What a Fed Rate Move Actually Does to Delta-Neutral Funding Yields. We Checked 17 FOMC Decisions Against 22,000 Hours of Perpetual Funding.

The Committee released its decision at 2:00 PM Eastern this afternoon, with the press conference half an hour behind it. This was a meeting without a Summary of Economic Projections, so there was no dot plot to argue over, only the statement and the questions after it. The target range now stands at 3.50 to 3.75 percent, the effective federal funds rate last printed 3.63 percent, and the three-month Treasury bill was yielding 3.96 percent going in.

Within a few hours, a good number of yield-bearing dollars will have posted something connecting this to their APY. We build one of those, and our published rate is a live function of perpetual funding, so the honest thing to do was measure the relationship rather than assert one. This post is the measurement. It was written before the decision and the analysis in it does not depend on which way the vote went, which is itself the finding.

The idea that Fed policy and the crypto basis trade are linked is not new, and we are not claiming to have found it. Ethena's own risk docs discuss it, research desks write it up around every meeting, and there is a serious academic literature on crypto carry. What is missing is someone measuring it on the funding rate itself, showing the numbers, and reporting the null when the null is what the data gives back. That is the narrow thing this adds.

Where a delta-neutral yield comes from

A delta-neutral synthetic dollar holds a crypto asset and shorts a perpetual future against it, so the price exposure cancels and what remains is carry. The carry has two legs. One is whatever the held asset pays: for an Ethereum liquid staking token that is consensus issuance plus priority fees and MEV, currently running near 2.18 percent on Lido's seven-day average. The other is the funding rate collected by the short leg of the perpetual position, which is paid by levered longs to shorts whenever the perpetual trades above spot.

Our own model, which you can read live at /api/apy and whose full derivation is at /docs/yield-methodology, takes exactly those two inputs: 2.18 percent from Lido and 4.18 percent from Hyperliquid's trailing 180-day ETH funding. Funding is the larger of the two, which is why a question about rates and this kind of yield is really a question about funding. So look at how funding is actually computed.

The interest rate inside the funding formula is set by hand, and it does not follow the Fed

Both venues that matter here compute funding the same way. Binance's USD-margined perpetuals set funding as the average premium index plus a clamp of interest rate minus premium index, bounded at plus or minus 0.05 percent, with a per-contract cap sitting outside that and an interval divisor that equals one for the eight-hour ETH contract. Hyperliquid uses the same core construction, clamp at plus or minus 0.05 percent, payment split hourly. Two terms, then. The premium index is real market information: it measures how far the perpetual is trading from the oracle price, which is to say how much traders will pay to hold levered exposure right now. The interest rate term is where you would expect policy to enter.

It does not, and the tell is that each venue sets it by hand and by contract. On Binance the interest rate for the ETH and BTC contracts is 0.01 percent per eight-hour interval, but some contracts, including BNBUSDT and the ETHBTC pair, sit at exactly zero, and Binance says in its own documentation that it reserves the right to change any of them. Hyperliquid uses the same 0.01 percent, charged hourly at 0.00125 percent, which its documentation describes as roughly 11.6 percent a year paid to the short. The value traces back to the original BitMEX perpetual, where it stood in for the interest-rate differential between the base and quote currency of the pair; BitMEX's own borrow indices behind it have sat parked at 0.03 and 0.06 percent through the entire cycle in which the Fed went from near zero to 5.5 percent and back. Venues do revise the convention around it, OKX changed its funding formula in June 2026, but not one of them has tied the term to a central-bank rate. It is a number an exchange picks, not a number the Fed sets.

You can check this in one call and we would rather you did than take our word for it. Binance publishes the value it is using: request https://fapi.binance.com/fapi/v1/premiumIndex?symbol=ETHUSDT and read the interestRate field. It returns 0.00010000. That is 0.01 percent per eight hours, or about 10.95 percent a year on a simple annualization. Between June 2024 and today the Federal Reserve moved its target range from 5.50 percent down to 3.75 percent, a shift of 175 basis points across six cuts. Across those six cuts, the number in that field changed by zero.

