The Stock-to-Flow Trade: Would S2F Have Made You Money?

PlanB’s stock-to-flow model still carries a number, and it is a big one. Through the back half of 2026, with Bitcoin trading in the low $80,000s (CoinGecko put it at $83,183 on 14 October, about 34% below the $126,080 record set on 6 October 2025), the pseudonymous analyst kept reaffirming that scarcity points to roughly $500,000 on average for this halving cycle, inside a range of $250,000 to $1 million (CoinGape). Critics have spent years picking apart the statistics behind that figure. We have run those arguments before, and they are largely settled.

This piece asks a narrower and more honest question. Set aside whether stock-to-flow is philosophically sound. Would it have made you money? A price model earns its keep in exactly one way, as a decision rule that beats the alternatives after costs. And we finally have the thing the debate lacked for years: a complete out-of-sample cycle to test it against. The model was published in March 2019. Bitcoin has since run through the 2020 to 2021 bull market, the 2022 collapse, the April 2024 halving, the October 2025 all-time high and the 2026 decline. That is enough history for a proper backtest. The verdict is not kind to the model.

Two lives: the story and the signal

Stock-to-flow leads two lives, and keeping them apart is the whole game. One life is a story: Bitcoin is the hardest money ever made, issuance halves roughly every four years, and growing scarcity will drag the price higher over time. That is a worldview, and you cannot backtest a worldview. The other life is a signal, a specific curve with specific bands and floor prices that claims to tell you what Bitcoin is worth at any moment and, by extension, when it is cheap and when it is expensive. That you can test, and it is the only part a trader can act on.

The test is easy to state. Take the model as published, turn it into rules a disciplined person could actually follow, apply those rules to the price history that came after publication, and compare the outcome to what you would have earned doing something almost mindlessly simple instead. If the signal does not beat buying and holding, or dollar-cost averaging, or a crude moving-average rule, then as a trading tool it is worthless, no matter how clean the chart looks. Worse, if it loses to those alternatives it has negative value, because following it cost you money you would otherwise have kept.

Turning a narrative into rules

To backtest stock-to-flow you first have to make it concrete. In its published form the model outputs a fair-value line driven by the stock-to-flow ratio, wrapped in colored bands. PlanB and the people trading off his work used it in three ways:

• As a fair-value anchor. Price should gravitate toward the model line, roughly $55,000 on average for 2020 to 2024 in the original paper, lifted to $288,000 in the April 2020 cross-asset revision.
• As a buy zone. When price sits in the lower bands, beneath model value, the asset is cheap and you accumulate.
• As a floor. In 2021 PlanB published explicit monthly floor prices that Bitcoin supposedly could not close beneath.

A trader following stock-to-flow would therefore have held through the cycle expecting reversion up toward the model line, added aggressively in the lower bands, and treated the published floors as hard support. Those are concrete, testable behaviors, not vibes. The beauty of a backtest is that it does not care how compelling the story was in the moment; it only cares what the rules did to a portfolio. So let us watch each behavior play out with real money on the line.

In-sample versus out-of-sample, the only split that matters

Every honest backtest lives or dies on one distinction: in-sample versus out-of-sample. The model was fit on data from 2009 to early 2019. A curve fit to history will always hug that history; that is simply what curve-fitting does, and a high fit on the data you trained on proves nothing at all. The only evidence that counts is how the model behaved on data it had never seen, the period after the paper went public in March 2019.

So make the cut there. Everything before March 2019 is the model admiring its own reflection. Everything after is a real experiment. The day the paper was published, Bitcoin traded at $4,002 (casebitcoin); that is when the honest clock starts, and every claim that follows uses only the post-publication record. Conveniently, that record now spans a full cycle, from a bear market through a halving to a new high and back down, which is exactly the variety a backtest needs to avoid flattering any single regime.

