Bayesian approach to portfolio selection

Details

Supervisors

Vrins, Frédéric

Faculty

Louvain School of Management

Degree label

Master [120] en ingénieur de gestion

Abstract

Portfolio selection and optimization is one of the most important problems in modern finance. Markowitz's mean-variance framework, introduced in the 50s, has been the theoretical standard framework for academic and practitioners. However, research has shown that it is particularly sensitive to estimation risk in the parameters, on top of other known defaults (concentration, non-intuitiveness of wealth allocation, sensitivity...). The Bayesian approach addresses this issue by specifying prior beliefs on the behavior of asset returns and on portfolio weights. This thesis starts by reviewing two classes of Bayesian techniques: (i) with prior on assets' returns' behavior and (ii) with prior on portfolio weights. We implement a set of different Bayesian techniques using simulated data with random return jumps and show that implementing Bayesian techniques systematically leads to lower sensitivity and turnover, as expected, and may lead to superior performance in terms of out-of-sample variance and sharpe ratio.