Analysis and Optimization of aPortfolio of Catastrophe Bonds

Abstract: This Master's Thesis in mathematical statistics has the two major purposes; (i) to model and measure the risk associated with a special type of reinsurance contract, the catastrophe bond, and (ii) to analyze and develop methods of portfolio optimization suitable for a portfolio of catastrophe bonds. Two pathways of modeling potential catastrophe bond losses are analyzed; one method directly modeling potential contract losses and one method modeling the underlying contract loss governing variables. The first method is simple in its structure but with the disadvantage of the inability to introduce a dependence structure between the losses of different contracts in a simple and flexible way. The second modeling method uses a stochastic number of stochastic events representation connected into a multivariate dependence structure using the theory of copulas. Results show that the choice of risk measure is of great importance when analyzing catastrophe bonds and their related risks. As an example, the measure Value at Risk often fails to capture the essence of catastrophe bond risk, which in turn means that portfolio optimization with respect to the same might lead to a systematic obscurity of risk. Two coherent risk measures were investigated, the spectral risk measure and the Expected Shortfall measure. Both measures provides good representation of the risk of a portfolio consisting of catastrophe bonds. This thesis extends and applies a well-known optimization method of Conditional Value at Risk to obtain a method of optimization of spectral risk measures. The optimization results show that expected shortfall optimization leads to portfolios being advantageous at the specific point at which it is optimized but that their characteristics may be disadvantageous at other parts of the loss distribution. Portfolios optimized for the spectral risk measure were shown to possess good characteristics across the entire loss distribution. Optimization results were compared to the popular mean-variance portfolio optimization approach. The comparison shows that the mean-variance approach handles the special distribution of catastrophe bond losses in an over-simplistic way, and that it has a severe lack of flexibility towards focusing on different aspects of risk. The spectral risk measure optimization procedure was demonstrated to be the most flexible and possibly the most appropriate way to optimize a portfolio of catastrophe bonds.