JPMorgan Added Bitcoin to the LTCMA. The Numbers It Left Out Are the Actual Signal.

The document is five sentences long. It contains no expected return, no volatility assumption, no correlation coefficient, no recommended allocation weight, and no version year. It has already generated more commentary this quarter than any research note published by a systemically important bank.

I have audited proof systems with that shape.

In 2020 I spent four months inside a Groth16 circuit for a privacy lending protocol. Three developers, 500,000 constraint gates, one arithmetic circuit. The defect we eventually isolated was not a broken gate. It was a mismatch in the public input encoding. The circuit computed the correct witness. The verifier checked it against the wrong statement. A proof that verifies against malformed public inputs is not a weak proof. It is a lie carrying a valid signature.

The JPMorgan Long-Term Capital Market Assumptions news has that exact geometry. The claim that circulated — bitcoin outperforms US equities and gold over ten years — verifies. Somebody said it. The clause that gave the claim operational meaning — that volatility and unstable correlation limit its role in a portfolio — did not circulate.

The headline is a proof with a truncated public input. It verifies against a statement nobody made. Zero knowledge, maximum proof — and in this case, maximum ambiguity too.

What the LTCMA Actually Is

The Long-Term Capital Market Assumptions are not a price target. They are an input table.

Every year, JPMorgan's research group publishes expected returns, volatilities, and correlations for roughly sixty asset classes on a ten-to-fifteen-year horizon. Registered investment advisers feed those numbers into model portfolios. Pension consultants use them to build the policy portfolio inside an investment policy statement. Endowments reference them when the finance committee asks why the endowment holds 12% in private credit.

The table is not read. It is computed with.

That distinction is the entire reason this announcement matters more than a normal research note. A price target is an opinion that a human being chooses to act on. An LTCMA entry is a parameter that a mean-variance optimizer ingests automatically. When an asset class appears in the table with a non-zero expected return, it becomes buyable by every systematic allocation process downstream. When it appears at 0%, it is not merely underweight — it is structurally unheld. A 0% asset cannot be purchased by an optimizer at all. It is outside the feasible set.

So the correct reading of the event is not "JPMorgan likes bitcoin." The correct reading is: bitcoin has been moved from outside the feasible set to inside the feasible set. That is a change in the shape of the constraint region, not a change in anyone's mood.

The institutional history matters here because it is monotonic. In 2017 the CEO called bitcoin a fraud and worse than tulips. By 2020 the research desk was publishing digital-gold framing and long-horizon targets. In 2024 the bank was reportedly acting as an authorized participant for spot bitcoin ETFs and running a deposit-token settlement network. In 2025 the asset enters the assumptions table.

That sequence runs in one direction only. Once an asset class is inside a policy portfolio template, it is rarely removed entirely — the same institutional inertia that has kept emerging-market debt at a fixed weight through four separate crises. Call it a ratchet.

There is also a constraint almost nobody discussing this news has mentioned. Under the Basel Committee framework, a bank's crypto-asset exposures carry a 1,250% risk weight. With an 8% minimum capital ratio, that means $100 of bitcoin exposure consumes $100 of capital. Not $8. One hundred. A bank cannot warehouse this asset on its balance sheet at scale. It is arithmetically impossible.

Which tells you what the research posture actually is. JPMorgan's bullishness cannot be monetized through proprietary positioning. It can only be monetized through flow — custody, ETF servicing, collateral management, FX, and the advisory relationships that follow a published allocation. Trust is a bug, not a feature. Assume the position is sized to the business model before you assume it is sized to the model.

The Optimizer Does Not Read Headlines

Here is the mechanical reason the two halves of the JPMorgan sentence belong together.

In mean-variance optimization with a risk-free asset, the tangency portfolio maximizes the Sharpe ratio. An asset enters that portfolio with positive weight if and only if its alpha relative to the tangency portfolio is positive:

(μᵢ − r) − βᵢ(μ_T − r) > 0, where βᵢ = ρ σᵢ / σ_T

Rearrange, and you get the entry condition in terms of correlation:

ρ < [(μᵢ − r) / (μ_T − r)] × [σ_T / σᵢ]

Plug in defensible LTCMA-style numbers. Bitcoin excess return of 6% over cash, equity excess return of 3%, equity volatility of 16%, bitcoin volatility of 55%. The threshold correlation is 2.0 × 0.291, or roughly 0.58.

