SIGNATURE WORK
CONFERENCE & EXHIBITION 2022

Nowadays, cryptocurrencies have emerged and created enormous impacts on investment markets. In addition to investing in traditional securities such as stocks, people are increasingly likely to add cryptocurrencies to their portfolios to pursue higher investment returns and balance investment risks. Even though researchers have investigated portfolio management problems using the Deep Reinforcement Learning (DRL) framework over a pure stock or pure cryptocurrency portfolio, DRL applications over a hybrid stock-cryptocurrency portfolio remain unexplored. This study intends to utilize the Proximal Policy Optimization (PPO) model under the DRL framework to automatically optimize asset allocation in a portfolio containing 30 energy-sector stocks and 16 cryptocurrencies. The asset allocation in this portfolio is updated on a daily basis throughout the investment process. To evaluate the portfolio’s performance, we simulate the investment process compared with three different securities or portfolios as the baselines: S&P 500 Energy Sector, Bitcoin, and the constant equal-weighted portfolio of 46 stocks or cryptocurrencies mentioned above. Through backtesting, we conclude that the PPO model can help investors gain stable and relatively high investment rewards. It created a 186.4% increase in the investment capital of the hybrid stock-cryptocurrency portfolio during the backtest. Beyond that, with the PPO model, this hybrid portfolio shows high stability of 0.74 and low volatility of 0.38. It also has a high Sharpe Ratio of 2.94, suggesting its robustness in accumulating investment returns at relatively low risks.