Kernel alternatives to approximate operational severity distribution: an empirical application

One of the main topics in operational risk is the estimation of loss severity distribution. Numerous parametric estimations have been suggested, although very few work for both high frequency small losses and low-frequency big losses. In this paper the most used parametric models, kernel alternatives, are explored to approximate operational severity distribution. The good performance of the double transformation kernel estimation in the context of operational risk severity is worthy of special mention. This method is based on the work of Bolancé and Guillén (2009). It was initially proposed in the context of the cost of claims insurance, and it means an advance in operational risk research.