The Mack model and the Bootstrap Over-Dispersed Poisson model have long been the primary modeling tools used by actuaries and insurers to forecast losses. With the emergence of faster computational technology, new and novel methods to calculate and simulate data are more applicable than ever before. This paper explores the use of various Bayesian Monte Carlo Markov Chain models recommended by Glenn Meyers and compares the results to the simulated data from the Mack model and the Bootstrap Over-Dispersed Poisson model. Although the Mack model and the Bootstrap Over-Dispersed Poisson model are accurate to a certain degree, newer models could be developed that may yield better results. However, a general concern is that no singular model is able to reflect underlying information that only an individual who has intimate knowledge of the data would know. Thus, the purpose of this paper is not to distinguish one model that works for all applicable data, but to propose various models that have pros and cons and suggest ways that they can be improved upon.

Regulation in the insurance market has increased greatly over the past four decades, and recent regulatory frameworks such as Solvency II have made simulations increasingly important. Monte Carlo simulations are often too inefficient to be used by themselves, and these Monte Carlo simulations begin to struggle when the complexity of insurance contracts increases. For that reason, there have been numerous suggested improvements to traditional MC methods such as the sample recycling method and a neural network method. This thesis will review various risk measures, the methods used to calculate them, and a detailed analysis of the neural network method and the sample recycling method. The sample recycling method and the neural network method will then be analyzed in detail, and a comparative analysis of the sample recycling method and the neural network method will be given. It was discovered that both the sample recycling method and the neural network method provide a large improvement in computational cost and overall run time with minor impacts on the accuracy. Thus, it was concluded that the sample recycling method is best suited for contracts where the inner loop estimations are particularly complex and the neural network is a general method that pairs well with complex input portfolios.