After Stock-to-Flow: Which Bitcoin Price Model Survives?

One year ago, on 6 October 2025, Bitcoin printed an all-time high near $126,080. On 12 October 2026 it trades around $83,460, roughly a third below that peak. The 2024 halving arrived exactly on schedule, cut new issuance in half, and the price then did the opposite of what the most famous scarcity model said it would. For the predictions cluster, the interesting question is no longer whether stock-to-flow was wrong. That debate is over. The interesting question is what, if anything, deserves to replace it.

Stock-to-flow (S2F) did not simply miss a target. It missed at the one moment it was designed to nail: the first full cycle after a halving, when programmed scarcity was supposed to do the heavy lifting. In its place a whole shelf of successor models now competes for attention, the power law, MVRV, realized price, NVT, Metcalfe’s Law, the rainbow chart and the Pi Cycle Top among them. Each claims to have learned from S2F’s failure. This piece runs them through one consistent test and asks a blunt question: do any of them pass it, or are they S2F wearing different math? Our companion audit graded the original model a year after the top in Stock-to-Flow on Trial; here we grade the heirs.

The model that promised a destination

Stock-to-flow is a ratio borrowed from commodities. Stock is the quantity already above ground; flow is the amount produced each year. Divide one by the other and you get the number of years of current production it would take to reproduce the existing pile. Gold scores high because annual mining adds little to a vast inventory. The pseudonymous analyst PlanB applied the idea to Bitcoin in a March 2019 article, when a coin cost about $4,000, and reported a 95% R-squared, meaning the ratio alone appeared to explain 95% of Bitcoin’s historical price, according to CoinGecko’s explainer.

The pitch was seductive because it was mechanical. Every halving cuts flow in half, which roughly doubles the ratio, which the regression then maps to a far higher price. By that logic the destination was not a guess; it was baked into the issuance schedule. The same source puts gold’s ratio near 62 and silver’s near 22, with Bitcoin climbing from roughly 27 before the 2020 halving to about 119 after the 2024 halving. A 2020 cross-asset revision, which PlanB called S2FX, pushed the implied average toward $288,000 for 2020 to 2024. His current figure is an average near $500,000 for the 2024 to 2028 cycle, inside a $250,000 to $1 million band, a stance he still repeats.

Where the promise broke

The first visible crack came in late 2021. PlanB had published a short-term floor model that called for a Bitcoin close near $98,000 by the end of November 2021 as a worst case. The month closed near $57,000, and the price never returned to the band. CoinGecko’s explainer describes that $98,000 miss as the first major crack in the model’s armour, followed by a 2022 bear market that dragged the price below $16,000 while S2F still implied a figure well above $100,000.

PlanB’s response set the pattern that still defines the debate. He conceded a “FLOOR MODEL FAIL” while insisting the main S2F model stayed intact. The distinction mattered to believers and to almost nobody else, because the main model’s own schedule had just run its cleanest test and lost. The 2024 halving cut issuance to roughly 450 coins a day on time, the ratio doubled on cue, and the price peaked near $126,000 in October 2025 before sliding about a third. The supply shock arrived precisely as coded. The price shrugged.

That is the one-sentence diagnosis of every scarcity model: it measures supply and calls the result price. It has no term for demand. When demand is the variable that moves, as it was through 2024 to 2026, a supply curve can only watch.

Why the 95% was never real

The number that sold S2F was its 95% R-squared, a figure that sounds like proof and means very little here. Two problems sink it. The first is autocorrelation. Market value equals supply multiplied by price, and the stock-to-flow ratio is itself built from supply, so the model effectively puts supply on both sides of its own equation. A series regressed partly against itself will always look like a tight fit; the correlation is baked in before Bitcoin does anything at all. Strip that overlap out, as the analysis published in Bitcoin Magazine did, and the explanatory power falls toward zero.

The second problem is spurious regression. Any two series that both rise over time will correlate strongly whether or not one causes the other, which is why econometricians run stationarity and cointegration tests before trusting a fit. Bitcoin’s price and its stock-to-flow ratio both trend upward with the calendar, so a high R-squared between them is the expected output of the arithmetic, not evidence of a causal link. The relationship does not pass the tests that would be needed to call it real, a point academic reviews of the model have made repeatedly since 2020.

Then there is sample size. The entire model rests on four halvings, and because the earliest cycle looks nothing like the most recent one, the honest count of comparable observations is closer to two or three. No credible forecast stakes a destiny on three data points. S2F did exactly that, wrapped the result in a decimal-heavy R-squared, and invited a generation of buyers to treat a curve fit as a law of nature.

