Corporate tax-planning effectiveness: the role of compensation-based incentives.
ABSTRACT: This study investigates whether compensating chief executive officers and business-unit managers using after-tax accounting-based performance measures leads to lower effective tax rates, the empirical surrogate used for tax-planning effectiveness. Utilizing proprietary compensation
Keywords: tax planning; performance measures; endogenous treatment effects.
Data Availability: Compustat data are publicly available. Survey data are proprietary in nature.
I. INTRODUCTION
Effective tax planning, defined by Scholes et al. (2002) as tax planning that maximizes the firm s expected discounted after-tax cash flows, requires managers to consider their decisions' after-tax consequences. In this paper, I investigate whether after-tax accounting-based performance measures lead to lower effective tax rates (ETRs), my empirical surrogate for tax planning effectiveness. (1) The ETR, an income-statement-based outcome measure calculated as the ratio of total income tax expense to pre-tax income, generally measures the effectiveness of tax reduction strategies that lead to higher after-tax income. A lower ETR, however, can only proxy for tax savings and does not always imply that after-tax income and/or cash flows have been maximized. (2) Despite this limitation, the ETR has been used to measure the effectiveness of spending on the tax function (Mills et al. 1998) and corporate tax department performance (Douglas et al. 1996). Also, lowering the ETR is frequently cited as a way to increase earnings (e.g., Ziegler 1997) and increase share price (e.g., Mintz 1999; Swenson 1999). (3)
Accounting research has addressed the relation between accounting-based compensation and managers' actions (e.g., Larcker 1983; Healy 1985; Wallace 1997). This paper is the first to address whether after-tax accounting-based performance measures motivate managers to take actions that help lower their firms' ETRs and does so at both the chief executive officer (CEO) and business-unit (BU) manager levels. Prior after-tax performance measure research has focused only on the determinants of compensating CEOs using pre-versus after-tax earnings (e.g., Newman 1989; Cames and Guffey 2000; Atwood et al. 1998; Dhaliwal et al. 2000) and provides no evidence concerning after-tax compensation's effectiveness in lowering a firm's tax liability. Extending this investigation to the BU level is motivated out of the apparent conflict between arguments that taxes should be allocated to BUs for incentive compensation purposes (e.g., McLemore 1997) with empirical observations that a majority of firms do not do so (e.g., Douglas et al. 1996). (4) The current investigation provides evidence concerning the incremental effectiveness of explicitly motivating CEOs and BU managers to incorporate tax consequences into their operating and investment decisions.
A common issue in cross-sectional studies that attempt to link a particular management accounting choice to an outcome measure is that all sample firms may be optimizing with respect to the choice being investigated (Ittner and Larcker 2001). Without addressing the endogeneity of a firm's choice, it is difficult to provide evidence consistent with this choice leading to an improved outcome. To address this issue, the relation between ETRs and CEO and BU-manager after-tax performance measures is estimated using a two-step approach that helps correct for the potential endogeneity bias associated with these two choice variables. As a first step in implementing this approach, the Antle and Demski (1988) controllability principle is used to model a firm's decisions to adopt after-tax CEO and BU-manager performance measures. To include a particular measure in a manager's compensation contract, this principle requires that the expected benefits from holding a manager responsible for a measure must be greater than the additional wage that must be paid to compensate the manager for the resulting additional risk and effort. Accordingly, an after-tax performance measure should be used as a contracting variable in a manager's incentive compensation contract only if the manager's involvement in tax-planning efforts leads to a difference between pre-tax and after-tax accounting results, which is generally reflected in the ETR. Consistent with prior research, the pre- versus after-tax CEO and BU-manager selection models include variables that control for a firm's tax-planning opportunities because the presence of such opportunities reflect the extent to which a manager's actions can be expected to lower the ETR.
Even if a manager's efforts are expected to lead to a lower ETR, a firm will use an after-tax performance measure only if the expected benefits exceed the expected costs of doing so. An after-tax performance measure is expected to lead to a lower ETR because it motivates the manager's increased cooperation with tax professionals to help identify, develop, and execute tax-planning strategies. McLemore (1997, 1) cites Hewlett Packard's tax director to support the need for BU-manager involvement in tax-planning efforts:
Tax planning is only as good as being involved in the early stages of such things as business planning, strategic planning, and merger and acquisition work.... Your tax department has to be represented at the table when those decisions are made. The evolving model for the future is the tight integration of tax people with business unit planning.
Costs associated with using after-tax performance measures include the additional wage that must be paid to compensate the manager for the increased risk due to potential tax law changes and the increased effort that results from including income tax expense in the compensation contract. Other potential costs associated with after-tax compensation include the administrative cost of allocating tax expense to a firm's BUs, increased tax examination costs, and increased tax authority scrutiny. Contrary to measuring after-tax compensation's benefits via observed ETRs, there are no clear empirical surrogates for after-tax performance measures' costs. This study thus focuses on the realized benefits of compensating managers on an after-tax basis but does not provide evidence of the associated costs' magnitude.
Proprietary data obtained in a survey of corporate executives are used to construct certain test variables, including those indicating whether CEOs and BU managers are compensated using after-tax accounting-based performance measures. Publicly available data are used to construct ETRs and other test variables. The results are consistent with the hypothesis that compensating BU managers, but not CEOs, on an after-tax basis leads to lower ETRs, resulting in an estimated median tax savings of $13.3 million annually. Sensitivity tests performed on a subsample of firms with high simulated MTRs (Graham 1996) provide further evidence that low-MTR firms' potential ETR-lowering actions that could have ambiguous effects on cash flows and after-tax profits are not driving this result. Further sensitivity tests help rule out the proportion of tax function outsourcing as an alternative explanation for the statistically and economically significant negative relation between after-tax BU-manager compensation and ETRs.
The results contribute to the accounting-based compensation literature by linking after-tax accounting-based performance measures to BU-manager involvement that is incrementally effective in lowering firms' ETRs. Consistent with Guidry et al. (1999) who document bonus-induced earnings management at the BU level, this finding provides additional insight into the effect that BU-manager accounting-based incentives have on managers' actions. Also, the estimated explicit tax savings resulting from after-tax performance measures provide corporate decision makers with information relevant to the design of BU-manager incentive compensation plans.
The paper proceeds as follows. The next section sets forth the hypotheses tested in this study. Section III outlines the empirical models and estimation procedures used in testing these hypotheses. Section IV provides a discussion of the data and sample, including a brief overview of the survey used to obtain proprietary compensation data. Results are presented in Section V. The final section provides the conclusion and a discussion of the study's limitations.
II. HYPOTHESIS DEVELOPMENT
Newman (1989), Carnes and Guffey (2000), and Atwood et al. (1998) investigate firms' choices of after-tax earnings as the contracting variable in CEO bonus plans. These studies hypothesize that firms with greater tax-planning opportunities, consistent with the Antle and Demski (1988) controllability principle, are more likely to use after-tax performance measures. Using proxies for tax-planning opportunities, these studies collectively find that multinational status, number of operating segments, firm size, and capital intensity are positively associated with after-tax CEO compensation. Atwood et al. (1998) also presents evidence that leverage is negatively associated with this choice.
Dhaliwal et al. (2000) also focuses on the relation between a firm's tax-planning opportunities and the decision whether to use after-tax earnings in CEO bonus plans, but measures planning opportunities using only realized tax credits and absolute values of firms' permanent book-tax differences. They exclude temporary book-tax differences such as those resulting from accelerated tax depreciation because these differences do not impact a firm's after-tax accounting earnings, the performance measure used in after-tax accounting-based bonus plans. Similarly, the ETR used in the current study is computed using total tax expense in the numerator, thereby excluding the effects of temporary book-tax differences.
Consistent with prior literature, Dhaliwal et al. (2000) find that firms with greater tax credits and absolute values of permanent book-tax differences are more likely to compensate their CEOs on an after-tax basis. Because permanent differences can either increase or decrease taxable income relative to book income, they provide no direct evidence of an association between after-tax CEO compensation and lower ETRs. The current study investigates this relation between after-tax CEO compensation and ETRs. The first hypothesis, stated in alternative form, is:
H1: Using after-tax performance measures in CEO accounting-based bonus plans leads to lower ETRs.
This paper also extends the pre- versus after-tax compensation analysis to BU-manager incentive compensation plans. The need for BU-manager involvement in tax-planning efforts is consistent with the coordination task that arises when specialists require input and cooperation from nonspecialists (Milgrom and Roberts 1992). The firm's tax professionals not only need help from the firm's BU managers in the identification of tax-planning opportunities, but also their cooperation in developing and executing strategies to capitalize on such opportunities.
In the context of lowering the ETR and increasing after-tax accounting income, how do BU managers participate in efforts that increase tax credits and lead to permanent differences that lower taxable income? Colgate-Palmolive's CFO states that "Colgate managers routinely watch for ways to time remittances from overseas [which leads to higher foreign tax credits], maximize research and development tax credits...." (Mintz 1999, 2.) BU managers can also help tax professionals formulate transfer prices (Ernst & Young 1999), which helps lower the ETR by decreasing (increasing) income reported in countries with tax rates in excess of (lower than) the U.S. tax rate. (5) Other examples of BU-manager involvement that helps lower the ETR include (1) determining whether a potential employee qualifies for a payroll-based tax credit before the hiring decision is made, (2) consulting with the firm's tax professionals prior to domestic location decisions to incorporate varying state and local tax rates as a factor in such decisions, and (3) helping tax professionals gather data necessary to document tax return reporting positions. (6)
The above discussion suggests that BU-manager involvement is helpful in lowering the ETR. It is possible that the pre- versus after-tax nature of the CEO's accounting-based bonus plan signals tax planning's importance and thus provides implicit incentives sufficient to motivate BU managers to assist in efforts that lower the firm's ETR. If this were so, then explicit incentive pay would not be necessary to motivate such efforts. Wilson (1995, 12), however, cites an interview with a corporate executive to support his conclusion that after-tax measures motivate BU managers to become involved in tax-planning efforts:
We instituted after-tax divisional measures that we have continued to refine. This puts the business units in the position that they think about taxes as they make decisions.
Similarly, Mintz (1999, 2) cites the Colgate-Palmolive CFO's argument for after-tax performance measures:
Were pre-tax income to govern incentives, then employees with bonuses on their minds would pay far less attention to favorable tax implications.... But because incentive compensation reflects after-tax earnings ... managers are rewarded when taxes are lowered.
