Taxes as a determinant of managerial compensation in privately held insurance companies.

I. INTRODUCTION

The objective of this study is to empirically investigate how taxes affect managerial compensation in privately held firms whose managers own a large percentage of the firm's stock (hereafter I refer to these firms as management-owned firms). I focus on management-owned firms

Despite the theoretical argument for the importance of taxes in compensation arrangements of management-owned firms, empirical evidence has been limited because detailed firm-level data are seldom publicly available. Instead, the compensation literature has largely focused on publicly traded firms, where share ownership is highly dispersed and compensation data are accessible. This literature finds that taxes are not a significant determinant of managerial compensation in publicly traded firms. For example, Goolsbee (2000) documents that the significant increase in individual tax rates in 1993 encouraged corporate executives to accelerate the exercise of stock options from 1993 to 1992, but had little impact on the level of managerial compensation (see also Huddart 1998). (1) Hall and Liebman (2000) find that despite numerous changes in the tax advantages of stock options relative to cash compensation over the past two decades, the fraction of annual compensation from stock options is virtually unrelated to tax changes and is more closely related to changes in corporate governance (e.g., the increasing role of institutional investors). Rose and Wolfram (2000) further find that IRC [section] 162(m), which capped publicly traded firms' tax deductibility of nonperformance-based compensation paid to the top five executives at $1 million, had little effect on the level and structure of executive compensation. These findings suggest that agency considerations dominate tax minimization in setting managerial compensation of publicly traded firms. However, there is no evidence on whether managerial compensation is more responsive to taxes when managerial agency costs are low.

The effect of taxes on compensation design also has public policy implications. Because shareholder/managers control significant amounts of economic resources, their responses to taxes affect Treasury revenue receipts and economic efficiency (Feldstein 1999; Slemrod 1995). (2) Recent presidential debates on tax policy have hinged on the way and magnitude in which high-income individuals respond to tax changes (see, e.g., Laffer and Moore 2000).

I use a sample of management-owned, privately held property-liability insurers (PL insurers) during 1989-1996 to test whether shareholder/managers adjust their own compensation to minimize the taxes on their total income from the firm. I focus on the insurance industry because firm-level data are available for privately held insurers. Similar information is not disclosed for privately held firms in most other industries.

Shareholder-managers receive two types of income from the firm they own: compensation income as employees, and investment income (i.e., dividends and capital gains) as shareholders. In the absence of tax influences, these two payments are independently determined. The labor market determines the level of compensation, whereas the capital market determines investment income.

The U.S. federal income tax system taxes compensation income and investment income differently. Compensation income is fully taxable to individual employees, while it is deductible by employing firms. Investment income from a corporation is subject to so-called "double taxation." First, it is subject to corporate income tax; second, income after corporate taxes is further taxed at the shareholder level when individual shareholders receive dividends from the firm or sell their shares and realize capital gains. Thus, shareholder/ managers have an incentive to reduce the double taxation on their investment income by paying themselves more tax-deductible compensation (e.g., salaries and bonuses). (3) This incentive is stronger when corporate tax rates increase or individual tax rates decrease.

Ordinary individual income tax rates were generally lower than corporate marginal tax rates during 1989-1992, but the 1993 Omnibus Budgetary Deficit Reconciliation Act significantly increased individual tax rates after 1992, while keeping corporate statutory tax rates essentially unchanged. Although the difference between individual and corporate tax rates suggests that shareholder/managers have an incentive to use tax-deductible compensation to mitigate double taxation throughout the entire 1989-1996 period, I predict that shareholder/managers paid themselves less deductible compensation in 1993-96 than in 1989-1992, ceteris paribus.

Compensation could change for both tax and nontax reasons. To isolate the effect of the 1993 tax rate change on the amount of deductible compensation shareholder/managers used to avoid the double taxation on their investment income, I use deductible managerial compensation of nonmanagement-owned insurers (i.e., insurers whose managers are not shareholders of the firm) as a benchmark. Because shareholders and managers are independent parties in nonmanagement-owned insurers, shareholders cannot use managerial compensation to reduce the double taxation on their investment income. Thus, nonmanagement-owned insurers represents a useful control group. However, the reader should interpret the study's empirical results with caution because this may not control for all determinants of managerial compensation correlated with ownership structure.

Consistent with my prediction, I find that relative to nonmanagement-owned insurers, the level of tax-deductible employee compensation that management-owned insurers paid declined significantly from 1989-1992 to 1993-96. The mean (median) reduction in deductible compensation over the two periods is $134,000 ($114,000) per year, representing 4.10 percent (2.42 percent) of the management-owned insurers' net premiums written. Furthermore, the reduction in shareholder/managers' compensation after 1992 was not limited to the two years adjacent to the tax law change, suggesting that their responses were persistent.

The rest of this paper is organized as follows. Section II briefly summarizes the changes in federal income taxation during 1989-1996. Section III discusses how privately held insurers are expected to respond to the 1993 Tax Act, and presents the research hypothesis. Section IV describes the sample selection procedure and provides descriptive statistics. Section V presents the research design and variable definitions. Section VI reports the main regression results and sensitivity checks, and Section VII concludes.

II. RECENT HISTORY OF U.S. FEDERAL INCOME TAXATION

Corporations and individual taxpayers face different statutory tax rates in the United States. Panel A of Table 1 shows a chronological history of the top individual and corporate statutory federal tax rates since 1981. The top individual statutory tax rate was significantly higher than the top corporate statutory tax rate before 1987. For example, in 1985, the top individual statutory tax rate was 50 percent, while the top corporate statutory tax rate was 46 percent. The Tax Reform Act of 1986 lowered the top statutory tax rates for both individuals and corporate taxpayers and made the top statutory tax rate lower for individuals (28 percent) than for corporations (34 percent) for the first time in U.S. history. In 1991, the top individual tax rate increased to 31 percent, but it was still lower than the top corporate tax rate. Under the 1993 Budgetary Deficit Reconciliation Act, the top individual and corporate tax rates increased to 39.6 percent and 35 percent, respectively, thus restoring the pre-1987 ordering of the top individual and corporate tax rates.

