MAXIMUM-SHARPE-RATIO ESTIMATED AND SPARSE REGRESSION (MAXSER): PORTFOLIO OPTIMIZATION FOR LARGE STOCK UNIVERSES WITH FACTOR INVESTING

Abstract

The traditional mean-variance portfolio conceptualised by Markowitz (1952) is a constrained optimization problem that does not perform well with large asset universes due to the curse of dimensionality. Ao et al. (2019) proposed a novel unconstrained regression representation of the mean-variance portfolio problem and devised a model, maximum-Sharpe-ratio estimated and sparse regression (MAXSER), that utilizes penalized regression to obtain the optimal portfolio. MAXSER also has a specification to allow for factor investing. This thesis (1) evaluates the performance of the MAXSER model using the Least Absolute Shrinkage and Selection Operator (LASSO); and (2) extends the factor set in the MAXSER with factor investing model and the data set to December 2022. We conclude that the MAXSER model does not consistently outperform the market index and the naïve 1/N-rule in a 20-year investment horizon, and the MAXSER with factor investing model is highly sensitive to the moments of factors.