Reinsurance to close - stochastic method

Document description

This method is based on a general underlying model of the claims process. Risk theory provides a basis for both the random and systematic components of the run off. The method is free from dependence on any particular form of error structure (for example Log Normal). The model is fitted by Fishers scoring method (this is the method used in the GLIM package) which has the advantage that it is fitted directly to the original data without transformation (for example taking logs). Thus negative incremental data do not present a problem. The variation in development pattern across years of origin is estimated from the data and modelled by use of the Kalman filter. The method is able to project the run off beyond the observed range of the data and, where available, may be fitted to data whose development period is less than annual (for example quarterly). The model is also able to model the non uniform build up of exposure during early development periods.