The 2020 to 2021 half-win

For about eighteen months, stock-to-flow looked like genius. Through 2020 and into 2021 Bitcoin tracked the model line closely, and PlanB’s monthly floor calls from the spring of 2021 were, for a while, uncanny. August 2021 near $47,000 and September near $43,000 were, in his own words, spot on, and the October call near $63,000 landed close to the roughly $61,000 monthly close (CryptoPotato). For a stretch, the signal printed money, and it is the main reason anyone still defends it.

Two things deserve attention. First, a model that calls for up and to the right during the strongest bull market in the asset’s history is not doing much work; almost any bullish heuristic nailed 2020 to 2021, from the halving narrative to a simple trend-following rule. The floor calls looked precise, but they were precise about a market that was going up anyway. Second, and more corrosive, part of the success was self-fulfilling. Stock-to-flow was so widely shared that it became a Schelling point: traders bought because the chart told them to, which pushed price toward the chart. Reflexivity, not scarcity, did much of the lifting.

That distinction matters enormously for a backtest, because reflexivity does not generalize. The mechanism depended on a retail crowd all watching the same meme, and it has not come back in the institutional era, when the marginal buyer is an ETF desk that has never heard of a rainbow band and would not trade on one if it had. A signal that works only while a crowd believes in it is not measuring the asset; it is measuring the belief, and belief is the one input that evaporated after 2021.

The floor that arrived above the top

Then came the call that defines the entire backtest. PlanB’s floor model projected that Bitcoin would close November 2021 at or above $98,000, and December at or above $135,000. Bitcoin’s all-time high at that moment was about $69,000, set on 10 November 2021. Read that again: the floor, a price the model said Bitcoin could not close beneath, sat roughly 40% above the highest price Bitcoin had ever reached.

A signal that tells you an asset cannot fall below a level it has never closed above is not a floor. It is a top-buying instruction dressed as support. A trader who trusted the $98,000 floor in November 2021 would have held, or added, within a few percent of the cycle peak. November instead closed near $57,000; Bitcoin then broke below the model’s lower band in June 2022 near $17,600 and bottomed around $15,500 after the FTX collapse that November. That is a drawdown of roughly 77% from the top the floor had implicitly endorsed. The single most actionable timing call stock-to-flow ever produced was an instruction to buy the top with conviction and then hold through an 80% decline.

To his credit, PlanB owned the miss on 1 December 2021: “Floor model first miss (after nailing Aug, Sep, Oct). No model is perfect, but this is a big miss and the first in 10y!” (CryptoPotato). The candor is real and worth respecting. The profit-and-loss statement is still a disaster. Here is the full signal scorecard, every documented call lined up against what the market actually did.

The chameleon you can never be wrong with

Watch what happened to the number itself. The original 2019 model implied roughly $55,000 on average for 2020 to 2024, against a realized average nearer $34,000 (Yahoo Finance). The April 2020 cross-asset revision raised it to $288,000 (PlanB’s own S2FX paper, which reported an eye-watering R-squared of 99.7%). The current target is $500,000, with a $250,000 to $1 million band. Every time reality came in low, the model was refit and the target moved.

Nico Cordeiro of Strix Leviathan named this in 2020: stock-to-flow is a “chameleon model” that changes color to match whatever the data demands, and he judged it about as successful “as the astrological models of the past” (CoinDesk). For a backtest this is fatal. A model you can always re-fit is a model you can never be wrong with, and a model you can never be wrong with cannot be traded, because there is no falsifiable call to act on and no honest way to keep score. The $250,000 to $1 million band makes it concrete. That is a four-times spread; no one can size a position against a forecast that wide, and no outcome inside it could ever count as a failure that forces a rethink.

