Pricing Of Catastrophe Bonds Based On Benchmark Model

With global warming, the frequency and loss amount of catastrophes is rising. Catastrophic risk can be transferred to strong capital markets through the issuance of catastrophe bonds (CAT bonds) to respond to such a change. The studies on catastrophe bonds are mostly based on actuarial model, no arbitrage or equilibrium model. No arbitrage or equilibrium pricing methods are required to meet the existence of risk-neutral measure; however, it may not exist. Therefore, this thesis model with Benchmark Theory to solve the problem that risk-neutral measure does not exist.Firstly, the thesis sorts out the literature about the study of catastrophe bonds pricing models and Benchmark theory and introduces its transaction structure, trigger mechanism, advantages and disadvantages; then, the benchmark pricing model of CAT bonds on the hypothesis that the cumulative numbers of catastrophe are subject to complex non-homogeneous Poisson process is derived; Next, assuming loss function is lognormal distributed, PCS index is used to fit losses distribution and intensity function; Finally, in the single and multi-machine parallel computing environment, this thesis using Monte Carlo simulation method to solve the pricing model and make sensitivity analysis for the results. Numerical results show that the catastrophe bond pricing model constructed in this thesis has a good applicability and feasibility, and the corresponding results can provide a theoretical basis and technical support for issuance and rational decision for the investors.There are four directions to be explored:Firstly, the Benchmark model can be extended to the pricing of other complex derivatives, such as power derivatives, long-term bonds. Secondly, using the extreme value theory to explain the loss function. Thirdly, using numerical discretization method with better convergence, such as Milstein approximation method or Taylor methods. Some variance reduction methods, such as importance sampling or control variables, can be used to reduce the statistical error,. Fourthly, using GPU parallel calculation method instead of multi-CPU parallel calculation method.