Open interest

It is the second number every derivatives screener puts next to price, and the one most often read as though it were a measurement of conviction. It is a count of contracts, in units the venue chose, aggregated by a vendor who made a decision you cannot see from the chart.

How it works

The regulator's definition is the clean one. The CFTC's explanatory notes for the Commitments of Traders report define open interest as the total of all futures and option contracts entered into and not yet offset by a transaction, by delivery, by exercise or otherwise, and state that the aggregate of all long open interest equals the aggregate of all short open interest. Both halves matter. It is a stock, not a flow — a level that persists until somebody closes a position — and because every contract has two sides, a market-wide total is one number counted once, not the sum of the longs and the shorts.

That is where the usual confusion with volume starts. Consider the four combinations of a single trade. Both sides opening — open interest rises by one contract. Both sides closing — it falls by one. One opening and one closing, in either direction — it does not move at all, because the position changed hands rather than being created or destroyed. Volume records a trade in all four cases. A day of frantic turnover between existing holders leaves open interest flat, and a quiet day in which a few large positions are built moves it a long way.

The units are not standard, and they are not always stated. Binance's USDⓈ-M futures endpoint for open-interest statistics returns sumOpenInterest and sumOpenInterestValue with no unit annotation on either field. The endpoint of the same name on its COIN-M futures product annotates both: sumOpenInterest is marked "unit: cont", meaning contracts, and sumOpenInterestValue is marked "unit: base asset", meaning bitcoin on a BTCUSD contract. Two products at one venue, two identically named fields, and three different quantities between them — a count of contracts, a quantity of the base asset, and a notional in the quote currency.

This follows from how the contracts are built rather than from a documentation lapse. A coin-margined contract is denominated in dollars and collateralised and settled in the coin, which is why OKX computes the position value of a crypto-margined contract by dividing by the mark price where a USDT-margined one multiplies by it. A count of contracts is only convertible to a notional with the contract multiplier, and only convertible to a coin quantity with a price — and the price used is the vendor's choice as well.

What an aggregate is actually summing

"Total BTC open interest" is therefore a sum over instruments that do not share a unit. To add a coin-margined contract to a USDT-margined one, a vendor has to convert at least one of them at some price at some moment, and the result is a series that moves when the price moves even if not one contract was opened. A dollar-denominated open-interest chart and a coin-denominated chart of the same market can point in opposite directions on the same day, and neither of them is lying.

The vendors here differ in whether they normalise and in whether they say so. Coinglass's API publishes open interest as OHLC candles per pair and aggregated across exchanges, with stablecoin-margined and coin-margined kept separate. Coinalyze lists by contract rather than by coin, so quarterly and bi-quarterly expiries and the coin-margined and stablecoin-margined pairs appear as their own rows instead of being folded into one aggregate — and its per-coin aggregated charts carry the vendor's own notice that they are not real-time, which the chart itself does not show. Velo makes the refusal to merge the product: its figures are broken down by the venue they came from, so "open interest went up" becomes "open interest went up on Bybit while Binance was flat". Its catalogue, counted from unauthenticated endpoints, holds 717 Binance USDT-margined futures products against two coin-margined ones each on Binance, Bybit, OKX and Deribit — which is the shape of the problem in one line, because the two coin-margined rows are not a rounding error in dollar terms.

Why it matters here

Thirty-three of the ninety-two cards in this catalogue mention open interest: fourteen market-data APIs, nine charting platforms, five on-chain analytics products, four terminals and one bot framework. It is one of the most widely claimed fields here, and a claim to carry it says almost nothing about what arrives.

Coverage is thinner than the headline in a way that is visible when a vendor publishes a matrix. Amberdata documents a start date per dataset per venue, and on that matrix open interest exists on 21 rows against funding rates on 16 of 20 futures venues, liquidations on 10 and long/short ratio on 4. The product carries open interest; a particular venue on a particular date may not.

So three questions before a number is used as a number rather than as a shape.

In what unit, and converted at what price. If the vendor cannot answer, the series is only comparable with itself, and only while its denominator is stable.

Which contracts are inside the aggregate. Perpetuals only, or dated futures too; coin-margined folded in or kept out; which venues, and whether delisted ones are still in the archive.

At what moment it was sampled. Open interest is a level, so a series of it is a series of snapshots, and the snapshot interval is a property of the feed rather than of the market.

The reading everybody quotes — rising open interest with a rising price as new longs, rising open interest with a falling price as new shorts — is a convention, not a measurement. No exchange document assigns a direction to the count, and the count cannot distinguish a new position from a hedge against one. It is a reasonable prior and it belongs in the same sentence as its caveat. The other half of the perpetuals pair, funding rate, has the same problem in a different form: a number whose interval the venue chose.