The 2024 to 2026 out-of-sample verdict

The cleanest test of all just finished. Scarcity models make one unambiguous prediction: the halving cuts new supply, so price should rise. The April 2024 halving reduced issuance from 6.25 to 3.125 BTC per block, exactly on schedule. The code did precisely what it was written to do. Then Bitcoin did close to the opposite of what the model required. It climbed to a record $126,080 in October 2025, roughly double the previous cycle’s peak, then fell about 34% over the following year to the low $80,000s (CoinGecko). The supply shock landed on time; the price drifted sideways to down, nowhere near the $288,000 the cross-asset model implied for the cycle, let alone $500,000.

Our sibling coverage audited the model exactly one year after the top and arrived at the same place: stock-to-flow on trial, a year after the high. For a backtest the lesson could not be blunter. The one event stock-to-flow treats as its engine fired exactly as designed, and the wheels turned the wrong way. When the independent variable does precisely what you predicted and the dependent variable ignores it, the relationship you drew on the chart was never causal in the first place.

Being early is not the same as being right

The last refuge of a failed forecast is the phrase “I am early, not wrong.” It deserves a serious hearing, because on a long enough timeline a Bitcoin permabull has mostly been right about direction. But direction is not a trade, and early is not a position you can hold. A thesis (Bitcoin rises over decades) and a model (Bitcoin is worth $288,000 right now) are different objects. The first is a reasonable long-horizon view. The second was a specific, datable, falsifiable claim, and it was wrong.

Here the receding horizon becomes the tell. PlanB still frames $500,000 to $1 million as the current halving cycle through 2028 (U.Today), and now suggests the 2025 record was not the real peak and that the climb toward target is still to come later in the window. A target with no firm deadline is not a forecast you can backtest; it is a horizon that moves one halving further out each time price fails to meet it. For a trader, that is the costliest kind of comfort. Being anchored to $288,000 while Bitcoin fell 77% was not being early. It was a 77% drawdown that may or may not be recovered, financed by a conviction the model manufactured and never earned.

Beat buy-and-hold, or it has negative value

Here is the bar every price model has to clear, and almost none do. A signal is worth following only if it beats the laziest available alternative after costs. For Bitcoin that bar is punishing, because the asset went up enormously. Someone who bought on the day the stock-to-flow paper was published in March 2019 at $4,002 and then did absolutely nothing would be up about 1,979%, roughly 20x, by October 2026 (casebitcoin). Buy-and-hold is the number to beat, and it requires no model, no chart and no opinion about scarcity.

What did trading the signal add on top of that? Nothing good. Stock-to-flow never issued a sell in its life. The fair-value line told followers Bitcoin was cheap relative to $288,000 for the entire 2022 bear market, all the way down to $15,500. Its one hard timing call told you to hold with conviction at the top. A model that is permanently bullish is indistinguishable from buy-and-hold on the way up and strictly worse on the way down, because it talks you out of trimming and out of ever respecting a drawdown. It did not add return; it added confidence precisely where confidence was most expensive. Compare it to rules so simple they are almost insulting:

The point is not that a moving-average rule is sound investment advice. It is that a crude, price-only rule with no theory of scarcity at all would have sidestepped the exact drawdown stock-to-flow walked its followers into, and it would have done so knowing nothing but the price. The scarcity input did not merely fail to help; it actively hurt, by anchoring believers to a six-figure number while the price fell into five. Cory Klippsten of Swan Bitcoin, who is no Bitcoin bear, put the mechanism bluntly: the model “is autocorrelated” and “does not need S2F at all. Just price vs time” (Bitcoin Magazine). If the only working part of your scarcity model is a disguised time trend, you never needed the scarcity.

There is a quieter catch inside the buy-and-hold benchmark, too. Earning that 20x required actually holding the coins across a brutal cycle without losing them, which is a custody problem as much as a conviction problem. Readers weighing the hold-forever path should budget for real self-custody discipline, from air-gapped hardware wallets to defending against the SIM-swap attacks that have drained wallets belonging to people who read the macro correctly and still lost everything. A model that keeps you holding is only ever as good as your ability to keep what you hold.

