Identifying and Testing for Moral Hazard in French Automobile Insurance Market

    Project: Research project

    Project Details


    Asymmetric information is well exemplified in every insurance market in the form of moral hazard and adverse selection. The former is manifested when agents buying more insurance become more risky since more coverage discourages cautious behavior. In contrast, adverse selection occurs when agents deliberately hide their unfavorable characteristics from insurance companies. Theoretical literature has already shown the existence of these two phenomena. However, empirical literature often shows the opposite.

    To avoid agents’ dishonest reports, French government introduced experience rating system in early 90s - premium increases if an accident claim is filed and decreases if there is none. Moral hazard is present when he becomes more cautious for avoiding further premium increase after an accident had occurred. Thus, it imposes a negative dependence structure on the claim intensity. In contrast, adverse selection, captured by agents’ hidden characteristics, say, bad driving behavior, generates a positive dependence structure. If it is not carefully controlled, these two phenomena will be confounded together.

    This observation, which is also implied by standard utility maximization model, suggests that claim intensity can be modeled in a dynamic binary choice framework. If moral hazard exists, the coefficient of the lagged dependent variable, claim intensity, will be negative. Traditionally, dynamic binary choice model is estimated by either fixed or random effect model. The former treats the unobserved individual characteristics as fixed. However, when number of individuals is large, estimating the heterogeneity involves incidental parameter problem. If it is treated as random, consistency of parameters hinges on the correctness of its conditional distribution specification and the treatment of the initial value of the dependent variable - initial condition problem.

    This project is the first one to apply estimation and inference procedure for parameter sets to test for moral hazard in the presence of heterogeneity. This methodology avoids both incidental parameter and initial condition problem. The sacrifice of this relaxation is that we can only partially identify the parameters. Since our goal is to test whether there is moral hazard in the data by adequately controlling heterogeneity, that is, the sign of the coefficient of the lagged dependent variable, absence of point identification does not impose any problem in this analysis. To test for its significance, a subsampling test proposed by Chernochukov, Hong and Tamer (2007) will be adopted.
    Effective start/end date1/09/1131/08/13


    Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.