Measuring moral hazard using insurance panel data

This paper describes two new methods for testing and measuring moral hazard that are applicable to a variety of insurance settings where panel data is available. The methods are able to measure moral hazard separately from adverse selection. The first method uses the variation in risk over different objects with different levels of insurance owned by the same person to isolate the average effect that insurance has on precautionary effort. For the second method, I show that many insurance contracts imply different incentives for precautionary effort for both the beginning and end of the contract period. These differences can be used to test moral hazard. I demonstrate the methods using a detailed insurance data set of virtual space ship insurance from a large virtual world called EVE Online. Because of moral hazard, the daily hazard rates for insured ships are sixfold compared to uninsured ships. Moral hazard is especially high at the beginning of each contract with newly insured ships having thirteen times higher daily hazard rates compared to ships with only 15 days left in their contract.