The test every price model has to pass

To judge the successors fairly, it helps to fix the rules in advance rather than grade each one on its own marketing. Four questions do most of the work, and they are the questions a statistician would ask of any forecast.

• Does it predict out of sample? A curve fitted to past prices will always look good on past prices. The test is whether it called anything useful after it was published, not before.
• Does it escape the price-versus-time trap? If the only input is the calendar (days since launch, halving number, a four-year clock), the model is really saying price rises because time passes.
• Does it model demand? Price is where buyers and sellers meet. A model with no buyer in it is half an equation.
• Is it falsifiable on a horizon you care about? A band from $250,000 to $1 million over four years is so wide that almost any outcome confirms it. A model you cannot be wrong about cannot be right either.

These are not academic niceties. Cory Klippsten, chief executive of Swan Bitcoin, has argued for years that S2F is autocorrelated, that it uses supply on both sides of its own equation and therefore reduces to just price versus time. Vitalik Buterin put the cultural cost more bluntly in 2022, writing that financial models giving people a false sense of certainty that number will go up are harmful and deserve the mockery they get. Both critiques land on the same spot: a model that passes question one but fails the other three is a story, not a forecast. The same instinct, trust what can be checked rather than what merely fits the past, runs through crypto’s wider push toward verifiable machine-learning output, where a claim counts only once it survives out of sample.

The power law: better geometry, same determinism

The heir with the most momentum is the Bitcoin power law, popularised by the physicist Giovanni Santostasi and the analyst Harold Christopher Burger. It fits price to a straight line on a log-log chart: price is proportional to the age of the network, measured in days since the genesis block, raised to a power of about 5.82, per the specification quoted by CryptoSlate. In plain terms, price grows as a steep but slowing function of age. It has aged better than S2F, hugging the long-run trend across several cycles, which is why it inherited the scarcity crowd.

Then it meets the test. The independent variable is days since genesis, which is time, so it fails question two for the same reason S2F does. CryptoSlate is candid that the model does not mechanistically incorporate known drivers such as ETF demand or liquidity cycles, so it fails question three. And its outputs are quoted as ranges rather than calendar calls, with the same analysis describing broad windows in which Bitcoin would not sustain prices below six figures after 2028 and could touch seven figures at some point between 2028 and 2037. A window nine years wide struggles badly with question four.

The experts split along exactly that line. The consultant Adrian Morris called the power law a magic trick, accusing its promoters of putting a statistics bunny into the hat and pulling a physics bunny back out, and noting that under the model Bitcoin in 2045 might be worth $200,000 or $10 million, which is closer to a horoscope than a prediction. In the same report, the network economist Timothy Peterson dismissed both the power law and S2F as historical relationships rather than true models because they are based on time, which is not an independent variable. Santostasi, for his part, insists the pattern is undeniable, that one only needs eyes to see it, and files it under the young discipline of social physics. Readers can decide for themselves whether that is a defence or a confession.

The deeper issue is that calling something social physics does not make it physics. A genuine physical law states a mechanism: this force produces that acceleration, and you can intervene on the cause. The power law states only that price has tracked a curve drawn against time, with no account of why a buyer would pay the number the curve names on any given day. That is a pattern, and patterns hold until they do not. A model with no mechanism cannot tell you whether 2026 is a temporary deviation from the line or the moment the line stopped working, which is exactly the question an investor needs answered.

The rainbow chart and the Pi Cycle: timing dressed as value

Two popular tools belong to the same time-regression family but are at least honest about it. The Bitcoin rainbow chart draws a logarithmic regression through price history and paints coloured bands from fire sale at the bottom to maximum bubble at the top. It is a mood ring, and its makers are careful to call it a bit of fun rather than financial advice. It fails question two by construction, because the bands are drawn against time, but it rarely pretends otherwise.

The Pi Cycle Top indicator, built by Philip Swift, is narrower and more interesting. It fires when the 111-day moving average crosses above twice the 350-day moving average, a condition that landed within about three days of the 2013, 2017 and 2021 cycle tops. That is a striking record for a timing signal. It is also not a price model: it suggests the market may be overheating, not what number comes next, and it degraded in the exchange-traded-fund era, giving a far less clean read on the 2025 cycle. Both tools pass the honesty test precisely because they never promise a destination.

MVRV and realized price: the honest thermometers

The most useful successors are the ones that stopped forecasting. They come from on-chain data rather than a time axis. The foundation is realized capitalisation, devised at Coin Metrics by Nic Carter and Antoine Le Calvez, which values every coin at the price it last moved rather than today’s price. Divide realized cap by supply and you get realized price, the aggregate cost basis of the entire network.

