Insurance market risk modeling with hierarchical Fuzzy Rule Based Systems

The continued development of large, sophisticated, repositories of knowledge and information has facilitated the accessibility to vast amounts of data about complex objects and their behavior. However, in spite of the recent renewed interest in knowledge-discovery techniques (or data mining), the usefulness of these databases is partially limited by the inability to understand the system-related characteristics of the data. Some applications from the financial or insurance market-such the ones concerned with risk analysis-require to meet solutions that emphasize precision while aiding to understand and validate their structure and relations. We present results about an ongoing project being carried out by the Argentinian State Insurance Agency for tracking the status of the insurance companies, i.e., for screening and analyzing their condition through time. Specifically in this paper, we will tackle with the modeling of the mathematical reserves of the premiums, or risk reserves, of the insurance companies in the local insurance market. To do so, we propose the use of Linguistic Modeling which is one of the most important applications of Fuzzy Rule-Based Systems. Particularly, we apply Hierarchical Linguistic Modeling with the aim of obtaining the desired trade-off between accuracy and interpretability of the system modeled, i.e., decomposing such nonlinear systems into a number of simpler linguistically interpretable subproblems. The achieved results will be also compared with global hierarchical methods and other system modeling techniques, such as classical regressions and neural networks.