So there is no direct channel, and the clamp makes that sharper than it first looks. Because funding is the premium plus a clamp of interest minus premium bounded at plus or minus 0.05 percent, the moment the premium sits more than about that far from the interest term, the interest term is clamped out of the rate entirely. In an actively trading market, which is most of them most of the time, the clamp drops the constant out of the funding number altogether. None of this means the Fed cannot reach funding. It means the Fed cannot reach funding through the term the formula labels interest. If policy gets there at all it goes through the premium, and the premium is set by the traders who want leverage and by the desks that arbitrage the perpetual against spot with dollar-priced capital. That is a real channel, and it is the one worth thinking about. It is also looser and slower than the word "mechanical" implies, and, as the next two sections show, hard to see in the data at all.

Half the time, that constant is the whole funding rate

The constant turns out to matter more than a footnote about formula design would suggest. We pulled every hourly ETH funding observation Hyperliquid has published, more than 22,000 of them running from early February 2024 to this afternoon, and asked how often the printed rate equalled that constant exactly.

The answer is 51.5 percent of all hours. This is the clamp doing its work, not an emergent property of ETH: whenever ETH tracks spot closely enough that the premium sits within about 0.05 percent of the interest term, the two premium pieces cancel and funding equals the constant exactly. For a bit over half the sample the market was that calm. It follows, and adds nothing to say, that the constant is also the median reading, because a value that fills more than half the hours is the median by arithmetic. The point worth keeping is the mechanism: for most of two years the single most common value of ETH funding was a number an exchange sets by hand and has held at 0.01 percent per eight hours since the perpetual was designed in 2016, entirely independent of anything the Fed did in the same window.

Two caveats belong with that, not in a footnote. This is one venue and one asset, and Hyperliquid's premium construction makes the pinned state especially visible; a venue that builds its premium differently would show a different share, and even here the 51.5 percent reflects this window's particular mix of calm and trending stretches rather than a structural constant. And the pinned state is neither the average state nor the profitable one. Because ETH funding is skewed positive, the full-sample mean sits near 13 percent annualized, well above the median, dragged up by a fat right tail where crowded-long stretches pushed funding into the double and triple digits, with the 95th percentile above 50 percent. It is also not always positive: roughly one hour in seven printed negative funding, where the short leg paid rather than collected. What a carry book actually earns lives in that dispersion, not at the mode, and quoting the constant as the typical rate would understate what a long really pays.

All of that is the two-year picture. What makes this particular week worth separating out is that the pinned state broke in the last few days, and the direction it broke in is the one nobody would guess. Over the seven days to publication, 50 percent of Hyperliquid's settled hourly prints were exactly the constant and the other 50 percent came in below it. Not a single hour printed above it. The seven days before that were pinned in every hour of the window. So the constant has stopped being the value funding sits at and become a ceiling funding is falling away from: the last settled hour read 6.11 percent annualized against a constant worth 10.95 percent, twelve of the 168 hours were outright negative, and the seven-day mean is 7.87 percent. The premium index is negative as we publish, meaning the perpetual is trading below the index price, which is the market saying it does not especially want to be long.

Read that fall in the pinned share carefully, because the intuitive reading of it is backwards. A premium index recovering, traders paying up for leverage again, would push prints above the constant and unpin the series upward. These went the other way. What unpinned the series was the premium turning negative enough to escape the clamp on the downside, which is the market withdrawing rather than returning. A week ago the honest description of Hyperliquid's headline was that it was the venue's hand-set number with almost no market on top. Today the honest description is worse than that, because the market is now subtracting from the constant rather than adding nothing to it.

Take the one-venue caveat seriously and the picture gets worse rather than better. Binance runs the same nominal 0.01 percent per eight hours on its ETH contract and the same clamp bounds, but it builds its premium index from its own order book rather than from the gap between a mark price and an oracle, and the printed rates come out somewhere else entirely. Compare the identical trailing 180 days on the identical asset. Hyperliquid sat exactly on the constant for 40.5 percent of its hourly prints and averaged 4.18 percent annualized. Binance sat exactly on it for 2.4 percent of its eight-hourly prints, had 41.9 percent of them come in negative, and averaged 0.25 percent. Same asset, same window, same nominal constant, and a gap of nearly four percentage points that says nothing about the price of ETH carry and a great deal about how each venue computes its premium. A single funding number quoted as though it were a market rate is a quote about a venue.

