Umeå universitets logga

This essay is written in consultation with SPP (a Swedish insurance company), where we will investigate if it is possible to improve the settlement function for disability insurance by taking more variables into account. The settlement function describes the probability that an individual remains subject to compensation t months after the start of the illness and is an important component in the pricing of the disability insurance as well as in the allocation to sickness reserves. To be able to measure if the settlement function improves, we compare the settlement function with the estimated settlement obtained from Kaplan-Meier curves from historical illness cases. We use cases from SPP between 2009-2024 and cases from the Society of Actuaries (USA) between 2009-2017. Often, the age of the individual is used as a variable in the settlement function. Using SPP’s data, we test whether the settlement function improves when the gender of the individual is also taken into account. Nine different variables have been tested on the US data, including region, diagnosis, gender, industry and salary. In total, four models are tested on SPP’s data and six models on the US data. The result of the study is that the Random survival forest model, which is a Machine learning model, is the model that is by far best at reflecting the Kaplan-Meier curves. In addition, we developed our own model, which also reflects the Kaplan-Meier curves well. We also found that both gender and diagnosis improve the prediction of settlement. Region is a variable that is significant for the settlement of cases in the United States. In this study we have not dug deeper into that variable. However, it may be interesting in the future to investigate if region also affects the settlement of cases in Sweden.