Market value to realized value, or MVRV, compares the live price to that cost basis. It was first framed by Murad Mahmudov and David Puell in October 2018, and a normalised version, the MVRV Z-Score, was built shortly after by a pseudonymous analyst known as Awe and Wonder, as Glassnode documents. When MVRV runs hot, holders are sitting on large unrealised gains and history says distribution tends to follow; when it runs cold, the average coin is underwater and bottoms have historically formed. Network value to transactions, or NVT, popularised by the analyst Willy Woo, does something similar on the activity side, a rough price-to-earnings ratio for a monetary network.

Notice what none of these produce: a target. They tell you where the market sits relative to its own history, not where it is going next. That is not a weakness dressed up as modesty. It is the whole point. A thermometer that refuses to invent a future temperature is exactly the tool you want when the future is being set by variables the model does not contain.

In practice that is how desks actually use them. A deeply negative MVRV Z-Score has coincided with every major cycle low, so it frames risk: when the average coin is far underwater, forced selling is usually closer to exhaustion. Realized price acts as a rough floor, the level at which the median holder breaks even. Neither reading tells you the price next quarter, and neither pretends to. They answer a smaller, more honest question, how stretched is the market right now, and leave the forecasting to whoever is brave enough to guess the demand.

Metcalfe’s Law: the one that looks at demand

Only one model in common use tries to put demand on the main axis. Metcalfe’s Law holds that a network’s value scales with the square of its users, and the economist Timothy Peterson formalised it for Bitcoin in a paper titled Metcalfe’s Law as a Model for Bitcoin’s Value, using active addresses as the proxy for users. This is the only candidate that passes question three, because its driver is adoption rather than the calendar or the coin schedule.

It is also the most instructive failure of the group. Addresses are not people; one user can hold dozens, and one custodian can sit behind a single wallet holding coins for millions. In the ETF era the marginal buyer is often a BlackRock creation ticket that never touches an on-chain address at all, so the demand that actually moved price in 2026 is partly invisible to an address-count model. The same distortion shows up in Bitcoin’s own block space, where the inscription and data-layer activity that revived on-chain fees inflates address and transaction counts without a matching change in monetary demand. Metcalfe at least asks the right question. It just cannot see most of the answer.

The scorecard

Lined up against the four questions, the family sorts into three groups. The predictive scarcity-and-time models (S2F, the power law, the rainbow chart) share one flaw: their driver is the clock or the issuance schedule, so they cannot price demand and their bands are too wide to fail. The descriptive on-chain gauges (realized price, MVRV, NVT) pass by refusing to forecast. Metcalfe sits alone as the only model that targets demand, and the only one whose failure is about measurement rather than philosophy.

What actually moved Bitcoin in 2026

The clearest way to expose a supply model is to list what set the price this year and ask how many of those forces the model can even see. The 2024 halving is the only driver S2F contains. Everything else that mattered in 2026 sits outside it.

Spot ETFs changed the marginal buyer. Corporate and sovereign treasuries kept accumulating. The Federal Reserve, under chair Kevin Warsh, raised rates 25 basis points to a 3.75% to 4.00% target on 16 September 2026, its first hike since 2023, with markets leaning toward a hold at the 27 to 28 October meeting and keeping December live. And on 10 October 2025, a single leverage cascade erased about $19 billion in positions in a day, the largest liquidation event on record, as Bitcoin fell from roughly $122,000 toward $105,000. None of that is in a stock-to-flow ratio.

Believers point to 2020 and 2021, when price did climb toward the model’s path, as proof the framework works. The likelier explanation is reflexivity. S2F went viral at the exact moment a wave of retail buyers wanted a reason to believe, and a widely shared chart with a huge number on it became a coordination device, a Schelling point that helped pull price toward the line it had drawn. That is not the model being correct; it is the model briefly moving the thing it claimed to measure. The mechanism cannot repeat in 2026, because the marginal buyer is now an institution allocating against rates and mandates, not a retail trader screenshotting a chart.

Even PlanB reaches past his own model

The most telling evidence against the scarcity family comes from the person with the most to lose by admitting it. In September 2026, with Bitcoin trading below the band miners need to break even, PlanB argued that mining difficulty better reflects market price than models such as S2F or power laws, because difficulty tracks miner hash rate and production cost. He put difficulty near 127 trillion and framed a renewed rise in it, not a scarcity ratio, as the thing that would confirm a bottom.

Read that again. The author of the most famous scarcity model now reaches past both his own model and its leading heir, toward a production-cost metric, to read the market in real time. Estimates of all-in mining cost clustered around $80,000 to $90,000 a coin earlier in 2026, which is why a sub-cost tape stressed the public miners we compared in Marathon versus Riot. PlanB has not recanted the $500,000 target; he still repeats it. But his working toolkit has quietly migrated to exactly the price-versus-something-real metrics his critics always preferred, which is the loudest possible concession that a wall-poster number and a trading signal are different objects.

