Research on Investment Strategy Based on LSTM Neural Network Prediction

A transaction decision model will be built with quantitative investment as the core. The moving average (MA) and the moving average convergence and divergence (MACD) is taken as the opening level signal. Based on Markowitz’s classical portfolio theory, the optimal portfolio with the weight on the effective frontier will be constructed, and the weight structure of each risk asset in the portfolio will be dynamically adjusted. A neural network model based on Long Short-Term Memory (LSTM) will be established to predict the prices of gold and bitcoin, so as to modify the trading strategy.The results show that on the 40th day, trend tracking indicators (MA and MACD) and weight calculation methods can provide scientific guidance for transaction decision-making; on the 350th day, the LSTM model can predict price changes in a good manner. Finally, after 46 trading activities, the investment value will become $105,131.5864 five years later.