Seventeen decisions, and no detectable response

Mechanism is one thing and evidence is another, so we ran the event study. For each FOMC decision since June 2024 we stamped the release at 18:00 UTC, which is 2:00 PM Eastern, and compared mean annualized ETH funding over the 168 hours before against the 168 hours after. The script that produces this table is in our repository at scripts/fomc_funding_event_study.py and takes no private inputs, so you can rerun it and check us.

The average change across all 17 is minus 1.81 percentage points, with a standard deviation of 12.42 and a range running from minus 23.67 to plus 28.79. Seven went up and ten went down. Split by outcome, the six cuts averaged plus 0.18 points and the eleven holds averaged minus 2.89, a difference of 3.08 points against a standard error of 7.55, which is a t-statistic of 0.41. Put a confidence interval on that difference and it runs from roughly minus 12 to plus 18 points. The honest reading of that is not that cuts do nothing; it is that a sample this size is equally consistent with cuts lifting funding a lot, cutting it a lot, or leaving it alone. The two largest moves in the whole table both follow cuts and point in opposite directions thirteen months apart, which is what a non-result looks like up close.

Be careful about what this can and cannot establish, because the weaknesses matter more than the point estimate. The power comes from 17 events, not from the 22,000 hours; funding an hour apart is almost the same number, so the hourly count is data density, not independent evidence, and seventeen events cannot detect a modest effect. There is a deeper problem than sample size. A market is supposed to react to the surprise in a decision, not the decision itself, and by the time the Committee speaks most of what it will do is already priced. This test compares realized cuts against realized holds, not surprises against expectations, so a fully anticipated cut shows up here as a non-event and biases the estimate toward zero by construction. Averaging over seven days rather than the minutes when a surprise actually moves a market pushes the same way. Every decision in the window was a cut or a hold, so this says nothing about a hiking cycle. And a fortnight around each announcement sweeps up everything else that happened in crypto in those two weeks. Read the null as a floor on how hard the effect is to see, not as proof there is nothing to see.

The level relationship does not rescue the popular story either, though it is the wrong test to run. A claim that cuts lift funding is a claim about changes, so correlating the level of the policy rate against the level of funding answers a different question from the one being asked. Run it anyway and the correlation is positive, around plus 0.5, which if anything is the opposite sign to "lower rates, higher funding." Put no weight on that. It is not statistically distinguishable from zero once you notice how few genuinely independent rate regimes the window holds, with most months parked at one of three flat values, and both series ran through the same single crypto cycle, which will manufacture a correlation on its own. We mention it because leaving it out would be the more flattering choice, not because it decides anything.

What a rate decision does change

Something has to be true here, and there is a real answer. It is smaller than the marketing and it runs through channels worth separating.

The clearest one to state is not a yield effect at all, and it is only true with everything else held still. A Treasury bill is the alternative to any dollar-denominated yield, and its yield tracks policy closely. Hold the carry fixed and move the policy rate 25 basis points, and the gap between a carry yield and the risk-free rate moves by roughly that much in the opposite direction while the carry earns exactly what it earned yesterday. That is arithmetic on the benchmark, and the cut version of it is the effect most likely to be shown around a meeting as though a strategy improved. It is worth being clear which direction is actually live here, because the reflex in this category is to write about cuts whatever the calendar says. Going into this meeting the market gave a cut close to no weight at all, the standing alternative to a hold was a rise, and the three-month bill was yielding 3.96 percent against a target range topping out at 3.75 percent and an effective rate of 3.63 percent. A bill trading above the top of the range is the cash market leaning toward tightening. Run the identity in that direction and it is unflattering: a higher benchmark narrows the spread every dollar in this category is quoting, without any of them earning less. Two further things keep the identity from being the whole story. Everything else is not held still, because the funding leg moves too, usually by more, which the rest of this section is about. And some dollar products here hold actual Treasury bills or money-market assets in their backing, so a policy move drags part of their yield along with it, and the spread they quote has the same moving bill rate on both sides of it. The identity is real. It is not a windfall in either direction.

