When Is the “Kennedy Correction” Appropriate in Estimating Overcharges?

In regressions using a semi-logarithmic functional form that include a dummy variable, Kennedy (1981) showed that instead of interpreting the dummy coefficient directly, one needs to “correct” it to estimate the percentage effect of the dummy variable on the dependent variable. In the context of an antitrust application, we show that when using a dummy variable to estimate the overcharge as a percentage of the actual price, one should not apply the correction proposed by Kennedy because doing so will lead to an overcharge estimate with a larger bias.