Panel B provides the progressive tax rate schedules for married individuals filing joint returns for 1989 and 1994, representing the periods 1989-1992 and 1993-1996, respectively. Reflecting changes in the 1993 Tax Act, individual tax rates were uniformly higher in 1994 than in 1989 for taxable income above $89,150 (nominal dollars). (4) The difference in the individual tax rates over the two years ranges from 3 percent to 11.6 percent for taxable income above $89,150. Panel C shows the progressive tax rate schedules for taxable corporations for 1989 and 1994. Except for the one percentage point increase for taxable income over $15 million, the corporate tax rates were the same for the two years. (5)

The information in Panels B and C leads to three conclusions. First, with a taxable income threshold of only $75,000 for the 34 percent rate, most corporations faced a marginal tax rate of at least 34 percent during 1989-1996. (6) Second, individual marginal income tax rates should be lower than corporate marginal tax rates during 1989-1992. Third, because individual tax rates in 1993-1996 exceeded those in 1989-1992 for taxable income of more than $89,150, individual marginal tax rates should have increased after 1992 for managers in my sample of management-owned insurers (relatively high-income individuals). (7) To the extent that shareholder/managers' marginal tax rates did not increase after 1992, I would not observe a decrease in their deductible compensation after 1992 as predicted.

III. RESEARCH HYPOTHESIS

In this section I discuss various tax strategies that management-owned and nonmanagement-owned insurers could take in response to changes in individual and corporate tax rates over the 1989-1996 period. I also explain why nonmanagement-owned insurers can serve as a useful control group to identify the effect of tax changes on the amount of deductible compensation shareholder/managers used to avoid the double taxation on their investment income from the firm. Finally, I discuss the benefits and costs of using deductible compensation to shift income, and present the research hypothesis.

Strategies Common to Both Management-Owned and Nonmanagement-Owned Insurers

Assume that the amount of compensation that managers of privately held insurers can receive is limited to the true value of their personal services to the firm, which I describe as "reasonable compensation." Due to low managerial agency costs, all privately held firms could alter the form and timing of their managers' reasonable compensation to minimize taxes (Scholes and Wolfson 1992, Chapter 7). For example, because individual tax rates increased from 1989-1992 to 1993-96 while corporate tax rates remained almost unchanged, all privately held insurers could shift part of their managers' compensation from 1993 to 1992. In addition, all privately held insurers could increase the ratio of taxable compensation (e.g., salaries and bonuses) to tax-favored compensation (e.g., fringe benefits) during 1989-1992 and reduce this ratio during 1993-96 (see also Gordon and Slemrod 2000).

Strategy Unique to Management-Owned Insurers

In addition to reasonable compensation, shareholder/managers can receive additional income from the firm in the form of compensation (hereafter "unreasonable compensation"). This is because shareholder/managers receive both compensation and investment income from the same firm. Although the sum of the two received by shareholder/managers will likely be fixed due to competitive market forces, one could alter the mix of the two to minimize taxes. In contrast, managers of nonmanagement-owned insurers receive only reasonable compensation as employees. Therefore, their tax-deductible compensation can serve as a control to identify the amount of unreasonable compensation that shareholder/ managers of management-owned insurers use to reduce the double taxation on their investment income.

Hypothesis

I now illustrate how a shareholder/manager's incentive for using unreasonable compensation to avoid double taxation on investment income changes with individual and corporate income tax rates. To highlight the difference in after-tax returns attributed solely to the different tax treatment of compensation and investment income, I assume the before-tax annual rate of return that the firm or individual shareholders can earn is a constant R. I denote the individual and corporate marginal tax rates [[tau].sub.i] and [[tau].sub.c], respectively. Compensation income is taxed at the rate [[tau].sub.i].

If the shareholder/manager used unreasonable deductible compensation to shift all before-tax rate of return R to her personal account, then she would avoid corporate tax and her annual rate of return after all taxes would be:

(1) R(1 - [[tau].sub.i]).

If the shareholder/manager retained the before-tax rate of return R within the firm, then the rate of return after corporate taxes would be R(1 - [[tau].sub.c]). However, the retained earnings would be subject to further shareholder taxes when the shareholder/manager received dividends directly from the firm or realized capital gains through the sale of her stock. Generally, shareholder taxes on dividends or capital gains depend on both the timing of the dividends or capital gains and the individual tax rate when the dividends or capital gains are received. I capture the actual amount of tax burden that the shareholder/manager must pay on the future dividends and capital gains using a hypothetical effective annualized tax rate (denoted [[tau].sub.a]) that she would pay each year on the returns after corporate taxes (see Scholes and Wolfson 1992, 73). Thus, if the firm retained corporate earnings, then the shareholder/manager's annual rate of return after both corporate and shareholder taxes would be:

(2) R(1 - [[tau].sub.c])(1 - [[tau].sub.a]).

Absent any costs of income shifting, the shareholder/manager would use deductible compensation to shift corporate income to her personal tax base if the following condition held:

(3) (1 - [[tau].sub.i]) > (1 - [[tau].sub.c])(1 - [[tau].sub.a]).

Using reasonable assumptions for [[tau].sub.c] and [[tau].sub.a] in equation (3), I calculate the values of [[tau].sub.i] at which the shareholder/manager is indifferent between retaining corporate income within the firm and shifting it to her personal account for the sample period 1989-1996 (see Table 2). The corporate tax rate [[tau].sub.c] is assumed to be a constant 34 percent because it remained relatively unchanged during 1989-1996. I allow the annualized effective tax rate [[tau].sub.a] to vary from 0 percent to 28 percent. Generally, [[tau].sub.a] depends on both the shareholder/manager's investment horizon and future tax rates on dividends and capital gains. However, because dividends are taxed more heavily than long-term capital gains, I assume that management-owned insurers do not distribute corporate earnings in the form of dividends. As a result, the maximum rate for [[tau].sub.a] is 28 percent, the maximum long-term capital gains tax rate during 1989-1996. Allowing a higher value for [[tau].sub.a] increases the shareholder/manager's incentive to shift corporate income to her personal tax base but will not change any of the following discussions. (8)

The maximum individual marginal tax rate was always lower than 34 percent (the assumed corporate tax rate in Table 2) during 1989-1992. Thus, it is clear from Table 2 that the shareholder/manager had an incentive to shift corporate income to her personal tax base during 1989-1992, absent costs of income shifting. Further, the shareholder/ manager might have found it optimal to shift corporate income to her personal tax base in 1993-96 because the maximum marginal tax rate was only 39.6 percent. However, the tax benefits of shifting income to shareholders was reduced from 1989-1992 to 1993-1996 because the individual tax rate increased significantly after 1992.

Tax rule restrictions and transaction costs constrain shareholder-managers' ability to convert investment income to compensation income. For example, the IRS has the power to disallow the deduction of unreasonable compensation on the corporate tax return, but still requires the shareholder/manager to recognize it as ordinary income (see Wilkie et al. 1996). Thus the income-shifting strategy is inferior to the strategy of retaining investment income within the firm if the probability of an IRS challenge is high.

