Quantitative trading strategy and portfolio optimization analysis based on LSTM model

Abstract

This paper focuses on the optimal risk portfolio problem. Based on the principle of return maximization, an LSTM neural network model is used to predict the price movements of gold and bitcoin; then an objective planning method is used to find the optimal trading strategy for each day. Finally, the final amount of the initial asset is calculated for the end date. To accommodate the huge price changes of Bitcoin, an LSTM neural network model was built for prediction in this paper, which produced well-fitted Bitcoin price data. After obtaining the predicted prices of gold and bitcoin, an objective programming model is built with daily returns as the objective function, constraints are set, and market days (gold can be traded) and non-market days (gold cannot be traded) are discussed to derive the optimal investment strategy; a linear programming solution method is used to derive the optimal solution under the constraints to ensure the maximum amount of daily returns. The LSTM neural network forecasting model used in this paper has a long memory function and good predictability. Meanwhile, the heuristic algorithm greatly accelerates the convergence of the model for the optimal solution.