This project asks a narrow question: after accounting for information already visible in market prices, do onchain metrics improve forecasts of future Bitcoin and Cardano returns? The study uses daily data from January 2020 to June 2026. It compares price-based variables with network activity, valuation ratios, miner data, exchange flows, sentiment, and staking information (for Cardano).
The analysis was divided into two steps. The first tested whether the data contained a relationship with future returns. The second tested whether statistical and machine learning models could use any relationship that remained. This separation matters because a weak model and a weak signal can produce the same poor result. The analysis also examined volatility, long-horizon cycle timing, price momentum, and market regimes.
For the daily prediction tasks tested here, the answer was consistent. On-chain variables did not provide a reliable improvement over price alone for either asset. Some results looked positive before statistical corrections, but they disappeared after accounting for overlapping return windows, repeated testing, time trends, leakage, and simple baselines. On-chain data was still useful for describing market conditions, and a long-horizon Bitcoin valuation result was promising, but the available history contained too few independent market cycles to treat it as established evidence.
The main contribution of this project is not a trading rule. It is a clear example of how financial prediction claims should be evaluated, so that a good-looking result is not mistaken for real forecasting value.