The Impact of State Taxes on Self-Insurance: Sensitivity tests

The Impact of State Taxes on Self-Insurance: Sensitivity testsOne control variable whose coefficient is significantly different from zero is the maximum statutory state corporate income tax rate. An attraction of purchasing coverage for business risks is that the insured party can deduct premiums against its income taxes. Business losses arising under self-insurance also are deductible. Losses under self-insurance, however, are deducted when incurred while premiums are deducted when paid. Moreover, because selfinsurance excludes insurer profits, it provides a smaller tax deduction than purchased coverage.
Thus, self-insurance of business risks arguably is greater in states with lower income tax rates. To test this proposition, we reestimated the regression, including the maximum statutory state corporate income tax rate as reported in Commerce Clearinghouse’s State Tax Handbook. The mean and median corporate income tax rate is 7 percent, ranging from zero to 12 percent with a standard deviation of 3 percent. It is negatively correlated with TAX, though not significantly different from zero.
Data limitations potentially introduce measurement error in this test in at least two ways. First, it is impossible to determine whether the losses (or the purchased coverage) are business-related and thus deductible. Second, the test presumes that the deduction offsets income in the state where the loss is incurred. In reality, determining the state in which a business can deduct expenses is governed by complex laws (e.g., apportionment factors and nexus) that are beyond the scope of this paper and not ascertainable using these data (see Klassen and Shackelford, 1998, for further details). Nevertheless, we find the coefficient on the maximum statutory corporate income tax rate is positive and significantly greater than zero in every year, consistent with self-insurance decreasing in the value of the state tax deduction. Inferences drawn from the TAX coefficients, however, remain qualitatively unaltered by the inclusion of an income tax rate. The TAX coefficients (t-statistics) are -0.21 (-1.91) in 1993, -0.23 (-2.18) in 1994, and -0.18 (-1.89) in 1995.
Finally, conclusions are unaffected if the dependent variable is premiums earned. Recall that insured losses is selected as the dependent variable because Petroni and Shackelford’s (1999) findings imply that using premiums as the dependent variable would bias toward rejecting the null hypothesis. Consistent with the expectation that results would be stronger using premiums, when premiums is the dependent variable, the t-statistics for the TAX coefficients are slightly more negative: -2.36 in 1993, -2.90 in 1994, and -1.82 in 1995.