Estudio de la siniestralidad en el seguro de automóvilesidoneidad de los modelos inflados de ceros y otras extensiones

Abstract

Knowing the factors that have influence on the accidents of the automobile insurance industry is an issue of a great relevance for insurance companies. The study of these factors could eventually allow insurers to fix more precisely the premiums their clients should pay for the policies and to reach higher levels of efficiency. Traditionally, research oriented to analyze the number of declared accidents by the policy holders has been based on the use of count data models, basically Poisson and negative binomial models. However, policy holders have the option of not declaring the accidents they suffer which leads to consider the zero-inflated models as a very appropriate option, from a theoretical point of view. The present work is focused on the application of this type of models to determine the main features of vehicles, insured clients and contracted policies that are relevant to explain why accidents happen. For this purpose we use a database provided by a Spanish private insurance company. The study is completed with the estimations of the number of clients that have accidents but decide not to declare them, and the number of these accidents. We conclude with a final theoretical discussion on the possible advantages given by other alternative models, as the so-called thinned models.