A Modified Multi-objective Differential Evolution Algorithm with Application in Reinsurance Analytics

摘要

In the reinsurance marketplace, the risk of financial loss in the event of natural catastrophes (such as earthquakes, hurricanes and floods) is exchanged between market participants for a premium. Here, prudent risk management takes the form of a hedge against the risk of a contingent uncertain loss in exchange for a payment. Reinsurance contracts that define the terms of the transfer are elaborated multi-layered financial treaties that represent complex trade-offs between expected return and risk. Formulating an effective risk transfer strategy depends on a careful multi-objective optimization process. In this paper, we study from the perspective of an insurance company the Reinsurance Contract Optimization problem in which, given the structure of a multi-layered reinsurance contract, we are required to discover specific contractual terms that capture the best trade-offs between expected return and risk for the insurer. Our approach is based on an adaptation of Multi-Objective Differential Evolution. In searching for the best mutation operators, we performed an experimental analysis on large-scale real problem instances using industrial datasets and evaluated five different mutation operators. Our experimental results indicate that those mutation operators based on selecting non-dominated individuals from the archive tend to produce better outcomes. Since speed is critical in this application, we also developed a parallel version achieving a speedup up to 9.3 on a 16 core machine.