Quantitative Trading Portfolio Optimization-Based Stock Prediction Using Long-Short Term Memory Network

Author: Ruizhi Hao
Degree: M.S. in Business Intelligence and Analytics
Year: 2021
Advisory Committee: Dr. Dragos Bozdog, Dr. Ionut Florescu

Abstract: Stock prediction is critical in quantitative trading for creating an efficient trading strategy that yields a high return. The ability to predict outcomes is also needed for successful portfolio construction and optimization. Stock prediction, on the other hand, is a difficult task due to the numerous factors involved, such as uncertainty and instability. Deep learning techniques, especially the recurrent neural network (RNN), have recently been developed for sequence prediction. A long short-term memory (LSTM) network is proposed in this paper to predict market movement using historical data. Multiple portfolio optimization techniques, such as equal-weighted modeling (EQ) and optimization modeling maximizing Sharpe ratio, are used to optimize portfolio efficiency in order to build an effective portfolio. The results showed that our proposed LSTM prediction model is effective in predicting stock prices with high accuracy. In addition, using maximizing Sharpe ratio method to rebalance the allocation strategy every three months showed a significant improvement in the cumulative return of the constructed portfolios. Furthermore, our constructed portfolios beat the benchmark Sector ETF index in both XLU and XLB.