Portfolio Optimization with Return Prediction on Vietnam Stock Market

Abstract

Portfolio optimization has been widely studied over the past decades and has found numerous applications in finance and economics. In this paper, we investigate the portfolio optimization problem in the Vietnamese stock market using deep learning methods based on two datasets: (1) a dataset on technical analysis and (2) a data set on technical analysis supplemented with data extracted from quarterly financial reports disclosed by companies. These data sets were collected from the Vietnam Stock Exchange from early 2011 to the end of 2023. This paper aims to develop an efficient algorithm to identify a portfolio with the highest Sharpe ratio in the coming weeks. We selected 100 stocks with the highest average weekly trading value in the market to construct the data set for training instead of including all stocks in the Vietnam market. In addition, we compared various deep learning models, such as Residual Networks (ResNet), Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and the Transformer model (4 layers). Experimental results show that the Transformer model outperforms other methods regarding the Sharpe ratio, delivering promising outcomes for portfolio optimization problems in Vietnam and other markets.