This paper is inspired by the work of Wüthrich, who introduced the use of machine learning techniques in non-life claims reserving. Machine Learning techniques, born to channel data complexity and able to deal with highly non-linear dependencies, are presently expected to provide a valid alternative to traditional reserving techniques. Moreover, in a framework where insurance undertakings are collecting an increasing amount of data on policyholders and claims, it seems natural investigating the potentialities of these algorithms. We focused on salary-backed loan insurance, a peculiar branch of credit insurance providing coverage for the outstanding debts in case of policyholder’s unemployment. A particular feature of these contracts is that they are characterised by a claim frequency very sensitive to credit cycle trends, with strong variability in time. The aim of this work is to investigate whether machine learning techniques can deal efficiently with such variability, exploiting information of macro-economic indices. We offer a comparison of three different statistical techniques: Generalized Linear Models (GLM), which represent the benchmark of the analysis, Artificial Neural Networks (ANN), whose popularity is spreading in actuarial sciences, and Support Vector Machines (SVM), that are known for their good generalisation capabilities but are new in this field.
Individual claims reserving, granular reserving, machine learning, GLM, Artificial Neural Networks, Support Vector Machines, credit insurance
Tipo di pubblicazione: