Development and optimization of reserving models in actuarial science : a python-based approach

Abstract

Estimating claim reserves is a relevant but challenging topic in insurance. Several laws and guidelines define the need for technical provisions where an accurate calculation of the claims reserves is needed. However, making accurate estimations is challenging, especially for claims that include personal damage and potential long-term effects. In this thesis, we address the optimisation of the Chain Ladder method, analysing the impact of various factors, including outlier exclusion, using simple max /min exclusion, Reverse Nearest Neighbour and Interquartile Distance as outlier detection methods. Other adjusted parameters are the periods considered, tail adjustment, inflation adjustment, and using a weighted average of both the paid and incurred triangle. The algorithm was implemented using Python. In this analysis, one key finding occurred, which is that optimal model parameters differ significantly between incurred and paid losses and between short-tail, long-tail, and volatile insurance branches. More straightforward methods, such as excluding the maximum value or limiting the number of periods, often perform better than more complex approaches. Although according to our analysis, the tail and inflation analysis have no positive impact on the accuracy of the estimations, they should be further examined in future studies. This thesis shows the potential for automated optimisation to improve claims reserving accuracy while reducing actuaries' manual workload. However, it highlights that expert judgment remains essential, primarily when sudden changes or external trends that historical data alone cannot capture, occur.