Portfolio Management Strategy Based on LSTM

Abstract

Mean-variance theory emphasizes the consideration of returns and risks simultaneously. However, traditional optimization of the risk-return trade-off may not always align with the preferences of individual investors. Given that outcomes about daily LSTM predictions have been inconclusive, this study uses weekly price dataset for prediction aiming to improve the signal-to-noise ratio. The objective is to develop a portfolio management strategy that prioritizes achieving reasonable returns while effectively managing risk. To achieve this, a selection process identifies 12 stocks from different sectors. LSTM models are trained using historical stock price data, and these models predict future stock prices. The predictions guide the weekly rebalancing of the portfolio based on a designed principle and the performance of the strategy-based portfolio is evaluated by comparing it to two benchmarks: the NASDAQ index and a 1/N investment portfolio. The results demonstrate that the strategy portfolio outperforms both benchmarks, achieving significant improvements in returns. The strategy portfolio exhibits a remarkable return of 116.21%, a Sharpe ratio of 2.944 and a maximum drawdown of 13.1%. Although there is a slight increase in volatility, the strategy effectively manages risk while delivering superior returns.