The second channel is slow and points in the direction nobody advertises. The short side of a cash-and-carry trade is supplied by capital that could otherwise sit in bills, so the risk-free rate is that capital's hurdle. Lower it and the trade should clear at a lower funding rate over time, because more capital is willing to run it for less. Raise it and the same logic runs backwards, with arbitrage capital demanding more to show up, which is the one channel in this whole piece that would argue for wider funding after a tightening. Present either direction as a direction, not a measured force. Our 17 meetings cannot separate it from everything else moving, and the one serious piece of academic work here argues against leaning on it hard: a Bank for International Settlements study of crypto carry finds that interest-rate differentials explain very little of the variation in the crypto basis, with leverage demand and the frictions on arbitrage capital doing almost all of the work. If the risk-free rate were the lever the popular story treats it as, the basis would show it, and it mostly does not.

The third is fast and points the other way. A surprise, meaning a decision or a tone different from what was priced, moves risk appetite within hours, and levered long demand is exactly what the premium index measures. This is why a dovish shock can lift funding for a few days while the arbitrage-supply effect drags it lower over a few quarters, and why a hawkish one can knock funding down on the day while arguing for the opposite later. The two are not in conflict; they operate on different clocks. Note what the trigger is: the surprise relative to what was priced, not the level of rates. A cut that was fully priced a month ago is not news to anybody's position.

And the staking leg, roughly a third of the gross input in our own model, has close to no rate exposure at all. Ethereum's issuance is a function of how much ETH is staked, and priority fees plus MEV are a function of how busy the chain is. Neither has a term for the federal funds rate. Policy can touch on-chain activity the way it touches all risk-taking, but there is no transmission mechanism worth modelling, and anyone claiming a Fed decision moved their staking yield should be asked to draw the arrow.

The comparison that actually matters this week

Set the decision aside, because the more important thing about this week is a fact we would rather state ourselves than have someone else point out. Perpetual funding is compressed right now. It is not elevated, it is not healthy, and a post from a funding-linked protocol that implied otherwise on a day this many people were reading would be exactly the sort of convenient framing we spend our time criticizing in other people.

Here is the state of it, read on the day. The headline number on Hyperliquid is 6.11 percent annualized, which is not a working carry environment, and it now sits below the venue's own hand-set constant rather than on it. The premium index is negative as we write. On Binance, where the premium has more room to speak, the current print annualizes to 4.89 percent and the two most recent settled prints came in at 7.52 percent and 2.35 percent. Both of those are positive, which is a change from earlier in the week and we would rather say so than quote the version of this sentence we drafted six days ago. They are also small enough that the change carries very little: a negative print sits three days back at minus 4.00 percent annualized, and across the trailing 180 days 41.9 percent of Binance's settled prints have been negative. Negative funding is not a curiosity for a delta-neutral book. It means the short leg pays instead of collecting, so in those windows the hedge costs money rather than earning it, and a structure whose entire pitch is carry runs the carry backwards.

The category leader shows the same thing at a scale no one can call venue-specific. Ethena publishes sUSDe's yield on its own API, and it reads 4.00 percent with a thirty-day average of 3.92 percent, against an average since inception of 10.86 percent. One caveat on those three, because this post's whole claim is that its numbers were read on the day: Ethena stamps that payload with its own last update of July 22, so they are the issuer's most recent published figures rather than a same-day read, and we would rather flag the stamp than let it pass. That is the largest delta-neutral dollar in the market yielding a bit over a third of its own lifetime average. Three-month bills were at 3.96 percent going into the meeting, so the biggest product in this category is yielding roughly four basis points more than cash, which is not a premium anyone would take risk for. When the leader and the venue tape agree, the compression is structural rather than anybody's individual problem.

None of that has anything to do with Wednesday. Funding has been compressing for over a year as the basis trade got crowded, which is visible in the same public series this post runs on: the trailing two-year funding carry on ETH annualizes to 9.4 percent, the trailing year to 6.45 percent, and the trailing 180 days to 4.18 percent on Hyperliquid and 0.25 percent on Binance. Ethena spent this year moving the bulk of USDe's backing out of the basis trade toward lending and real-world assets, which is a rational response to that compression and which we wrote up in our piece on the category shrinking at par. A 25 basis point move in the policy rate is a rounding error next to a funding regime that has come down that far. If you want one number to take away from today, it is not the target range. It is that the carry this whole category is built on is currently thin, and the Committee has very little to do with why.

