Mean-variance and mean-ETL optimizations in portfolio selection: an update

Abstract

In this research update, we apply the Mean-Variance (MV) and Mean-Expected Tail Loss (ETL) portfolio optimization techniques on earnings forecasting and robust regression-based composite models. A time series model with multivariate normal tempered stable (MNTS) innovations is applied to generate the out-of-sample scenarios for the portfolio optimization. We report that (1) a composite variable of analysts’ forecasts, revisions, and direction of analysts’ revisions continues to produce value in portfolio construction; (2) robust regression-based models continue to produce meaningful active returns; and (3) the Mean-Variance and Mean-ETL portfolio optimizations produce statistically significant active returns, passing the Markowitz and Xu (Journal of Portfolio Management 21:1–60, 1994) data mining corrections test.