Most Bitcoin Price Forecasting Models Fail to Beat Simple Benchmark
Bitcoin price forecasting has become an elaborate competition between various methods, but most of them struggle to beat a simple benchmark.
The benchmark in question is naive forecasting, which uses only current market information. This can include today's price for a price forecast, zero for a return forecast, and a random walk for a direction forecast.
Researchers have examined hundreds of papers on Bitcoin prediction research, but few have demonstrated durable superiority over the naive benchmark at horizons of one to six months across several market regimes.
A 2026 preprint by Carlos Baquero found that no model had shown consistent results when tested outside its design period. The review itself is still awaiting peer review, which highlights the importance of stronger evaluation in forecasting claims.