The headline number is 10%. The number that matters is -3.2%.

JPMorgan's 2027 Long-Term Capital Market Assumptions handed Bitcoin a 10% baseline annualized return — the highest projected return on the entire asset table. Buried a few paragraphs deeper sits the only model in the ensemble that actually touches the Bitcoin network. It prints negative. Minus 3.2%. The engine built on transaction volume, fee revenue, and active addresses — the metrics that describe whether anyone is actually using the chain — concludes that the network's fundamentals do not support the current price.

That is not a rounding error. That is a verdict, and it was issued by the same document currently being reported as bullish.

I have spent enough years inside risk models to know how this happens. You build three engines, weight them, and the weighted average conceals the argument the engines are having with each other. The output looks like consensus. It is actually a disagreement that has been averaged into silence.

The code doesn't lie, but the spreadsheet that summarizes it will.

Context: what this document actually is

Set the mechanics before breaking them.

The LTCMA is JPMorgan's annual long-horizon capital market assumptions publication — 31st edition, data cutoff September 30, 2026. Institutions do not read it for entertainment. They read it to build target-date glidepaths, multi-asset strategic allocations, and the long-run return inputs inside mean-variance optimizers. When an LTCMA number moves, model portfolios move with it. That is the real weight of this document. It is a standard-setting artifact, not a research note.

The Bitcoin chapter is a first for the publication. That alone is a structural event — an institution that spent years declining to model the asset has now been forced to model it, because clients keep asking and the questions no longer go away. The chapter runs a three-model weighted ensemble:

• 65% weight: supply cycle versus global M2 — a macro liquidity model.
• 30% weight: gold substitute — a relative valuation model.
• 5% weight: on-chain usage — the only engine anchored to the network itself.

Inputs: a spot anchor of $83,000, a volatility assumption of 55% annualized, three scenarios (bull +16.6%, base +10.0%, bear -11.2%). The stated conclusion: Bitcoin has no strategic role in a portfolio beyond a bet on fiat debasement, and it is not entered into the covariance matrix that drives the optimizer.

Now the forensics.

Core: dismantling the ensemble

Start with the weighting, because the weighting is the argument.

95% of the model's explanatory mass sits on factors that have nothing to do with the Bitcoin network. 65% on a macro liquidity comparison, 30% on a gold analogy. The one engine that reads the chain — the one that could tell you whether blockspace demand is growing or decaying — gets 5% of the vote and produces a negative number.

Read that structurally and the conclusion is unavoidable: JPMorgan's 10% baseline is a macro liquidity forecast wearing a Bitcoin label. It is a prediction about dollar and global M2 expansion, filtered through a supply-rigidity lens, with a gold cross-check bolted on. If you disagree with the M2 path, the 10% dissolves. Nothing in the number is derived from how the network is actually being used.

I flagged this exact failure mode in 2020 when I reverse-engineered Compound's cToken interest rate curves on a local Hardhat fork. The rate model looked sophisticated. Under stress simulation the collateral factor adjustments behaved nothing like the market they claimed to represent. The parameters were arbitrary — chosen, not discovered — and the elegance of the curve hid that arbitrariness. Here the parameter is the 65% weight. Nobody has published a robustness test justifying it. It is a number someone picked, and the document does not defend the pick.

The circular reasoning at the center of "no portfolio role"

Here is the part the coverage keeps missing. Bitcoin is excluded from the covariance matrix. In mean-variance optimization, the covariance matrix is the input that determines weights. An asset that never enters the matrix cannot receive an allocation. The output — zero weight — is not an analytical finding. It is an arithmetic consequence of the setup.

You cannot conclude that an asset has no portfolio role after removing it from the only machinery capable of assigning one. That is not analysis. That is a design choice presented as a conclusion. The document has effectively said: we did not model it, therefore the model says not to hold it.

