Extent of Performance Gains from Incorporating On-Chain Data into RL-Based Crypto Portfolio Management

Determine the extent to which incorporating blockchain on-chain metrics into reinforcement learning-based cryptocurrency portfolio management systems enables these systems to outperform baseline approaches in terms of return performance.

Background

On-chain metrics provide transparent, real-time measurements of blockchain activity and are analogous to company fundamentals. Despite their potential relevance for valuation, comprehensive integration of on-chain metrics into reinforcement learning-based cryptocurrency portfolio management systems has been largely unexplored.

The paper introduces CryptoRLPM, an RL-based system that incorporates on-chain data for end-to-end cryptocurrency portfolio management and reports backtesting results against baselines. The motivating question explicitly stated in the introduction concerns the magnitude of performance improvement achievable by using on-chain data within RL-based systems, which the authors note remained unanswered prior to this work.