Optimal selection of simulated years of catastrophe activity for improved efficiency in insurance risk management

Catastrophe risk models used in the insurance industry consist of complex Monte Carlo simulations of events such as earthquakes or hurricanes. Due to the large uncertainty in the characteristics and severity of these events, the number of samples needs to be large enough to capture the spectrum of possible consequences. This creates operational challenges in terms of computational time. The tendency in the industry has been to reduce the computational burden by selecting a subset of samples of years of simulated activity. The subsampling process usually takes place after the larger sample has been built and, therefore, it is not possible to apply traditional sampling variation reduction techniques within the Monte Carlo process. In this paper, an algorithm based on evolutionary computation is presented to construct an optimal subset of samples that minimizes the statistical variation between the larger and smaller sets for a specific portfolio of risks. Numerical testing has shown a tenfold improvement in computational time with a 90% reduction of sampling error with respect to a random subsampling process.