A preregistered analysis by Carlos Baquero and Raquel Menezes applies power-law models to Bitcoin's price history, projecting future values based on the cryptocurrency's relationship between price and age. The study examines 820 weekly observations from January 2011 through September 2019, concluding that Bitcoin's long-run growth rate has slowed under their testing framework.
The power-law model serves as a framework for forecasting. For December 28, 2030, the model projects a median price of $383,000, with a 90% confidence interval ranging from $150,000 to $1,015,000. Across five alternative trend models, median forecasts range from $85,000 to $1,011,000.
Critical limitations of the approach:
The study's conclusions depend heavily on assumptions about how price deviations behave over time. The authors found that when they allowed for more persistent changes to prices, their main finding of a slowing growth rate disappeared entirely. This sensitivity reveals how dependent the results are on unproven persistence assumptions.
Additionally, the researchers found that a Gompertz curve—a mathematical function that approaches a ceiling—fits the historical price record almost as well as the power-law model, suggesting the historical data does not uniquely support one approach over another.
The recovery forecasts assume prices revert toward their long-run trend, yet the historical analysis cannot establish this tendency within the models tested. As the authors acknowledge, a forecast that assumes a pull toward trend cannot use its projected rebound as evidence that the pull exists.
The paper also warns that its predictive intervals are likely too narrow and not properly calibrated to reflect the full range of uncertainty. Broader uncertainty about how persistent price shocks are remains only partially represented.
The authors have committed to testing their forecasts against actual prices for the last Saturday of each year from 2027 through 2030, using a log-density score to measure prediction accuracy. They preregistered their protocol on October 6, applied the historical test on October 7, and registered results that morning.

