Empirical comparison of chain-ladder and generalized linear modeling techniques in claim reserving
Rantanen, Jouni (2017-08-09)
Empirical comparison of chain-ladder and generalized linear modeling techniques in claim reserving
Rantanen, Jouni
(09.08.2017)
Tätä artikkelia/julkaisua ei ole tallennettu UTUPubiin. Julkaisun tiedoissa voi kuitenkin olla linkki toisaalle tallennettuun artikkeliin / julkaisuun.
Turun yliopisto
Kuvaus
Siirretty Doriasta
Tiivistelmä
This thesis studies the benefits of using generalized linear modelling methods in claim reserving instead of traditional actuarial methods. Assumptions required by the traditional reserving methods can cause difficulties in the usage in case changes occur in the claim compensation and the homogeneity assumption is not filled.
The claim payment data 2008-2013 of an insurance product is used for the modelling in this study. Claim years 2014-2016 is used as a validation data. The claim reserve estimates calculated with the Chain-ladder method are compared to the estimates produced with GLM method. For the GLM model selection residuals, AIC/BIC and deviance are used. Also the parameter estimate significance is checked. Validation data is used to compare the realized claim payment estimates to the estimates produced by the 2 models.
Equal accuracy of the reserve estimates can be achieved with both selected methods when equal amount of data information is used. In the GLM reserve estimation adding more information of claim behavior is easier than with in traditional actuarial methods. The GLM estimated reserves converges closer to the reality when more claim information is added to the model. Thus the accuracy of estimates improves.
Adding available claim information into GLM reserve modelling can be used to provide alternative reserve estimates and to improve the accuracy of the claim reserve estimates.
The claim payment data 2008-2013 of an insurance product is used for the modelling in this study. Claim years 2014-2016 is used as a validation data. The claim reserve estimates calculated with the Chain-ladder method are compared to the estimates produced with GLM method. For the GLM model selection residuals, AIC/BIC and deviance are used. Also the parameter estimate significance is checked. Validation data is used to compare the realized claim payment estimates to the estimates produced by the 2 models.
Equal accuracy of the reserve estimates can be achieved with both selected methods when equal amount of data information is used. In the GLM reserve estimation adding more information of claim behavior is easier than with in traditional actuarial methods. The GLM estimated reserves converges closer to the reality when more claim information is added to the model. Thus the accuracy of estimates improves.
Adding available claim information into GLM reserve modelling can be used to provide alternative reserve estimates and to improve the accuracy of the claim reserve estimates.