Weak: fits history beautifully, fails out-of-sample — the textbook spurious regression
ORIGINATED BY PlanB (pseudonymous), 2019
*In 2019 a pseudonymous analyst plotted Bitcoin's price against its scarcity and produced a chart so clean it looked like physics. The dots marched up a straight line across nine orders of magnitude, and the fitted line promised prices in the hundreds of thousands — even the millions. For two years it was the most-shared model in the asset class — and then price walked away from the line and never came back. This is not a takedown for sport. It is a dissection, because the most instructive failure in on-chain analysis* teaches more than a dozen working indicators.
Stock-to-Flow measures scarcity as a ratio. Stock is the existing supply — every coin already mined. Flow is the annual new issuance. Divide one by the other and you get the number of years, at the current issuance rate, it would take to reproduce the entire existing supply. A high number means new supply is a trivial fraction of what already exists — the intuition of hard money, expressed as a single figure. Bitcoin's S2F roughly doubles at every halving as the block subsidy is cut in half, marching from the low tens upward on a known schedule.
The theory borrowed from commodities. The pitch was that scarcity, so measured, is the dominant driver of monetary value: that gold and silver command their prices because their stock-to-flow is high, and that Bitcoin, whose S2F is not only high but rises on a known schedule, should command a price you can read straight off the ratio. It is a seductive story because it makes the halving — an event hard-coded in the protocol — into a price prophecy. If scarcity sets value and scarcity is scheduled, then price is scheduled too.
The model itself was a regression. PlanB fitted a line through the logarithm of price against the logarithm of S2F across Bitcoin's history and reported a correlation so tight it looked deterministic. That single fitted line, extrapolated forward through the 2020 halving, produced the six-figure targets that made the model famous. Everything rode on one claim: that the historical relationship between the ratio and the price was stable enough to project.
But scarcity alone cannot see demand. A pure S2F model has no term for adoption, liquidity, regulation, rate cycles, competing assets, or the reflexive psychology that actually moves a young speculative market. It says price is a function of supply schedule and nothing else. A commodity's price is set where supply meets demand; S2F silently assumes demand will always show up to clear whatever scarcity implies. When it fitted history, that assumption looked free. When history changed, the bill came due.
THE MATHS
S2F = existing_supply / annual_issuance ln(price) = a + b · ln(S2F) + ε ← the fitted regression price_hat = exp( a + b · ln(S2F) ) ← extrapolated to get the targets
THE HONEST READ · LIMITATIONS
Where this indicator lies to you.
It fit the past and missed the future — the signature of overfitting. A 2024 peer-reviewed study (Shelton, Journal of Risk and Financial Management) tested S2F alongside Metcalfe's Law regressions and reached the verdict that matters: these models help explain Bitcoin returns in-sample but have "limited to no ability to predict" them out-of-sample. That gap is the whole autopsy. Any curve with enough freedom can be bent through a decade of dots; the only honest test is whether it forecasts data it never saw. S2F did not. A model that fits history beautifully and fails forward is not a model — it is a description of the past wearing a lab coat.
The high R² was largely an artefact — the spurious-regression trap. Both price and S2F are non-stationary series that trend upward over time: S2F rises mechanically at each halving, price rose over the same decade for reasons of its own. Regress any two trending series against each other and you will get an impressive correlation whether or not they are causally related — Granger and Newbold showed this in 1974, and it is first-year econometrics precisely because it fools so many people. Without a cointegration test to establish a genuine long-run relationship, the tight fit tells you almost nothing.
The sample was tiny and the stakes were dressed as certainty. The model spanned only a handful of halving epochs — a few distinct values of the ratio doing the heavy lifting across nine orders of magnitude of price. From that thin evidence came specific dollar targets and a schedule. When the 2021–22 cycle diverged sharply from the line, the response was not to retire the model but to invoke longer horizons. That is the tell of a framework built to be unfalsifiable. We include S2F here not to mock a pseudonymous analyst but because its failure is load-bearing for everything else this school teaches: correlation is not a model, in-sample fit is not evidence, and a scheduled input cannot manufacture a scheduled price.
THE CITATIONS
Read the sources. Check our work.
FIELD TEST · SIGN-IN REQUIRED
Take the quiz. Track the curriculum.
The module text is open to everyone — no account required to read. The 4-question field test and the curriculum-wide graduation stamp need a free account so we can record your progress.
SIGN IN TO TAKE THE QUIZFREE · NEW ACCOUNT TAKES 30 SECONDS