Portfolio Construction Based on Combination of Individual Stocks and LSTM

Abstract

With the development of the idea of portfolio optimization, it has become one of the important topics in the field of modern finance for achieving the balance between maximizing the return of assets and minimizing the risk. This study selected ten stocks from WIND and used LSTM Neural Network as the fundamental forecasting model. Based on the method of cyclic prediction, we predicted the asset’s return forecast value of the next day by rolling multi-dimensional historical features. Five stocks with the best outcomes were selected for optimization, using both the return component and the volatility, covariance matrix of the stocks’ returns as the covariance matrix to build the efficient frontier and to optimize the weighting. The Monte Carlo Simulation was used to build the maximum Sharpe ratio portfolio and minimum volatility portfolio. The maximum Sharpe ratio portfolio showed that the Midea and Mindray had the potential to profit dramatically at the risk of high volatility. However, relatively high volatility means the possibility of profiting is questionable, so a Minimum Volatility Portfolio was also conducted, which indicated conservative investors usually reduce the risk of their investment as much as possible. Whereas the max Sharpe ratio portfolio showing that risk-seeking investors are more willing to take risks if it can generate more profit. Overall, a hybrid intelligent algorithm based on cyclic prediction, Monte Carlo Simulation and artificial neural network was proposed, and the new applications of machine learning and mathematical methods in asset allocation were promoted.