Model Averaging for Volatility Forecasting, Option Pricing and Asset Allocation

Abstract

In this thesis the problem of model uncertainty is under scrutiny along with its implications in attaining optimal forecastability. To account for that averaging techniques are adopted including Bayesian model averaging, Bayesian Approximation and Thick Modelling. After an introductory chapter and a second one where some of the most celebrated conditional-volatility modelling proposals are discussed the third chapter investigates volatility forecasting and its direct association to option pricing. Some novel approaches to perform averaging are suggested here including variations of the predetermined methods together with more sophisticated algorithmic propositions such as Neural Networks. The fourth chapter extends the focal point of averaging to the whole predictive volatility density as this can be inferred first from derivatives on the underlying volatility index and second directly from the asset class under consideration (here the equity index) using bootstrap based - GARCH type models. The fifth chapter introduces some widely used variable selection techniques to the Finance continuum while averaging schemes once more are used in order to avoid model misspeci cation risk. Extensions to a nonlinear regression framework are also suggested while investment strategies are implemented in all chapters substantiating the ultimate supremacy of averaging schemes against single model alternatives. The last chapter concludes the research and makes some future suggestions for additional investigation.