The above arguments are consistent with evidence that mid-level managers respond to explicit bonus-based incentives (e.g., Kahn and Sherer 1990; Guidry et al. 1999). Compensating a firm's BU managers on an after-tax basis is thus expected to promote efforts that lead to a lower ETR. The second hypothesis, stated in alternative form, is as follows:
H2: Using after-tax BU-manager accounting-based performance measures leads to lower ETRs.
III. EMPIRICAL MODELS
To test the hypotheses stated in the previous section, the following equations are estimated cross-sectionally:
(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
(2) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
(3) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Hypotheses 1 and 2 predict negative signs on [[beta].sub.1] and [[beta].sub.2], the coefficients on CEOATAX and BUATAX, respectively, in Equation (1). The CEO pre- versus after-tax compensation choice model, Equation (2), is based on models estimated in prior studies (e.g., Atwood et al. 1998; Dhaliwal et al. 2000). The BU manager compensation pre- versus after-tax compensation choice model, Equation (3), is also predominantly based on prior studies that investigate the determinants of CEO after-tax compensation and utilizes firm-level, rather than business-unit-level, measures. (7)
Taken together, Equations (1)-(3) represent a double treatment effects model, extending the treatment effects model (e.g., Heckman 1990; Greene 2003, 787) to a setting in which there are two related and potentially endogenous treatments: after-tax CEO and BU-manager compensation. Without investigating the determinants of these two choice variables and correcting for potential endogeneity bias, estimating the ETR model using OLS produces inconsistent coefficient estimates. As discussed in more detail below in the "Estimation" subsection, this double treatment effects model is estimated using a two-step approach that captures the benefits of after-tax CEO and BU-manager compensation as intercept shifts, ([[beta].sub.1],[[beta].sub.2]), in the ETR model. The remainder of this section defines the variables included in the above equations, discusses specification issues, and describes the estimation procedures.
Variable Definitions
Endogenous Variables
The effective tax rate (ETR) is defined as the ratio of total tax expense (TTE) to pre-tax income. An average annual ETR for the three-year period 1995-1997 is used to control for unexplained year-to-year fluctuations in the annual ETR. TTE is used in the numerator because only efforts that lower TTE decrease the ETR and increase after-tax earnings. Accordingly, the ETR as currently measured represents the appropriate theoretical construct for evaluating the effectiveness of using after-tax accounting-based bonus plans.
There are at least two potential issues, however, regarding measurement of the ETR using TTE in the numerator. First, TTE excludes the effects of temporary book-tax differences, so tax-planning actions that result in a tax deferral are not reflected in the ETR. For example, a more favorable purchase price allocation in an asset acquisition, which leads to greater book-tax depreciation differences, is not reflected in the ETR. Similarly, investments in special purpose entities could lead to temporary differences that are not reflected in the ETR (e.g., Hanlon 2002). Locating a foreign subsidiary in a low-tax country and postponing the repatriation of the foreign earnings is another way to defer taxes and does not impact the ETR unless the firm takes the financial reporting position that the foreign earnings are permanently or indefinitely reinvested (Krull 2001). To the extent that a firm's CEO is compensated on an after-tax basis and is involved in decisions that involve the potential deferral of tax, which is likely to be. the case in foreign location decisions and the formation of special purpose entities, the ETR will not capture the CEOs' tax-planning efforts.
The second measurement issue regarding the ETR is that TTE excludes the tax benefit of deductions resulting from the exercise of nonqualified employee stock options (ESOs). Firms that have little or no tax liability because of their ESO deductions could have high ETRs, which can be problematic in studies that attempt to use the ETR to measure a firm's tax burden or to evaluate a firm's incentive to engage in tax-planning actions (e.g., Hanlon and Shevlin 2002). Arguably, these firms have little incentive to engage in costly tax-planning actions and are unlikely to use after-tax performance measures. Adjusting the ETR to reflect the tax benefit of the ESO deductions would result in lower ETRs for these firms, despite the appropriate use of pre-tax compensation, and would bias toward failing to reject the null hypotheses of no association between after-tax performance measures and ETRs.
CEOATAX is an indicator variable equal to 1 if the CEO receives an annual bonus based, at least in part, on an accounting measure determined on an after-tax basis, and 0 otherwise. BUATAX equals 1 if the firm uses a unit-based accounting measure in the compensation plans of a majority of its BU managers and this measure reflects the allocation of at least one unit-specific income tax item. (8) If a firm does not use a business-unit-based accounting measure, then BUATAX equals 1 if the firm uses an after-tax firm-wide accounting measure to compensate its BU managers.
The ETR model also includes two selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], which represent the double-selection analogs to the Heckman (1976) single-equation inverse Mills' ratio. These variables, which are calculated based on the results of the bivariate probit estimation of the CEOATAX and BUATAX models, are included in the ETR model to achieve consistent coefficient estimates using OLS (Tunali 1986). These variables are described in more detail below in the "Estimation" subsection.
Exogenous Variables Common to the ETR and Compensation Equations
All three equations contain variables that proxy for a firm's tax-planning opportunities. The rationale for including these variables in the ETR equation is that a firm's ETR differs from the statutory rate only to the extent it has opportunities to take actions that either increase or decrease taxable income or tax credits (Mills et al. 1998). Consistent with prior research (e.g., Newman 1989; Atwood et al. 1998; Dhaliwal et al. 2000) and the Antle and Demski (1988) controllability principle, the tax-planning opportunity variables are also included in the CEOATAX and BUATAX equations. Firms with multinational operations are expected to have greater tax-planning opportunities due to the tax effects of location, repatriation, and transfer pricing decisions. MULTI, the proxy for multinational operations, is computed as the ratio of foreign assets to the book value of total assets as of year-end 1997. (9) CAPINT is computed as the ratio of property, plant, and equipment to the book value of total assets (BVA) as of year-end 1997 and is included to control for opportunities related to investments in fixed assets. Larger firms, through the scope and scale of their operations, are expected to have greater opportunities to engage in tax-planning efforts. SIZE is computed as the natural logarithm of the BVA as of year-end 1997. Firms with greater leverage are also expected to have greater opportunities to engage in tax-planning efforts related to financing activities (Atwood et al. 1998). LEV is the ratio of long-term debt to BVA as of year-end 1997. Finally, tax laws vary on an industry-by-industry basis. IND is a vector of industry indicator variables for the mining and construction, transportation and communication, utilities, sales, financial institutions and insurance, and services industries.
Other explanatory variables are also included in all three equations. ROA, the ratio of the average pre-tax income for the three years ending in 1997 to BVA as of year-end 1997, is included in the ETR equation because prior research (e.g., Wilkie 1988; Gupta and Newberry 1997) has linked ETRs to profitability. ROA is also included in the CEOATAX and BUATAX equations consistent with the Dhaliwal et al. (2000) suggestion that more profitable firms are more likely to compensate the CEO on an after-tax basis because such firms have more income subject to tax. Dhaliwal et al. (2000) also suggests that firms with greater variability in profitability have a greater risk of making business decisions that are inappropriate for their tax status and thus have greater incentives to engage in tax planning. VARROA, the variance in ROA for the three-year period ended in 1997, is included in both compensation choice equations. VARROA is also included in the ETR equation because firms with greater incentives to engage in tax planning are expected to have lower ETRs. Bankman (1994) finds that high-growth firms typically place less emphasis on tax-planning efforts, suggesting that such firms are expected to have higher ETRs and are less likely to use after-tax performance measures. GROWTH, calculated as the geometric mean annual growth rate of the market value of assets (MVA) for the three years 1995-1997, is included in the ETR equation and both pre- versus after-tax performance measure choice equations. (10)
Finally, all three equations contain proxies for stock-based compensation. Consistent with the argument that firms with little or no tax liability because of their ESO deductions have little incentive to use after-tax performance measures, Dhaliwal et al. (2000) find that firms with higher proportions of current CEO compensation comprised of stock option grants are less likely to use after-tax earnings in CEO incentive compensation plans. Proxies for stock-based compensation are thus included in both pre- versus after-tax performance measure choice equations. CEOESO, the percentage of the CEO's compensation received in 1997 attributable to stock option grants calculated as the ratio of the Black-Scholes value (BSV) of stock option grants to the sum of the CEO's 1997 salary, annual bonus, and BSV of stock option grants, is included in the CEOATAX equation. (11) Similarly, BUSTK, an indicator variable equal to 1 if stock price is used as a performance measure in a majority of the firm's business-unit managers' incentive compensation plans, is included in the BUATAX equation. Both CEOESO and BUSTK are included in the ETR equation to control for the possibility that stock-based compensation motivates CEOs and BU managers to focus on after-tax earnings and engage in tax-planning efforts that result in lower ETRs. (12)
Exogenous Variables Appearing In Only One Equation
Following the Mills et al. (1998) finding that ETRs are negatively related to spending on the tax function, LSPEND is included in the ETR equation. LSPEND is computed as the natural logarithm of 1 plus total 1997 spending on the corporate tax function scaled by total SGA expense. LSPEND is not included in the two compensation choice equations because a firm's decisions to compensate its CEO and BU managers on an after-tax basis are not decisions typically made on a year-by-year basis and are thus not expected to be based on the tax department's annual budget. (13)
Bushman et al. (1995) and Keating (1997) find that the weight placed on firm-wide performance measures increases as interdependencies among a firm's BUs increase. Such interdependencies could lead to a greater need for a firm's BU managers to coordinate tax-planning efforts, such as those relating to transfer pricing. Accordingly, when a firm uses a firm-wide, and not a unit-based, accounting-based performance measure in its BU-manager incentive compensation plans, this performance measure is more likely to be computed on an after-tax basis. FWONLY is defined as an indicator variable equal to 1 if the firm uses a firm-wide accounting-based performance measure, but not a BU-based measure, in its BU-manager incentive compensation plans, and 0 otherwise. FWONLY is included in the BUATAX, but not the CEOATAX equation, because the need to coordinate efforts among a firm's BU managers does not apply to firm-level decisions.
Specification
Several specification issues arise in connection with this three-equation double treatment effects model. First, the ETR in the absence of compensating managers on an after-tax basis (hereinafter ET[R.sup.*]) does not appear in either the CEOATAX or BUATAX selection model. A higher ET[R.sup.*] neither impacts the potential benefit of ETR-lowering efforts nor enhances the opportunities for managers to engage in tax-planning efforts. Instead, following the Antle and Demski (1988) controllability principle, it is the manager's ability to take actions that lower the ETR that is critical to the decision to use pre- versus after-tax compensation. Arguably, this ability is unrelated to a firm's ETR in the absence of using after-tax performance measures.
