The researchers claim that CryptoRLPM is the first AI system based on reinforcement learning that manages a portfolio using on-chain measurements.
A cryptocurrency portfolio management system dubbed CryptoRLPM powered by artificial intelligence was recently developed by two academics from the University of Tsukuba in Japan. This ground-breaking system, which is regarded as the first of its type in the realm of bitcoin portfolio management, uses on-chain data for training.
To incorporate on-chain data into its model, CryptoRLPM uses reinforcement learning (RL), a training method. The optimization paradigm of reinforcement learning involves an AI system interacting with its environment, in this case, a portfolio of cryptocurrencies, and updating its training based on reward signals.
The system is divided into five main sections to make it easier to operate: the data feed unit, the data refinement unit, the portfolio agent unit, the live trading unit, and the agent updating unit. To evaluate information, manage structured portfolios, and make wise investment decisions, these components collaborate.
The researchers gave the system three distinct portfolios to test the effectiveness of CryptoRLPM. The first portfolio includes Storj (STORJ) and Bitcoin (BTC), the second Storj and Bluzelle (BLZ), and the third Storj and Chainlink (LINK).
The trials carried out by the researchers included three distinct phases: training, validation, and backtesting. They ran from October 2020 to September 2022. These phases made it possible to assess and improve CryptoRLPM’s performance.
Three crucial metrics—accumulated rate of return (AAR), daily rate of return (DRR), and Sortino ratio (SR)—were used by the researchers to compare CryptoRLPM’s performance to a benchmark assessment of typical market performance. These indicators shed light on the profitability and risk-adjusted returns of the CryptoRLPM-managed cryptocurrency portfolios.
The usefulness of the AI-powered system in optimizing cryptocurrency portfolio management and developing improved investment strategies is highlighted in the researchers’ pre-print research paper, which reports that CryptoRLPM achieved considerable gains over the baseline performance.
The creation of CryptoRLPM serves as a demonstration of the expanding potential of artificial intelligence in the world of cryptocurrencies and investment management. Systems like CryptoRLPM have the potential to improve investment outcomes by utilizing reinforcement learning and on-chain data in the volatile and often changing cryptocurrency market.