Scientists from the University of Tsukuba in Japan have successfully created an AI-powered cryptocurrency portfolio management system that leverages on-chain data for training. Termed CryptoRLPM (Cryptocurrency reinforcement learning portfolio manager), this cutting-edge artificial intelligence system marks a significant milestone in the field.
Utilizing Reinforcement Learning with On-Chain Metrics
CryptoRLPM, by incorporating a training technique called reinforcement learning, has achieved a major milestone. This is being the first AI system to actively utilize on-chain metrics for portfolio management. With this innovative approach, it has revolutionized the field of portfolio management.
Reinforcement learning involves the interaction of an AI system with its environment, in this case, a cryptocurrency portfolio, enabling it to adapt its training based on reward signals.
The Architecture and Experimental Testing of CryptoRLPM
Notably, CryptoRLPM operates by employing reinforcement learning feedback throughout its architecture. The system is comprised of five primary units that work in tandem to process information and effectively manage structured portfolios. These units include the data feed unit, data refinement unit, portfolio agent unit, live trading unit, and agent updating unit.
Furthermore, to assess the performance of CryptoRLPM, the researchers conducted experiments involving three portfolios. The initial portfolio consisted of Bitcoin and Storj (STORJ), while the second portfolio incorporated Bitcoin, Storj, and Bluzelle (BLZ). The final portfolio included all three cryptocurrencies alongside Chainlink (LINK).
These experiments were carried out from October 2020 to September 2022, encompassing three distinct phases: training, validation, and backtesting.
Measuring Success with Key Metrics and Impressive Results
To gauge the effectiveness of CryptoRLPM, the researchers employed three metrics to compare its performance against a baseline evaluation of standard market performance.
These metrics included the accumulated rate of return (AAR), daily rate of return (DRR), and Sortino ratio (SR). AAR and DRR provide concise measurements of the gains or losses incurred by an asset within a specific timeframe, while SR assesses an asset’s risk-adjusted return.
The pre-print research paper by the scientists revealed that CryptoRLPM displayed significant improvements over baseline performance. Specifically, compared to the baseline performance of Bitcoin, CryptoRLPM demonstrated a remarkable 83.14% improvement in AAR, a notable 0.5603% improvement in DRR, and a substantial 2.1767 improvement in SR.
Read More:
CFTC Investigators’ Findings: Celsius and Former CEO Found in Violation of Rules