Quantitative trading decision model based on LSTM neural network

Abstract

As trading markets are vulnerable to national policies and economies, some investors choose to leave their assets in the hands of market traders for management. To help market traders find the best portfolio, the best trading strategies were dynamized to obtained by predicting the daily price movements of gold and bitcoin. BP neural network, time series analysis, and LSTM neural network were chosen for modeling quantitative trading decisions and modify the dynamic model of optimal trading. The LSTM neural network was used to indicate the time series because of the best fit of 0.991. And the investment weights for gold and bitcoin were calculated based on the particle swarm algorithm to be 0.09:0.91 and over the five-year trading period an initial $1,000 investment would yield $66,588.987. In addition, the optimal daily trades dynamically were re-solved by changing the values of the investment weights by 2:3, 2:1, 3:2, and 1:2 and obtained that the total number of assets of gold and bitcoin is smaller than the total number of assets with an allocation ratio of 0.09:0.91 for either allocation ratio. Thus, the model in the present paper provides the optimal strategy.