Furthermore, the income-shifting strategy becomes more complicated when there are multiple shareholder/managers with different marginal tax rates, or a mix of shareholder/ managers and non-shareholder/managers. When there is only one shareholder who is also a manager of the firm, the shareholder/manager should have virtually full flexibility to use deductible compensation to change the mix of his compensation and investment income because both types of income accrue to the same person. However, if there are multiple independent shareholders, then the cost (benefit) of an increase (decrease) in compensation paid to a shareholder/manager would be borne by all shareholders. As a result, to make the income-shifting strategy work, shareholder/managers must find costly means to adjust the payoffs to other shareholders simultaneously so that they are not made worse off after the income-shifting strategy. (9) Thus, the tax rule restrictions and transaction costs may be sufficient to prevent management-owned insurers from garnering the tax benefits of income shifting.

The cost function of income shifting is unlikely to have changed while the tax benefits of income shifting decreased from 1989-1992 to 1993-96 due to the significant increase in individual tax rates. Therefore the shareholder/manager should have used deductible compensation to shift less corporate income to her personal tax base after 1992. Using deductible managerial compensation of nonmanagement-owned insurers to control for the common nontax changes affecting managerial compensation, the above analysis leads to the following prediction: (10)

HI: Relative to nonmanagement-owned insurers, management-owned insurers paid less deductible compensation to their shareholder/managers during 1993-96 than during 1989-1992, ceteris paribus.

IV. SAMPLE SELECTION AND DESCRIPTIVE STATISTICS

Data and Sample Selection

I drew the firm-level data used in this study from the NAIC property and liability (PL) insurance database during 1989-1996. (11) The NAIC database contains PL insurers' annual statutory accounting statements filed with state insurance regulators.

PL insurers operate either independently or as a group whose members are owned by a common parent and usually share the same management team. Whereas the majority of stock insurance groups are publicly traded, this study drew its sample of privately held insurers from independent PL stock insurers (see Table 3). Of the initial sample of 408 independent stock insurers incorporated before 1990, I deleted 138 insurers due to lack of ownership information. I excluded an additional 69 insurers due to either ownership change or financial distress. (12) Mayers and Smith (1992) indicate that stock insurers owned by mutual insurers behave differently from other stock insurers. As a result, I excluded 8 additional mutually owned stock insurers. I deleted 21 additional insurers owned by publicly traded companies, and 32 others because the same firm does not have at least one observation for each of the two periods 1989-1992 and 1993-1996. The final sample includes 64 management-owned insurers and 76 nonmanagement-owned insurers.

Management-Owned vs. Nonmanagement-Owned

I classify an insurer as management-owned if top management collectively owns more than 25 percent of the stock. If top management owns no stock, then I classify the insurer as nonmanagement-owned. The ownership information was hand-collected from A. M. Best's Insurance Report: Property and Casualty (1989-1997 editions). As indicated by this report, there are no insurers in the sample whose managers own between 0 and 25 percent of the stock.

For the sample of 64 management-owned insurers, top management owns 100 percent of the stock in 47 insurers, between 50 percent and 99 percent in 12 insurers, and between 25 percent and 49 percent in 5 insurers. None of the following regression inferences change if the management-owned insurer sample includes only those firms whose top management owns at least 50 percent or 100 percent of the stock.

Descriptive Statistics

Table 4 reports descriptive statistics by ownership structure. Except for TAXINC (see below for an explanation), I express all variables in 1989 constant dollars. The variables in Table 4 are means over 1989-1996 for each insurer. There are 444 (551) firm-years for the sample of management-owned insurers (nonmanagement-owned insurers). Of the 140 unique firms, 122 (106) firms have at least two (three) years of observations in each of the two periods 1989-1992 and 1993-96. (13)

Management-owned and nonmanagement-owned insurers differ on several dimensions. Management-owned insurers are significantly smaller, and younger, and operate in fewer states than do nonmanagement-owned insurers. (14) However, return on assets before deductible compensation (ROA) is similar for the two groups. Workers' average annual pay (COST_OF_LIVING) in the insurers' states of domicile is similar for the management-owned and nonmanagement-owned insurers.

COMPENSATION is total tax-deductible compensation paid to all company employees (i.e., salaries and bonuses) obtained from schedule "Part 4--expenses" (line 8a) in PL insurers' annual statements. Because PL insurers do not disclose managerial compensation separately in their annual statements, the later regression analysis uses COMPENSATION as the dependent variable (see Section V for further discussion). Because management-owned insurers are smaller, COMPENSATION is smaller than that of nonmanagement-owned insurers (p < .001).

Table 4 also provides an estimate of total taxable income before COMPENSATION, averaged over 1989-1996 for management-owned and nonmanagement-owned insurers. Taxable income is the sum of taxable investment income and underwriting gain. Because I do not index the corporate tax rate schedule for inflation in Table 1, I express corporate taxable income in nominal dollars in Table 4. Untabulated calculation indicates that taxable income before COMPENSATION over 1989-1996 is more than $75,000 (the threshold for the 34 percent tax rate) for 89 percent of the management-owned insurers and 88 percent of nonmanagement-owned insurers. This result suggests that most of the sample firms faced high corporate tax rates during 1989-1996.

V. RESEARCH DESIGN

Regression Model

To test whether shareholder/managers of management-owned insurers use deductible compensation to shift corporate income to their personal tax bases to avoid double taxation on investment income (i.e., dividends and capital gains), I estimate the following regression model:

(1) [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],

where:

             i = firm index;
             t = year index;
  COMPENSATION = total tax-deductible employee compensation;
     MGR_OWNED = 1 for management-owned insurers, and 0 otherwise;
       YR93_96 = 1 for the 1993-96 period, and 0 otherwise;
      PREMIUMS = net premiums written (in millions);
       LICENSE = the number of states in which an insurer is licensed
                 to underwrite insurance;
           ROA = return on assets, measured by earnings before taxes
                 and total tax-deductible employee compensation,
                 scaled by the average of beginning and ending total
                 admitted assets;
COST_OF_LIVING = workers' average annual pay in an insurer's state of
                 domicile; and
             e = error.

Except for YR93_96, MGR_OWNED, and YR93_96 x MGR_OWNED, I express all variables in natural logarithms to control for the skewness of several variables and to facilitate the interpretation of the regression coefficients as elasticities. (15) The inferences are similar for nontransformed data. Because the descriptive statistics in Table 4 suggest that most privately held insurers had high corporate marginal tax rates during 1989-1996, regression model (1) does not model the variation in the true corporate marginal tax rate.

