Social Sciences Brown Bag Seminar

Abstract

Quantitative asset allocation is severely hindered by sampling noise and nonlinear optimization, which are magnified by the curse of dimensionality in investment universes with many assets.  Sampling error's influence on portfolio weights is typically ignored when evaluating investors' interim subjective expected utility, but may have an arbitrarily bad influence on their ex-ante objective expected utility.  To address this issue, I propose a new algorithm that optimizes over small, randomly selected, subsets of securities rather than the full investment universe.  By construction, these "subset portfolios" are ex-ante exchangeable and weighted equally in an ex-ante efficient portfolio, so diversifying across a large number of subset portfolios bounds the influence of sampling error on portfolio performance.  Balancing the utility lost to subset selection with that of sampling error, the implemented algorithm's performance dominates existing asset allocation strategies in a broad array of simulation and backtest experiments.