Second, although these three equations are not modeled as a three-equation simultaneous-equation system, assignment of firms to subsamples based on whether firms use after-tax performance measures is not random. As a result, the CEOATAX and BUATAX variables are potentially endogenous to the ETR because the expected ETR equation error term given the use of after-tax compensation may not equal zero, e.g., if any tax-planning opportunities that influence both the ETR and the pre- versus after-tax compensation decisions are omitted from the three equations. Accordingly, the ETR equation includes the two selectivity correction variables that help correct for the possible bias in all ETR model coefficient estimates resulting from this potential endogeneity. In the ETR equation, the coefficients on CEOATAX and BUATAX represent these variables' incremental effect on the ETR after controlling for their endogeneity, i.e., after removing the effect of the exogenous determinants of CEOATAX, BUATAX, and the ETR.
A third specification issue relates to the potentially sequential nature of a firm's decisions to compensate its CEO and BU managers on a pre- versus after-tax basis. Arguably, a firm's directors first decide whether to compensate its CEO using after-tax earnings. The CEO then decides whether to base BU manager compensation on after-tax results. The data in Table 2 show that firms typically do not compensate their BU managers on an after-tax basis unless the CEO is also compensated after-tax, suggesting that these two decisions are sequential in nature. CEOATAX is thus included as an explanatory variable in the BUATAX equation, but not vice versa. Taken together, the CEOATAX and BUATAX equations comprise a recursive, simultaneous binary choice model (Greene 1998a; 2003, 215). (14)
The final specification issue relates to the identification of the coefficients and covariance parameters in Equations (1)-(3). Based on the discussion in Tunali (1986, 245), the CEOATAX and BUATAX selection models, Equations (2) and (3), are identified because all four possible combinations of the two choice variables (after-tax compensation for both CEO and BU managers, after-tax for CEO and pre-tax for BU managers, pre-tax for CEO and after-tax for BU managers, and pre-tax for both CEO and BU managers) are observed. This complete 2 x 2 classification results in 3 degrees of freedom, which allows the two means and correlation coefficient to be identified. The ETR model parameters are identified because only this model includes LSPEND, the measure of tax function spending, and the selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], which are nonlinear functions of the unknown parameters in the CEOATAX and BUATAX selection models.
Estimation
In general, OLS coefficient estimates will be inconsistent if the error terms of the outcome (treatment) equation and the selection equation are correlated (Heckman 1976). Tunali (1986) formulates a general double-selection model in which the typical Heckman (1976) selection model is extended by including a second selection equation. This paper follows Tunali's (1986) two-step approach in which the CEOATAX and BUATAX selection equations are first estimated using maximum likelihood bivariate probit.
Estimation of the ETR, CEOATAX, and BUATAX models proceeds as follows. First, this three-equation double treatment effects model is expressed as:
(1) ETR = [beta]'[X.sub.1] + [mu],
(2) CEOATA[X.sup.*] = [gamma]'[X.sub.2] + [epsilon],
(3) BUATA[X.sup.*] = [alpha]'[X.sub.3] + [phi],
where [X.sub.1], [X.sub.2], and [X.sub.3] are the vectors of explanatory variables in the respective structural equations, [beta]', [gamma]', and [alpha]' are the vectors of coefficients, and [mu], [epsilon], and [phi], are error terms that are assumed to have a trivariate normal distribution with variances equal to [[sigma].sup.2], 1, and 1, respectively, and correlations [[rho].sub.12], [[rho].sub.13], and [[rho].sub.23]. CEOATA[X.sup.*] and BUATA[X.sup.*] are unobserved latent variables. Instead, the following variables are observed:
CEOATAX = 1 if CEOATA[X.sup.*] > 0, 0 otherwise; and BUATAX = 1 if BUATA[X.sup.*] > 0, 0 otherwise.
Note that CEOATAX is included in [X.sub.3], the vector of explanatory variables in the BUATAX equation, and [X.sub.1], the vector of explanatory variables in the ETR equation, includes both CEOATAX and BUATAX.
Bivariate probit jointly estimates the CEOATAX and BUATAX models utilizing the correlation in error terms, [[rho].sub.23], and produces the selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], which represent the double-selection model analogs to the Heckman (1976) single-equation inverse Mill's ratio. (15) The selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], are calculated as:
[[lambda].sub.CEO] = [phi](-[gamma]'[X.sub.2])[PHI][(-[alpha]' [X.sub.3] - [[rho].sub.12]CEOATAX)/[(1 - [[rho].sub.12.sup.2]).sup.1/2]] /[[PHI].sub.2], and [[lambda].sub.BU] = [phi](-[alpha]'[X.sub.3])[PHI] [(-[gamma]'[X.sub.2] - [[rho].sub.13]BUATAX) /[(1 - [[rho].sub.13.sup.2]).sup.1/2]]/[[PHI].sub.2]
where:
[phi](*) = PDF; [PHI](*) = CDF; and [[PHI].sub.2] = bivariate normal CDF, [PHI](-[gamma]'[X.sub.2], -[alpha]', [X.sub.3], [[rho].sub.23])'
Finally, OLS regression of the ETR on CEOATAX, BUATAX, other explanatory variables, [[lambda].sub.CEO], and [[lambda].sub.BU] produces consistent coefficient estimates. (16)
The above estimation approach is a restricted control function (CF) approach because the outcome equation includes the selectivity correction variables and the treatment effect is captured only in a shift in the outcome equation's intercept (e.g., Vella and Verbeek 1999). In contrast to the restricted CF approach, an instrumental variables approach imposes the restrictions that either the treatment effect does not vary cross-sectionally or that a firm's idiosyncratic benefit is not known ex ante (Heckman 1997), neither of which is likely satisfied in the current setting. Vella and Verbeek (1999) find that the CF approach is more efficient and robust to specification errors, but that it is more dependent on the normality assumption. They argue that the restricted CF approach is more appropriate in settings such as the current one in which a treatment's associated costs are excluded from the choice model. (17)
IV. SAMPLE AND DATA
Proprietary survey data are used to construct CEOTAX and BUATAX, the variables indicating whether the firm uses after-tax performance measures in its CEO and BU-manager incentive compensation plans, respectively. Survey data are also used to construct BUSTK, the indicator of BU-manager stock-based compensation, LSPEND, the amount spent on the corporate tax function, and to determine whether a firm has multiple BUs. The Appendix contains the survey's general instructions and survey questions relevant to these variables. All other variables, including the ETR, are constructed using Compustat data.
Questionnaires were sent to corporate executives selected from two sets of firms that met preliminary data eligibility requirements. The first set was drawn from a list provided by the Financial Executives Institute. The second set was drawn from Fortune 500 firms not included in the first set and for which a tax executive's name and address was listed in the Tax Analysts corporate tax manager directory. Tax executives were targeted because they were presumably more interested in this study's results and were thus more likely to respond to the questionnaire (Dillman 1978). Survey instruments were sent to 829 publicly traded firms that met preliminary Compustat data availability requirements and had positive pre-tax income before equity in subsidiaries earnings in both 1995 and 1996. (18) All potential respondents were sent postcard reminders three weeks after the initial mailing. Thereafter, second and third reminders along with the questionnaire and a postage-paid return envelope were sent to nonrespondents. This process yielded 249 survey responses, resulting in a 30 percent response rate. Nine responses from firms that did not meet final data requirements and three unidentified responses were excluded, leaving 237 usable instruments.
To investigate how the 829 firms surveyed (survey firms) compare to the remaining population of firms listed in Compustat (non-survey firms), descriptive statistics comparing the two groups were computed. These comparative statistics (not tabulated) indicate that survey firms are larger, more capital intensive, more leveraged, and have higher ETRs than non-survey firms. (19) Targeting larger firms with a tax executive was necessary to produce a response rate adequate to test this study's hypotheses. Accordingly, the results may not be generalizable to smaller firms. To the extent, however, that small firms are less likely to have multiple business units and more likely to have net operating losses (Wang 1991), and thus less need for tax-planning efforts, the inability to generalize the results to small firms may not be an issue.
To investigate whether nonresponse bias is an issue, descriptive statistics were computed to compare firms responding to the survey (respondent firms) and those firms that did not respond (nonrespondent firms). This comparison (not tabulated) reveals no statistically significant mean differences in the ETR, pre-tax profitability (ROA), size (BVA), leverage (LEV), capital intensity (CAPINT), and the market-to-book ratio, suggesting that nonresponse bias is not a concern.
Table 1, Panel A summarizes the sample selection process that results in a final sample of 209 firms. Fourteen firms without multiple BUs were excluded. Eight firms that did not base the CEO's bonus on an accounting measure were also excluded along with five firms that used neither a unit-based nor a firm-wide accounting measure to compensate its BUs managers. One outlier with an ETR in excess of 300 percent was also deleted. (20) Table 1, Panel B summarizes the final sample's industry composition.
V. RESULTS
Descriptive Statistics
Table 2, Panel A presents descriptive statistics for sample firms. The mean ETR (.356) approximates the top statutory corporate tax rate. The means for CEOATAX and BUATAX, respectively, indicate that approximately 61 percent of the sample firms compensate the CEO on an after-tax basis while approximately 32 percent use an after-tax performance measure to compensate their BU managers. (21)
Table 2, Panel B presents means for the two subsamples based on whether the CEO is compensated on a pre-tax (n = 81) versus after-tax (n = 128) basis. The ETR mean when CEOATAX equals 1 (.348) is significantly lower than the mean when CEOATAX equals 0 (.368), consistent with H1. Consistent with prior research, the means for BVA (firm size) and CAPINT (capital intensity) are greater for the after-tax firms. The BUATAX mean difference indicate that few sample firms compensate BU managers on an after-tax basis when the CEO is compensated using a pre-tax measure, and this percentage increases substantially when the CEO's bonus is based on an after-tax measure. This result provides evidence that a firm is more likely to compensate its BU managers on an after-tax basis if the CEO is compensated after-tax, consistent with these decisions being sequential in nature. Contrary to Dhaliwal (2000), the mean percentage of CEO compensation derived from stock option grants is higher, not lower, for the after-tax firms.
Table 2, Panel C presents means for the two subsamples based on whether the firm's BU managers are compensated on a pre-tax (n = 143) versus after-tax (n = 66) basis. Consistent with H2, the mean ETR for the after-tax firms is significantly lower than the mean ETR for the pre-tax firms. Firms using after-tax BU performance measures are larger and more capital-intensive than pre-tax firms. Consistent with expectations, the mean difference in FWONLY indicates that firms using a firm-wide accounting measure, and not a unit-based measure, to compensate its BU managers are more likely to base this measure on after-tax results.