I have watched this pattern on institutional risk desks for years. The covariance matrix is where politics hides. An asset that is hard to estimate — thin history, fat tails, regime shifts — gets dropped, and the drop is later narrated as prudence. In the 2017 ICO cycle I audited the Waves platform's IDEX contracts and found the same move in miniature: the liquidity pool mechanism looked robust until you isolated it from the surrounding assumptions. The bug was never in the code that ran. It was in the code that was assumed to run. Integer overflow in a trading engine is easy to find once you stop trusting the abstract layer. Circular reasoning in a covariance matrix is the same category of defect, one level up.

The scenario math nobody ran

The document publishes three annualized rates. It does not publish the aggregate. So I ran it.

Equal-weighted across the three scenarios: roughly +5.1% annualized. Geometric mean, which is what actually compounds a portfolio: roughly +4.4%. Both figures sit below the 5.6% the same table assigns to the US aggregate bond index.

Sit with that. The asset being marketed at a 10% baseline has a geometric expected return lower than investment-grade bonds, once you weight the scenarios the way a distribution actually behaves. The 10% is not the median. It is one of three paths, and the paths are not symmetric in the way readers will assume. The bear case at -11.2% is not a distant tail; it is a full scenario with its own probability mass.

Extrapolating ten-year terminal values from the $83,000 anchor:

• Bull: ~$386,000, ~$7.7T market cap, 26–31% of gold's market cap.
• Base: ~$215,000, ~$4.3T, 14–17% of gold.
• Bear: ~$25,000, ~$0.5T, ~2% of gold.

All terminal values are my calculations, assuming ~19.9M circulating supply and a $25–30T gold market cap. Confidence: moderate.

The risk-adjusted number is the real indictment

The 10% baseline is the highest projected return on the entire table — above US large cap at 7.5%, small cap at 7.7%, high yield at 6.6%, gold at 5.5%. It also carries a 55% volatility assumption, roughly three times equity volatility.

Compute the Sharpe ratio. Assume a 3.75% risk-free rate and ~16% equity volatility. Bitcoin lands near 0.11. US large cap near 0.23. Gold near 0.12.

Bitcoin's return per unit of risk is roughly half that of US equities and indistinguishable from gold — without gold's crisis correlation. That is the actual technical basis for "no strategic role," and the document does not state it. It states the conclusion and leaves the arithmetic for someone else to do.

And the 55% volatility input is itself optimistic. Realized Bitcoin volatility has spent most years in the 60–80% band. If the true figure is 70%, every risk-adjusted conclusion in the chapter deteriorates further. The model is calibrated to understate the risk it is measuring — a small choice with compounding consequences over a ten-year horizon.

The supply side is nearly exhausted as an input

Here is where the 65% macro model quietly depends on a contested assumption. It encodes continued validity of the four-year halving cycle. Current block subsidy has fallen low enough that annual new issuance is now under 1% of circulating supply. The marginal supply shock is close to informationless.

If the four-year cycle has been weakened or dissolved by ETF flow dominance — one of the live debates since 2024 — then the model carrying 65% of the weight loses its explanatory engine, and the confidence interval on the entire forecast blows out. No robustness test is disclosed. The assumption is load-bearing and unexamined.

The on-chain model's -3.2% also points at something the document leaves undeveloped: the long-run security budget. If fee revenue does not scale with price, miner income depends on subsidy, subsidy decays, and security spending comes under pressure. A -3.2% reading on a usage-based model is a quantitative expression of that structural problem. It is a far harsher judgment than the 10% headline, and it occupies one sentence. Hash power that cannot be paid for by fees concentrates by necessity — the pools that survive the subsidy decay are the ones with the cheapest energy and the deepest balance sheets. Decentralization at the consensus layer becomes a narrative about a set of operators that can be counted on one hand.

ETF absorption broke the transmission chain

The 65% model implicitly assumes supply rigidity converts into price elasticity. That transmission ran through on-chain activity: scarcity, then address growth, then visible demand, then price. Spot ETFs interrupted it. Demand now settles as custodied balances and never appears as on-chain address growth. The chain that connected scarcity to observable usage to price has been severed at the middle link.

