AN ANALYSIS OF PROPERTY-CASUALTY INSURERS: Robustness Tests

Columns D and E report the results from two tests of the robustness of the results to different specifications. Conclusions are qualitatively unaltered by the results in these two regressions. Column D uses a separate intercept for each insurer instead of insurer-specific controls (DIRWRITE, SIZE, CAPITAL, PORTRISK, and lines of business). The tax measures remain negative and highly significant. The coefficient on ETR is -4.07 with a t-statistic of -3.07. The coefficient on INCDUM is -0.063 with a t-statistic of -3.73.

Column E presents summary statistics from an alternative estimation procedure undertaken to assess whether the t-statistics in the earlier regressions are overstated. Because our sample includes multiple observations for every insurer, the OLS residuals likely are contemporaneously correlated, potentially biasing the standard errors (Bernard 1987). To assess the extent of the bias, we estimate standard errors by applying the Froot (1989) technique. The Froot (1989) standard errors account for intra-insurer correlations in the residuals of each insurer and heteroskedasticity. They are consistent and asymptotically efficient. same day payday loans online

To limit computational demands, we apply the Froot technique to the regression reported in Column A, excluding CAPITAL, PORTRISK, and the eight lines of business variables.18 The OLS coefficient estimates from this modified model are essentially identical to those reported in Column A. The Froot adjusted standard errors are generally higher than the OLS standard errors, but the significance of the t-statistics is qualitatively unchanged. In short, any interdependencies caused by multiple observations for insurers neither overstate the statistical significance of our results nor affect the inferences drawn from OLS regressions.