Bitcoin price forecasting models struggle to beat naive benchmarks

A recent preprint by Carlos Baquero of the University of Porto suggests that complex Bitcoin price forecasting models often fail to outperform simple, naive benchmarks. Despite a wide variety of methodologies—including scarcity models based on halving schedules, on-chain activity analysis, power-law charts, and advanced machine-learning systems—academic research has struggled to demonstrate durable superiority over naive forecasts at one-to-six-month horizons.

Baquero’s review examined 23 influential papers that utilized genuine out-of-sample evaluation. The findings indicate that while short-horizon order flow and daily return forecasts may provide predictive value, they are often conflated with long-term valuation models in public discourse. The study emphasizes that historical path formulas and daily direction models serve different purposes and that current forecasting claims require more rigorous evaluation.