Testing Portfolio Construction Methodologies Out-of-Sample

Bernd Scherer

10.1 INTRODUCTION

Various portfolio optimisation and data refinement techniques have been proposed in the academic literature to address estimation error and its impact on portfolio construction.¹¹An excellent review can be found in Connor et al (2010). While the academic literature remains firmly rooted in Bayesian decision theory, Michaud (1998) deviated from this tradition and suggested a portfolio construction heuristic²²Note that Bayesian analysis does not allow any additional adjustments (“hand waving”) after parameter uncertainty has been addressed using priors. Resampling the predictive distribution is inconsistent with Bayesian theory. In other words, we cannot bring together Bayesian methods and RE. using resampling techniques (called resampled efficiency or RE),³³US patent #6003018. based on the earlier work by Jorion (1992).⁴⁴Jorion (1992) used resampling to visualise the effect of estimation error on the distribution of portfolio weights. Simply averaging across this weights for a given risk aversion will lead to RE. Even though the in-sample properties of RE have already been subject to substantial critique,⁵⁵For a comprehensive review of resampling see Scherer (2002