While Fargher et al. (2014) document a positive association between audit fees and vega, they report a significant negative relationship between delta and audit fees. They explain this unexpected finding by arguing that large deltas discourage managerial risk-taking, which is contrary to the argument by Kannan et al. (2014) that large deltas pose greater financial misreporting risk. Citing the inconclusive evidence on delta in the literature, Chen et al. (2015) control for this factor and analyze how vega is related to audit fees. Consistent with the other findings, they show firms with high vega pay significantly higher audit fees, although the significance weakened in the post-SOX period. The authors find no evidence of an association between delta and audit fees.

Billings et al. (2014) describe the pervasive role of managerial incentives in accounting scandals and examine delta and vega. They find that CEO delta and vega are not related to audit fees, but CFO delta and vega are themselves related. They also report their CFO results are more pronounced in firms with weaker controls, indicating that auditors perceive greater audit risk from compensation structure in environments where internal controls are less effective. An interesting insight is provided more recently by Albrecht et al. (2018). They find that excess compensation paid to named executives is associated with greater risk of financial misreporting when these executives possess accounting competence. When they examine how auditors respond, they find that auditors are discounting the “dark-side” of executives with accounting competence when these executives receive excess compensation. Overall, this small but growing body of literature remains unclear as to how CEO compensation incentives are related to audit fees and implies more research is warranted. One aspect not considered by this prior research is a firm's internal governance, which is pivotal in monitoring management behavior, compensation, and the audit. We extend this literature by incorporating firms' internal governance as a moderating factor.

2.3 Hypotheses development

3.1 Sample selection

Our sample is selected from the population of firms listed on the New Zealand Stock Exchange (NZX) from fiscal years 2004 to 2012.⁶ The financial data for all companies are obtained from the Global Vantage Database. Data for audit fees, corporate governance, executive compensation, and ownership percentage are manually extracted from the annual reports filed with the NZX. We begin with an initial sample of 2,042 firm-years. To avoid the effect of foreign audit and corporate regulations, we exclude 606 firm-years that are not headquartered in New Zealand and are dual listed on the NZX. We exclude 225 and 72 firm-years of finance and utility firms, respectively, due to their unique governance and reporting features. We exclude 290 firm-years due to data unavailability for all years. We then exclude 39 firm-years due to less than five observations per industry as we need sufficient industry samples to derive our industry-adjusted CEO compensation measures that we employ in additional analyses. Thus, our overall sample consists of a balanced panel of 810 firm-years comprising 90 unique firms for each of the nine years (2004–2012). Table 1 summarizes our sample selection procedure, and Table 2 provides the sample distribution by the NZX industry descriptors. All continuous variables are winsorized at the 1% and 99% levels.

3.2 Empirical model –Audit fee regression

To test Hypotheses 1 to 4, we specify and estimate our ordinary least squares (OLS) fee regressions based on prior audit fee research (e.g., Dunmore & Shao, 2006; Ghafran & O'Sullivan, 2017; Griffin et al., 2008, 2009; Hay & Knechel, 2010; Hay, Knechel, & Li, 2006; Kannan et al., 2014; Mitra et al., 2019; Smith et al., 2018; Venkataraman, Weber, & Willenborg, 2008;). Equation 1 is a baseline model that regresses audit fees on total CEO compensation. Equation 2 tests Hypotheses 1 and 2. All our regressions are based on firm-level clustered robust (heteroscedasticity-consistent) standard errors since our analysis includes repeated firm measures over time (Petersen, 2009). The control variables include CEO base compensation, and various firm, auditor, and governance factors that we explain below.

(1)

(2)

To test Hypotheses 3 and 4, we introduce governance interaction terms and estimate the OLS regressions specified in Equations 3 and 4. In these equations, the variables of interest are the interaction terms whereby a significant interaction term would indicate the moderating effect of corporate governance on the association between the CEO compensation variables and audit fees. Following prior N.Z. and other governance and audit fee research (e.g., Albrecht et al., 2018; Carcello, Hermanson, Neal, & Riley, 2002; Ghafran & O'Sullivan, 2017; Griffin et al., 2008; Hay, Knechel, & Ling, 2008; Mitra et al., 2019; Smith et al., 2018; Vafeas & Waegelein, 2007), we focus on the following governance variables. These variables have been extensively studied and consistently relate to audit fees over time and across countries, and have been confirmed by Hay (2013) through a meta-analysis of audit fee research: board and audit committee independence (measured as the proportion of independent directors), audit committee accounting expertise (measured as the proportion of experts on the committee), board and audit committee meetings (measured as the number of meetings during the year), and board and audit committee size (measured as the number of directors).⁷

