What the carry pays, and against what

The most common market neutral construction in digital assets does not try to pick a direction. It holds the asset and sells a future or a perpetual against it, and earns the difference between the two prices as that difference converges toward expiry. The position is built so that whether the asset rises or falls is largely irrelevant to the outcome.

The spread exists for a reason worth stating plainly: somebody is paying for leveraged long exposure, and the carry is what they pay. It is not a yield in the sense of a coupon, and nothing issues it. It is the price of someone else's leverage, and it moves with the demand for that leverage rather than with the value of the underlying.

That has a consequence which is obvious once said and is frequently left unsaid. The carry can compress, and it can compress to nothing, without any part of the strategy breaking. There is no default, no failed hedge and no operational incident. The return source simply pays less — and at some level it pays less than leaving the money in a short-dated government bond.

In late July 2026, Glassnode's weekly on-chain report observed that the "three-month futures basis, the yield on the cash-and-carry trade that anchors institutional participation in crypto, has been paying less than a 2-year Treasury since February".

The comparison is the useful part of that sentence. A cash-and-carry position commits capital for a period and earns a spread; a two-year Treasury commits capital and earns a yield. Setting one against the other is the comparison an allocator is making in any event, whether or not it is made explicitly.

The same report set out how unusual the stretch is, and it is worth quoting exactly rather than in summary: "Only one other stretch on record has run this long, from August 2022 into January 2023, and it ended at the cycle low."

Note what that sentence says and what it does not. It says only one other stretch has run this long — a statement about duration. It does not say the basis has fallen below the two-year on only one other occasion, which would be a different and stronger claim. The distinction matters because the stronger version is the one that tends to survive into a headline, and it is not the one the underlying report makes.

Two weeks later a different reading was published from a different source. Working from the CME settlement bulletin for 7 August 2026, CryptoSlate reported that "August 7 gross futures carry ranged from 5.69% to 7.89%, above the 4.19% Treasury yield" — the August contract at 7.89%, September at 6.25% and December at 5.69%.

The Treasury side of that comparison checks out independently. The Federal Reserve's published series for the two-year constant maturity records 4.19% on 7 August 2026, and the same series sits between 4.15% and 4.25% through the first three weeks of the month.

So one credible source reports that the carry has been below the two-year since February, and another reports that on a specific day in August it was between one and four percentage points above it. Neither is wrong. They are not measuring the same thing.

A carry figure is not a fact on its own. It is the output of a measurement, and three choices inside that measurement move the answer by more than the difference these two readings appear to be arguing about.

None of that is resolved here, and it is not the reader's job to resolve it either. The point is narrower: two accurately reported figures that appear to contradict each other usually differ in how they were constructed, and the construction is something anybody can ask about.

This is the discipline the term "market neutral" already demands. The label establishes nothing until the version is stated — dollar-neutral, beta-neutral or factor-neutral, on what measure, how often and within what tolerance. A carry number is the return side of precisely the same problem, and it has tended to attract less scrutiny than the hedge because it arrives as a percentage and percentages look self-explanatory.

The strategies that earn the basis are not the only thing affected when it compresses. The desks running those positions are also, in the ordinary course, a substantial part of what supplies depth, leverage and volume to the market as a whole. When the trade stops paying more than a risk-free alternative, their reason to be present weakens.

The July report described the downstream effect in the same week. Measured in coins rather than dollars, so that a falling price does not flatter it, "spot volume has fallen to its lowest since 2019", and "the offer side has thinned considerably, with resting sell orders in the same band near their thinnest of the past month".

Its own conclusion is the one worth keeping, because it cuts both ways rather than one: "Thin books cut both ways, and they are often how a quiet market becomes a fast one." A thin book is not a directional signal. It is a statement about how far a given quantity of buying or selling will move the price, in either direction.

For an allocator that is a second-order exposure which appears in no strategy's exposure report. A position sized against normal liquidity is a different position when liquidity is at a multi-year low, and the change has happened to the market rather than to the portfolio.

The questions follow from the construction, and every one of them can be asked before an allocation rather than reconstructed after one.

This note presents no performance data. It contains no return, no drawdown, no correlation and no attribution for any strategy, manager or vehicle, including any of Block Asset Management's, and nothing in it describes how any of them has behaved or would behave.

It is also not a view on the basis. Nothing here predicts whether the carry will widen or compress. The figures quoted are dated at the point they were published — late July and early August 2026 — for the specific reason that a carry number without its date and its measure cannot go stale visibly. They will age, and they should be read as of those dates rather than as current levels.

The two readings above are reported as they were published. They are not reconciled here, because reconciling them would require the construction behind each, and that is not stated in either source.