What funding rates say about positioning, and how crowded trades unwind

Perpetual futures are the dominant instrument in crypto, and funding is the mechanism that makes them work. It is also the most misread number on any dashboard. "Funding is high, so everyone is long, so it's going down" is a sentence that has been correct often enough to be believed and wrong often enough to end accounts.

The useful version of the argument is narrower and more mechanical. Funding is the price of leverage. Prices tell you about supply and demand for leverage, not about future returns. Everything worth extracting from a funding series comes from reading it against something else — open interest, spot basis, its own history, or the same rate on another venue.

The mechanism, precisely

A perpetual future never expires. Without an expiry there is no settlement to force convergence with spot, so the contract needs another anchor. That anchor is the funding payment: at fixed intervals, one side pays the other a fraction of position notional.

• —When the perpetual trades above its index, funding is positive and longs pay shorts.
• —When it trades below, funding is negative and shorts pay longs.

The payment flows between traders. The venue is the plumbing, not the counterparty, and it does not collect the funding.

Most venues compute the rate from two components: a premium measuring how far the perp has traded from its index over the interval, and an interest-rate term standing in for the cost of the underlying versus the quote currency. The sum is clamped to a maximum, and the clamp matters — in violent markets funding often sits at its cap, which means the true imbalance is larger than the printed rate can express.

Two contract parameters must be read from the venue's own specification rather than assumed: the interval (eight hours on most large venues, one hour on some others) and the cap. Comparing a rate from an hourly venue with a rate from an eight-hour venue without normalising is the single most common error in funding analysis, and it produces differences of nearly an order of magnitude out of thin air.

Normalise to an annual rate before comparing anything:

annualised = rate_per_interval x intervals_per_day x 365

An eight-hour venue has three intervals a day, an hourly venue twenty-four. Once both series are annualised they are comparable — and the comparison itself becomes informative, which we come back to below.

Funding is a price, not an opinion

The cleanest way to think about a positive funding rate: leveraged long exposure is in demand, and the market is charging for it. That is a statement about positioning and the cost of carry, not about where price goes next.

There is a second, quieter reason funding sits positive most of the time in crypto: the cash-and-carry trade. A participant who holds spot and shorts the perpetual is flat on direction and collects funding while it is positive. That trade is available to anyone with capital on both venues, and it is the main force compressing funding back toward the interest component. When funding stays elevated for long stretches, the interesting question is not "why is everyone long" but "why is the arbitrage not being put on" — the usual answers are capital constraints, venue or custody risk, and the cost of holding collateral, all of which are real and all of which tell you something about market stress.

Funding alone says almost nothing. Funding with open interest says a lot

Open interest is the total notional of contracts currently open. Funding tells you the price of leverage; open interest tells you the quantity. Reading them together gives four states, and the states are genuinely distinguishable:

The table describes how the two series relate mechanically; it is a reading framework, not a measured result.

The distinction between the first and third rows is the one that matters most, and it is invisible if you watch price alone. A rally on rising open interest and rising funding is a rally being financed by new leverage. The same rally on falling open interest is short covering — the fuel runs out when the shorts are gone, and it does not need a catalyst to stop.

Levels are meaningless; percentiles are not

An annualised funding rate of, say, some double-digit percentage means nothing in isolation, because the same number is unremarkable in one regime and extreme in another. The only reading that survives regime change is a distributional one: where does today's value sit in this asset's own recent history, on this venue?

That is what a percentile gives you as a reading — and it is also why we do not print one. A percentile needs a window, and the perpetual-metrics history is not long enough for the window a percentile would have to claim. So the market-structure screen publishes no percentile, no rank and no position in a range, and it prints the measurement that refuses them rather than leaving the absence unexplained. What it does publish for every reading is the window that reading covers and the age of the reading itself. The same discipline governs the public register — every measurement is published with its coverage, including the parts that do not flatter the result.

Two more comparisons carry information:

Cross-venue spread. Once annualised, funding on the four major venues should track closely; persistent divergence means flow is segmented — one venue's participants are positioned differently, or capital cannot move freely between them. That is a structural observation about where the crowding lives, and it is far more specific than a single blended rate.

Funding versus basis. Funding is the perpetual's anchor; the calendar basis on dated futures is the term structure of the same demand. When they disagree, one of the two markets is being pushed by flow rather than carry.

How a crowded trade actually unwinds

The word "unwind" hides the mechanism, and the mechanism is where the risk is.

Leveraged positions carry a maintenance margin: the minimum equity the position must hold. When equity falls below it, the venue liquidates. Crucially, liquidation is triggered against a mark price derived from an index of spot venues, not from the last trade on the perpetual — this exists specifically to stop a single thin wick from liquidating the book, and it means the liquidation trigger lives partly outside the venue you are trading on.

Once triggered, the position is taken over by the liquidation engine and closed aggressively — a market order that takes whatever the book offers. Now connect that to the order book: resting depth is revocable, and it is withdrawn first in exactly the conditions that produce liquidations. The forced order arrives at the moment the book is thinnest.

The loop is then obvious. Forced selling moves price. Moving price crosses the next cluster of maintenance margins. Those liquidate, forcing more selling. This is why liquidation events are not spread evenly across price — they cluster, and the clusters are where leverage was built, which is precisely what rising open interest with rising funding was recording days earlier.

When the price moves faster than the engine can close positions, the losses exceed the posted margin. Venues absorb that shortfall with an insurance fund, and when the fund is insufficient they fall back on auto-deleveraging: profitable traders on the opposite side have their positions closed against the bankrupt ones. Being right about direction does not protect you from being closed out at the venue's convenience — a risk that lives entirely in market structure and never appears in a backtest.

The trap

Here is the trap the opening sentence set up. Elevated funding means leveraged longs are crowded. Crowded positioning raises the conditional severity of a move against that side — the cascade mechanism above is real and the clusters are real. It does not raise the probability that the move happens now.

Positioning is a statement about fragility, not about timing. Funding can stay at its cap for days while price grinds higher, and every trader who shorted the crowding pays the same funding to hold their view. The asymmetry is in the size and speed of the eventual move, not in its arrival date. Treat funding as a risk input — sizing, stop placement, how much leverage you are willing to hold overnight — and it is one of the most useful series in crypto. Treat it as an entry signal and you are paying, every eight hours, for a forecast it never made.

What to do with it

• —Normalise before comparing. Annualise every venue to the same basis. Half the funding "signals" circulating are interval mismatches.
• —Never read it alone. Pair it with open interest to separate new leverage from short covering, and with basis to separate carry from flow.
• —Rank it, don't level it. Percentile within the asset's own history, per venue.
• —Watch the cap. A rate pinned at its maximum understates the imbalance; the printed series compresses exactly when the information is highest.
• —Use it for size, not for entry. Crowding tells you what a move would cost you, not when it starts.
• —Check the clusters, not the average. Liquidation risk is concentrated where leverage was added, and open-interest history tells you where that was.

Funding is one of the few genuinely honest series in this market: it is a price, set by participants, settled in cash, with no room for interpretation about what it measured. The interpretation problem is entirely on the reader's side — and it is solved by refusing to read the number without its context.

Related reading: how to read a level-2 order book for the liquidity side of the same picture, and why most prediction-market traders lose money for what happens when costs are ignored on the other terrain we cover.