Dependent Variable

The dependent variable, COMPENSATION, is total tax-deductible compensation paid to all company employees (i.e., salaries and bonuses). Although the research hypothesis in Section III focuses on deductible compensation paid to shareholder/managers and nonshareholder/managers, due to data limitations I test the research hypothesis using total employee compensation. Including lower-level employee compensation in the dependent variable reduces the power of my hypothesis tests, but should not induce any bias. First, the marginal tax rates for lower-level employees were unlikely to change much during 1989-1996. Second, lower-level employees' utility functions are likely too diverse to allow the firm to design an efficient and uniform tax-minimizing strategy. Third, because both management-owned insurers and nonmanagement-owned insurers employ lower-level employees, any feasible tax strategy for lower-level employees should be common to both groups. In addition, lower-level employees are unlikely to be shareholders of the firm, limiting the amount of taxes avoided using their compensation.

Control Variables

I include several variables to control for the exogenous nontax determinants of COMPENSATION. PREMIUMS controls for firm size and LICENSE controls for the number of states in which an insurer is licensed to conduct business. (16) Because firms that are larger and operate in more states are expected to incur more compensation expense (Rosen 1992), the coefficients on PREMIUMS and LICENSE are predicted to be positive. COST_OF_ LIVING controls for the cost of living in an insurer's state of domicile and is predicted to be positive. (17) ROA controls for performance-related compensation. Although Ke et al. (2000) find that CEO compensation is not sensitive to firm performance for privately held insurers, I include ROA as a control because my dependent: variable COMPENSATION includes deductible compensation paid to all employees. The coefficient on ROA should be positive if there is a pay-for-performance relation in my sample firms. (18) I do not allow the coefficient on ROA to vary with MGR_OWNED because share ownership is highly concentrated in both types of insurers, and thus the pay-for-performance sensitivity (if there is any) should not vary across the two groups. Regression results are qualitatively the same if I allow the coefficient on ROA to vary with MGR_OWNED. Further, the coefficient on MGR_OWNED x ROA is never significant.

Predictions Related to H1

If all exogenous nontax determinants of COMPENSATION are controlled, then the coefficient on MGR_OWNED captures the tax-induced difference in COMPENSATION between management-owned insurers and nonmanagement-owned insurers for the period 1989-1992. Similarly, the sum of the coefficients on MGR_OWNED and MGR_OWNED x YR93_96 captures the tax-induced difference in COMPENSATION between management-owned insurers and nonmanagement-owned insurers for the period 1993-1996. The coefficient on MGR_OWNED x YR93_96 measures the incremental tax-induced change in COMPENSATION from 1989-1992 to 1993-1996 for management-owned insurers relative to nonmanagement-owned insurers. Hypothesis 1 predicts the coefficient on MGR_ OWNED x YR93_96 to be negative. Assuming all nontax determinants of COMPENSATION are controlled, a positive coefficient on MGR_OWNED (a positive sum of the coefficients on MGR_OWNED and MGR_OWNED x YR93_96) would be consistent with shareholder/managers of management-owned insurers using unreasonable deductible compensation to shift corporate income to their personal tax bases during 1989-1992 (1993-1996).

Realistically, the regression model is unlikely to control for all nontax determinants of COMPENSATION, but to the extent that these omitted variables are uncorrelated with MGR_OWNED, the regression coefficients should still be consistent. One potential correlated omitted variable is number of employees. Although PREMIUMS and total assets are highly correlated with number of employees, they may not be adequate proxies for the number of employees. Similarly, the ratio of managers to lower-level employees could systematically differ across management-owned insurers and nonmanagement-owned insurers.

Unfortunately, data on the number of employees or on the ratio of managers to lower-level employees are unavailable. However, both the number of employees and the ratio of managers to lower-level employees (or any other firm fixed effects) are likely to be stable over time; thus I use fixed-effects regression to control for these correlated omitted variables. Despite numerous controls for the nontax determinants of employee compensation and alternative specifications of the dependent variable in later sensitivity checks, one should interpret the following empirical results with caution. As indicated above, I omitted some important determinants of compensation from the regression model and simply treated them as firm fixed effects. To the extent that these omitted variables change over time and my existing time-varying control variables do not adequately control for these omitted variables, the estimated regression coefficients could be biased.

VI. EMPIRICAL RESULTS

Pooled Regression Results

Table 5 reports the pooled regression results for the full sample using both OLS regression and fixed-effects regression methods. (19) I cannot estimate the coefficient on MGR OWNED in the fixed-effect regression because it is time-invariant.

Column A of Table 5 reports the pooled OLS regression result for the full sample. YR93_96 is significantly positive, suggesting that there was an upward nontax trend in employee compensation from 1989-1992 to 1993-96. This finding highlights the importance of using a control sample to separate the effect of tax changes from that of nontax changes.

The positive coefficient on MGR_OWNED is consistent with the prediction that shareholder/managers used deductible compensation to avoid the double taxation on investment income during 1989-92. Consistent with H1, the coefficient for MGR_OWNED x YR93_96 is significantly negative (p = 0.03, one-tailed), suggesting that shareholder/ managers of management-owned insurers used deductible compensation to shift less corporate income to their personal accounts when individual tax rates increased from 1989-1992 to 1993-96. The sum of the coefficients on MGR_OWNED and MGR_OWNED x YR93_96 is only 0.043 and not statistically significant (p = 0.65, two-tailed); thus there is no clear evidence that shareholder/managers used deductible compensation to shift corporate income to their personal tax bases during 1993-96.

The coefficients on the control variables from the OLS regression are all significant and consistent with expectations. Firms that are larger and that operate in more states incur more employee compensation expense. Insurers domiciled in states with higher costs of living incur more compensation expense. Total employee compensation is also significantly associated with firm performance.

The fixed-effects regression coefficients appear in Column B of Table 5. Consistent with the OLS regression result, the coefficient on MGR_OWNED x YR93_96 remains significantly negative (p = 0.001, one-tailed), although the absolute magnitude is reduced slightly. The coefficients on the control variables also remain significant except for the coefficient on LICENSE. The insignificant coefficient on LICENSE is not surprising because LICENSE does not vary much over time and thus loses significance in a fixed-effects regression. (20)

To assess the economic magnitude of the coefficient on MGR_OWNED x YR93_96, I also calculate the dollar value of the changes in deductible compensation that shareholder/ managers of management-owned insurers made in response to the 1993 individual tax rate increase. Using the fixed-effects regression coefficients from Column B of Table 5, untabulated calculation indicates that shareholder/managers (as a whole) of management-owned insurers reduced their deductible compensation (in 1989 dollars) by a mean (median) of $134,000 ($114,000) per year after 1992. The mean (median) reduction in deductible compensation represents about 4.10 percent (2.42 percent) of the annual insurance premiums written.