Simple Correlations
Table 3 contains simple correlations among the test variables with significance levels noted at the .05 level. Consistent with H1 and H2, the ETR is significantly negatively correlated with CEOATAX and BUTATAX. The only other significant correlations involving the ETR are those with LSPEND (spending on the tax function), LEV (leverage), and FWONLY. The correlations involving CEOATAX and BUATAX are consistent with the subsample mean differences reported in Table 2, Panels B and C.
ETR Model Estimations
Table 4 presents the ETR model estimation results. OLS results, presented in Column (1), do not support either H1 or H2. Without correcting for potential endogeneity bias, the coefficients on CEOATAX ([[beta].sub.1]) and BUATAX ([[beta].sub.2]) are both negative but insignificant.
Table 4, Column (2), reflects the results of estimating the ETR model augmented with the selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], which are calculated based on the results of the bivariate probit estimation of the CEOATAX and BUATAX models. The coefficient on CEOATAX ([[beta].sub.1]) is insignificant and thus fails to support H1. The coefficient on BUATAX ([[beta].sub.2), however, is -.071 and significant (p = .005). This result is consistent with H2 and provides evidence of a link between after-tax BU-manager performance measures and lower ETRs. (22)
The economic significance of the association between after-tax BU-manager compensation and lower ETRs is based on the interpretation of the coefficient on BUATAX, [[beta].sub.2], as the "experimental average treatment effect" (Heckman 1990). Although firms are not randomly assigned this treatment, the estimated intercept shift, [[beta].sub.2], captures the average effect on the ETR of a random assignment of the after-tax BU-manager compensation treatment. Using the estimated coefficient on BUATAX, -.071, per-firm tax savings are computed, resulting in a mean (median) tax savings of $41.9 million ($13.3 million), relative to mean (median) total tax expense of $211.2 ($64.8) million. (23) This evidence is consistent with the argument that compensating BU managers on an after-tax basis is not only statistically associated with lower ETRs, but also results in economically significant tax savings.
The result that after-tax compensation for BU managers, but not CEOs, is associated with lower ETRs leads to at least two questions. First, given the economic significance of the tax savings that result from compensating BU managers on an after-tax basis, why do a majority of firms compensate their BU managers on a pre-tax basis? Following the Antle and Demski (1988) controllability principle, the most plausible explanation for this behavior is that for firms compensating their BU managers on a pre-tax basis the expected costs associated with after-tax compensation outweigh the expected benefits. Second, why is CEO after-tax compensation not associated with lower ETRs? It is likely that CEOs have other incentives that sufficiently motivate a focus on after-tax, rather than pre-tax, earnings. For example, a CEO's job security could depend on after-tax earnings if the firm's shareholders expect the CEO to maintain a certain profitability level. To the extent, however, that CEOs compensated on an after-tax basis are more likely to implement after-tax BU-manager performance measures that are directly linked to lower ETRs, the evidence is consistent with an indirect link between after-tax CEO compensation and lower ETRs.
Other ETR model estimation results reported in Table 4 reveal that the relation between the ETR and firm spending on the tax function (LSPEND) is significantly negative, consistent with Mills et al. (1998). The coefficient on the variation in ROA (VARROA) is also significantly negative. In contrast to previous studies (e.g., Gupta and Newberry 1997; Mills et al. 1998) the coefficients on other explanatory variables are insignificant. There are at least three possible explanations for these results. First, prior ETR studies typically investigate determinants using the ETR specification that has only current tax expense in the numerator, whereas the ETR in this study is computed as the ratio of total tax expense to pre-tax income. Explanatory variables used in prior studies could possibly explain only variation in current tax expense; e.g., the capital intensity measure, CAPINT, could proxy solely for temporary depreciation book-tax differences. Second, the coefficients on these explanatory variables in the ETR model estimation only capture their incremental direct effects above and beyond their indirect effect on the ETR via their inclusion in the CEOATAX and BUATAX selection equations (Greene 2003, 783). Accordingly, the insignificance of the ETR determinants could be due to the separation of these direct and indirect effects. Finally, as reported in Table 3, the significant correlations between MULTI, CAPINT, LEV, SIZE, and ROA could lead to higher standard errors and lower t-statistics associated with these variables' coefficient estimates.
The coefficients on the selectivity correction variables, [[lambda].sub.CEO] and [[lambda].sub.BU], are -.035 (p = .758) and .044 (p = .053), respectively. The selectivity correction variables help correct for the possible bias in all ETR model coefficient estimates, not just the coefficients on CEOATAX and BUATAX, resulting from the omission of unobserved variables from the ETR model that are correlated with unobserved variables from the CEOATAX and BUATAX equations. A t-test of these coefficients represents a test of exogeneity (e.g., Vella and Verbeek 1999), suggesting that only BUATAX is endogenous to the ETR equation.
CEOATAX and BUATAX Selection Equation Estimations
Table 5, Column (1) presents the results of estimating the CEOATAX model using probit regression. Column (2) of Table 5 presents the results of estimating the CEOATAX model in a bivariate probit estimation of the CEOATAX and BUATAX models. Similarly, Table 6, Columns (1) and (2), present the results of estimating the BUATAX model using probit and bivariate probit, respectively. The correlation in error terms between the two selection equations is -.331 but insignificant (p .813). This insignificance is due to the inclusion of CEOATAX in BUATAX equation, consistent with the findings in Greene (1998).
As reported in Table 5 and consistent with prior research, capital intensity (CAPINT) is positively associated with the choice to compensate the CEO on an after-tax basis in both the single equation and bivariate probit estimations. The coefficients on firm size (SIZE) and profitability (ROA) are also positive, though only weakly significant, consistent with previous findings. Contrary to prior research, the coefficients on multinational status (MULTI), leverage (LEV), and the percentage of the CEO's compensation derived from stock option grants (CEOESO) are insignificant.
In the single-equation BUATAX model, probit estimation results reported in Table 6, Column (1), the coefficient on CEOATAX (.945) is significant (p = .000), consistent with the argument that firms compensating their CEOs on an after-tax basis are more likely to use after-tax BU-manager compensation. The coefficient on CAPINT (capital intensity) is significantly positive, consistent with expectations. Multinational status (MULTI) and firm size (SIZE) are also positive, as expected, but their associated coefficients are only weakly significant. The probit results also provide strong evidence that firms using a firm-wide accounting measure (FWONLY), and not a unit-based measure, are more likely to compensate BU managers on an after-tax basis. (24)
Sensitivity Tests Focusing on High-MTR Firms
Evidence consistent with the hypothesis that accounting-based after-tax BU-manager performance measures lead to lower ETRs does not necessarily imply that such measures promote tax-planning effectiveness. Although low-MTR firms are expected to engage in tax-disadvantaged activities that would generally increase the ETR (Scholes et al. 2002), if after-tax performance measures lead these firms to engage in naive tax-minimization strategies that lower the ETR, then these measures are not effective. To help rule out this behavior as a potential explanation and provide additional evidence concerning the effectiveness of after-tax BU-manager compensation, the tests are performed using only high-MTR firms, identified using the average Graham (1996) MTR for the three years ending in 1997. Using a subsample of firms with average MTRs in the top 75 percent of the MTR distribution (MTRs of 33.5 percent and higher), the ETR model estimation results are substantially the same as those reported in Table 4. (25) These results provide evidence that the negative relation between after-tax BU-manager compensation and ETRs is not a result of low-MTR firms' tax minimization strategies.
Sensitivity Tests Involving Firm Spending on the Tax Function and Outsourcing
To investigate differential effects of internal and external spending on the ETR, the LSPEND variable in the ETR model is first decomposed into two variables, LSPENDOUT and LSPENDIN, which represent outside and inside spending, respectively. (26) When the ETR model is re-estimated with these variables replacing LSPEND, the coefficient on LSPENDIN (-.782) is weakly significant in a one-tailed test (p = .096), while the coefficient on LSPENDOUT is insignificant, suggesting that only internal tax function spending is associated with lower ETRs. The coefficients on BUATAX and CEOATAX are substantially identical to those reported in Table 4.
Dunbar and Phillips (2001) find a negative relation between the use of after-tax operating-manager compensation and the proportion of corporate tax function expenditures made to external service providers. To investigate whether the proportion of outsourcing is a correlated omitted variable that could bias the BUATAX coefficient estimates reported in Table 4, OUTSOURCE, defined as the proportion of 1997 outside-to-total spending on the tax function, is added to the ETR model. When this revised ETR model is re-estimated, the coefficient on OUTSOURCE is .032 but insignificant (p = .127). The coefficient on BUATAX remains significantly negative (-.072, p = .004).
VI. CONCLUSION
This study investigates the relation between firms' effective tax rates (ETRs) and after-tax CEO and BU-manager accounting-based performance measures. The evidence is consistent with after-tax BU-manager performance measures leading to lower ETRs and economically significant tax benefits. This result highlights the importance of BU-manager involvement in tax-planning efforts that lower firms' ETRs and suggests that explicit accounting-based incentives are incrementally effective in promoting such efforts. The evidence does not support the hypothesis that after-tax CEO performance measures lead to lower ETRs. These results suggest that other incentives such as job retention are sufficient to motivate CEOs to focus on after-tax results, whereas annual after-tax accounting-based bonuses provide incremental motivation only for BU managers to consider their decisions' tax consequences. After-tax CEO performance measures could have an indirect negative effect on ETRs, however, because CEOs compensated after-tax are more likely to compensate their BU managers on an after-tax basis.
The results contribute to our overall understanding of the role that accounting-based incentive compensation plays in motivating managers' efforts. In particular, this finding adds to prior evidence (Guidry et al. 1999) that BU managers respond to unit-based incentives. This study also identifies the pre- versus after-tax nature of BU-manager compensation as a factor in firms' propensities to engage in tax-planning efforts. This evidence should assist future researchers in their investigations of effective tax planning strategies. Finally, the estimate of potential explicit tax benefits associated with compensating BU managers on an after-tax basis should be useful to corporate decision makers in their design of incentive compensation plans.
