– Researchers from the University of Tsukuba in Japan have developed CryptoRLPM, the first reinforcement learning-based AI system for cryptocurrency portfolio management that utilizes on-chain data.
– CryptoRLPM applies reinforcement learning, where the AI system interacts with its environment and updates its training based on reward signals.
– The system is composed of five primary units: Data Feed Unit, Data Refinement Unit, Portfolio Agent Unit, Live Trading Unit, and Agent Updating Unit.
– The researchers tested CryptoRLPM with three portfolios over a two-year period and measured its success against standard market performance using metrics such as accumulated rate of return, daily rate of return, and Sortino ratio.
– According to the researchers’ pre-print research paper, CryptoRLPM demonstrated significant improvements over baseline performance, including an 83.14% improvement in accumulated rate of return and a 2.1767 improvement in Sortino ratio compared to Bitcoin.