However, the use of multiple single indicators of governance can produce contradictory results and become difficult to interpret. This is because corporate governance is a complex construct and using single or arbitrary indices (e.g., summing up the governance attributes) can lead to measurement errors resulting in inconsistent regression parameter estimates (Larcker, Richardson, & Tuna, 2007). Therefore, following the approach by Larcker et al. (2007), we factor analyze these seven governance attributes using the “eigen value greater than one” criterion with varimax rotation, and use the factor scores in our analyses. The factor analysis produces three factors explaining 61% of the total variance, which we label as follows (factor loadings are in parentheses): (1) audit committee effectiveness (ACEFF) that comprises audit committee independence (0.750), audit committee accounting expertise (0.751), and audit committee size (−0.717); (2) board resources (BOD) that comprises board independence (−0.640) and board size (0.751), and (3) diligence (DILI) that comprises board meetings (0.817) and audit committee meetings (0.709). The factor loadings are quite high and suggest strong communality among the variables in a factor, and the signs on the loadings are as expected based on the prior literature cited above. For instance, because audit committee independence and accounting expertise promote effective governance, they load positively, while audit committee size loads negatively because size can reduce audit committee effectiveness (Carcello et al., 2011). Similarly, board independence and board size have opposite signs but load together on the board factor, and for the final factor, both board and audit committee meetings load positively as both of these are related to effective governance (Carcello et al., 2011).

(3)

(4)

4.1 Descriptive statistics

We present descriptive statistics for our audit fee variables in Table 3. The mean and median audit fees are $258,340 and $82,030, respectively. The mean and median total CEO compensation is $692,700 and $331,140, respectively. Disaggregating this into base compensation and short-term incentive compensation, the means (medians) are $554,470 ($331,120) and $138,220 ($0), respectively. The mean (median) stock options awarded to CEOs is 0.41 (0.25), suggesting that on average, 41% of firms are offering and disclosing this form of equity incentive compensation. The mean and median total assets of our sample firms are $489,000,000 and $106,510,000, respectively. Summary statistics on all other variables are provided in Table 3.

4.2 Correlation analysis

Table 4 presents Pearson and Spearman correlations for our independent variables. We find a significant positive correlation (0.85) between total compensation (TOTCOMP) and base compensation (BASE), as the former contains the latter, and they are thus not included in the same regressions. Our variance inflation factors (VIFs) are in the range of 2 to 3, well below the multicollinearity threat thresholds of 10 (Gujarati, 2003). Since the data involve the same companies over a 9-year period, we also run a time-series test for auto-serial correlation and find the Durbin-Watson coefficients are around 1.914 across all our regressions. Moreover, we employ firm-level clustered robust standard errors in all our regressions to address heteroscedasticity (Petersen, 2009) and include year fixed-effects.

4.3 Regression results

In Table 5, we present the results of our analyses. We first report a baseline model that regresses audit fees on total CEO compensation, followed by the audit fee regressions testing Hypotheses 1, 2, 3, and 4. All our audit fee regressions have explanatory power comparable to prior N.Z. (Griffin et al., 2008, 2009; Hay, Knechel, & Li, 2006) and other studies. The baseline results in Column 1 of Table 5 show the coefficient on total CEO compensation (TOTCOMP) is positive and significant (p < 0.01). The results in Column 2 of Table 5 test Hypotheses 1 and 2 that examine the association between audit fees and the incentive components of total compensation, short-term incentive (STIC), and stock option (STOCK) compensation paid to the CEO, with base salary (BASE) serving as a control. The results show the coefficient on short-term incentive compensation (STIC) is positive and significant (p < 0.01). This suggests that CEO compensation derived from short-term performance incentives such as annual bonus, spot rewards, and/or retention bonus presents risks that are considered and priced by the auditor. This finding rejects the null association proposed in Hypothesis 1. Similarly, the results in Column 2 of Table 5 show that the coefficient on stock option compensation (STOCK) is positive and significant (p < 0.05). This finding suggests rejection of the null association in Hypothesis 2 and implies that audit fees are higher when CEOs receive stock options. Taken together, the findings on short-term incentive and stock option compensation suggest that auditors consider and price performance-based compensation paid to the CEO. However, our CEO compensation control variable results suggest that auditors do not price base (BASE) compensation paid to the CEO, as the coefficient on this control variable is not significant, which is consistent with the fact that this component of compensation is not contingent on firm performance and is fairly static, thus it may not pose financial reporting risks for the audit.

