SIGNATURE WORK
CONFERENCE & EXHIBITION 2022

Many pieces of research have shown the machine learning algorithms including deep learning and reinforcement learning on stock portfolio management. However, there is only limited information about deep reinforcement learning (DRL) methods in cryptocurrencies portfolio, and it’s crucial to apply DRL methods to cryptocurrencies since it’s a market that is rapidly expanding with potential possibilities. This paper proposed to apply a stable DRL method, Advantage Actor-Critic (A2C), with the tickers’ historical prices as its input, to address the portfolio management problem based on a portfolio containing cryptocurrencies and energy stocks. Backtest of the A2C model in the one-year period has a great performance with the 2.5-fold return. Three related baselines are also implemented to perform the same backtest, in order to compare with the results of A2C, and A2C achieves a much higher Sharpe ratio than the other three baselines.