That is the whole argument. Bitcoin gets a positive weight in the policy portfolio only if its correlation to the equity-heavy tangency portfolio stays below approximately 0.58. Empirically, in calm regimes, that number sits somewhere between 0.3 and 0.5. It clears. Barely. With a margin of safety that would embarrass anyone who has ever sized a position in a market with fat tails.

Now the sentence "volatility and unstable correlation limit its role in the portfolio" stops reading like a hedge. It is the optimizer's output, stated in words. Given a volatility of 55% and a correlation that is not stable, the maximization returns a weight near the floor. The qualifier is not caution. It is arithmetic.

And notice what that arithmetic requires. Three inputs: expected return, volatility, correlation. Two of the three were not published. An optimization problem with two of three inputs redacted is not a judgment. It is a redaction with a conclusion attached.

This is not an unusual failure mode in this industry. I have watched the same defect in lending protocols for years. The utilization kink in an Aave or Compound interest rate model — the 80% threshold where borrowing costs go vertical — is not discovered by the market. It is a governance parameter, chosen by committee, dressed in the vocabulary of equilibrium. A number that looks defensible in a governance forum and behaves arbitrarily in a stress state. The LTCMA correlation assumption is the same species of artifact. It is a parameter selected to be publishable, not a constant measured to be robust.

The Correlation Is the Load-Bearing Wall, and It Fails Under Load

The entire inclusion case rests on one estimated parameter. So the audit question is not "is bitcoin scarce." It is "does that parameter survive regime change."

I ran this class of test in 2022 while dissecting optimistic rollup dispute games — five months simulating malicious sequencer behavior across the thirty-day challenge window to see whether the economic security assumptions held when the sequencer was adversarial rather than honest. The finding was that bond sizes looked adequate in the specification and inadequate in the simulation. The parameter was fine on paper and thin under load.

Correlation has the same property, and it is trivially reproducible. Anyone with a price series can run it:

import pandas as pd, numpy as np

# px: daily closes, columns ['BTC', 'NDX'] r = np.log(px).diff().dropna()

# 90-day rolling correlation: the input an optimizer would see roll = r['BTC'].rolling(90).corr(r['NDX'])

# conditional correlation: equity down days only stress = r[r['NDX'] <= -0.02] rho_crisis = stress['BTC'].corr(stress['NDX'])

# drawdown contribution of a 1% sleeve port = 0.01 r['BTC'] + 0.99 r['NDX'] print(roll.describe(), rho_crisis, port.min()) ```

Run that against March 2020 and against the 2022 tightening cycle. In March 2020, bitcoin fell roughly 39% in a single session while equities collapsed — the correlation did not go to zero, it went to one, at the exact moment diversification was being paid for. Through 2022, the ninety-day rolling correlation to the Nasdaq printed above 0.6 for sustained stretches.

The gap between the full-sample ρ and the conditional ρ is the model risk. Same asset. Two numbers. Two portfolios. An optimizer fed the full-sample value produces a weight. A risk committee fed the conditional value produces a different one, usually zero. The diversification property is estimated in the regime where you do not need it and falsified in the regime where you do.

There is a second-order effect here that almost nobody has priced. The holder this announcement creates is an optimizer, not a believer. Optimizers are mechanical. They rebalance on a calendar. They sell into strength to restore the policy weight. They buy into weakness for the same reason. And they cut to zero when an input parameter changes.

That is a different holder profile from the 2021 cohort, which was reflexive and trend-following. Converting reflexive holders into mechanical holders does something specific to the tape: it reduces trend persistence and increases mean reversion, and it introduces a class of seller whose behavior is triggered by a research publication rather than a price level. That is not obviously bullish. It is a change in the autocorrelation structure of flows.

The One Number in the Thesis That Can Be Proven

Almost every input to a ten-year bitcoin forecast is an estimate. One is a constraint.