The results presented in Columns 3 and 4 test Hypotheses 3 and 4 that proposed there is no moderation effect of governance on the association between CEO compensation and audit fees. In Column 3, the coefficient on STIC, which captures the association between CEO short-term incentive compensation and audit fees in the absence of governance, is positive and significant (p < 0.01). With regard to the governance interactions, coefficients on two of the three are significant. Specifically, the coefficient on STIC*ACEFF is negative and significant (p < 0.05) and the coefficient on STIC*DILI is also negative and significant (p < 0.01). These results suggest that the association between short-term incentive compensation and audit fees is moderated by the governance factors of audit committee effectiveness (ACEFF) and diligence (DILI), but not by board resources (BOD). This may be due to the audit committee's direct oversight of the financial reporting process, which the auditor recognizes and considers when assessing financial reporting and fraud risks arising from short-term incentives paid to the CEO (PCAOB, 2012, 2018). Since the short-term incentives in our setting comprises mainly cash bonuses, a relatively more straightforward component of compensation, an appropriately qualified and diligent audit committee is more likely to fairly and easily monitor this type of compensation without the greater board needing to get involved.

Moving on to the results in Column 4, the coefficient on STOCK, which captures the association between CEO stock option compensation and audit fees in the absence of governance, is positive and significant (p < 0.05). The interactions between STOCK and both ACEFF and DILI are insignificant, but STOCK*BOD is positive and significant (p < 0.05). Taken together, these results suggest that strong governance at the board level heightens the demand for additional audit effort when the CEO is paid with stock options.⁸ Stronger boards may demand greater auditor assurance over stock options because of their inherent risk and complexity. Also, the lack of prescribed N.Z. CEO stock option disclosure requirements may intensify the board's concern over the clarity of stock options. Additionally, the larger board (not audit committee) is responsible for setting executive compensation in New Zealand, so given the relatively more opaque incentive-based component of stock options, the audit committee may not be in as strong a position to assess it relative to financial reporting and thus may rely more on the board. As our board factor captures independence and size, more independent and larger boards face greater exposure to market scrutiny, which would likely motivate such boards to protect their reputation by seeking greater assurance from the auditor given increasing N.Z. investor concern over stock option compensation (Taylor, 2018).

With respect to the control variables, our results are similar to prior research. For example, the governance control variables of audit committee effectiveness (ACEFF) and board diligence (DILI) are positive and significant. This is similar to prior research that generally finds individual governance variables such as accounting expertise on the audit committee, audit committee independence, and board and audit committee meetings to be positively related to audit fees, while board and audit committee size tend to have mixed associations with audit fees (see Hay, 2013, for a meta-analysis of the audit fee literature). Similar to prior research (see Hay, 2013), we find several firm characteristics (SIZE, SUBS, ARINV, LEV, MERGER, and BIGFOUR) are positive and significantly related to audit fees.

Taken together, the results suggest that auditors pay attention to CEO performance-based compensation and price this into audit fees. Given that CEO short-term incentive and stock option compensation are linked to firm performance, and such CEO compensation is known to incentivize earnings management, our results are consistent with the view that auditors are considering potential financial reporting risks stemming from short-term incentive compensation and stock options. Moreover, our results suggest that stronger corporate governance mitigates auditors' upward pricing of short-term incentive CEO compensation. This governance mitigating effect is consistent with auditors adopting a risk-based perspective. Conversely, auditors' consideration of stock option compensation is positively influenced by the strength of firm governance, in particular, board resources, which may be due to the risk in options that is consistent with the view that stock option compensation is inherently more complex, thus requiring additional audit assurance. Stronger boards may be even more demanding of such assurance given the complexity and lack of consistent stock option disclosure in New Zealand.

