The 65% Problem Inside JPMorgan's 10% Bitcoin Assumption

Over the past seven days the $83,000 level has done more narrative work than any single price print this quarter. It has been cited as a floor, as a capitulation target, and — most usefully — as the launchpad for a forecast that nobody at JPMorgan actually wrote.

On October 20, one post from an analyst whose tweets move allocation committees put the Long-Term Capital Market Assumptions back into circulation. Within forty-eight hours the number had metastasized into a headline: JPMorgan sees Bitcoin at $215,000 by 2036. The spot reference was $83,000 at the end of September. The math checks out to the dollar. The attribution does not. Hype is the signal; silence is the warning — and the silence here is the part where nobody asked who multiplied what by what, and whether the bank's own framework would survive that multiplication if you unwound it.

I have spent the last week unwinding it. What follows is not a price call. In a market where the average drawdown from cycle high is now deep enough that survival has replaced upside as the primary question, a ten-year terminal value is entertainment. What matters is the structure of the assumption — because that structure is what a family office in Riyadh, a pension committee in Toronto, and a treasury desk in Singapore will all copy into a memo they will not read again for three years.

Context: What the LTCMA Is, and Why It Binds

The Long-Term Capital Market Assumptions are not research. They are plumbing. JPMorgan Asset Management publishes them annually across hundreds of asset classes with ten-to-fifteen-year horizons, and they enter the institutional world through a very specific door: the Investment Policy Statement. A CIO cannot justify a 1% digital-asset sleeve to a board by pointing at a chart. They justify it by citing an assumption set that has a bank's name on the cover and a decade of editions behind it. The 2027 edition is now in circulation. A 2026 family-office survey landed alongside it. The two documents will be stapled together in more decks than either author will ever see.

The edition carries three frameworks for Bitcoin. A monetary-supply framework at a 65% weight, with a 10.7% base-case annual return. A gold-substitution framework at a 30% weight, also at 10.7%. A network-adoption framework at a 5% weight, at negative 3.2%. The blended base case is 10.0%, with a bull case of 16.6% and a bear case of minus 11.2% annualized.

That is the entire factual payload. Everything else in circulation — the $215,000, the $386,000, the $25,000 — is downstream arithmetic performed by someone outside the bank. I want to be precise about this because it is the single most important structural fact in the story, and it is the one fact that every headline buried.

The Arithmetic Is Correct. That Is the Problem.

Let me verify the compounding before I attack it. At $83,000 and 10.0% annualized, the ten-year factor is 1.10^10, or 2.594, giving $215,280. At 16.6%, the factor is 4.647, giving $385,700. At minus 11.2%, the factor is 0.305, giving $25,307. All three reconcile to the published figures within rounding.

So the arithmetic is clean. The arithmetic is clean and the number is still fiction, because the starting price is an editorial choice. JPMorgan's framework produces a rate of return. The media applied that rate to a spot price the bank never specified, compounded it over a decade the bank frames as a range, and published a terminal value with five significant figures. That is not a forecast. That is a spreadsheet with a headline.

I have seen this exact failure mode before, from the inside. In late 2017 I was auditing ICO whitepapers for a Riyadh-based venture fund — more than forty of them in a single quarter — and the tell was always the same. The models that knew the least carried the most decimal places. Three-decimal stoichiometric supply curves. APY projections to the second decimal, built on emission schedules that assumed a demand function nobody had tested. My recommendation to halt three of those launches was not based on finding an error in the arithmetic. It was based on noticing that the precision was load-bearing where the assumptions were not. I saved the fund $2.5 million by treating decimal places as a warning sign rather than a credential.

The $215,000 figure has the same fingerprint. It is arithmetically flawless and structurally meaningless, because it launders a rate assumption into a price target through a starting value that carries no analytical weight at all.

Reverse-Solving the Weights: 95% of the Model Is One Bet

Here is where the structure gets interesting. The framework-level numbers are published, which means the weights can be reverse-engineered. Run the blend:

0.65 × 10.7% = 6.955%

0.30 × 10.7% = 3.210%

0.05 × (−3.2%) = −0.160%

Sum: 10.005%, which rounds to the published 10.0%.

The weights reconcile exactly. And that reconciliation is the most damning thing in the document, because look at what it reveals: two of the three frameworks produce an identical 10.7% base case. The monetary-supply framework and the gold-substitution framework are not independent inputs. At the base case they are the same number wearing two hats. Which means 95% of the model's weight — 65 plus 30 — is a single proposition: that Bitcoin's long-run return is a function of monetary debasement and hard-asset substitution. The three-framework architecture is decorative. It is one bet, expressed in triplicate, with a 5% hedge bolted to the side.

