Gated Neural Network-Based Mean-EVaR Skewness Portfolio Optimization under Uncertain Environment

摘要

Numerous empirical studies show that portfolio returns are generally asymmetric, and investor would prefer a portfolio return with larger degree of asymmetry along with risk and return. In this paper, a concept of skewness is defined as the third central moment and studied its mathematical properties. To predict the stock prices, a novel recurrent neural network with gated recurrent unit (GRU) cell is preferred. Based on these predictions, stock returns, entropic value at risks and skewness are calculated. A mean-EVaR-skewness multi-objective portfolio optimization model is devised to account for market uncertainty. Cardinality, bounding restrictions, and liquidity are considered in addition to risk and return to make the model more effective. Uncertain goal programming is used to solve the proposed model. Finally, an example portfolio is presented to display the efficacy and the feasibility of the model suggested in this paper.