Interpretation of the results is subject to the following limitations. First, it is possible that for those firms not currently using an after-tax BU-manager performance measure, the associated costs of doing so may be greater than its expected benefits. Accordingly, the results do not lead to an inference that all firms should adopt this practice. Second, the financial-statement-based ETR does not reflect implicit taxes, defined as the decrease in pre-tax returns associated with tax-advantaged investments (Scholes et al. 2002). Tax-planning strategies that lower the ETR could also result in implicit taxes, reducing, at least in part, the explicit tax benefits obtained from using after-tax BU-manager compensation.
APPENDIX
SELECTED SURVEY INSTRUCTIONS AND QUESTIONS
General Instructions
This questionnaire includes questions about the compensation plans of your company's chief executive officer, tax department personnel, and business-unit managers. These questions relate to how these persons are compensated. There are no questions that request confidential compensation amounts. There are also questions concerning the use of certain corporate tax management practices. If you do not believe that you can answer questions that pertain to your company's tax department, please ask your company's top tax professional to complete this questionnaire.
For purposes of this study, two key terms are defined as follows:
1. The definition of a business unit follows the definition of an operating segment set forth in FAS No. 131, Disclosures about Segments of an Enterprise and Related Information. Accordingly a business unit is defined as a "component of an enterprise about which separate financial information [assets, liabilities, income, and expenses] is available that is evaluated regularly by the chief operating decision maker in deciding how to allocate resources and in assessing performance."
2. Your company is broadly defined and refers to the combined entity that is required to prepare and issue a consolidated financial statement; i.e., your company encompasses a parent company and all of the entities in which it has a 50 percent or more ownership interest. Some questions refer to specific individuals within your company (e.g., the CEO, CFO, the company's top tax professional, etc.). You should refer to individuals of the parent company in answering these questions. For example, "your company's CEO" refers to the CEO of the parent company, not the CEO of any of the company's particular business units or divisions.
All the information you provide is confidential and will be published only in summary, statistical form. You or your company will not be identified in any way. Accordingly, there are no foreseeable risks to you or your company. In order to get accurate information about corporate tax management practices and to properly assess the relationship between the use of these practices and companies' tax liabilities, we need information from all financial executives selected to participate in this study.
Q-4 How many business units are there in your company's worldwide organization? Estimates of this number are acceptable.
Q-5 For the fiscal year ending in 1997, how much did your company spend, on a worldwide basis, for salaries and other compensation-related costs within the company and fees to outside consultants for tax compliance and tax-planning/consulting work? Please indicate amounts, in 000s, and the estimated percentages allocable to tax planning/consulting in the spaces provided. Estimates of these amounts and percentages are acceptable.
Outside Consults
Estimated %
Related to Tax
Total Expenditures Planning/Consulting
Within Company -- --
Outside Consults -- --
Q-12 The CEO's annual bonus is based, at least in part, on an accounting measure (e.g., net income or loss, operating cash flow, etc.) computed (Circle one):
1 ON A PRE-INCOME TAX BASIS
2 ON AN AFTER-INCOME TAX BASIS
3 THE CEO'S ANNUAL BONUS IS NOT BASED, EVEN IN PART, ON AN ACCOUNTING MEASURE
4 NA--THE CEO DOES NOT RECEIVE AN ANNUAL BONUS
III. Business-unit managers generally receive incentive compensation based, at least in part, on four alternative "types" of performance measures: (1) the company's stock price (e.g., through the issuance of stock-based compensation such as stock options or stock appreciation rights, through cash/stock bonuses tied to the achievement of stock price levels and/or increases, etc.), (2) company-wide accounting measures (e.g., measures based on net income/loss, cash flow, etc.), (3) accounting measures computed at the business-unit level and/or group level (i.e., at a level other than on a company-wide basis), and (4) nonfinancial goals or measures (e.g., a performance appraisal rating). The following questions relate to the current use of the first three of these performance measures in the incentive compensation plans of a majority of your company's business-unit managers. In your responses to these questions, please consider the following:
1. If your company does not have two or more business units, please refer to the incentive compensation plans of a majority of your company's operating managers.
2. If a question asks whether a particular performance measure is being used in the incentive compensation plans of your company's business-unit managers, then please answer YES even if this measure is only one of two or more performance measures used in such plans.
3. The accounting-based measures addressed in these questions should be interpreted in a broad manner. Specific measures, such as Return on Assets (ROA) and Return on Equity (ROE), may be derived from these broadly defined accounting-based measures. For example, if your company uses ROA and computes this measure using pretax operating cash flow in the numerator, then this measure would be considered an accounting measure based on cash flow for purposes of this study.
Q-13 Is your company's stock price used as a performance measure in the incentive compensation plans of a majority of your company's business-unit (operating) managers (e.g., through the issuance of stock-based compensation such as stock options and stock appreciation rights, through cash/stock bonuses tied to the achievement of stock price levels and/or increases, etc.)? (Circle one)
1 YES 2 NO
Q-14 Is a company-wide accounting measure (e.g., a measure based on net income/loss, operating cash flow, etc.) used as a performance measure in the incentive compensation plans of a majority of your company's business-unit (operating) managers? (Circle one)
1 YES 2 NO
Q-15 The company-wide accounting measure(s) used in the incentive compensation plans of a majority of your company's business-unit managers is determined: (Circle one)
1 ON A PRE-INCOME TAX BASIS 2 ON AN AFTER-INCOME TAX BASIS
Q-17 Is a business-unit level and/or group level (i.e., at a level other than on a company-wide basis) accounting measure used as a performance measure in the incentive compensation plans of a majority of your company's business-unit managers? (Circle one)
1 YES 2 NO 3 NA--THE COMPANY DOES NOT HAVE TWO OR MORE BUSINESS UNITS.
Q-18 Which of the following statements best describes the extent to which the company's income tax expense affects the determination of the business-unit and/or group level accounting measure(s) used in the incentive compensation plans of a majority of your company's business-unit managers? (Circle one)
STATEMENT 1 THIS MEASURE IS DETERMINED STRICTLY ON A PREINCOME TAX BASIS. INCOME TAX EXPENSE IS NOT ALLOCATED TO THE COMPANY'S BUSINESS UNITS.
STATEMENT 2 THE COMPANY'S INCOME TAX EXPENSE IS ALLOCATED TO THE BUSINESS UNITS BASED SOLELY ON A CONSTANT COMPANY-WIDE PERCENTAGE (e.g., THE EFFECTIVE TAX RATE). THE EFFECTS OF INCOME-TAX-RELATED ITEMS ATTRIBUTABLE TO SPECIFIC BUSINESS UNITS ARE NOT CONSIDERED IN DETERMINING THIS MEASURE.
STATEMENT 3 THE EFFECTS OF A FEW (POSSIBLY ONLY ONE) INCOME-TAX-RELATED ITEMS ATTRIBUTABLE TO SPECIFIC BUSINESS UNITS ARE ALLOCATED TO THE BUSINESS UNITS IN DETERMINING THIS MEASURE.
STATEMENT 4 THE COMPANY'S TOTAL INCOME TAX LIABILITY IS EXTENSIVELY ALLOCATED AMONG THE BUSINESS UNITS IN DETERMINING THIS MEASURE.
TABLE 1
Sample Determination and Industry Classification
Panel A: Sample Determination
Number of firms responding to survey 249
Less: Responses that did not identify the firm 3
Less: Observations that do not have data available 9
on either the 1997 Compustat or SEC EDGAR database
Usable Instruments 237
Less: Firms without multiple business units 14
Less: Firms not basing the CEO's annual bonus on 8
accounting results
Less: Firms not using either a firm-wide or BU accounting 5
measure in the incentive compensation plans of its BU
managers
Less: Outlier 1
Final Sample 209
Panel B: Industry Classification
Industry SIC Codes Ranging From:
Mining and Construction 1000 to 1999 6
Manufacturing 2000 to 3999 114
Transportation and Communication 4000 to 4899 9
Utilities 4900 to 4999 20
Sales 5000 to 5999 24
Financial Institutions and Insurance 6000 to 6699 24
Services 7000 to 9099 12
Final Sample 209
TABLE 2
Descriptive Statistics
Panel A: Overall Descriptive Statistics
Continuous Std.
Variables Mean Dev. 10% 25% Median 75% 90%
ETR .356 .062 .279 .325 .360 .385 .416
SPEND .006 .011 .001 .001 .003 .006 .011
MULTI .171 .185 .000 .050 .072 .278 .429
CAPINT .322 .225 .017 .152 .285 .485 .653
LEV .178 .133 .006 .066 .162 .275 .344
BVA 7,371 12,962 265 730 2,029 6,694 21,453
ROA .107 .070 .024 .054 .092 .148 .196
VARROA .001 .002 .000 .000 .001 .002 .003
GROWTH 1.213 .223 .999 1.082 1.173 1.293 1.498
CEOESO .315 .243 .000 .086 .305 .468 .671
Dichotomous Std.
Variables Mean Dev.
CEOATAX .612 .488
BUATAX .316 .466
FWONLY .120 .325
BUSTK .603 .490
Panel B: Subsample Means--Pre-Tax versus After-Tax CEO Annual Bonus
Mean for Subsample in Mean for Subsample in t-test of
which CEOATAX = 0 which CEOATAX = 1 Differences
Variable (n = 81) (n = 128) in Means
ETR .368 .348 2.268 *
BUATAX .111 .445 5.927 *
SPEND .006 .014 0.945
MULTI .151 .184 0.200
CAPINT .281 .348 2.147 *
LEV .177 .179 0.126
BVA 4,817 9,013 2.541 *
ROA .103 .109 0.637
VARROA .001 .001 0.517
GROWTH 1.216 1.211 0.138
CEOESO .267 .345 2.387 *
Panel C: Subsample Means--Pre-Tax versus After-Tax BU-Manager
Compensation
Mean for Subsample in Mean for Subsample in t-test of
which BUATAX = 0 which BUATAX = 1 Differences
Variable (n = 143) (n = 66) in Means
ETR .363 .340 2.494 *
CEOATAX .497 .864 6.143 *
SPEND .013 .005 1.051
MULTI .160 .195 1.195
CAPINT .298 .375 2.233 *
LEV .178 .177 0.070
BVA 5,779 10,848 2.351 *
ROA .111 .096 1.543
VARROA .001 .001 0.517
GROWTH 1.216 1.211 .138
FWONLY .056 .258 3.503 *
BUSTK .559 .697 1.948
* Significant at the .05 level (two-tailed test).
