A fully Bayesian approach to financial forecasting and portfolio selection

Abstract

The movements of share prices has long been of interest to both academic researchers as well as market practitioners. The statistical research in this field dates back to the work of Bachelier (1900) and there have been many approaches adopted subsequently. This thesis considers a Bayesian approach to multivariate forecasting of financial time series based on dynamic linear models. We will also consider the forecasting of the returns distribution using stochastic volatility models. We will then look at combining these two model structures. We will also demonstrate how the posterior forecast distribution can be simulated and how this may be used directly in order to implement a fully Bayesian decision theoretic approach to selection of optimal stock portfolios. These methods are first illustrated on simulated data and then applied to real data for selected shares from the Standard and Poor 500.