What actually moved Bitcoin, and why stock-to-flow could not see it

If scarcity was not the driver, what was? The post-2019 record points squarely at demand and liquidity, the exact variables a supply-only model cannot represent. The 2024 to 2026 cycle was shaped by spot Bitcoin ETFs, approved by the SEC in January 2024, which turned Wall Street allocators into the marginal buyer almost overnight; by corporate and sovereign treasuries hoarding coins; by the Federal Reserve’s rate path, which turned hawkish again when the Fed raised its target range to between 3.75% and 4.00% in September 2026, its first hike since 2023; and by leverage. The most violent move of the year was the 10 October 2025 crash, when roughly $19 billion of positions were liquidated in a single day and Bitcoin fell about 14% in hours (CoinShares), a pure deleveraging event at the cycle hinge that a stock-to-flow chart has no variable for.

Stock-to-flow sees one force out of six, and it is the one force that behaved mechanically anyway because it is written into the code. A model blind to five of the six variables that actually set the price is not a forecasting tool. It is a chart of the single input that was never in doubt. Readers trying to make sense of the competing six-figure calls floating around in 2026 will get further from the analysts who model demand and flows than from anyone still extrapolating supply off a four-point sample.

Why the fit looks so good, and why that is the trap

The obvious objection is the chart. Stock-to-flow’s historical fit looks spectacular, with an R-squared around 95% in the original paper and 99.7% in the cross-asset version. How can something that fits that tightly be useless? Three reasons, and together they are the heart of the matter.

First, autocorrelation. Both sides of the model embed Bitcoin’s own price. Market value is stock multiplied by price, while the stock-to-flow ratio climbs mechanically over time as issuance falls. Plot one rising series against another rising series and you get a beautiful fit that means nothing. The analyst writing as Level39 showed that once you adjust for autocorrelation, the explanatory power collapses toward zero (Bitcoin Magazine). The chart is not evidence of a law; it is two time trends holding hands.

Second, spurious regression. Both series are non-stationary and trend upward, the textbook setup for a spurious relationship; two unrelated things that both grow over time will correlate strongly. Proving a real link requires cointegration testing that the model’s promoters never cleared. Third, sample size. Bitcoin has had four halvings. The model’s entire causal claim rests on four data points, and the single clean post-publication test of that claim, the 2024 to 2026 cycle, went the wrong way. You cannot build a durable trading edge on n equals four, no matter how confident the regression line looks.

Vitalik Buterin said the quiet part out loud in June 2022, as Bitcoin broke the lower band: models “that give people a false sense of certainty and predestination that number-will-go-up are harmful and deserve all the mockery they get” (The Block). For a trader, false certainty is the single most expensive thing you can own, because it is the thing that keeps you sized wrong at the worst possible moment.

The steelman, and the creator’s own retreat

The fair version of the bull case goes like this: stock-to-flow was never meant to time tops and bottoms, only to show direction across multi-year cycles, and $500,000 is a long-horizon value signal rather than a trade. PlanB himself now makes exactly this argument, reiterating that the stock-to-flow value sits around $500,000 while emphasizing that it is “not designed to identify tops or bottoms” (Bitbo). Taken on its own terms, that is a reasonable thing to say about a long-run valuation anchor.

But count the cost of the concession. A signal that cannot time entries or exits, that ships with a four-times band, and that silently re-fits whenever it is wrong is, by its own description, not tradeable. And notice what the author reaches for when he actually needs to read the market. In his 2026 commentary PlanB leans on the 200-week moving average, realized price and the RSI, and he conceded that the usual post-halving cycle peak “didn’t happen after 2024 halving” (Bitbo). In September 2026 he went further, telling followers that mining difficulty “better reflects market price” than models such as stock-to-flow or power laws (KuCoin). When the creator of the scarcity model navigates by moving averages and difficulty, the backtest has effectively been settled by the one person with the most to lose from admitting it.