That matters for a specific reason. Institutional readers scan for diversification of method. Three frameworks reads as three independent lines of evidence. What is actually there is one line of evidence, tested twice with the same answer, plus a token allocation to the only framework that actually models Bitcoin as a network rather than as a macro asset. And that token allocation — the 5% sliver — prints at minus 3.2%.

Now push on it. Raise the network-adoption weight from 5% to 10% and rebalance proportionally, holding every framework's base case constant. The new blend is 0.60 × 10.7% + 0.30 × 10.7% + 0.10 × (−3.2%), which is 9.31%. The terminal factor drops from 2.594 to 2.436, and the $215,000 becomes roughly $202,000.

Read that again. The model gets more bearish the more it looks like Bitcoin. Every step you take toward weighting the framework that actually describes the asset — adoption curves, transaction demand, network effects — moves the answer down. The 10% is not a Bitcoin forecast. It is a dollar forecast with a Bitcoin wrapper, and the wrapper is the only part with a negative sign.

This is the same structural error I flagged during the Curve wars in 2020, when 3CRV's stablecoin dominance was being read as a signal of protocol health rather than a signal of what the incentive schedule was paying for. The lesson from that cycle was mechanical: narratives in this asset class are downstream of tokenomics, and a number's composition tells you more than its magnitude. Apply that here. The composition of JPMorgan's 10% is 95% monetary theory and 5% network reality, and the network reality is negative. Anyone quoting the 10% without quoting that split is quoting a conclusion they have not audited.

The Volatility Drag Nobody Quoted

There is a second layer of silence, and it is more technically embarrassing than the first.

The 10.0% base case is a geometric return — a compound annual growth rate. Geometric returns and arithmetic expected returns are not the same thing, and the gap between them is the variance drag, approximately σ²/2. Bitcoin's realized volatility across any multi-year window you care to sample sits between 55% and 75%. Take 60%. The drag is 0.60²/2, which is 18 percentage points. Take 70%. The drag is 24.5 points.

Which means a 10% geometric base case implicitly asserts an arithmetic expected return somewhere between 28% and 34.5% per year. That is the actual claim embedded in the document. It is a far more aggressive statement than "10% annualized," and not one outlet in the coverage chain reported it, because the drag never appears in a summary table. The geometric number is the one that fits in a chart. The arithmetic number is the one that would have forced a conversation about whether the assumption is defensible.

I ran this decomposition repeatedly for institutional clients after 2022, when the algorithmic-stablecoin unwind taught everyone the same lesson in a single week: a model's survival depends entirely on whether its assumptions are stated in the form in which they will be tested. Terra's peg was not broken by an unforeseeable event. It was broken by a reflexive demand assumption that had been present in the design from day one, disclosed in prose, and ignored in every dashboard because the dashboard showed a number that only went up. The drag calculation is the prose that gets ignored. The 10% is the dashboard.

The Bear Case Is the Only Honest One

Minus 11.2% annualized compounds to a factor of 0.305. On an $83,000 start, that is $25,307 in 2036 — a cumulative drawdown of roughly 69.5% over ten years, with the pain front-loaded in a way the annualized figure conceals.

Most coverage treated this as a tail. It is not a tail. The bear case is what the model produces when the 5% framework is the correct one — when Bitcoin's return is determined by network adoption and adoption plateaus, and the gold-substitution story never converts because the marginal institutional allocator keeps treating it as a risk asset rather than a reserve asset. That is not a catastrophic scenario. That is a boring one. It is the scenario where Bitcoin becomes a permanently high-beta Nasdaq proxy with a decaying narrative premium, and it produces minus 11.2% annualized for a decade.

The reason this matters right now, at $83,000 in a market where survival has replaced upside as the operative question, is that the bear case is the only one of the three that maps onto observable conditions. ETF flows have concentrated into two issuers. Basis trades have compressed. The marginal buyer of the last eighteen months was an allocator who bought a narrative, and allocators who buy narratives do not average down — they rebalance on a calendar and sell on a drawdown threshold. The bear case is not a prediction of failure. It is a description of what happens if nothing new happens.

