Short-term Portfolio Trading Strategy Based on Deep Learning and Simulated Annealing Algorithm Taking Bitcoin and Gold for Example

Abstract:
As the foundation of modern finance, portfolio theory was proposed by Harry Markowitz in 1952, with mean-variance analysis to determine the efficient frontier for a variety of tradable assets. However, Markowitz mainly relies on past statistical data to describe the impending changes of assets and to determine the optimal allocation of assets, making it difficult to capture price changes in time. Additionally, there are many limitations in Markowitz's theory, which makes it difficult for the theoretical optimal solution obtained by the model to be directly used in real-life trading scenarios. Based on this, this paper selects gold and bitcoin as two typical tradable assets, and uses LSTM to predict the prices of them. Then, according to the portfolio theory, two models that can be utilized for short-term trading are proposed. Finally, the simulated annealing algorithm is utilized to obtain the solutions of the two models respectively, and the optimal trading strategy is formulated.