Monte Carlo simulation techniques : The development of a general framework

Algorithmica Research AB develops software application for the financial markets. One of their products is Quantlab that is a tool for quantitative analyses. An effective method to value several financial instruments is Monte Carlo simulation. Since it is a common method Algorithmica is interesting in investigating if it is possible to create a Monte Carlo framework. A requirement from Algorithmica is that the framework is general and this is the main problem to solve. It is difficult to generate a generalized framework because financial derivatives have very different appearances. To simplify the framework the thesis will be delimitated to European style derivatives where the underlying asset is following a Geometric Brownian Motion. The definition of the problem and delimitation were defined gradually, in parallel with the review of literature, this to be able to decide what purpose, and delimitations that is reasonable to treat. Standard Monte Carlo requires a large number of trials and is therefore slow. To speed up the process there exist different variance reduction techniques and also Quasi Monte Carlo simulation, where deterministic numbers (low discrepancy sequences) is used instead of random. The thesis investigated the variance reduction techniques; control variate technique, antithetic variate technique, and the low discrepancy sequences; Sobol, Faure and Halton. Three test instruments were chosen to test the framework, an Asian option and a Barrier option where the purpose is to conclude which Monte Carle method that performs best, and also a structured product; Smart Start, that is more complex and the purpose is to test that the framework can handle it. To increase the understanding of the theory the Halton, Faure and Sobol sequence were implemented in Quantlab in parallel with the review of literature. The Halton and Faure sequences also seemed to perform worse than Sobol so they were not further analyzed. The developing of the framework was an iterative process. The chosen solution is to design a general framework by using five function pointers; the path generator, the payoff function, the stop criterion function and the volatility and interest rates. The user specifies these functions by him/her given some obligatory input and output values. It is not a problem-free solution to use function pointers and several conflicts and issues are defined, therefore it is not recommended to implement the framework as it is designed today. In parallel with the developing of the framework several experiments on the Asian and Barrier options were performed with varying result and it is not possible to draw a conclusion on which method that is best. Often Sobol seems to converge better and fluctuates less than standard Monte Carlo. The literature indicates that it is important that the user has an understanding of the instrument that should be valued, the stochastic process it follows and the advantages and disadvantages of different Monte Carlo methods. It is recommended to evaluate the different method with experiments, before deciding which method to use when valuing a new derivative