Naive Forecasting Proves Tough to Beat in Bitcoin Price Predictions

Bitcoin price forecasting has become an elaborate game of cat and mouse between complex models and naive forecasts. The latter uses only current market information, making it a straightforward yet surprisingly effective benchmark. In fact, research suggests that no model has consistently beaten the naive forecast at horizons of one to six months across multiple market regimes.

A study by Carlos Baquero found that 23 peer-reviewed papers failed to demonstrate durable superiority over the naive benchmark. The issue lies in non-stationarity, the relationships between variables change over time, making it difficult for models to learn stable patterns. This is particularly evident in Bitcoin's evolving user base and liquidity.

Many researchers use walk-forward evaluation or multiple non-overlapping holdout windows to strengthen their results. However, even these methods can be flawed if not implemented correctly. For instance, information leakage can occur when a feature calculated with future data gives the model a faint view of the answer.