Linear Models for Portfolio Selection with Real Features

Abstract

An efficient investment portfolio would have maximum return or minimum risk. Several approaches based on the “expected returns - variance of returns” rule seek for a good balance between yield and risk. These approaches may differ in either how to measure risk or how to estimate expected yields. In this work we consider linear programming models found in the literature to estimate risks, like mean absolute deviation and Gini’s mean difference. Thus, two mixed integer programming models are investigated in a portfolio optimization problem for a given expected return. For such, we add real features, including transaction lots, cardinality, and investment threshold. Experiments using data from the Dow Jones stock market demonstrate the superiority of the investigated models in the presence of these real features when compared with a market average indicator of return.