Variable Definitions:
ETR = average annual ETR, defined as the ratio of total tax
expense to pre-tax income, for the three years
ending in 1997;
CEOATAX = 1 if the CEO's annual bonus is based on after-tax
accounting results;
BUATAX = 1 if the firm's BU managers are compensated on an
after-tax basis;
SPEND = 1997 spending on the corporate tax function scaled by
total SGA expense;
BVA = book value of total assets as of year-end 1997
(000s omitted);
MULTI = the ratio of foreign assets to BVA as of year-end 1997;
CAPINT = the ratio of property, plant, and equipment to BVA as of
year-end 1997;
LEV = the ratio of long-term debt to BVA as of year-end 1997;
ROA = the ratio of the average pre-tax income for the three
years ending in 1997 to BVA as of year-end
1997;
VARROA = the variance in ROA for the three-year period ended in
1997;
GROWTH = the geometric mean annual growth rate of the market value
of assets (MVA) for the three years 1995-1997;
FWONLY = 1 if the firm uses a firm-wide accounting measure, but
not a unit-based measure, in its BU-manager
incentive compensation plans;
CEOESO = the percentage of the CEO's compensation received in 1997
attributable to stock option grants calculated as the
ratio of the value of stock option grants to the sum of
the CEO's salary, annual bonus, and value of stock option
grants; and
BUSTK = 1 if stock price is used as a performance measure in the
incentive compensation plans of a majority of the firm's
business-unit managers.
TABLE 3
Correlations
ETR BUATAX CEOATAX LSPEND
ETR 1.000
BUATAX -.169 * 1.000
CEOATAX -.157 * .350 * 1.000
LSPEND -.158 * -.004 -.010 1.000
MULTI .053 .088 .087 -.061
CAPINT .042 .160 * .146 * -.021
LEV .147 * -.005 .009 -.048
SIZE -.070 .197 * .162 * -.130
ROA .053 -.102 .046 -.047
VARROA -.070 -.019 -.035 -.016
GROWTH -.029 -.087 -.010 .126
FWONLY -.163 * .289 * .051 -.031
CEOESO -.040 .132 .158 * -.130
BUSTK .044 .131 .137 * -.156 *
MULTI CAPINT LEV SIZE ROA
ETR
BUATAX
CEOATAX
LSPEND
MULTI 1.000
CAPINT -.058 1.000
LEV -.076 .387 * 1.000
SIZE .130 .008 .221 * 1.000
ROA .255 * .045 -.288 * -.289 * 1.000
VARROA .048 -.074 -.202 * -.304 * .246 *
GROWTH .081 -.124 -.118 -.043 .293 *
FWONLY -.060 .112 -.047 -.124 .127
CEOESO .165 * -.063 -.021 .317 * .090
BUSTK .030 .034 .119 .287 * .056
VARROA GROWTH FWONLY CEOESO
ETR
BUATAX
CEOATAX
LSPEND
MULTI
CAPINT
LEV
SIZE
ROA
VARROA 1.000
GROWTH .160 * 1.000 *
FWONLY .111 .079 1.000
CEOESO .003 .081 .092 1.000
BUSTK -.016 -.061 .028 .111
* Significant at the .05 level.
Variable Definitions:
ETR = average annual ETR, defined as the ratio of total
tax expense to pre-tax income, for the three years
ending in 1997;
BUATAX = 1 if the firm's BU managers are compensated on an
after-tax basis;
CEOATAX = 1 if the CEO's annual bonus is based on after-tax
accounting results;
LSPEND = the natural logarithm of 1 plus total 1997 spending
on the tax function scaled by total SGA expense;
MULTI = the ratio of foreign assets to the book value of total
assets (BVA) as of year-end 1997;
CAPINT = the ratio of property, plant, and equipment to BVA as
of year-end 1997;
LEV = the ratio of long-term debt to BVA as of year-end 1997;
SIZE = the natural logarithm of BVA as of year-end 1997;
ROA = the ratio of the average pre-tax income for the three
years ending in 1997 to BVA as of year-end 1997;
VARROA = the variance in ROA for the three-year period ended in
1997;
GROWTH = geometric growth rate of the market value of assets for
the three years 1995-1997;
FWONLY = 1 if the firm uses a firm-wide accounting measure, but
not a unit-based measure, in its BU-manager compensation
plans;
CEOESO = the percentage of the CEO's compensation received in 1997
attributable to stock option grants calculated as the ratio
of the value of stock option grants to the sum of the CEO's
salary, annual bonus, and value of stock option grants; and
BUSTK = 1 if stock price is used as a performance measure in the
incentive compensation plans of a majority of the firm's
business-unit managers.
TABLE 4
ETR Model Estimations
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
(1) (a)
Independent Pred. Coeff. p-value
Variable Coeff. Sign Est. (c)
Intercept [[beta].sub.0] ? 0.391 .000
CEOATAX [[beta].sub.1] - -0.010 .152
BUATAX [[beta].sub.2] - -0.011 .129
LSPEND [[beta].sub.3] - -0.617 .067
MULTI [[beta].sub.4] ? -0.022 .393
CAPINT [[beta].sub.5] ? -0.031 .247
LEV [[beta].sub.6] ? 0.028 .491
SIZE [[beta].sub.7] ? -0.002 .471
ROA [[beta].sub.8] + -0.012 .559
VARROA [[beta].sub.9] - -5.708 .010
GROWTH [[beta].sub.10] + 0.005 .398
CEOESO [[beta].sub.11] - 0.009 .838
BUSTK [[beta].sub.12] - 0.009 .915
[[lambda].sub.CEO] [[theta].sub.CEO] ? NA
[[lambda].sub.BU] [[theta].sub.BU] ? NA
Model Chi-Squared 2.34
(190 d.f.)
Adj./Pseudo [R.sup.2] .104
(2) (b)
Independent Coeff. p-value
Variable Est. (c)
Intercept 0.386 .000
CEOATAX 0.063 .660
BUATAX -0.071 .005
LSPEND -0.675 .048
MULTI -0.017 .614
CAPINT -0.032 .667
LEV 0.007 .863
SIZE -0.003 .583
ROA -0.110 .774
VARROA -5.519 .036
GROWTH 0.004 .429
CEOESO 0.000 .504
BUSTK 0.013 .920
[[lambda].sub.CEO] -0.035 .708
[[lambda].sub.BU] 0.044 .053
Model Chi-Squared 2.43
(188 d.f.)
Adj./Pseudo [R.sup.2] .121
(a) Column (1) presents the results from estimating the ETR model
using OLS with all variables assumed exogenous and excluding
[[lambda].sub.CEO] and [[lambda].sub.BU] from the model.
(b) Column (2) presents the results from estimating the ETR model
augmented with the selectivity correction ([lambda]) variables from
the CEOATAX and BUATAX selection equations. The correlation between
the error terms of the ETR equation and the CEOATAX (BUATAX) selection
equation is -.495 (.638).
(c) p-values are based on a one- (two-) tailed test where the
coefficient sign is (not) predicted.
Variable Definitions:
ETR = average annual ETR, defined as the ratio of total tax expense
to pre-tax income, for the three years ending in 1997;
CEOATAX = 1 if the CEO's annual bonus is based on after-tax accounting
results;
BUATAX = 1 if the firm's BU managers are compensated on an after-tax
basis;
LSPEND = the natural logarithm of 1 plus total 1997 spending on the
tax function scaled by total SGA expense;
MULTI = the ratio of foreign assets to the book value of total
assets (BVA) as of year-end 1997;
CAPINT = the ratio of property, plant, and equipment to BVA as of
year-end 1997;
LEV = the ratio of long-term debt to BVA as of year-end 1997;
SIZE = the natural logarithm of BVA as of year-end 1997;
ROA = the ratio of the average pre-tax income for the three years
ending in 1997 to BVA as of year-end 1997;
VARROA = the variance in ROA for the three-year period ended in 1997;
GROWTH = the geometric mean annual growth rate of the market value of
assets (MVA) for the three years 1995-1997;
CEOESO = the percentage of the CEO's compensation received in 1997
attributable to stock option grants calculated as the ratio
of the value of stock option grants to the sum of the CEO's
salary, annual bonus, and value of stock option grants;
BUSTK = 1 if stock price is used as a performance measure in the
incentive compensation plans of a majority of the firm's
business-unit managers;
IND = vector of industry indicator variables for the mining and
construction, transportation and communication, utilities,
sales, financial institutions and insurance, and services
industries (results not tabulated); and
[[lambda].sub.CEO] ([[lambda].sub.BU]) = the selectivity correction
variable from the CEOATAX (BUATAX) equation resulting from
the bivariate probit estimation of the CEOATAX and BUATAX
equations.
TABLE 5
CEOATAX Model Estimations
CEOATA[X.sub.i] = [[gamma].sub.0] + [[gamma].sub.1]MULT[I.sub.i]
+ [[gamma].sub.2]CAPIN[T.sub.i] + [[gamma].sub.3]LE[V.sub.i]
+ [[gamma].sub.4]SIZ[E.sub.i] + [[gamma].sub.5]RO[A.sub.i]
+ [[gamma].sub.6]VARRO[A.sub.i] + [[gamma].sub.7]GROWT[H.sub.i]
+ [[gamma].sub.8]CEOES[O.sub.i] + [[gamma].sub.9-14]
IN[D.sub.i] + [[epsilon].sub.i]
(1) (a)
Independent Pred. Coeff.
Variable Coeff. Sign Est. p-value (c)
Intercept [[gamma].sub.0] ? -1.056 .160
MULTI [[gamma].sub.1] ? 0.405 .242
CAPINT [[gamma].sub.2] + 1.285 .013
LEV [[gamma].sub.3] + -0.271 .621
SIZE [[gamma].sub.4] + 0.094 .091
ROA [[gamma].sub.5] + 2.398 .087
VARROA [[gamma].sub.6] + 37.118 .257
GROWTH [[gamma].sub.7] - -0.256 .281
CEOESO [[gamma].sub.8] - 0.638 .936
Model Chi-Squared
(14 d.f.) 25.92
Pseudo [R.sup.2] .093
(2) (b)
Independent Coeff.
Variable Est. p-value (c)
Intercept -1.058 .174
MULTI 0.430 .247
CAPINT 1.247 .019
LEV -0.210 .588
SIZE 0.092 .115
ROA 2.360 .098
VARROA 37.126 .278
GROWTH -0.246 .286
CEOESO 0.650 .930
Model Chi-Squared
(14 d.f.) NA
Pseudo [R.sup.2] NA
(a) Column (1) presents the results from a probit estimation of
the CEOATAX model.