What happened today

The Committee held, which is what the market had priced. Note which way the other side of that pricing pointed, because most of the commentary today will get it backwards out of habit: the live alternative going into this meeting was a hike, not a cut. The vote settles that rather than leaving it to inference. The statement was approved 9 to 3, and all three dissents, from Beth Hammack, Neel Kashkari and Lorie Logan, were votes to raise the target range by a quarter point at this meeting. Not one member dissented toward easing. The statement also kept the line that inflation remains elevated relative to the Committee's 2 percent goal and added that the Committee will deliver price stability. The three-month bill had been signalling the same thing for weeks by trading above the top of the target range, and it went in at 3.96 percent against a 3.75 percent ceiling, a gap that widened over the past week rather than closing. So a hold is the least interesting outcome for a carry yield twice over. Nothing about the two legs changed at 2:00 PM, and the benchmark those legs get measured against did not move either. If anything the hawkish delivery argues mildly against levered long demand, and levered long demand is what actually prices funding. If the next seven days look like the eleven other holds in our sample, the honest expectation is a funding change somewhere inside a band well over ten points wide in either direction with no reliable sign to it. That is not a forecast to be proud of. It is the width the data supports.

Here is what we will be checking over the next seven days, written so you can hold us to it. ETH funding on Hyperliquid read 6.11 percent annualized at publication, below the venue's own constant, on a seven-day window that was half pinned to that constant and half underneath it. Binance read about 4.89 percent with its last two settled prints small and positive. Our prior, from the table above, is that the seven-day average lands within about 12 points of where it started with no reliable direction, and that whatever it does will be better explained by liquidation flow and spot direction than by the Committee. The single most useful thing to watch is Hyperliquid's pinned share, because it fell from every hour to half of them inside one week with nothing at all printing above the constant, and whether it keeps falling is a cleaner read on whether anyone wants leverage than the funding headline itself is. That is also the number a rate decision has the least claim on. If funding moves sharply and cleanly in the direction the consensus take predicted, that is evidence against the position in this post, and we will say so on the same page rather than quietly not mentioning it. The data to settle it is public and the script is in our repository.

Where Kerne sits, and how to check it

We say where we stand so this is not an abstraction, and nothing here is investment advice or an offer of any token. Our published rate is 2.92 percent, and every qualifier on that number matters. It is a model of the book we actually run rather than a record of what has been distributed: the engine sizes its short one for one against on-chain spot, and the leverage it posts margin at reduces the margin required rather than multiplying the carry earned on the underlying, so the multiplier on the carry sits below one. A higher levered figure was published beside it under its own label, a modeled target at scale, until July 28, 2026, when it was withdrawn because reaching it would have needed a spot leg levered through a borrowing facility we do not operate. The deployed figure is now the only forward rate we publish. Our own published feasibility analysis puts the sustainable through-cycle band for this design at 8 to 9.4 percent, and that band sits above the published rate rather than under it, because it assumes leverage and a funded insurance buffer this deployment has not reached; the derivation of both bounds, with the dated record of how the basis was corrected, is at the feasibility post. The first on-chain skUSD yield distribution happened on July 8, 2026, and realized distributions to date are small against the modeled rate. All of that is served by the same endpoint the model runs on, at /api/apy, and we would rather you read it there than take a number off a page we control the framing of.

There is a specific thing in that model this post obliges us to say out loud. The funding input is Hyperliquid's trailing 180-day ETH funding, currently 4.18 percent, and Hyperliquid is the venue this piece just spent two sections showing is the one where a hand-set constant does most of the work. The same 180 days on Binance annualized to 0.25 percent. We use the Hyperliquid series because Hyperliquid is where the hedge actually runs, so it is the funding our book actually earns rather than a number chosen to look better, and we would make the same choice again. It remains true that the venue we hedge on is the one that produces the friendlier input, that the gap between the two venues over this window, near four points, is wider than the three-point cut-versus-hold difference the event study could not tell apart from zero, and that a reader comparing our modeled rate against a competitor's should know which venue's funding sits underneath it. One detail cuts the other way and belongs here for exactly the same reason, which is that we checked it rather than that it helps: the trailing 180 days is the lowest of every window we ran on this series, below the trailing 30 days at 9.93 percent, the trailing year at 6.45 percent and the trailing two years at 9.4 percent, so the window our model publishes from is not the flattering one available to us. Our published rate is a model and the funding leg of that model is currently thin. Both of those statements are on the same endpoint, and we would rather write this paragraph than have someone else write it about us.

