What a long/short ratio actually counts

Somewhere on every derivatives dashboard there is a number labelled "long/short ratio", and it is usually read as though the market were leaning one way. It cannot be. Every futures contract is made by one party agreeing to buy and another agreeing to sell, so on any contract the longs and the shorts are the same quantity by construction. What the ratio counts instead is a decision the venue made, and the three largest venues made it three ways.

How it works

The Commodity Futures Trading Commission's notes on its own positioning reports say it in one sentence: "The aggregate of all long open interest is equal to the aggregate of all short open interest." Open interest is one total, not two.

So a ratio of longs to shorts, taken over everyone holding a contract, is always exactly one if it counts contracts. Any published ratio that is not one is therefore counting something else, and there are only two somethings available:

• Accounts, not contracts. Count the people who are net long and the people who are net short. An account with one contract weighs the same as an account with ten thousand. This number can be anything, and it has an arithmetic consequence: on one venue's contract, if more holders are long than short while the contract totals are equal, the average short is bigger than the average long.
• Contracts, but for a slice. Take a subset of traders the venue selects and compare their long contracts with their short contracts. This can differ from one because the subset's counterparties are outside it. If the selected group is net long by some quantity on a contract, everyone else on that venue's contract is net short by the same quantity.

Neither is a reading of the market's direction. One is a headcount and the other is a partition, and in both cases what you are looking at depends on how the venue drew the line.

What each venue counts

Bybit

One series, over everyone. The V5 endpoint GET /v5/market/account-ratio describes itself as "the net long and short positions as percentages of all position holders during the selected time" and gives the arithmetic: "Long account ratio = Number of holders with long positions / Total number of holders", the same for short, and the long-short account ratio as one divided by the other. The response carries buyRatio and sellRatio, not the ratio itself — you divide.

It covers linear and inverse contracts, at periods of 5min, 15min, 30min, 1h, 4h and 1d, up to 500 rows a page with a cursor, and states that "the earliest query start time is July 20, 2020". It also warns that during extreme volatility the interface "may experience increased latency or temporary delays in data delivery".

What the page does not say is how a holder with a long and a short open at the same time is counted. That is an unaddressed case, not a known answer.

Binance

Three series on USDⓈ-M futures, two populations, and a short memory.

• globalLongShortAccountRatio — described on Binance's developer portal only as "Query symbol Long/Short Ratio", returning longAccount, shortAccount and longShortRatio. The page does not define the population beyond the endpoint's name.
• topLongShortAccountRatio — "The proportion of net long and net short accounts to total accounts of the top 20% users with the highest margin balance. Each account is counted once only."
• topLongShortPositionRatio — "The proportion of net long and net short positions to total open positions of the top 20% users with the highest margin balance."

All three take periods from 5m to 1d, and all three carry the same note: "Only the data of the latest 30 days is available", with an IP rate limit of 1,000 requests per five minutes. The two top-trader descriptions are quoted from Binance's own published SDK on GitHub, which carries the endpoint documentation verbatim, because the portal page could not be read in full by automated means.

Note the selector: margin balance, which is the size of the account's collateral, not the size of its positions.

OKX

Four series, two populations, a different selector and a different history rule. All sit under /api/v5/rubik/stat/ and answer unauthenticated at five requests per two seconds.

• contracts/long-short-account-ratio — "the ratio of users with net long vs net short positions for Expiry Futures and Perpetual Futures", per currency rather than per contract. Its window depends on the period: five-minute data "within two days at most", hourly within 30 days, daily within 180 days.
• contracts/long-short-account-ratio-contract — the same account ratio for one contract, returning "the latest 1,440 data entries".
• contracts/long-short-account-ratio-contract-top-trader and contracts/long-short-position-ratio-contract-top-trader — the account and position ratios for top traders, where "top traders refer to the top 5% of traders with the largest open position value", again the latest 1,440 entries.

The documentation gives each contract-level endpoint a data time range as well — "up to March 22, 2024" for the two top-trader series, and for the contract account ratio "up to January 1, 2024" at daily resolution and "early February 2024" at other periods. The page states those dates without saying which end of the range they mark; read them against the data before relying on either reading.

There is also a margin/loan-ratio endpoint labelled "margin long/short ratio" in the navigation. It is "the ratio of cumulative amount of quote currency to base currency" borrowed on spot margin — a lending statistic, not a derivatives positioning one, under nearly the same name.

Why the same label gives different numbers

Put the three side by side and "top trader long/short ratio" turns out to be two different selections: the top 20 percent by collateral on Binance, the top 5 percent by position value on OKX. Bybit's V5 market endpoints list no top-trader series at all. A fund with a large balance and a small position is inside one slice and possibly outside the other.

The account ratios differ again: Bybit counts all holders, OKX counts users net long against net short, and Binance's global series does not define its population in the text at all. The history differs by an order of magnitude — 30 days on Binance, 1,440 rows on OKX, back to 2020 on Bybit. A chart that overlays them is overlaying three definitions.

Aggregators inherit this. A site that shows one cross-venue long/short figure has either picked one venue's series or combined several, and the combination is its own method. CoinGlass, Coinalyze and Hyblock Capital all carry long/short data; which series sits under a given chart is a question for each of them, not something the label answers.

What it costs

The data itself is free. On 23 October 2026 Bybit's account ratio, OKX's top-trader account ratio and all three Binance series each answered a request sent with no key at all; the venues' own data APIs are on their cards (Bybit, Binance, OKX). One quirk surfaced on the way: Binance's top-trader position ratio returns its two shares in fields named longAccount and shortAccount, the same names the account series uses, so a parser that keys on field names cannot tell the two apart.

What costs something is history. Binance keeps 30 days and OKX 1,440 rows per series; past that the only source is somebody who called the endpoint every period and kept the answer, which is a job you either run yourself or buy. A thousand requests per five minutes on Binance is generous for one symbol and tight for every symbol at five-minute resolution.

What you can do about it

Read the endpoint, not the label. Before using any long/short figure, find out three things: is it accounts or contracts, who is in the population, and which venue's contract it describes. If a chart cannot answer all three, it is showing you a number without a definition.

Never compare a top-trader series across venues as if it were one quantity. Binance's top traders are chosen by margin balance and OKX's by position value, at different percentages. Compare each venue's series with its own history, not with the other's.

Read an account ratio as a headcount. A high account ratio says many small accounts are on one side, and the arithmetic of equal totals says the other side is holding larger positions on average. That is all it says.

Start recording now if you need history. If your work needs Binance's ratios beyond 30 days, or OKX's beyond 1,440 rows, schedule the calls today; the venue will not give the past back later. The mechanics of pulling the related derivatives series are in how to get funding rates and open interest.

Ask an aggregator which series it uses. A vendor that combines venues should be able to name the endpoints and the method. If it cannot, its figure is not comparable with anything, including its own figure for another coin.