Scientists create a crypto portfolio management AI trained with on-chain data

Scientists create a crypto portfolio management AI trained with on-chain data


A pair of researchers from the University of Tsukuba in Japan recently built an artificial intelligence-powered cryptocurrency portfolio management system that utilizes on-chain data for training, the first of its kind, according to the scientists. Called CryptoRLPM, short for “Cryptocurrency reinforcement learning portfolio manager,” the artificial intelligence (AI) system utilizes a training technique called “reinforcement learning” to implement on-chain data into its model. Reinforcement learning (RL) is an optimization paradigm wherein an AI system interacts with its environment — in this case, a cryptocurrency portfolio — and updates its training based on reward signals.CryptoRLPM applies feedback from RL throughout its architecture. The system is structured into five primary units that work together to process information and manage structured portfolios. These modules include a data feed unit, data refinement unit, portfolio agent unit, live trading unit and agent updating unit. Screenshot of pre-print research. Source: Huang, Tanaka, “A Scalable Reinforcement Learning-based System Using On-Chain Data for Cryptocurrency Portfolio Management”Once developed, the scientists tested CryptoRLPM by assigning it three portfolios. The first contained only Bitcoin (BTC) and Storj (STORJ), the second kept BTC and STORJ while adding Bluzelle (BLZ), and the third kept all three alongside Chainlink (LINK). The experiments were conducted over a period lasting from October 2020 to September 2022 with three distinct phases (training, validation and backtesting).The researchers measured the success of CryptoRLPM against a baseline evaluation of standard market performance through three metrics: accumulated rate of return (AAR), daily rate of return (DRR) and Sortino ratio (SR). AAR and DRR are at-a-glance measures of how much an asset has lost or gained in a given time period, and the SR measures an asset’s risk-adjusted return.Screenshot of pre-print research. Source: Huang, Tanaka, “A Scalable Reinforcement Learning-based System Using On-Chain Data for Cryptocurrency Portfolio Management”According to the scientists’ pre-print research paper, CryptoRLPM demonstrates significant improvements over baseline performance:“Specifically, CryptoRLPM shows at least a 83.14% improvement in ARR, at least a 0.5603% improvement in DRR, and at least a 2.1767 improvement in SR, compared to the baseline Bitcoin.”Related: DeFi meets AI: Can this synergy be the new focus of tech acquisitions?