An agent-based model for lapse risk in life insurance

in paragraph 1.1 we first describe the various policyholder behaviors and options that can impact the liquidity and solvency of firm and then in paragraph 1.2 we focus on the surrender options and its determinants. In the second chapter we introduce agent-based models and their applications to life and non-life insurance. In particular, in paragraph 2.1 we highlight the importance of switching from aggregate level data to micro data relative to the single policyholder to model a complex system. In the third chapter we describe the results of six recent papers which used generalized linear regression to model lapse risk with data relative to the single policyholders. In particular we bring to the attention of the reader the logit regression by Milhaud et al. (2010) reviewed in paragraph 3.3 as we will use the coefficients provided in their paper to define the lapse probability of the last version of the model in paragraph 4.4. The model is described as a project diary to account for the modifications I applied in three steps: paragraph 4.2, 4.3 and 4.4. The first version of the model in paragraph 4.1 is not propaedeutic to understand the following three paragraphs. It is a different model whose purpose is to analyze cash outflows, accounting for lapse risk, allowing for the insurer to influence policyholders in their decision. This first version of the model has not been completed and suggestions on how to develop this model will be provided in the last paragraph. In paragraph 4.2 we consider lapse and liquidity risks for an insurance company having in its portfolio only pure endowments. In paragraph 4.3 we introduce other two types of contract and we facilitate the user definition of the relevant features of the contracts: duration, underwriting age and sum insured. In paragraph 4.4 we introduce a lapse probability of the contract particular for each policyholder dependent on his characteristics and the coefficients provided by Milhaud et al. (2010). In chapter 5 it follows a comparative analysis of the results of the three models. Further guidance on the results obtained can be found in the conclusions. In the last paragraph 'further developments' we suggest model extensions necessary to make it an operational tool for insurance companies to evaluate their liquidity and lapse risks.