Can Bitcoin be valued, and what does a fair-value model produce?

What methods produce a Bitcoin fair value?

Four model families produce a Bitcoin fair value: network-value models, on-chain cost-basis measures, money-supply regressions and power-law fits. They rest on different inputs and they disagree.

The peer-reviewed footing is narrower than the volume of published output suggests. Wheatley, Sornette and co-authors put a generalized Metcalfe's law estimate of Bitcoin's fundamental value into Royal Society Open Science in June 2019, combined with a bubble-detection model, and found that fundamental value heavily exceeded on at least four occasions. That is the strongest academic support available for the claim that Bitcoin has a computable fundamental value, and it is one paper.

The on-chain measures are the industry's own vocabulary rather than the academy's. The market-value-to-realised-value ratio and its z-score, introduced by David Puell and Murad Mahmudov, and the network-value-to-transactions ratio introduced by Willy Woo, are widely adopted and carry no peer review behind them. That is worth saying plainly rather than implying otherwise, because a reader who checks will find it out anyway.

The family this work belongs to is the money-supply regression, and it has institutional cover. CF Benchmarks, which administers the CME's crypto benchmarks, published research on the money supply relationship in March 2026 and used the term M2-implied fair value for the output. Fidelity Digital Assets published a fit statistic on the same relationship in the same month. Two of the most credible publishers in the field do the same thing, which is the strongest argument available that the approach is legitimate.

How well do the money-supply models fit Bitcoin?

Reported fits are high on levels and much lower on rolling windows, which is the same arithmetic that inflates every levels correlation in this topic.

Fidelity Digital Assets reported an r-squared of 0.87 between Bitcoin and global broad money over the past fifteen years, on Bloomberg data, published 25 March 2026, and stated in the same research that this is correlation and not necessarily causation. CF Benchmarks, working on monthly data from 2000 to February 2026, reported a rolling four-year correlation ranging between 0.4 and 0.6. Both are model estimates over the samples stated and past results do not indicate future results.

The gap between those two numbers is the point rather than a discrepancy. A single high fit statistic on two series that both trend upward is weak evidence, and a rolling window is the honest version of the same question. Anyone quoting the higher figure without the lower one is selecting.

Fidelity's number is also not a BlockPhi finding and is not restated as one anywhere on this domain. It is Fidelity's measurement, on global broad money, on levels, over fifteen years.

Why does BlockPhi publish no power-law price target?

Because the fit is not specification-robust. The exponent moves substantially when the time origin is shifted, and the authors who documented that reject the structural reading.

Baquero and Menezes fitted a power law to Bitcoin and reported an in-sample r-squared of 0.9595, then showed that the exponent varies from 5.65 to 16.49 across reasonable shifts of the time origin, and concluded that the specification is not robust while conceding that the forecasts work at longer horizons. Those are model estimates published 20 May 2026 and past results do not indicate future results.

A model whose central parameter triples depending on where you start counting is a model that describes the sample rather than the asset. It can still be useful. It cannot carry a price target, and a firm that puts one on it is selling the fit rather than the finding.

The stock-to-flow model is a separate case and this work does not use it. It has been criticised as a spurious regression on non-stationary series, with stock appearing on both sides of the equation, and using it costs credibility with exactly the readers this research is written for.

What would make a fair-value figure checkable?

A stated sample period, a stated method, and an out-of-sample statistic printed beside the in-sample one. Most published fair values carry none of the three.

An in-sample fit is the easiest number in this field to attack, because it is the number the model was fitted to produce. The statistic that carries information is the one measured on data the model never saw, and the second one is almost never published beside the first.

BlockPhi's own fit statistic is therefore not on this page. The sample period and the walk-forward result that would make it checkable are not settled, and a fit statistic published without them would be precisely the number this section argues against. The standard is easier to hold when it is applied first to your own work.