As we reflect on the results, we conclude that there is sufficient collective evidence to reject the null hypotheses in H1 (short-term incentive compensation is not associated with audit fees), H2 (stock option compensation is not associated with audit fees), and H3 and H4 (no moderation effect of governance on CEO compensation and audit fees). Our evidence is consistent with some of the prior literature that supports auditor pricing of executive equity-based compensation (e.g., Chen et al., 2015; Fargher et al., 2014; Kannan et al., 2014; Kim et al., 2015; Vafeas & Waegelein, 2007), with our study extending this body of knowledge by documenting that specific components of CEO compensation are related to audit fees, and that a firm's corporate governance moderates these associations. As the degree of public disclosure of CEO stock option compensation by N.Z. firms varies cross-sectionally, our study suggests that auditors' consideration of this compensation through access to private information provides some level of assurance to investors. This is likely due to auditors evaluating the financial reporting implications of the complexities and risks inherent in stock option compensation.

4.4 Additional analyses

For greater confidence in our results, we conduct several additional analyses. First, given that auditors may learn more about CEO compensation risks over the course of the current year audit that could impact the following year's audit pricing, we re-perform all of our regressions using current year compensation measures and next year audit fees. Phrased in a different way, auditors can price the current year's audit based on risks discovered during the prior year audit. Thus, we perform lead–lag analyses by estimating regressions of (1) current year fees on prior year compensation and (2) next year fees on current year compensation. While both sets of results are consistent with those reported in Table 5, for brevity, we report our results for prior year compensation effects on audit fees in Panel A of Table 6. We also consider client engagement tenure because a new auditor may not have the opportunity to learn about the client from prior audits. Thus, we restrict our sample to continuing audit engagements (INITIAL = 0) and reestimate our main and the lead–lag regressions and find that all of our results (untabulated) remain the same.

Second, we consider if our results differ between Big 4 and non–Big 4 audit firms. Generally, the expectation is that Big 4 audit firms will provide greater audit quality and thus may pay more attention to risks in CEO compensation compared to non–Big 4 auditors because larger audit firms have deeper pockets, and thus face a greater risk of potential litigation (Hay, Knechel, & Wong, 2006). However, we note our study is situated in the low auditor–penalty risk context of New Zealand (Sharma et al., 2011), which conversely suggests that both Big 4 and non-Big 4 auditors face a low risk of litigation and reputational consequences. When we reestimate our results for subsamples of the Big 4 and non–Big 4 auditors, our results are the same for both types of audit firms.

Third, we exclude firms that give spot or retention bonuses (42 firm-years) from our short-term incentive compensation (STIC) calculation and reestimate the regressions where STIC contains only cash bonuses. All our results are again consistent with our primary results (p < 0.01). Fourth, we previously noted excluding long-term incentive compensation from our analyses because it is rarely used in New Zealand (only nine companies [6.5%] in our sample sporadically report this type of CEO compensation). When we include long-term incentive compensation paid to CEOs in our regressions, we find it to be insignificantly related to audit fees. Results on our other CEO compensation variables are not sensitive to the inclusion of this variable. The insignificant result of long-term incentive compensation may be due to its low incidence (low power), or it may suggest auditors believe long-term incentives to the CEO pose less financial misreporting risk.

Fifth, in our tests we employ the natural logarithm values of CEO compensation that capture levels of compensation. Some studies also employ a percentage measure of CEO compensation and accordingly we test the robustness of our results to this measure, noting that a proportion measure only applies to CEO compensation variables for which we have dollar values. We derive the proportion of short-term incentive compensation in total compensation and use this in place of the natural logarithm value of STIC. Since we control for CEO base salary, we also derive a proportion measure of BASE. When we reestimate the relevant regressions from Table 5 (Columns 2 and 3), we find that the proportion measure of STIC remains positive and marginally significant (p < 0.10), and the interaction term STIC*DILI is negative and significant (p < 0.05), but we lose significance in both these measures and STIC*ACEFF. Thus, our results are not robust to the proportions of incentive compensation. We suggest two potential reasons for the loss of significance. First, there are higher correlations between the proportion measures of compensation (STIC, BASE, and the interaction terms containing STIC) that consequently results in shared variance.⁹ Second, proportion measures may not be very informative of potential audit risks because the level of compensation may be considered too low to pose material risks in financial reporting, which the proportion measure may not clearly convey.¹⁰