Persistence of Management-Owned Insurers' Responses

The pooled regression results reported so far do not indicate whether management-owned insurers' reduction in tax-deductible managerial compensation after 1992 was a temporary or a persistent response. Such distinction is important for understanding the effect of the tax change on future tax revenues. In addition, the distinction has implications for tax incidence (i.e., who bears the tax burden) and efficiency, the two most important issues in the economic analysis of taxation (see Slemrod 1995).

To address this question, I reran the fixed-effects regression after excluding years 1992 and 1993, the two years immediately before and after the 1993 tax change (see Column A of Table 6). As a comparison, I also report the fixed regression result in Column B of Table 6 for the years 1992 and 1993. The coefficient on MGR_OWNED x YR93_96 is significantly negative (p = 0.002, one-tailed) for the regression in Column A, suggesting that management-owned insurers' adjustment to managerial compensation represented a more persistent response to the 1993 tax rate increase.

As a robustness check, I also estimate the regression model by year using OLS (results not tabulated). This allows the regression coefficients on all independent variables to vary over time, although the coefficient on MGR_OWNED may also capture nontax effects. Consistent with the results in Table 6, the coefficient for MGR_OWNED is significantly positive in three of the first four years (p < 0.10, one-tailed), but insignificantly different from 0 in three of the last four years. More importantly, the coefficients on MGR_OWNED generally are larger in the first four years than in the last four years, consistent with the pooled regressions in Table 6.

Alternative Specifications

The regression results reported so far use COMPENSATION as the dependent variable. To check the robustness of the regression results, I used two alternative specifications to estimate the regression model. First, I estimated regression model (1) in the change specification (Panel A, Table 7). As I expect management-owned insurers' income-shifting incentive to decline from 1989-1992 to 1993-96 but not to change within each of the two periods, I examine changes in COMPENSATION only from 1989-1992 to 1993-96. The change of each variable in regression model (1) is defined as the difference in means from 1989-1992 to 1993-96.

In contrast to the COMPENSATION levels regression, the change specification directly estimates the changes in COMPENSATION over the two periods for each firm. As a result, the change model provides further assurance on the robustness of the levels regression. A limitation of the change model is that it is less powerful because the variables are averaged over several years. Consistent with the pooled regression results reported in Table 5, the coefficient on MGR_OWNED, which captures the changes in COMPENSATION for management-owned insurers relative to nonmanagement-owned insurers, is significantly negative (p < 0.03, one-tailed).

Second, I used the ratio of total deductible employee compensation to net premiums written (COMPENSATION_PREMIUMS) as the dependent variable (Panel B, Table 7). Because COMPENSATION_PREMIUMS is likely to be a function of firm size (Joskow 1973), the model controls for PREMIUMS. The inference from the fixed-effects regression for COMPENSATION_PREMIUMS is similar to those in Table 5. The coefficient on MGR_OWNED x YR93_96 is still significantly negative (p = 0.03, one-tailed).

VII. CONCLUSION

This study investigates the influence of taxes on managerial compensation in privately held firms whose managers own a high percentage of the firm's stock (I describe these as management-owned firms). Shareholders suffer double taxation on their investment income from taxable corporations. The firm first pays corporate taxes on corporate income; they pay additional individual taxes on the dividends or capital gains when shareholders receive dividends or sell shares. Shareholder/managers of management-owned firms can avoid the double taxation on their investment income (i.e., dividends and capital gains) by using deductible compensation to shift corporate income to their personal tax bases.

Using a sample of 64 management-owned, privately held insurers and 76 nonmanagement-owned, privately held insurers during 1989-1996, this study tests whether management-owned insurers structure their shareholder/managers' compensation to avoid double taxation. Because managers and shareholders are independent parties in nonmanagement-owned insurers, managers of nonmanagement-owned insurers can receive only reasonable compensation from the firm. Thus managerial compensation of nonmanagement-owned insurers provides a benchmark to identify the amount of deductible compensation that shareholder/managers of management-owned insurers use to avoid the double taxation on their investment income. Because individual marginal tax rates increased significantly from 1989-1992 to 1993-96 while corporate marginal tax rates remained essentially unchanged, I expect that shareholder/managers of management-owned insurers would use their compensation to shift less corporate income to their personal tax bases after 1992.

Because managerial compensation data are not available, I test the above prediction using total tax-deductible employee compensation. Consistent with my prediction, I find that management-owned insurers reduced the amount of deductible compensation by a mean (median) of $134,000 ($114,000) per year after 1992, representing 4.10 percent (2.42 percent) of the annual insurance premiums written. Furthermore, management-owned insurers' response to the 1993 tax change appears relatively persistent.

The results from this study are important for several reasons. First, management-owned firms are the base case in the agency theory of the firm (Jensen and Meckling 1976). In addition, Scholes and Wolfson (1992) use management-owned firms as a benchmark to illustrate how optimal tax planning deviates from tax minimization when there are incentive problems. In spite of the importance of management-owned firms in theory, the way these firms respond to taxes is little understood. This study addresses this gap in our knowledge of tax planning. The significant response of managerial compensation to tax changes in management-owned insurers suggests that incentive alignment does not dominate tax minimization in the design of managerial compensation in management-owned, privately held firms. The results also suggest that tax rule restrictions and transaction costs are not sufficient to prevent management-owned firms from structuring managerial compensation to minimize taxes.

Second, the estimated magnitude and persistence of management-owned insurers' responses to taxes also provide insight into likely revenue and efficiency effects of future tax changes. Because privately held firms are more responsive to taxes than publicly traded firms, tax policymakers may wish to consider the responses of privately held firms and publicly traded firms separately when analyzing possible effects of future tax policy changes.

An important limitation of this study is that it examined only one exogenous tax change in 1993; thus it is possible that the documented tax effect could be a result of differential nontax trends in compensation for management-owned and nonmanagement-owned insurers. As a result, an examination of the same issue using other time periods (e.g., the 1986 tax changes) seems warranted.

TABLE 1
Statutory Tax Rates for Individuals and Corporations

Panel A: Top Individual and Corporate Statutory
Tax Rates: 1981-1996 (a)

            Top Individual   Top Corporate
Year        Statutory Rate   Statutory Rate   Difference

1981-86          50                46             4
1987             38.5              40            -1.5
1988-1990        28                34            -6
1991-92          31                34            -3
1993-96          39.6              35             4.6

Panel B: Tax Rate Schedules for Married Individuals
Filing Joint Returns (b)

      Taxable Income

More than    But not over   1989   1994

         0         29,750    15    15
    29,750         36,900    28    15
    36,900         89,150    28    28
    89,150        140,000    28    31
   140,000        250,000    28    36
   250,000                   28    39.6

Panel C: Tax Rate Schedules for Corporations (b)

      Taxable Income

More than    But not over   1989   1994

         0         50,000    15    15
    50,000         75,000    25    25
    75,000     15,000,000    34    34
15,000,000                   34    35

(a)  Source: Gordon and Slemrod (2000, Table 8.1).

