Insights from fine-grained data for actuarial applications in mortality modelling and non-life reserving

Insurance and reinsurance companies provide financial protection to individuals and businesses against a wide range of risks, from car theft to health related expenses and life events such as death. These individuals and businesses transfer their risks to (re)insurers through so-called insurance contracts. Mathematical and statistical methods are key in this process to evaluate these risks, set appropriate premiums, and manage reserves. As such, the field of actuarial science naturally arises. Originating from early work by Johan de Witt and Jakob Bernoulli, actuarial science has quickly evolved through advances in data science and machine learning. Furthermore, because of the increasing availability of fine-grained data, insurance and reinsurance companies can enhance their pricing models, estimate reserves at a more fine-grained scale, and gain deeper insights into mortality statistics. This thesis consists of four chapters that explore the application of fine-grained data to actuarial practices, particularly in reserving for non-life insurance and mortality modelling. In the first chapter, we introduce a hierarchical reserving model that leverages both historical claim data and claim-specific covariates to improve non-life reserving practices. The resulting model is highly flexible in handling different reserving scenarios and incorporates many existing reserving models. We showcase our model with a case study on a real home insurance dataset and on portfolios generated along several claim development scenarios. In the second chapter, we explore the impact of a pandemic data point on the calibration of and projection with a stochastic multi-population mortality model. In particular, we use a weekly mortality database to investigate how the COVID-19 pandemic impacts future mortality rate and life expectancy projections. In the third chapter, we extend this research and focus on mortality shocks more generally. More specifically, we develop a multi-population mortality model that accounts for mortality shocks through switching volatility regimes. In the last chapter, we investigate the effect of environmental factors on mortality rates using fine-grained, publicly available data. To this end, we employ machine learning to analyse how weather and air pollution variables are associated with deviations from regional mortality baseline levels.