Researchers at the University of Tsukuba in Japan have developed a groundbreaking AI-powered cryptocurrency portfolio management system. Known as CryptoRLPM, short for “Cryptocurrency reinforcement learning portfolio manager,” this system utilizes on-chain data for training, making it the first of its kind.
Utilizing Reinforcement Learning
CryptoRLPM utilizes a training technique called “reinforcement learning” to incorporate on-chain data into its model. Reinforcement learning is an optimization paradigm where an AI system interacts with its environment, in this case, a cryptocurrency portfolio, and updates its training based on reward signals. By applying feedback from reinforcement learning throughout its architecture, CryptoRLPM ensures continuous improvement.
The Five Primary Units
CryptoRLPM is structured into five primary units that work together to process information and manage structured portfolios. These units include the Data Feed Unit, Data Refinement Unit, Portfolio Agent Unit, Live Trading Unit, and Agent Updating Unit. Each unit plays a crucial role in the system’s overall functionality.
Successful Testing and Performance
To test the effectiveness of CryptoRLPM, the researchers assigned it three portfolios. The first portfolio consisted of Bitcoin (BTC) and Storj (STORJ), the second included BTC, STORJ, and Bluzelle (BLZ), and the third contained all three plus Chainlink (LINK). The testing period spanned from October 2020 to September 2022 and consisted of three phases: training, validation, and backtesting.
The researchers evaluated CryptoRLPM’s performance against a baseline assessment of standard market performance. They used three metrics to measure success: “accumulated rate of return” (AAR), “daily rate of return” (DRR), and “Sortino ratio” (SR). AAR and DRR provide a quick overview of an asset’s gains or losses over a specific time period, while SR measures the asset’s risk-adjusted return.
According to the researchers’ pre-print research paper, CryptoRLPM demonstrates significant improvements compared to the baseline Bitcoin performance. The system shows a minimum 83.14% improvement in AAR, at least 0.5603% improvement in DRR, and a minimum 2.1767 improvement in SR.
In summary, the researchers’ development of CryptoRLPM represents a significant advancement in AI-powered cryptocurrency portfolio management. By utilizing reinforcement learning and on-chain data, the system outperforms the baseline Bitcoin performance in terms of accumulated rate of return, daily rate of return, and risk-adjusted return. These promising results highlight the potential of AI in revolutionizing the cryptocurrency investment landscape.