Price Prediction and Investment Decision Model Based on LSTM

Abstract

In today’s market trading, quantitative trading has benefited investors through frequent trans- actions. Our team will operate at $1,000, forecasting by processing data of previous years and calculating the final earnings. For Problem 1, we based the prediction model of the LSTM neural network, trained the previous bitcoin and gold transaction price of the network algorithm, and predicted the trading market price of the next day. By comparison with the actual value, the model’s accuracy in predicting gold reached 92.3%, and the accuracy of bitcoin reached 70.5%, which are good results. Next, our team used a dynamic programming model, with the bear-bull market index reasonable divergence rate as the decision variable, to establish a system of equations for analysis. At the same time, we conducted a linear regression analysis of risk and return and used the characterization of the residual between the predicted value and the actual value to calculate the weight as the target function to describe the investment strategy under different people’s investment personalities. For Problem 2, we performed sensitivity tests for previous models, and we randomly selected ten sets of perturbation data for prediction within a range of 5% near the decision parameters. Finally, we find that if the investment scheme is perturbated, the maximum return of the target function will decrease, so the given scheme can be considered an optimal scheme. This paper uses LSTM neural network model for price prediction with extended memory function and good predictive. At the same time, through the suitable formulation of the bear market and bull market model and investment risk model, the rationality of decision-making is increased, and the scheme is optimal.