(b) Source: Internal Revenue Code 1986, as amended.
TABLE 2
Determining the Individual Marginal Tax Rate ([[tau].sub.i]) at
which Shareholder/Managers of Management-Owned Insurers are
Indifferent between Shifting Corporate Income to Their Personal
Tax Bases and Retaining Corporate Income within the Firm (a)

                                The value for [[tau].sub.i] =
Assumptions for [[tau].sub.c]      1 - (1 - [[tau].sub.c])
and [[tau].sub.a]                    (1 - [[tau].sub.a])

[[tau].sub.c] = 34%,
  [[tau].sub.a] = 0%                        34%
[[tau].sub.c] = 34%,
  [[tau].sub.a] = 10%                       40.6%
[[tau].sub.c] = 34%,
  [[tau].sub.a] = 20%                       47.2%
[[tau].sub.c] = 34%,
  [[tau].sub.a] = 28%                       52.5%

(a) [[tau].sub.c] is the corporate marginal tax rate; [[tau].sub.a]
is a hypothetical effective annualized tax rate for the total
individual taxes that a shareholder actually pays on her dividends
and capital gains in the future; [[tau].sub.i] is the individual
marginal tax rate.
TABLE 3
Sample Selection Criteria

U.S.-owned stock insurers in the 1992 NAIC
    property and liability insurer database with
    group code zero, nonmissing data during any
    of the years 1989-1996, and incorporation
    date before 1990                                408

Less:
  With no ownership information from A.M. Best's
    Insurance Reports: Property and Casualty
    (1989-1997)                                     138
  Ownership changes during 1989-1996                 17
  Distressed insurers (a)                            52
  Owned by a mutual insurer                           8
  Owned by a publicly traded corporation             21
  No observations in both 1989-1992 and 1993-1996    32

Final sample                                        140
  Management-owned privately held insurers           64
  Nonmanagement-owned privately held insurers        76

(a) Distressed insurers are firms that were placed in
receivership, in conservationship, or were being liquidated
(as disclosed in Best's Insurance Reports) during the period
1989-1996.
TABLE 4
Descriptive Statistics by Ownership Structure (means, medians,
and standard deviations)

                                           Nonman-
                             Management-   agement-   Rank Sum
                                Owned       Owned       Test
                              Insurers     Insurers    z-value
Variables                     (n = 64)     (n = 76)   (p-value)

ASSET (millions)                29.53       106.36     -5.39
                                (8.50)      (36.26)    (0.001)
                              [100.34]     [223.01]
PREMIUMS (millions)             11.81        28.84     -4.84
                                (4.52)      (12.89)    (0.001)
                               [39.89]      [43.75]
AGE                             14.57        23.40     -4.06
                                (8.13)      (15.75)    (0.001)
                               [16.87]      [20.66]
LICENSE                          6.17        13.40     -1.95
                                (2.27)       (3.80)    (0.05)
                                [9.68]      [16.93]
ROA                              0.10         0.09      1.27
                                (0.10)       (0.08)    (0.10)
                                [0.05]       [0.06]
COST_OF_LIVING (thousands)      26.20        26.33     -0.37
                               (25.66)      (25.46)    (0.71)
                                [3.50]       [3.20]
COMPENSATION (millions)          1.18         2.59     -3.21
                                (0.45)       (1.01)    (0.001)
                                [3.23]       [5.28]
TAXINC (millions)                2.03         5.35     -3.79
                                (0.88)       (2.09)    (0.001)
                                [4.66]      [10.73]

The variables reported in the table are means over 1989-1996 for
each insurer.

ASSET is total admitted assets at year-end. PREMIUMS is net insurance
premiums written. COMPENSATION is the total employee compensation,
including salaries, bonuses, and other immaterial emoluments. ROA is
return on assets, measured by total earnings before taxes and
tax-deductible compensation as a percentage of the average of
beginning and ending total admitted assets. AGE is the number of
years since incorporation. LICENSE is the number of states in which
an insurer is licensed to do business. COST_OF_LIVING is workers'
average annual pay in a state. TAXINC is estimated taxable income
before employee compensation. Except for TAXINC, all values are
expressed in constant 1989 dollars.
TABLE 5
Regressions of Log of Total Deductible Compensation
(p-values in parentheses)

                      Prediction       A            B

YR93_96                   ?         0.221        0.187
                                   (0.012) *    (0.000) **
MGR_OWNED                 +         0.241
                                   (0.017) *
MGR_OWNED x YR93_96       -        -0.197       -0.166
                                   (0.029) *    (0.001) **
PREMIUMS                  +         0.768        0.375
                                   (0.000) **   (0.000) **
LICENSE                   +         0.061        0.004
                                   (0.035) *    (0.461)
ROA                       +         1.162        0.596
                                   (0.019) *    (0.011) *
COST_OF_LIVING            +         0.748        1.437
                                   (0.013) *    (0.000) **
Constant                  ?        -4.647       -5.840
                                   (0.000) **   (0.000) **
n                                     927          927

*, ** Denote significance levels of 0.05 and 0.01, respectively.
The significance tests are one-tailed if there is a prediction and
two-tailed otherwise.

Column A reports the OLS regression coefficients, while column B
reports the firm fixed-effects regression coefficients.

YR93_96 is 1 for the 1993-96 period, and 0 otherwise. MGR_OWNED is 1
for management-owned insurers, and 0 otherwise. LICENSE is number of
states in which an insurer is licensed to write insurance. PREMIUMS
is net premiums written. COST_OF_LIVING is workers' average annual
pay in a state.

All the other variables are defined in Table 4.

Except for the dummy variables, all the variables are expressed in
natural logarithm.

Influential observations are deleted using Cook's (1977) distance
criteria.

The p-vaLues are corrected for heteroskedasticity and time serial
correlation using STATA's cluster command (see Rogers 1993).
TABLE 6
The Effect of the 1993 Tax Rate Change on Total Deductible
Compensation: Persistence of the Responses
(p-value in parentheses)

                      Prediction       A            B

YR93_96                   ?         0.234        0.124
                                   (0.004) **   (0.054)
MGR_OWNED x YR93_96       -        -0.172       -0.116
                                   (0.002) **   (0.063)
PREMIUMS                  +         0.375        0.328
                                   (0.000) **   (0.001) **
LICENSE                   +         0.046       -0.015
                                   (0.180)      (0.453)
ROA                       +         0.687        0.515
                                   (0.015) *    (0.261)
COST_OF_LIVING            +         1.109        1.916
                                   (0.005) **   (0.141)
Constant                  ?        -4.854       -7.288
                                   (0.000) **   (0.209)
n                                    678          249

*, ** Denote significance levels of 0.05 and 0.01, respectively.
The significance tests are one-tailed if there is a prediction
and two-tailed otherwise.