Why descriptive beats predictive

Put the family back together and a pattern appears. The models that sell a destination (S2F’s $500,000, the power law’s eventual seven figures) all share the same engine: time or supply in, price out, demand nowhere. The models that earn their keep (realized price, MVRV, NVT, and the Pi Cycle as a timing overlay) all refuse to name a future price. They describe the present relative to the past and stop there.

That is not a coincidence. Certainty is the product a predictive model sells, and in a market whose 2026 tape was written by Fed meetings, ETF creations and a record liquidation cascade, certainty is the one thing no supply-or-time curve can honestly offer. Matt Hougan, chief investment officer at Bitwise, made the same argument from the cycle side when he declared the four-year cycle dead and told investors to expect a longer, flatter grind now that institutions, not retail, set the marginal bid. A four-year clock and a halving ratio are the same bet in different clothes, and both lost in the same year.

How to read any Bitcoin price model

The point of the exercise is not to crown a winner; it is to give readers a filter they can apply to the next model that goes viral. A few flags do most of the sorting.

• Green flag: the model states its one independent variable out loud and names what it ignores.
• Green flag: it gives a range and a horizon tight enough that it could be proven wrong.
• Green flag: it separates description (where are we) from prediction (where next) instead of smuggling one in as the other.
• Red flag: a single magic number with no error bars.
• Red flag: time or supply as the only input.
• Red flag: it gets refitted after every miss. Nico Cordeiro of Strix Leviathan called S2F a chameleon model for exactly this, adjusting its parameters to match whatever just happened, and judged it about as successful as the astrological models of the past.

The bottom line for the predictions cluster

Here is the practical translation. Treat the headline numbers, S2F’s $500,000 and the power law’s eventual seven figures, as narrative anchors, not forecasts; they tell you what a scarcity story implies, not what the market will pay. Use the descriptive on-chain gauges (realized price, MVRV, NVT) to judge whether Bitcoin is cheap or expensive relative to its own cost basis right now. Then watch the things that actually set the tape: ETF flows and treasury demand on the buyer side, Fed and Treasury liquidity on the macro side, and leverage on the fragility side.

As of 12 October 2026, Bitcoin trades near $83,460, about a third below the record it set a year ago, with the Fed in a hiking posture and the next decision due on 28 October. A scarcity model looking only at the 2024 halving would have told you to expect the opposite. That gap, between what the issuance schedule promised and what policy and flows delivered, is the entire lesson. The halving already happened. The models that saw only the halving have already said everything they can, and the market spent a year proving them wrong.

Frequently Asked Questions

Why did the Bitcoin stock-to-flow model fail?

Stock-to-flow measures supply and treats scarcity as the only driver of price, so it has no term for demand. The 2024 halving cut issuance on schedule and roughly doubled the ratio, yet Bitcoin peaked near $126,000 in October 2025 and then fell about a third, because ETF flows, Fed policy and a record leverage crash, none of which the model contains, set the price instead.

Is the Bitcoin power law model better than stock-to-flow?

It fits the long-run trend more closely, but it shares the same core flaw. Its only input is time (days since the genesis block), it does not incorporate ETF demand or liquidity, and its forecasts are quoted as ranges so wide (six figures after 2028, up to seven figures by 2037) that almost any outcome confirms them. Better geometry, same determinism.

What is the best Bitcoin price prediction model in 2026?

No model reliably predicts Bitcoin’s price, and the most useful tools do not try. Descriptive on-chain gauges such as realized price, MVRV and NVT tell you whether the market is hot or cold relative to its own cost basis without issuing a target. For direction, demand and liquidity signals (ETF flows, treasury buying, Fed policy) have mattered far more in 2026 than any supply-based curve.

Does PlanB still predict $500,000 for Bitcoin?

Yes. PlanB continues to cite an average near $500,000 for the 2024 to 2028 cycle, inside a $250,000 to $1 million band. Notably, though, by September 2026 he was arguing that mining difficulty reflects market price better than S2F or power laws, so his day-to-day read on the market now leans on production cost and difficulty rather than the scarcity ratio itself.

What is the difference between MVRV and stock-to-flow?

Stock-to-flow is a predictive model that outputs a future price target from the supply schedule. MVRV is a descriptive gauge that compares the current price to the network’s aggregate cost basis (realized value) and outputs no target at all. One tries to tell you where price is going; the other tells you where it sits relative to history, which is the more honest and more usable of the two.

Priya Reddy is a markets writer at HOGE Wire, covering Bitcoin, on-chain data and the models people use to predict crypto prices.