Smoothing of chain-ladder estimators

Specialeforsvar ved Veli Kaya

Titel:  Smoothing of chain-ladder estimators 

Abstract: For small companies the estimated chain-ladder factors can sometimes be rather erratic. In most cases there is no underlying reason for this, it is only caused by pure randomness. To increase smoothness, various methods can be used. In this thesis we will analyze a few methods that smooth the chain-ladder factors. We will start by introducing the general problem of reserving, which will lead us to the non-parametric approach to the chain-ladder model. We will also examine the parametric approach. Next, we will study the main problem with a penalty method and a Bayesian method, which will give the desired smoothness of the chain-ladder factors. The penalty method is a non-parametric approach, where the smoothing allows the model to follow the behavior in the data. In the Bayesian method, it is possible to involve a priori information, such that the priori factors have an effect on the smoothness. Both methods will be used to predict the future claims, and will be compared to each other

Vejleder:  Jostein Paulsen
Censor:   Mette M Havning