Column A (B) includes only firm years other than 1992 and 1993
(firm years 1992 and 1993).

See Tables 4 and 5 for variable definitions.

Regression coefficients from fixed-effects regressions.

Except for the dummy variables, all the variables are expressed in
natural logarithm. Influential observations are deleted using Cook's
(1977) distance criteria.

The p-values are corrected for heteroskedasticity and time serial
correlation using STATA's cluster command (see Rogers 1993).
TABLE 7
Sensitivity Analyses: Alternative Definitions of the Dependent
Variable

Panel A: The Dependent Variable is Changes in the Average of the
Log of Employee Compensation from 1989-1992 to 1993-1996 (a)

                                         OLS Regression
                        Prediction   Coefficient (p-value)

MGR_OWNED                   -               -0.202
                                            (0.024) *
[DELTA]PREMIUMS             +                0.499
                                            (0.000) **
[DELTA]LICENSE              +               -0.029
                                            (0.400)
[DELTA]COST_OF_LIVING       +               -1.634
                                            (0.268)
[DELTA]ROA                  +               -0.409
                                            (0.318)
Constant                    ?                0.581
                                            (0.047) *
n                                              140

Panel B: The Dependent Variable is Total Employee Compensation
over Net Premium Written (b)

                                     Fixed-Effects Regression
                        Prediction    Coefficient (p-value)

YR93_96                     ?                0.014
                                            (0.005) **
MGR_OWNED x YR93_96         -               -0.010
                                            (0.029) *
PREMIUMS                    -               -0.074
                                            (0.000) **
LICENSE                     +               -0.001
                                            (0.448)
ROA                         +                0.070
                                            (0.006) **
COST_OF_LIVING              +                0.134
                                            (0.000) **
Constant                    ?               -0.181
                                            (0.049) *
n                                              947

[DELTA]PREMIUMS is defined as changes in the average of the log of
PREMIUMS from 1989-1992 to 1993-1996.

[DELTA]LICENSE, [DELTA]COST_OF_LIVING, and [DELTA]ROA are defined in
the same way but for LICENSE, [DELTA]COST_OF_LIVING, AND [DELTA]ROA.
See Table 5 for the definitions of other variables. All values are in
1989 dollars.

* and ** denote significance levels of 0.05 and 0.01, respectively.
The significance tests are one-tailed if there is a prediction and
two-tailed otherwise.

(a) OLS regression coefficients. There are no influential observations
using Cook's (1977) distance criteria.

(b) Fixed-effects regression coefficients. Influential observations
are deleted using Cook's (1977) distance criteria.

I thank the members of my dissertation committee, Naveen Khanna, Ed Outslay, Peter Schmidt, and especially Kathy Petroni (chair), for their support and valuable comments. I am grateful to the Andersen Foundation for financial support. I appreciate the valuable comments the associate editor, two anonymous reviewers, Charles Enis, Steve Huddart, and Ira Weiss (the discussant at the 2000 American Taxation Association Midyear Meeting). The paper also benefited from the comments of workshop participants at the University of California, Irvine; Florida State University; University of Georgia; The University of Iowa; Michigan State University; University of Missouri-Columbia; Pennsylvania State University; University of Rochester; Stanford University; Texas Christian University; and the 2000 American Taxation Association Midyear Meeting.

Submitted May 2000 Accepted May 2001

(1) Hall and Liebman (2000) further show that even the short-term timing of stock option exercises documented in Goolsbee (2000) appears to be due to correlated omitted variables rather than taxes.

(2) As indirect evidence of the importance of management-owned, privately held firms in the economy, the Statistics of Income Bulletin (1998) reports that more than 99 percent of the 4.47 million corporate federal income tax returns filed in 1995 had total assets under $25 million, well below the mean total assets of my sample of management-owned insurers. Ang et al. (2000) provide direct evidence that high managerial ownership in privately held firms is common in the U.S.

(3) Increasing interest or rent payments to shareholders also mitigates double taxation. Because both debt financing and rental expenses are immaterial for property and liability insurers, this study does not consider them (see Cloyd et al. 1997; Ayers et al. 2001).

(4) Because this study focuses on shareholder/managers who are likely to earn more than $36,900, I ignore the individual tax rate difference between 1989 and 1994 for taxable income of less than $36,900.

(5) During 1989-1996, the lower tax rates of 15 percent and 25 percent for corporations were phased out by imposing a surtax of 5 percent for taxable income between $100,000 and $335,000. During 1993-96, the 1993 Budgetary Deficit Reconciliation Act phased out the 34 percent rate by imposing a surtax of 3 percent for taxable income between $15,000,000 and $18,333,333. For simplicity, the following discussion ignores the surtax.

(6) Section IV provides direct evidence that taxable income before taxable employee compensation exceeds $75,000 for most PL insurers.

(7) To check the reasonableness of this assumption, I obtained from Ke et al. (2000) the CEO compensation figures in 1994 and 1995 for ten of my management-owned insurers. For these ten insurers, the mean (median) CEO compensation was $133,000 ($129,000) in 1994 dollars. This suggests that the CEO's ex post marginal tax rate was just below the 36 percent bracket in 1994. Based on my hypothesis, had the individual statutory tax rates not increased in 1993, the CEO's compensation should have been higher due to income shifting.

(8) I assume that shareholders do not pay dividend tax and capital gains tax on the same income when they sell shares and realize capital gains (i.e., dividend tax and capital gains tax are substitutes). Collins and Kemsley (2000) suggest that shareholders who sell shares on a secondary market may pay both dividend tax (implicitly) and capital gains tax (directly) because dividend tax has already been capitalized in the stock price. Following their assumption will not affect my analyses.

(9) Although it is unknown exactly how shareholder/managers adjust the payoffs to other shareholders as a result of the income-shifting strategy using deductible compensation, Gordon and Slemrod (2000) discuss a few possibilities. One idea is to issue new shares to shareholders negatively affected by the income-shifting strategy. For example, consider the case in which a shareholder/manager retains a portion of her reasonable compensation within the firm due to high individual tax rates. Because the extra retained earnings accrue to all shareholders, not only to the shareholder/manager, the shareholder/manager could be compensated by receiving new shares. In principle, these shares should generate taxable income for the shareholder/manager based on their fair market value, and an equivalent tax deduction for the firm. In practice, however, the market value is difficult to ascertain because shares of privately held firms do not trade publicly, thus allowing considerable flexibility for undervaluation.