The supply schedule is defined in code. Twenty-one million cap. Post-April-2024 block reward of 3.125 BTC. At 144 blocks per day, that is 450 BTC of gross issuance daily, roughly 164,000 annually against a circulating base near 19.9 million — an inflation rate of about 0.83%. At the 2028 halving the reward drops to 1.5625 BTC and the rate falls toward 0.4%.

Gold's annual mine supply growth runs somewhere in the 1.5% to 2% band. Bitcoin's issuance is already below it and falling on a schedule that no committee can revise.

That is the only structural advantage over gold that can be verified rather than argued. It is also, in the flow equation, almost irrelevant. Annual issuance of 0.83% of supply is noise next to the demand-side variance, which is where the entire ten-year forecast actually lives. Code doesn't lie; audits do. And the one input in this model that cannot lie is the one input that carries the least weight in the output.

Which is why the technical roadmap is a distraction from this thesis, and correctly so. The Lightning Network's routing failure rates have been documented for seven years and have never entered an LTCMA-style argument — because the argument is not a technology forecast. Taproot, BitVM, Babylon staking, Ordinals fee spikes: none of them touch a monetary-premium model. The thesis is supply rigidity plus adoption curve plus currency debasement. It cannot be invalidated by a delivery failure and it cannot be strengthened by a delivery success. That makes it unusually clean and unusually unfalsifiable at the same time.

What can move it is flow. Miner issuance of 450 BTC daily is the only programmatic seller in the system — price-insensitive in the sense that miners must sell to cover power contracts regardless of spot. In an inflow regime that supply is absorbed invisibly. In the absence of ETF inflows it becomes the marginal price setter, and it does not care about your correlation assumption.

The Blind Spot: Everyone Is Auditing the Forecast Instead of the Model

A decade of commentary has been generated by the question "will bitcoin beat stocks and gold." That is the wrong question, and it is the question the document was engineered to provoke.

The right question is what the table did not contain.

An expected return without a volatility assumption cannot be converted into a Sharpe ratio. A Sharpe ratio without a correlation assumption cannot be converted into a weight. A weight without a horizon cannot be stress-tested. Strip out those three numbers and what remains is a statement that cannot be falsified by any future price path — because the confidence interval was never specified. An unfalsifiable forecast is not a forecast. It is a distribution instrument.

The DAO was a warning we ignored — not about reentrancy, but about the gap between a system's stated invariants and its actual ones. Every post-mortem of that failure eventually landed on the same observation: the code did what it said, and what it said was not what everyone believed. The LTCMA says something precise. The coverage says something else. The gap is the risk.

There is a second blind spot in the horizon itself. Ten to fifteen years spans at least two halvings. That is not a neutral choice of window. On a three-year horizon, the same model would output a confidence interval wide enough to be useless, because the supply argument would not have time to bite. The horizon is a parameter, and it was selected — the same way the 80% utilization kink was selected — to make the desired conclusion mechanically true.

And the third: "first inclusion" is a signal that cannot repeat. Its information content is consumed on publication. The next data point is "second inclusion," or "weight raised from 0.5% to 1%," and the marginal meaning decays from there. The 2021 cycle produced a $146,000 long-horizon target from the same institution. The market's reaction to that target is the relevant precedent, not the target itself.

What to Watch, and What It Is Worth

The trade is not in the conclusion. It is in the next table.

When the following LTCMA prints, it will carry numbers. At that point the threshold correlation is computable — the formula above, four inputs, no judgment required. If the published correlation assumption sits below the threshold, the implied weight becomes deterministic, and so does the flow: policy weight multiplied by the assets under management in the consulting channel that uses the table as a benchmark input. That number can be sized before the optimizer runs. That is a trade with an auditable edge.

If the numbers stay unpublished, the inclusion is a marketing event and should be priced as one — a slow ratcheting lift in the demand floor, not a catalyst.

Which leaves one question worth asking of the committee that signed the document: when the same table tells you an asset has the highest expected return and the smallest permitted weight, which of those two numbers did you actually believe?

Zero knowledge, maximum proof. Read the table, not the press release. The table is the only part that can be audited.