Which is precisely why the usage model prints negative while the macro model prints positive. They are measuring two different assets. One is a network. The other is a ticker. The ensemble averages them as if they were the same thing.

The $83,000 anchor is a comparability defect

An annual publication anchors to a point-in-time spot price. When spot drifts, the same "10%" implies a different terminal value, and readers will not adjust. They will read a stable long-run expectation into a number mechanically tied to a snapshot. This is the kind of methodological seam that produces misallocation at scale — the same class of error I traced in 2022 when Mercurial Finance's risk parameters, set against a benign volatility regime, went insolvent the moment the regime shifted. Anchors that look conservative in calm markets become leverage in stressed ones.

The $83,000 figure is itself a bear-market signal. If that is the September 2026 mark, Bitcoin has already fallen materially from its cycle high, and JPMorgan is publishing a 10% baseline from a depreciated base. That actually raises the credibility of the number. It is not momentum-chasing optimism. It is a restrained framework output from a lowered starting point.

Market structure read

The $83,000 anchor and a $2.5T total crypto market cap imply BTC dominance near 65% [my calculation]. That is a high-concentration, risk-off consolidation profile — capital retreating to the largest, most liquid name while altcoins bleed. In that regime, an institutional publication concluding "no strategic role" reinforces the flow rather than reversing it. Research does not move price at this horizon. It moves the plumbing that later moves price.

Family office penetration is the other number worth holding. Roughly one in ten family offices reports holding crypto, per the 2026 Global Family Office Report cited. That is the classic post-zero, pre-mainstream band — behavior has started but has not institutionalized. Self-reported allocation surveys skew upward. Treat 10% as a ceiling, not a floor.

Competitive positioning across the same table tells you what JPMorgan thinks Bitcoin is for. It projects Bitcoin at roughly twice gold's expected appreciation (10% vs 5.5%) while assigning it a "gold supplement" weight of only 30% inside the crypto model and zero outside it. The return assumption gives Bitcoin a premium; the allocation gives it nothing. That gap is not a modeling error. It is the seam between a research desk that is structurally optimistic and a compliance function that is structurally conservative, and the document is where those two forces get averaged.

Contrarian: the wrong read

The consensus read on this chapter is "bearish." I think that is the wrong read, and it misses the actual event.

The event is not the 10%. The event is that Bitcoin entered the assumptions table at all. For a decade, the institutional posture was abstention — no number, no opinion, no slot in the model. Abstention is the strongest form of exclusion because it requires no justification and generates no record to argue against. Once an institution publishes a formal long-run assumption, the asset has crossed into the category of things that must be priced. Even a negative number is a form of admission. The next edition has to revise the number, not invent it.

The second blind spot is holding-layer centralization. Bitcoin's decentralization holds at the protocol layer and is quietly eroding at the ownership layer. ETF shares, custodied balances, exchange reserves — a growing share of supply sits behind a shrinking set of institutional gatekeepers. That creates a single-point liquidation risk no on-chain model captures, because on-chain models read addresses, not beneficial ownership. The -3.2% usage model is blind to the most consequential structural shift in the asset's ownership profile.

The third is the pricing-power position. JPMorgan does not just analyze the LTCMA — it publishes the standard other institutions build glidepaths from. Leaving Bitcoin out of the assumption table is not a neutral analytical act. It is an exercise of standard-setting authority that keeps Bitcoin out of long-horizon allocation blueprints before any committee ever debates it.

The code doesn't care about your narrative — but the standard that wraps the code decides whether the narrative ever reaches a portfolio.

Takeaway

Watch the next edition's two buried numbers, not the headline. If the on-chain usage model moves toward zero, the framework is conceding that network fundamentals have stabilized, and the 65% macro weight becomes the only argument left standing. If the covariance matrix gains a Bitcoin row, the "no strategic role" conclusion loses its arithmetic foundation and the whole chapter has to be rebuilt from the ground up.

The question is not whether Bitcoin returns 10%. The question is whether an institution can keep publishing a number it derived from M2 and then claim it analyzed a network. The code doesn't negotiate. Neither, eventually, does the market.