(b) Column (2) presents results from the maximum likelihood
bivariate probit estimation of the CEOATAX and BUATAX models.
The correlation between the error terms of the CEOATAX and BUATAX
selection equations is -.331 (p = .813).
(c) p-values are based on a one- (two-) tailed test where the
coefficient sign is (not) predicted.
Variable Definitions:
CEOATAX = 1 if the CEO's annual bonus is based on after-tax
accounting results;
MULTI = the ratio of foreign assets to the book value of
total assets (BVA) as of year-end 1997;
CAPINT = the ratio of property, plant, and equipment to BVA
as of year-end 1997;
LEV = the ratio of long-term debt to BVA as of year-end 1997;
SIZE = the natural logarithm of BVA as of year-end 1997;
ROA = the ratio of the average pre-tax income for the three
years ending in 1997 to BVA as of year-end
1997;
VARROA = the variance in ROA for the three-year period ended
in 1997;
GROWTH = the geometric mean annual growth rate of the market
value of assets (MVA) for the three years 1995-1997;
CEOESO = the percentage of the CEO's compensation received in
1997 attributable to stock option grants calculated
as the ratio of the value of stock option grants to
the sum of the CEO's salary, annual bonus,
and value of stock option grants; and
IND = vector of industry indicator variables for the mining
and construction, transportation and communication,
utilities, sales, financial institutions and insurance,
and services industries (results not tabulated).
TABLE 6
BUATAX Model Estimations
[BUATAX.sub.i] = [[alpha].sub.0] + [[alpha].sub.1][CEOATAX.sub.i]
+ [[alpha].sub.2][MULTI.sub.i] + [[alpha].sub.3][CAPINT.sub.i]
+ [[alpha].sub.4][LEV.sub.i] + [[alpha].sub.5][SIZE.sub.i]
+ [[alpha].sub.6][ROA.sub.i] + [[alpha].sub.7][VARROA.sub.i]
+ [[alpha].sub.8][GROWTH.sub.i] + [[alpha].sub.9][FWONLY.sub.i]
+ [[alpha].sub.10][BUSTK.sub.i] + [[gamma]].sub.11-16][IND.sub.i]
+ [[phi].sub.i]
(1) (a)
Independent Pred. Coeff. p.value
Variable Coeff. Sign Est. (c)
Intercept [[alpha].sub.0] ? -1.418 .082
CEOATAX [[alpha].sub.1] + 0.945 .000
MULTI [[alpha].sub.2] + 0.821 .101
CAPINT [[alpha].sub.3] + 1.204 .048
LEV [[alpha].sub.4] + -1.371 .895
SIZE [[alpha].sub.5] + 0.099 .109
ROA [[alpha].sub.6] + -3.975 .959
VARROA [[alpha].sub.7] + 33.426 .305
GROWTH [[alpha].sub.8] - -0.592 .157
FWONLY [[alpha].sub.9] + 1.527 .000
BUSTK [[alpha].sub.10] - 0.252 .854
Model Chi-Squared (16 d.f.) 69.46
Pseudo [R.sup.2] .267
(2) (b)
Independent Coeff. p-value
Variable Est. (c)
Intercept -1.421 .278
CEOATAX 1.452 .230
MULTI 0.707 .212
CAPINT 0.928 .270
LEV -1.285 .828
SIZE 0.072 .324
ROA -4.374 .942
VARROA 24.341 .401
GROWTH -0.526 .277
FWONLY 1.484 .009
BUSTK 0.242 .188
Model Chi-Squared NA
Pseudo [R.sup.2] NA
(a) Column (1) presents the results from a probit estimation of the
BUATAX model.
(b) Column (2) presents results from the maximum likelihood bivariate
probit estimation of the CEOATAX and BUATAX models. The correlation
between the error terms of the CEOATAX and BUATAX selection equations
is -.331 (p = .813).
(c) p-values are based on a one- (two-) tailed test where the
coefficient sign is (not) predicted.
Variable Definitions:
BUATAX = 1 if the firm's BU managers are compensated on an after-tax
basis;
CEOATAX = 1 if the CEO's annual bonus is based on after-tax accounting
results;
MULTI = the ratio of foreign assets to the book value of total
assets (BVA) as of year-end 1997;
CAPINT = the ratio of property, plant, and equipment to BVA as of
year-end 1997;
LEV = the ratio of long-term debt to BVA as of year-end 1997;
SIZE = the natural logarithm of BVA as of year-end 1997;
ROA = the ratio of the average pre-tax income for the three years
ending in 1997 to BVA as of year-end 1997;
VARROA = the variance in ROA for the three-year period ended in 1997;
GROWTH = the geometric mean annual growth rate of the market value of
assets (MVA) for the three years 1995-1997;
FWONLY = 1 if the firm uses a firm-wide accounting measure, but not a
unit-based measure, in its BU-manager incentive compensation
plans;
BUSTK = 1 if stock price is used as a performance measure in the
incentive compensation plans of a majority of the firm's
business-unit managers; and
IND = vector of industry indicator variables for the mining and
construction, transportation and communication, utilities,
sales, financial institutions and insurance, and services
industries (results not tabulated).
I thank the following members of my dissertation committee for their help and guidance: Dan Collins (chairman), Murray Barrick, Ed Maydew, Forrest Nelson, Mort Pincus, and Rick Tubbs. I also appreciate the helpful comments and suggestions of Ramji Balakrishnan, Amy Dunbar, Bill Greene, Bruce Johnson, Lil Mills, Sonja Olhoft Rego, Tom Omer, George Plesko, Richard Sansing, Peter Wilson, and workshop participants at the University of Connecticut, The University of Iowa, The University of Texas at Austin, 1999 University of Illinois Tax Symposium, 2000 American Taxation Association Midyear Meeting, and the 2000 American Accounting Association Northeast Regional Meeting, along with two anonymous reviewers and the associate editor. I am indebted to the Financial Executives Institute, particularly Carol Garnant and John Porter, for sponsoring the survey. I also thank Maggie Jesse and Josh Kaine for their help in implementing the survey and Kristi Herbrechtsmeyer for her help in collecting data. The financial assistance of The University of Iowa Tippie College of Business, the University of Connecticut School of Business, and the Deloitte & Touche Foundation is gratefully acknowledged. This paper is based on my dissertation completed at The University of Iowa.
Editor's note: This paper was accepted by Terry Shevlin, Senior Editor.
(1) The difference in after-tax income between firms compensating its managers on a pre- versus after-tax basis is the preferable empirical surrogate for measuring the effectiveness of after-tax accounting-based compensation schemes. Using this measure is infeasible, however, due to the many unidentifiable and uncontrollable factors that affect a firm's profitability.
(2) ETR minimization does not imply a firm has engaged in effective tax planning. An extreme example where ETR minimization does not imply effective tax planning is when a firm invests solely in assets that generate a tax-free return (e.g., municipal bonds) and achieves a zero ETR when greater after-tax income and cash flows could have been achieved through investments in taxable assets. Managers attempting to maximize after-tax accounting profits, however, would not engage in this activity.
(3) Lev and Thiagarajan (1993) and Abarbanell and Bushee (1997) both provide evidence that ETRs are negatively related to future changes in earnings and future stock returns. Guenther and Jones (2002) provide evidence that ETR changes are positively related to stock returns, but only if the ETR change increases earnings. These studies suggest that a firm's ETR-lowering strategies may not always lead to a higher stock price.
(4) Thirty-two percent of this study's sample compensates its BU managers on an after-tax basis.
(5) Reporting income at the lower U.S. tax rate instead of a higher foreign tax rate always lowers the ETR. Reporting income in a foreign subsidiary located in a country with a lower tax rate lowers the ETR if the firm takes the position that unrepatriated earnings are permanently reinvested in the foreign subsidiary. Otherwise, the firm would be required to record deferred tax expense equal to the incremental U.S. tax due when the earnings are repatriated.
(6) Without adequate documentation, tax professionals could forgo taking ETR-lowering deductions and credits, or, if taken and subsequently audited, the tax authority could disallow deductions and credits not properly documented. Also, to the extent that the firm's tax professionals do not have to spend time gathering data, they have more time to devote to the identification of tax-planning strategies.
(7) BU-level data are not available, so firm-level data must be used to compute variables included in the BU-manager pre- versus after-tax compensation choice model. Moreover, Equation (3) models a firm's decision to compensate a majority of its BU managers on an after-tax basis and does not seek to investigate the determinants of this decision on a unit-by-unit basis. Informal discussions with pilot study participants indicated that firms generally compensate all units' managers on either a pre-tax or after-tax basis.
(8) Responses to Question 18 of the survey (see the Appendix) were used to construct this variable. A BU accounting measure is considered to be an after-tax measure if the respondent circled either Statement 3 or Statement 4. A BU accounting measure is thus after-tax if either (1) the effects of a few (possibly only one) income-tax-related items attributable to specific business units are allocated to the business units in determining this measure or (2) the company's total income tax liability is extensively allocated among the business units in determining this measure. A BU measure is considered pre-tax if (1) there is no tax allocation or (2) a pro rata allocation of the firm's tax liability is allocated to the BUs.
(9) Under SFAS No. 14, which was in effect in 1997, geographic segment disclosures are required only if a firm's multinational operations meet a 10 percent materiality threshold. To the extent a firm responded that it filed one or more foreign tax returns but reported no foreign assets in its 1997 10-K, a 5 percent ratio of foreign-to-total assets is assumed.
(10) MVA is computed as the book value of assets minus the book value of equity plus the product of year-end stock price times the number of shares of common stock outstanding at year-end.
(11) 1997 salary, annual bonus, and BSV of options granted were obtained from firms' proxy statements. When the BSV of options granted was not disclosed, the BSV of options granted was computed as the number of options granted to the CEO multiplied by the weighted average per share value of options granted to all employees as in the disclosed FAS No. 123 footnote contained in the firms' annual reports.
(12) CEOESO and BUSTK, along with LSPEND, could be modeled as a function of CEOATAX and as endogenous to the ETR equation. Arguably, however, modeling the CEOATAX and BUATAX variables as endogenous helps correct for much of the potential endogeneity bias associated with the explanatory variables included in the ETR equation.
(13) As a sensitivity test, Equations (2) and (3) are reestimated including LSPEND as an explanatory variable in both equations. LSPEND is insignificant in both estimations.
(14) Following Greene (1998a; 2003, 383), the CEOATAX and BUATAX equations are recursive because after-tax CEO compensation is determined solely by exogenous variables. Then, given CEOATAX, after-tax BU-manager compensation is also a function of exogenous variables. The two equations are jointly determined via the direct effect of including CEOATAX in the BUATAX equation and indirectly through the equations' error terms.
