Do We Drive More Safely When Accidents are More Expensive? Identifying Moral Hazard from Experience Rating Schemes

A major shortcoming of the growing empirical work on asymmetric information is the inability to separately identify moral hazard from adverse selection. Abbring et. al. (2003) point out that dynamic insurance data can help here, by asking whether consumers have fewer claims when they are at a place in the ?experience rating? scheme where additional claims are more expensive. However, in the French setting they study, this test boils down to asking whether there is negative state dependence in claims occurrence, and thus requires them to assume away all other forms of state dependence in claims. This paper overcomes this problem by considering U.S data, where claims fall off consumer records after three years, creating an ?insurance event? that changes a consumer?s position in the experience rating scheme with no simultaneous claim, and thus allowing identification of moral hazard even with fairly general controls for state dependence and unobserved heterogeneity. In addition, the U.S. data follow consumers for 10 years (vs. 1 in Abbring et. al) and contain a much wider range of claim price effects, both of which increase identification power. The paper?s core finding is a small, but statistically significant, moral hazard effect. This is made more convincing by the fact that the effect grows following a 1997 pricing change that increased the cost of additional claims relative to the first one. Finally, without the controls for state dependence in claims, this effect disappears. This suggests that the lack of evidence for moral hazard in previous work may have resulted from confounding the negative state dependence associated with moral hazard with some underlying source of positive state dependence in claims occurrence.