The role of heterogeneous parameters for the detection of selection in insurance contracts

This study re-examines standard econometric approaches for detecting adverse and advantageous selection in insurance contracts based on variables that are not used for calculating the insurance premium. We formally demonstrate that existing strategies for detecting selection based on such ‘unused characteristics’ can lead to incorrect conclusions if the estimated coefficients of interest are driven by different parts of the population. We show that this issue can empirically be accounted for by allowing for heterogeneous parameters. We compare existing approaches by using simulated data with different selection regimes and test for parameter heterogeneity within the data. We further provide empirical evidence about selection into the market for private health insurance in England. Both our simulations, and the findings using real data, suggest that parameter heterogeneity is a relevant issue that can confound the interpretation of standard ‘unused characteristics’ approaches. Our findings are important for analysing the efficiency of insurance markets. They are of interest to both the insurance industry and policymakers, and should be accounted for when selection based on specific characteristics needs to be detected or the effects of structural changes of insurance policies/markets are to be predicted.