Traditional Regression Methods in Loss Reserving

A basic feature of the loss reserve estimation process is the analysis of historical loss development data in order to project future loss payments and case reserves for incomplete accident/policy/report periods. A recently developed approach to forecasting reserves is to use regression techniques to estimate loss development factors, accident or calendar period trends and future loss payments. In this session, the panelists will present applications of traditional regression techniques to the loss reserving process. In session 5G, newer techniques involving time varying parameters will be presented. Chain ladder (i.e., development factor) method of development can be viewed as subsets of regression methods. This session will show how to evaluate the suitability of chain ladder for given data sets, utilizing session diagnostics and graphical procedures. The panelists will then discuss how to parameterize various models of loss development, including the smoothed chain ladder and hoerl curves. As actuaries must frequently work with incomplete data, this session will illustrate how to use regression methods to derive tail factors.