3.2. Dependent variable
The dependent variable is designed to measure insurance coverage. As discussed above, premiums are not an option for the dependent variable because insurers manage the premiums they report to each state to reduce overall premium tax liabilities (Petroni and Shackelford, 1999). In other words, because insurers underreport the premiums earned in high-tax states, premiums would induce a negative relation between the dependent variable and taxes, biasing toward rejecting the null hypothesis. To avoid this bias, the study employs total insured losses, which are expected to suffer less manipulation than premiums for two reasons. First, insurers have little incentive to allocate insured losses across states for tax purposes. Unlike premiums, losses are irrelevant for purposes of computing the premium tax base. Although losses are included in income tax bases, only seven states tax insurers’ income. If insurers do overstate losses in high income tax states, the tests are biased against rejecting the null hypothesis.
Second, although losses presumably affect regulators’ assessments of an insurer’s solvency and profitability, insurers are less likely to manage losses for regulatory purposes than to manage premiums. Regulators can observe the state in which a loss occurs more easily than the state in which a premium is earned. For example, hurricane damage in Florida may be impossible to report as a claim arising in another state. However, the policy covering property in Florida also may cover property in other states. As a result, premiums provide more sourcing ambiguity, which facilitates insurers’ opportunistic reporting. Losses provide less flexibility from the regulatory management perspective. If, however, insurers do manage losses across states to reduce regulatory pressure, the manipulation should not bias this study’s estimates unless taxes and regulatory restrictions are correlated, a relation that Petroni and Shackelford (1995) fail to detect.
Finally, insurers occasionally process claims, mail refunds, and provide other bookkeeping services for large corporations that self-insure. During the period examined in this study, some insurers reported losses arising through such arrangements as “insured” losses, even though the company retained all risks. We are unable to assess the extent to which our measure of coverage includes these self-insured losses. However, such measurement error should bias against rejecting the null hypothesis that taxes do not affect self-insurance.
3.2. Dependent variable