A May 2026 preprint reviewing Bitcoin forecasting research found that no model has shown durable superiority over simple naive benchmarks at one- to six-month horizons across market cycles. The review, by Carlos Baquero of the University of Porto, examined 23 papers selected from hundreds and points to overfitting, non-stationarity, and information leakage as recurring weaknesses. The preprint has not yet completed peer review.
Carlos Baquero of the University of Porto reviewed the academic record on Bitcoin forecasting and found that no model demonstrated durable superiority over the appropriate naive benchmark at horizons of one to six months across several market regimes. The literature contains hundreds of papers, but Baquero selected 23 for close examination based on their methods, influence, or use of genuine out-of-sample evaluation. The review itself is still awaiting peer review.
Naive forecasts use only current information: a price forecast can use today's price, a return forecast can use zero, and a direction forecast can use a random walk. Scarcity models built on the halving schedule, on-chain models, power-law charts, and machine-learning systems all compete against that shallow opponent, yet much of the literature has struggled to beat it once a model leaves the period in which it was designed.
Why the simplest forecast wins
Naive forecasting works because financial prices are persistent: a model predicting $100,100 tomorrow when Bitcoin trades at $100,000 today can produce a tiny percentage error even when it has learned almost nothing about direction or return. The benchmark grows more demanding as the horizon expands, because Bitcoin can move sharply over a month while the relationships a model learns decay as the market evolves. In a separate study applying 12 statistical, machine-learning, and deep-learning approaches to five major cryptocurrencies at one-day, seven-day, and 30-day horizons, Francesco Puoti, Fabrizio Pittorino, and Manuel Roveri found that simple naive models consistently outperformed ARIMA, Prophet, random forests, XGBoost, LSTM networks, and N-BEATS.
Backtests can flatter a model
David Bailey and his co-authors formalized the risk of backtesting overfitting, showing that trying more model variations raises the odds of finding a strong historical result by chance, while presenting only the winning result hides how many attempts failed. Among the peer-reviewed papers Baquero examined, none evaluated the same approach across several non-overlapping holdout windows covering different market regimes; the strongest used rolling or walk-forward evaluation over a single continuous out-of-sample period. Information leakage adds another risk, since a feature calculated with future data, or variables normalized across the full sample, can give a model a hidden view of the answer the market never actually supplied.
Popular valuation models fare no better
Stock-to-flow and Metcalfe-style variables helped explain returns in-sample but offered limited or zero predictive ability out of sample, according to Alexander Shelton's 2024 peer-reviewed study. Once time effects entered the stock-to-flow regression, its statistical force disappeared, since Bitcoin's supply ratio rises on a predetermined schedule while price also climbed for much of its history, making the two series look connected. The stock-to-flow model has traded below its projected path for years.
Savva Shanaev and his co-authors found a similar pattern for network-activity models: once they addressed autocorrelation and the two-way relationship between activity and price across six proof-of-work assets, the positive effects attributed to hashrate and transaction count disappeared. Power-law models face a related gap — Baquero's review found the literature has not yet tested how sensitive the curve's fit is to its starting date or checked it against future observations.
An honest forecasting standard, the review argues, would publish the naive benchmark alongside every model and report results separately for each market regime.
Source: CryptoSlate
Trading involves risk.
Risk Disclaimer