Sixth, we employ industry-adjusted measures for three main reasons. First, CEO compensation could vary by industry, each having its own unique structural characteristics. More often than not, a CEO's compensation and firm performance metrics are benchmarked at the industry level and against industry peers (Bizjak, Lemmon, & Naveen, 2008; Finkelstein, 2009). Second, auditors usually evaluate client performance and risks along industry lines, and their industry benchmarking of CEO compensation to evaluate reporting risks is normal (Taylor, 2000). Third, the threat of litigation against the firm and the auditor can also vary across industries as some industries are more litigious than others. Although we control for industry fixed-effects in all our analyses, for the reasons outlined here, we compute and analyze industry-adjusted average measures for short-term CEO incentive compensation (STIC).¹¹ We then sort the STIC industry-adjusted measures by quintiles and categorize CEOs as highly paid if they fall in the top two pay quintiles for their industry, and 0 otherwise (Cooper, Gulen, & Rau, 2016). Since we control for CEO base compensation, we also derive industry-adjusted measures for our BASE variable. In Panel B of Table 6, we reestimate our regressions using industry-adjusted average measures and find that all our results are consistent with the results in Table 5.

In our seventh supplementary analysis, we consider industry litigation risk as firms operating in more litigious industries could heighten the auditor's sensitivity to the risk of fraud and misstatement stemming from CEO compensation. Similarly, internal governance monitors may be more concerned when circumstances potentially increase the risk of litigation. Consistent with prior N.Z. research (Knechel et al., 2012; Sharma et al., 2011), we re-perform our regressions partitioned on high-litigation (N = 504) and low-litigation (N = 306) industries and present the results in Panel C of Table 6. The results are quite interesting as they reveal that (1) auditors price the audit significantly higher when the CEO's compensation comprises stock options (STOCK) for clients operating in more litigious industries, (2) boards demand significantly greater assurance from the auditor over stock option compensation (STOCK*BOD) only for clients operating in more litigious industries, (3) auditors respond to diligent governance by lowering audit fees related to stock options (STOCK*DILI) only for clients operating in less litigious industries, and (4) the audit fees are lower for short-term incentive compensation in highly litigious industries only when clients have an effective audit committee (STIC*ACEFF). Overall, these results suggest that even in a country with low institutional litigation risk, the variation in industry-level litigation risk has an impact on the audit, at least in relation to CEO compensation and audit fees.¹²

Finally, we consider potential endogeneity, although we believe endogeneity is unlikely in our setting. First, we test for the exogenous assumption of our STIC and STOCK variables by employing the Wu–Hausman test routine in STATA following a two-stage instrumental variable regression where the instruments are the following compensation committee characteristics: size, independence, and meeting frequency. The compensation committee variables serve as instruments as we expect these to influence CEO compensation (e.g., Boyle & Roberts, 2013; Roberts, 2005), but have no reason to believe they will influence audit fees as no audit fee research has included compensation committee effects generally, or the three instruments specifically, in their regressions (see Hay (2013) for a meta-analysis of the audit fee literature).¹³ Results indicate that test variables are exogenous as both the Durbin and Wu–Hausman test statistics are never significant even at marginal significance (p < 0.10).¹⁴ Furthermore, the three instruments are significant at conventional levels in the STIC and STOCK regressions. Given our test variables are exogenous, it is not surprising that we continue to find that STIC, STOCK, and their interactions with the governance factors remain significant in the second-stage audit fee regressions.

Lennox, Francis, and Wang (2012) highlight the limitations of using two-stage endogeneity analyses and recommend the use of supplementary procedures. Therefore, we employ entropy balancing analysis that is growing in use in accounting/audit research because it has the primary advantage of avoiding sample-size reductions compared to a propensity-score-matching analysis. Briefly, the entropy-balancing approach utilizes a maximum entropy reweighting scheme that estimates a set of weights directly from the chosen covariates in a way that produces no significant post-weighting differences in the mean, variance, and skewness of the covariates across the test and control firms (Hainmueller & Xu, 2013).

We apply the entropy balancing method on our sample by specifying three moment constraints (mean, variance, and skewness) that need to be balanced across our covariates by including all exogenous determinants of STIC/STOCK as the covariates.¹⁵ Before the entropy balancing procedure is applied, the covariates have (untabulated) marginally significant differences (p < 0.10), but after the balancing, there are no (untabulated) significant (p > 0.10) differences in the covariates between the treated and control samples. We then reestimate our regressions in Table 5 based on the treated balance and find our results in Table 5 are confirmed. We find that STIC is positive and significant (p < 0.01), its interaction terms (STIC*ACEFF) and (STIC*DILI) are negative and significant (p < 0.01), STOCK is positive and significant (p < 0.05), and the interaction term STOCK*BOD remains positive and marginally significant (p < 0.10). Thus, our findings are robust to potential endogeneity threats.