A Simulation Study comparing MCMC, QML and GMM Estimation of the Stochastic Volatility Model

Sammanfattning: The stochastic volatility (SV) model is an alternative to GARCH models to model time varying volatility. In this thesis the basic stochastic volatility model and three different estimation methods are described---namely, Bayesian Markov chain Monte Carlo (MCMC) methods, quasi maximum-likelihood (QML) and generalized method of moments (GMM). To compare these estimation methods a large scale simulation study is conducted where many different parameter values and sample sizes are compared. Since both the latter two methods are non-likelihood based, our hypothesis is that the likelihood based MCMC would perform better. The conclusion of the study is that this is the case, MCMC turns out to be more efficient than QML and GMM by quite a large margin, especially for estimating the latent volatilities.