La Siniestralidad en seguros de consumo anual de las entidades de previsión socialPerspectiva probabilística y econométrica. Propuesta de un modelo econométrico neuronal para Cataluña.

Zusammenfassung

The knowledge of how the insurance market behaves is a topic of great importance, according it future viability. The total losses associated with the company portfolio have a random component that should be kept in mind in it analysis. The principal objective of this work is to model the total claims amount of the mutual insurance sector in Catalonia (non life) according a probabilistic and econometric point of view. The structure of the study is clearly divided into two different parts. In the first part we present several methodological tools that can be applied to the analysis that we are carrying out; in the second one we present some results related with the application of the previous theory to a real insurance Catalonian database. In the methodological part, we highlight the definition of some ratios to summarize different financial analysis mechanisms; the effort to systematize one of the most famous methods of data analysis: the neural model, including its approach to the statistical field and econometrics. Concerning the empirical part, we emphasize the following aspects: the analysis of the basic characteristics of the Spanish insurance market (1991-1997) and the characteristics of the insurance mutual societies (by Autonomous Communities); the analysis of the non life total claims amount of the insurance mutual societies in Catalonia, and finally, the presentation of several applications of the neuronal methodology. The main empirical contributions are about the study of an economic sector not sufficiently studied before: the analysis of the total compensation starting from its random components, the frequency and severity of the claims; the definition of the minimum margins of solvency by using two methods: Method of MonteCarlo and the distribution of the ratio of the total non life claims amount; the specification of several statistical functions for this ratio; the formulation of some hypothesis contrasts, starting from the Generalized Functional form of Box-Cox and different applications of the neuronal methodology. We highlight the use of the neural model for the identification of the functional form of the ratio and the application of the "Multilayer feed-forward" model.