The Anchor Problem: $83,000 Is a Date, Not a Valuation

The terminal value is a linear function of the starting price, which is the least examined input in the entire chain. Multiply the base-case factor of 2.594 by any anchor you like. At $83,000, the answer is $215,000. At $110,000, the same model produces $285,000. At $60,000, it produces $156,000. Nothing in the framework changed. Only the date changed.

This is the mechanical defect in every "bank sees X by year Y" headline ever written about a rate-of-return assumption. The rate is the analysis. The price is the timestamp. And because the timestamp is set by whatever spot print happened to be convenient when the piece was filed, the published target inherits the volatility of that print rather than the stability of the model. A 30% move in spot rewrites the decade-ahead target by 30% while the underlying thesis is untouched.

Elasticity runs the other direction too. The 65% weight sits on the monetary-supply framework at 10.7%. Move that framework's assumption by 100 basis points and the blended base case moves by 65 basis points — from 10.0% to 10.65% — which pushes the terminal factor from 2.594 to 2.751 and the terminal value up by roughly $13,000. A single percentage point of disagreement about global money-supply growth is worth thirteen thousand dollars of 2036 price target, and no reader of the coverage chain was told that the model has a 65% concentration in a variable that has been mis-forecast by every major house since 2008.

What a 1% Allocation Actually Buys

Strip the rhetoric and the LTCMA has one operational function: it tells a policy portfolio whether a digital-asset sleeve clears the bar. So let me run that bar.

Take a standard balanced portfolio at 7% expected return and 10% volatility. Add a 1% Bitcoin sleeve at the 10% base case and 60% volatility, correlated at 0.3. The return contribution is 10 basis points. The risk contribution, computed through the covariance term, is roughly 10 basis points as well. The trade is approximately Sharpe-neutral at 1%.

Sit with that. At the bank's own base case, the allocation it implicitly endorses neither helps nor hurts a balanced portfolio. It is a rounding error with a governance cost. The sleeve only becomes accretive if the return clears roughly 12% or the correlation stays below about 0.35 — and correlation in a liquidity event goes to one, which is precisely when the sleeve is being marked.

This is the arithmetic that institutional allocators actually run, and it is the arithmetic that explains why the LTCMA numbers cluster where they do. The output is not a forecast of Bitcoin. The output is a number calibrated to survive a policy committee.

The Contrarian Read: This Is a Permission Slip, Not a Forecast

The consensus interpretation of the 2027 edition is that JPMorgan turned bullish. That reading is wrong, and the wrongness is instructive.

The bank did not turn bullish. It turned governance-compatible. Follow the incentives, because the incentives here are unusually legible. An assumption that is wrong in the middle of the distribution is unpunishable — nobody can adjudicate a ten-year base case in year three. An assumption that is wrong at the tails is career-ending, because tails are the only part of a forecast anyone remembers. The rational strategy for any institution publishing a decade-ahead number is therefore to cluster near the middle, borrow the credibility of adjacent asset classes, and leave the tails wide enough to cover the eventual outcome. Two frameworks landing on an identical 10.7% is not a coincidence of independent analysis. It is the signature of a model being tuned for defensibility rather than for accuracy.

The second half of the contrarian read is darker. The only framework in the document that actually models Bitcoin — the adoption curve, the network effects, the thing that would capture agent-driven transaction demand if the 2025 convergence thesis holds — carries 5% of the weight and a negative sign. That is the tell. The document is not silent about Bitcoin's network value; it is explicit about it, and it is explicit that the network value is negative. Everything above that line is a dollar call.

Silence, again, is the warning. Not the silence of the bank — the bank published its weights. The silence of everyone who read the weights and quoted the blend.

What to Watch in the 2028 Edition

Three things, and none of them are price.

Watch the weight on network adoption. If it moves from 5% to 10% while the headline base case holds at 10%, the model has quietly re-based and the arithmetic above tells you exactly how much terminal value that costs. A rising adoption weight with a stable base case is only possible if the monetary framework's assumption is raised to compensate — which would be an admission that the two are not independent after all.

Watch whether any edition discloses the arithmetic-versus-geometric distinction. A model that publishes a geometric base case without publishing the implied arithmetic expectation is asking to be misread, and it is being misread at scale right now.

And watch whether the $83,000 anchor gets replaced with a valuation framework. Because a target that moves one-for-one with the spot price at the moment of publication is not a target. It is a mirror.

If the only framework that actually models Bitcoin carries five percent of the weight and a negative sign, whose forecast are you buying when you quote the ten?