Difference-in-Difference models to estimate causal effects on auto insurers behavior

Abstract

The Difference-in-Difference (DiD) method is useful to test if an event has effects in a given outcome using non-experimental data. Based on DiD method, we propose alternative panel models to estimate the causal effects of the traffic accidents on driving behavior patterns: the total annual driving distance in km, the percent of km circulated above the speed limits, in urban areas and at night. We use a data set provided by an ”insurtech” company that uses car sensors to measure driving data over a period of three years. The estimation results show as the causal effects of accidents are different if we consider frequency of accidents, type of damages and whose fault is the accident. Furthermore, different profiles of policyholders in function of drivers and cars characteristics are associated with specific causal effects.