Estimation of Claim Severity in Non-Life Insurance: A Non-Parametric Approach

Abstract: In non-life insurance setting of the right premium for the customer at the beginning of the insurance contract is absolutely necessary for an insurance practice. For that the accurate and authentic estimate of the number of claim occurrences and claims size is extremely important. Different methods are available in the literature for predicting the claim size of a policy for forthcoming years such as Generalized linear models (GLMs), Poisson regression models, Credibility models, Bayesian Models etc. But due to some changes in exposure, classification of rating factors, migration in risk classes, the above mentioned classical methods will not provide a suitable model for prediction of future claim size. Hence a dynamic empirical model will address this problem. Recent studies shown that Artificial Neural Networks (ANN) is powerful tools for prediction by observing variation present in the data and predict future observations based on the characteristics of trained data sets. In this paper, we have shown that ANN will produce relatively better result compare to GLM. Keywords: Claim Severity, Generalized Linear Model, Artificial Neural Network