The heirs face the same test

If stock-to-flow fails the backtest, does its successor pass? The popular heir is the Bitcoin power law, which fits price to time since genesis raised to an exponent of about 5.8 (Giovanni Santostasi). It has aged better than stock-to-flow, and its log-log corridor is a genuinely useful way to picture the long arc of adoption. But it fails the backtest for a familiar reason: it is the same price-versus-time curve Klippsten described, just dressed in better geometry. It still cannot see demand, its multi-year bands are wide enough to be unfalsifiable on any horizon you could actually trade, and it still assumes the future will rhyme with a fitted past.

We ran the whole family of successors through one consistent test in a companion piece, asking which Bitcoin price model actually survives. The uncomfortable finding: the tools that backtest least badly are the ones that promise the least. Descriptive on-chain measures such as realized price, MVRV and NVT tell you whether the market is historically hot or cold, and they refuse to hand you a target at all. That refusal is not a weakness. It is the honesty a forecast machine like stock-to-flow can never afford, because a model whose entire appeal is a six-figure destination cannot also admit it has no idea when or whether you will arrive.

How to backtest any price model in five questions

You do not have to take our word for any of this, and you should not. Any Bitcoin price model, stock-to-flow, the power law, the rainbow chart, or whatever launches next cycle, can be run through the same five questions before you risk a dollar on it.

Run stock-to-flow through that grid and it fails four of the five. It fits pre-publication history, loses to buy-and-hold, re-fits when wrong, and ignores demand entirely, while its one arguably falsifiable call, a floor, pointed at $98,000 just below the top. Before you let any model, or any analyst citing one, set your position size, it is worth recalling the regulator’s oldest warning, the line the SEC attaches to every investment product: past performance does not guarantee future results. A model built entirely out of past performance is that warning rewritten as an equation.

Frequently Asked Questions

Did the stock-to-flow model ever actually work?

For roughly eighteen months during the 2020 to 2021 bull market, stock-to-flow tracked Bitcoin’s price closely and PlanB’s monthly floor calls landed. But that window coincided with the strongest bull run in Bitcoin’s history, when almost any bullish rule looked right, and part of the fit was self-fulfilling as traders bought the widely shared chart. On out-of-sample data since publication in March 2019, including the 2022 crash and the 2024 to 2026 cycle, the model’s specific price calls failed.

What was PlanB’s stock-to-flow price prediction, and did it hit?

The original 2019 model implied around $55,000 on average for 2020 to 2024, and the April 2020 cross-asset version raised that to $288,000. The realized 2020 to 2024 average came in closer to $34,000. PlanB now targets roughly $500,000 on average for the 2024 to 2028 cycle, inside a $250,000 to $1 million range, and with Bitcoin in the low $80,000s in October 2026 that target is far off-path.

Would trading the stock-to-flow model have made money?

Not relative to the obvious alternative. Someone who simply bought Bitcoin when the paper was published in March 2019 and held would be up about 1,979% by October 2026. The model never issued a sell, kept followers anchored to six-figure targets through the 2022 bear market down to about $15,500, and its one hard timing call, the November 2021 floor at $98,000, pointed to support above the all-time high just before a 77% drawdown.

Why does the stock-to-flow chart look so accurate if the model fails?

Because the fit is largely a statistical illusion. Both sides of the equation embed Bitcoin’s price and both trend upward over time, so they correlate automatically, a mix of autocorrelation and spurious regression. Adjusted for autocorrelation, the explanatory power collapses toward zero. The model also rests on just four halvings, far too few data points to support a trading edge.

What should you use instead of stock-to-flow?

The honest answer is that no single model reliably predicts Bitcoin’s price. Descriptive on-chain measures such as realized price, MVRV and NVT describe whether the market is historically hot or cold without promising a target, and they backtest better precisely because they claim less. Treat any model that hands you a confident future price, the power law included, as a story to stress-test rather than a signal to trade.

Priya Reddy covers market structure, price models and prediction markets for HOGE Wire.