hero working papers

Recursive 'thick' modeling of excess returns and portfolio allocation

Number: 197
Year: 2001
Author(s): Carlo Ambrogio Favero (Università Bocconi, IGIER), Marco Aiolfi (Università Bocconi, IGIER), Giorgio Primiceri (Princeton University)
This paper explores the extent to which predictability of asset returns could be exploited for dynamic portfolio allocation among several (seven) assets taking model uncertainty explicitly into account.We consider model uncertainty when solving the problem of a representative fund manager who allocates funds between stock and bonds in three geographical areas: Europe, USA and Japan. We consider explicitly model uncertainty by implementing thick modelling to derive the average portfolio allocation generated by the recursively selected top fifty per cent of models in term of adjusted R-squared The portfolio allocation based on this strategy leads to systematic over-performance with respect to optimal portfolio allocation among several assets is based on the predictions of the best model as selected by the adjusted R-squared . Such over performance is mainly attributable to a reduction in the volatility of the returns on the selected portfolios. Thick modelling leads also to systematic replication, but not to over-performance, of a typical benchmark\ portfolio for our asset allocation problem.