(10) Assuming that the total income (i.e., compensation and investment income) a shareholder/manager receives from the firm is fixed due to competitive market forces, an interesting implication of HI is that the shareholder/ manager's investment income should change inversely with their unreasonable compensation. I did not test this implication because shareholders' investment income (realized and unrealized) is more difficult to measure than compensation income in privately held firms. The only measure of shareholder return, accounting rate of return, is problematic because it does not include unrealized return and could be manipulated (e.g., for tax minimization) by both management-owned insurers and nonmanagement-owned insurers.

(11) Data source: National Association of Insurance Commissioners (NAIC), used by permission. The NAIC does not endorse any analysis or conclusions based on the use of these data.

(12) Distressed insurers are those that are in receivership, conservationship, or are being liquidated.

(13) The following regression results were unaltered if firms with only two years of observations over 1989-1996 were deleted.

(14) ASSET (total admitted assets) in Table 4 is total assets prepared under the statutory accounting principles (SAP) that all domestic insurers follow. A major difference between SAP and generally accepted accounting principles (GAAP) is that certain illiquid or risky assets (e.g., real estate investments not meeting insurance regulation) recognized under GAAP are not recognized under SAP to reflect the conservatism of insurance accounting. See Galloway and Galloway (1986, 23) for a detailed discussion.

(15) Because ROA is negative or 0 for some firm years, I added 1 to ROA before taking the logarithm.

(16) Two alternative size measures are number of employees and total admitted assets. Data on the number of employees are unavailable (but see below for an explanation of how the fixed-effects regression controls for the number of employees). The regression results using only total admitted assets are qualitatively the same, but because net premiums written explains substantially more variation in total compensation than total admitted assets, I use net premiums written in the reported regressions. Further, the regression results were qualitatively the same if I included both net premiums written and total admitted assets. Because the two size measures are highly correlated, I report the main results using net premiums written, alone.

(17) I hand-collected the data on COST_OF_LIVING from the Statistical Abstract of the United States (1989-1998 editions). I thank Andy Leone for suggesting the variable.

(18) Note that a positive relation between pay and performance does not invalidate my argument that managerial compensation is structured to reduce taxes in management-owned insurers. All I need for my research hypothesis is that agency conflicts do not dominate tax minimization in the design of managerial compensation in management-owned insurers.

(19) Influential observations are deleted using Cook's (1977) distance criteria for all regressions in the paper.

(20) As a robustness check, I also allowed the coefficients on all firm-specific control variables (i.e., PREMIUMS, LICENSE, and ROA) to vary with MGR_OWNED. This specification essentially allows differential nontax trends related to the control variables between the two types of insurers. The fixed-effects regression coefficient on MGR_OWNED * YR93_96 remained highly significant (coefficient = -0.15, p = 0.002, one-tailed).

REFERENCES

Ang, J. S., R. A. Cole, and J. W. Lin. 2000. Agency costs and ownership structure. Journal of Finance 55: 81-106.

Ayers, B. C., C. B. Cloyd, and J. R. Robinson. 2001. The influence of income taxes on the use of inside and outside debt by small businesses. National Tax Journal 54 (1): 27-56.

Best, A. M. 1989-1997. Insurance Reports: Property and Casualty. Oldwick, NJ: A.M. Best Co.

Cloyd, C. B., S. T. Limberg, and J. R. Robinson. 1997. The impact of federal taxes on the use of debt by closely held corporations. National Tax Journal 50: 261-277.

Collins, J. H., and D. Kemsley. 2000. Capital gains and dividend taxes in firm valuation: Evidence of triple taxation. The Accounting Review 75 (4): 405-428.

Cook, R. D. 1977. Detection of influential observations in linear regression. Technometrics 19: 15-18.

Feldstein, M. 1999. Tax avoidance and the deadweight loss of the income tax. The Review of Economics and Statistics 81 (4): 674-680.

Galloway, C. J., and J. M. Galloway. 1986. Handbook of Accounting for Insurance Companies. New York, NY: McGraw-Hill Book Company.

Goolsbee, A. 2000. What happens when you tax the rich? Evidence from executive compensation. Journal of Political Economy 108 (2): 352-378.

Gordon, R. H., and J. Slemrod. 2000. Are "real" responses to taxes simply income shifting between corporate and personal tax bases? In Does Atlas Shrug? The Economic Consequences of Taxing the Rich, edited by J. Slemrod, 240-288. New York, NY: Russel Sage Foundation.

Hall, B. J., and J. B. Liebman. 2000. The taxation of executive compensation. NBER Working Paper No. 7596. Cambridge, MA: National Bureau of Economic Research.

Huddart, S. 1998. Tax planning and the exercise of employee stock options. Contemporary Accounting Research 15 (2): 203-216.

Jensen, M. C., and W. H. Meckling. 1976. Theory of the firm: Managerial behavior, agency costs and ownership structure. Journal of Financial Economics 3: 305-360.

Joskow, P. 1973. Cartels, competition and regulation in the property-liability insurance industry. Bell Journal of Economics and Management Science 4: 375-427.

Ke, B., K. Petroni, and A. Safieddine. 2000. Ownership concentration and sensitivity of executive pay to accounting performance measures: Evidence from publicly and privately held insurance companies. Journal of Accounting and Economics 28: 185-209.

Mayers, D., and C. W. Smith, Jr. 1992. Executive compensation in the life insurance industry. Journal of Business 65: 51-74.

Rogers, W. H. 1993. sg17: Regression standard errors in clustered samples. Stata Technical Bulletin 13: 19-23.

Rose, N. L., and C. Wolfram. 2000. Regulating executive pay: Using the tax code to influence CEO compensation. NBER Working Paper No. 7842. Cambridge, MA: National Bureau of Economic Research.

Rosen, S. 1992. Contracts and the market for executives. In Contract Economics, edited by L. Werin, and H. Wijkander, 181-217. Oxford, U.K.: Basil Blackwell.

Scholes, M., and M. A. Wolfson. 1992. Taxes and Business Strategy: A Planning Approach. Englewood Cliffs, NJ: Prentice Hall.

Slemrod, J. 1995. Income creation or income shifting? Behavioral responses to the Tax Reform Act of 1986. American Economic Review (May): 175-180.

Statistical Abstract of the United States. 1989-1998. Lanham, MD: Bernan Press.

Statistics of Income Bulletin. 1998. Corporation income tax returns: a collection of text, figures and tables. Summer. Washington, D.C.: Government Printing Office.

Wilkie, T. J., J. C. Young, and S. E. Nutter. 1996. Corporate business activity before and after the Tax Reform Act of 1986. Statistics of Income Bulletin 16: 32-45.

Bin Ke
Pennsylvania State University