Portfolio analysis with a large universe of assets
Covariance matrix optimization algorithms are applied to a large number of assets. A previous paper by Burgess and Bey (1988) suggests that attempting to optimize a large number of securities with the traditional covariance matrix model is not practical. An alternative approach ranks the securities with the reward to volatility (reward to beta) ratio and then optimizes a smaller subset of securities with the covariance matrix model. This study proposes additional screening methods such as stochastic dominance, reward to variability (R/V) ratios, reward to lower partial moment (R/LPM) ratios, and the optimization of subgroups, and provides an empirical test of the various screening methodologies. The results indicate that the full covariance critical line optimization algorithm is surprisingly robust compared to the other techniques.