(15) If the CEOATAX equation were the only selection equation, the selectivity correction variable, [[lambda].sub.CEO], would be calculated as [phi](-[gamma]'[X.sub.2])/[PHI](-[gamma]'[X.sub.2]).
(16) Estimation of the augmented Equation (1) using LIMDEP produces corrected standard errors. See Greene (1998b, 740) for details on how the asymptotic covariance matrix is computed.
(17) One limitation of the two-step estimation approach is that it could lead to a loss of efficiency relative to using maximum likelihood procedures (Wales and Woodland 1980). In a single-selection setting, Nelson (1984) shows that efficiency loss is greatest when the exogenous variables in the selection and outcome equations are highly correlated. Because most of the exogenous ETR equation explanatory variables are also included in the CEOATAX and BUATAX selection equations, efficiency loss could result.
(18) 1997 Compustat data were not available at the time survey firms were selected.
(19) Survey (non-survey) firms have an ETR of .353 (.314), BVA of 7,769 million (1,149 million), LEV of .179 (.160), and CAPINT of .311 (.227). All mean differences are significant at the .05 level in a two-tailed test.
(20) The income tax footnotes of firms with extreme ETR observations were reviewed to identify the sources of extreme ETRs. Extreme ETRs were recomputed to remove the effects of large, nonrecurring statutory reconciliation items such as the effects of a business dispositions and asset impairments, and decreases in the deferred tax asset valuation allowance. The results based on estimating the model without these eight observations with extreme ETR observations prior to correction (n = 201) are substantially the same as those reported in Tables 4 through 6.
(21) Seventy percent of the sample firms in Atwood et al. (1998) use after-tax earnings in the CEO's incentive compensation plan. Samples in other studies (e.g., Newman 1989; Dhaliwal et al. 1999) have a considerably lower percentage of after-tax firms. Atwood et al. (1998), unlike these other studies, code a bonus plan as after-tax when the firm's proxy statement uses "net income" or related terminology that is silent regarding the pre-tax versus after-tax nature of this amount. They assume terms such as "net income" means after-tax income unless otherwise stated.
(22) As previously discussed, the ETR does not reflect the tax benefits of ESO deductions. To the extent that firms with large ESO deductions are less likely to use after-tax compensation, adjusting the ETRs to reflect these tax benefits would bias toward failing to reject the null hypotheses of no association between ETRs and after-tax performance measures.
(23) A firm's tax savings from using after-tax BU performance measures was computed as the product of the coefficient on [[beta].sub.2], -.071, and that firm's pre-tax income. Mean (median) per-firm tax savings reported above represent the mean (median) tax savings for the 209 sample firms.
(24) When the BUATAX model is estimated using bivariate probit, the insignificant correlation between the CEOATAX and BUATAX models' error terms leads to a loss of efficiency consistent with the estimations reported in Greene (2003, 717). Following Greene (1998; 2003, 717), inferences concerning the effect of the BUATAX determinants on the probability that a firm compensates its BU managers on an after-tax basis should be based on the single equation estimation of the BUATAX model, which is achieved by restricting the correlation in error terms between the BUATAX and CEOATAX equation to zero. As a sensitivity test, the ETR model was reestimated with the selectivity correction variables recomputed based on a bivariate estimation of the CEOATAX and BUATAX models in which the correlation parameter was restricted to zero. The results of this sensitivity test are substantially the same as those reported in Table 4.
(25) In the estimation that includes firms with average MTRs greater than .335 (n = 157), the coefficient on BUATAX is -.080 and significant (p = .005).
(26) Mean (median) internal spending on the tax function, as a percentage of total SG&A, is .008 (.002). Mean (median) outsourcing expenditures as a percentage of total SG&A is .003 (.001). In t-tests of mean differences, neither internal nor external spending differs across types of bonus plans.
REFERENCES
Abarbanell, J., and B. Bushee. 1997. Fundamental analysis, future earnings, and stock prices. Journal of Accounting Research 35 (Spring): 1-24.
Antle, R., and J. Demski. 1988. The controllability principle in responsibility accounting. The Accounting Review 63: 700-718.
Atwood, T. J., T. Omer, and M. Shelley. 1998. Before versus after-tax earnings as performance measures in compensation contracts. Managerial Finance 11: 30-44.
Bankman, J. 1994. The structure of Silicon Valley start-ups. UCLA Law Review 41 (September): 1737-1768.
Bushman, R., R. Indjejikian, and A. Smith. 1995. Aggregate performance measures in business unit manager compensation: The role of intrafirm interdependencies. Journal of Accounting Research 33 (Supplement): 101-128.
Carnes, G., and D. Guffey. 2000. The influence of international status and operating segments on firms' choice of bonus plans. Journal of International, Accounting, Auditing, and Taxation 9 (1): 43-57.
Dhaliwal, D., J. Sneed, and R. Trezevant. 2000. Factors associated with the choice to use before-tax or after-tax earnings as the contracting variable in annual bonus plans. Working paper, University of Southern California.
Dillman, D. 1978. Mail and Telephone Surveys: The Total Design Method. New York, NY: John Wiley.
Douglas, R., P. Ellingsworth, and J. McAndrews. 1996. What management expects: Evaluation of the tax department and its executives. The Tax Executive (Jan/Feb): 29-32.
Dunbar, A., and J. Phillips. 2001. The outsourcing of corporate tax function activities. The Journal of the American Taxation Association 23 (Fall): 35-49.
Ernst & Young. 1999. Transfer pricing 1999 global survey: practices, perceptions, and trends. Tax Notes (November 22): 1073-1100.
Graham, J. 1996. Proxies for the corporate marginal tax rate. Journal of Financial Economics 42: 187-221.
Greene, W. 1998a. Gender economics courses in liberal arts colleges: Further results. Journal of Economic Education 29 (Fall): 291-300.
--. 1998b. LIMDEP Version 7.0 User's Manual. Plainview, NY: Econometric Software, Inc.
--. 2003. Econometric Analysis. 5th edition. Upper Saddle River, NJ: Prentice Hall, Inc.
Guenther, D., and D. Jones. 2002. Valuation implications of changes in firms' effective tax rates. Working paper, University of Colorado at Boulder.
Guidry, F., A. Leone, and S. Rock. 1999. Earnings-based bonus plans and earnings management by business-unit managers. Journal of Accounting and Economics 26:113-142.
Gupta, S., and K. Newberry. 1997. Determinants of the variability in corporate effective tax rates: Evidence from longitudinal data. Journal of Accounting and Public Policy 16: 1-34.
Hanlon, M. 2002. The persistence and pricing of earnings, accruals, and cash flows when firms have large book-tax differences. Ph.D. dissertation, University of Washington.
-- , and T. Shevlin. 2002. Accounting for tax benefits of employee stock options and implications for research. Accounting Horizons 16 (1): 1-16.
Healy, P. 1985. The effect of bonus schemes on accounting decisions. Journal of Accounting and Economics 7: 85-107.
Heckman, J. 1976. The common structure of statistical models with truncation, sample selection and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement 5: 475-492.
--. 1990. Varieties of selection bias. American Economic Review (Papers and Proceedings) 80: 313-318.
--. 1997. Instrumental variables. A study of implicit behavioral assumptions used in making program evaluations. Journal of Human Resources 32: 441-462.
Ittner, C., and D. Larcker. 2001. Assessing empirical research in managerial accounting: A value-based management perspective. Journal of Accounting and Economics 32 (December): 349410.
Kahn, L., and P. Sherer. 1990. Contingent pay and managerial performance. Industrial and Labor Relations Review 43 (February): 107S-120S.
Keating, S. 1997. Determinants of divisional performance evaluation practices. Journal of Accounting and Economics 24: 243-273.
Krull, L. 2001. The relation between permanently reinvested earnings, repatriation taxes, and earnings management incentives of U.S. multinationals. Ph.D. dissertation, The University of Arizona.
Larcker, D. 1983. The association between performance plan adoption and corporate capital investment. Journal of Accounting and Economics 5: 3-30.
Lev, B., and R. Thiagarajan. 1993. Fundamental information analysis. Journal of Accounting Research 31 (2): 190-215.
McLemore, I. 1997. World-class tax planning. Controller Magazine (April): 41-44.
Milgrom, P., and J. Roberts. 1992. Economics, Organization and Management. Englewood Cliffs, NJ: Prentice Hall.
Mills, L., and M. Erickson, and E. Maydew. 1998. Investments in tax planning. The Journal of the American Taxation Association 20 (Spring): 1-20.
Mintz, S. 1999. A taxing challenge. CFO Magazine (November).
Nelson, F. 1984. Efficiency of the two-step estimator for models with endogenous sample selection. Journal of Econometrics 24: 181-196.
Newman, H. 1989. Selection of short-term accounting-based bonus plans. The Accounting Review 64: 758-772.
Scholes, M., M. Wolfson, M. Erickson, E. Maydew, and T. Shevlin. 2002. Taxes and Business Strategy: A Planning Approach. 3rd edition. Upper Saddle River, NJ: Prentice Hall, Inc.
Swenson, C. 1999. Increasing stock market value by reducing effective tax rates. Tax Notes (June 7): 1503-1505.
Tunali, I. 1986. A general structure for models of double-selection and an application to a joint migration/earnings process with remigration. Research in Labor Economics 8B: 235-282.
Vella, F. and M. Verbeek. 1999. Estimating and interpreting models with endogenous treatment effects. Journal of Business and Economic Statistics 17 (4): 473-478.
Wales, T and A. Woodland. 1980. Sample selectivity and the estimation of labor supply functions. International Economic Review 21: 437-468.
Wallace, J. 1997. Adopting residual income-based compensation plans: Do you get what you pay for? Journal of Accounting and Economics 24: 275-300.
Wang, S. 1991. The relation between firm size and effective tax rates: A test of firms' political success. The Accounting Review 66 (January): 158-169.
Wilkie, P. 1988. Corporate average effective tax rates and inferences about relative tax preferences. The Journal of the American Taxation Association 10 (Fall): 75-88.
Wilson, G. P. 1995. Corporate tax effectiveness: Adding value through the tax function. Working paper, Harvard University.
Ziegler, B. 1997. IBM's profit tops analyst expectations. Wall Street Journal (July 22): A3.
Submitted April 2001
Accepted February 2003
John D. Phillips
University of Connecticut