The rest of the honest boundary is the one we disclose everywhere. The hedge runs on Hyperliquid, a single venue, as a live pilot-scale short sized against a small disclosed founder-custodied float, and that leg is self-reported and signature-bound rather than independently re-derivable on-chain; an independent attestation of it is being scoped. Our first external audit, by Hexens, delivered its initial report on July 20, 2026, with remediation underway and the final report still pending, so we are not an audited protocol yet and do not describe ourselves as one. The on-chain collateral is a different story and you can check it without us: it lives on Base, reads with raw calls, and we publish an hourly proof of reserves signed with an EIP-191 key at /verify and /api/por. Because yield claims decay faster than reserve claims, the advertised-versus-realized board across the whole field is at the honesty index, with our own numbers listed first and, at the moment, worst.

The reason we wrote this rather than a post about our APY is that the funding rate is the load-bearing input in our model and we would rather be the ones who told you it is weakly connected to the thing everyone will be watching this afternoon. A yield that depends on perpetual funding depends on how crowded the basis trade is and on how badly people want leverage. Those are worth tracking. The federal funds rate is worth tracking for what it does to the alternative you are measuring us against.

Figures are as of the publication date and nothing here is investment or economic advice. FOMC meeting dates, the 2:00 PM Eastern release convention, and which meetings carry a Summary of Economic Projections are per the Federal Reserve's published calendar; the effective federal funds rate is the H.15 release and the target range history is FRED series DFEDTARU, with the three-month bill from DGS3MO. The Binance funding formula, the plus or minus 0.05 percent clamp, and the 0.01 percent per eight-hour interest rate applied to its crypto-major contracts are per Binance's own funding-rate documentation and are readable live at its /fapi/v1/premiumIndex endpoint; that interest rate is set per contract and administratively, sits at zero on some pairs including ETHBTC and BNBUSDT, and Binance reserves the right to change it. Hyperliquid's formula, its 0.01 percent per eight hours charged hourly at 0.00125 percent, and its own description of that as about 11.6 percent a year paid to the short are per the Hyperliquid documentation; 11.6 percent is that constant compounded hourly, while the 10.95 percent used above is the same constant annualized simple. The funding series is Hyperliquid's public fundingHistory endpoint, hourly, more than 22,000 observations from early February 2024 through the publication date and growing as the series does, annualized simple rather than compounded; the event study stamps each decision at 18:00 UTC and compares 168-hour windows either side, and its power comes from the 17 decisions, not the hourly count; the full method, its stated limitations, and the code that produces every figure are in scripts/fomc_funding_event_study.py in our repository, which takes no private inputs. The correlation between the target range and monthly mean funding is a level relationship on trending series, is not statistically distinguishable from zero, and is the wrong test for a claim about changes; we report it as descriptive only, not causal. The finding that interest-rate differentials explain little of the crypto basis is per the Bank for International Settlements working paper on crypto carry (Schmeling, Schrimpf and Todorov). The sUSDe current, thirty-day and since-inception figures are Ethena's own published numbers, read from its public yields endpoint rather than from a third party, and cross-checked against DefiLlama as carried in our /api/apy market context; where the two disagree we quote the issuer's own figure. The Binance settled funding prints and the trailing 180-day venue comparison come from /fapi/v1/fundingRate?symbol=ETHUSDT, annualized simple as the printed rate times three times 365, against the same 180 days of Hyperliquid fundingHistory annualized as the printed rate times 24 times 365; the pinned shares quoted for each venue count prints equal to that venue's interest constant exactly, and the two series are compared over the identical window and asset. The Lido staking figure is the seven-day SMA from Lido's public API. Kerne's own claims resolve to live endpoints: the modeled yield and its framing at /api/apy, the hourly signed Proof of Reserves at /api/por/signed, its on-chain leg at /api/por, and the live risk surface at /api/risk-status. A /verify pass proves an attestation is authentic and fresh; it is not an audit and not a solvency opinion.