Using On-Chain Data to Predict Cryptocurrency Cycles

This thesis investigates whether on-chain data—transparent, publicly available data recorded directly on blockchains—can be used to measure investor sentiment and forecast cryptocurrency market cycles. The research focuses on three core on-chain indicators: Net Unrealized Profit/Loss (NUPL), MVRV Z-score, and Cumulative Value Days Destroyed (CVDD). Through manual analysis of Bitcoin’s historical price data and a Monte Carlo simulation, the study evaluates the performance of each indicator as a contrarian signal across three major market cycles. The findings support two hypotheses: (1) that on-chain metrics capture investor sentiment, and (2) exhibit a statistically significant relationship with price movements. The results indicate that NUPL ratio and MVRV Z-score can generate long signals that significantly outperform a buy-and-hold benchmark, with cumulative returns exceeding 326,000% in the strongest cases and annualized returns up to 175,6 %. CVDD entry strategies, tested using simulations with random control entries within ±50 to ±100-day ranges around the cycle bottom, consistently showed superior returns relative to chance. The study contributes to the literature by demonstrating that blockchain-based behavioral data offers predictive value in decentralized markets lacking intrinsic valuation anchors. It also complements recent advances in deep learning models trained on on-chain data, highlighting rule-based metrics’ blend of interpretability and statistical strength. The results carry practical implications for investors, traders, and regulators, and challenge traditional notions of market efficiency by providing evidence of repeated behavioral patterns embedded in public blockchain activity.