Essays on empirical contract theory : evidence from car insurance data

In this thesis the central aim is to identify and appreciate a fuller understanding of possible market imperfections in the insurance market: I have focused on contributing to the existing empirical tests to determine the presence of asymmetric information, such as moral hazard and adverse selection. Further to this the intention is to explore and improve plausible policy implications in order to make adjustments to market welfare. In particular, the empirical analysis is enriched using 'car insurance panel data sets' obtained from two major Korean car insurance companies. In the first chapter I review the detailed descriptions of the Korean car insurance market, and the data sets from the Korean car insurance companies are presented. Then, using the Korean data sets, I implement three pioneering empirical models for application within the field of empirical insurance economics. Namely: the conditional correlation approach (probit model/bivariate probit model) the occurrence dependence approach (duration model) and the Granger causality approach (dynamic bivariate probit model). In the second chapter I have sought to detect the presence of moral hazard via the introduction of a regulatory change that occurred in the Korean car insurance market in the year 2000. Then using logit and nonparametric estimation I have investigated whether there was any change in accident rates between, before, and then after the introduction of a stronger incentive system. In the last chapter - with an aid of an improved, more substantive, data set - I have investigated the presence of asymmetric information that is from a different direction from that previously employed within the literature. Firstly, with regards to the 'moral hazard problem', I have worked on the relationship between the purchase of coverage for damage to the policyholder car and the stated car value. Due to the presence of a 'missing data problem', I have implemented a bounds approach in estimating car value distribution. Secondly, regarding the identification of adverse selection I have introduced a simple conditional variance test to see whether there is a difference in risk level across policyholders.