Long/Short Ratio

One input drives the chart: the venue’s daily snapshot of the account-weighted long-to-short ratio for its top 20% of traders by margin balance, taken at the 00:00 UTC close. The y-axis is linear, anchored on a balanced book at 1.0, with dashed rails at 0.7 and 1.3 marking the canonical crowding thresholds and a muted spot line behind for cycle context. Provenance is on the data sources page; the formula is published verbatim in the venue’s API documentation:

Long Account % = Accounts of top traders with net long positions / Total accounts of top traders with open positions
L/S = Long Account % / Short Account %

The word account is the whole story. This metric counts wallets, not size. The venue publishes two surfaces — the account-weighted ratio plotted here, and a position-weighted sibling that sums notional. They often disagree: a market with seventy small long accounts against thirty leveraged short accounts can print account-weighted 2.3 and position-weighted 0.6 on the same day. We plot the account-weighted version because the contrarian signal — retail crowding — lives in head counts, not notional. The position-weighted divergence is a caveat we return to below, not a competing chart.

The chart shows one venue, and that is not laziness — it is the only honest construction. Aggregating long/short across venues to a single “global” reading is impossible upstream. Each venue defines its own ratio surface (top-X-percent by margin on one, all open-position holders on another, top traders by recent volume on a third), snapshots on its own cadence, and exposes only the precomputed result. There is no shared account-classification spec to sum across, so a number that looks aggregated would in fact be apples stacked on oranges. The cleanest read is to pick the venue carrying the largest perp-volume share — through this series, Binance — and live with the single-venue caveat plainly stated.

That choice has a cost. The dominant venue is itself a moving target. The first Bitcoin perpetual swap launched on BitMEX in May 2016; that venue dominated through 2018, a different one rose through 2020–2022 and then collapsed, and a handful of offshore books split the post-2022 market with one venue consistently on top. A long/short series spliced across that history is not a like-for-like comparison — which is why our daily series begins in 2023, after the venue mix had settled.

The middle of this chart is noise; its edge, such as it is, lives in the tails. Two prints in this series stand out. On 21 November 2025 the ratio spiked to 3.99 — top-trader accounts almost four-to-one long, nearly 80% on the long side — into a local high that promptly unwound, the textbook crowded-long flush. At the other extreme, on 3 July 2025 the ratio bottomed at 0.44, accounts seven-to-three short, into a base that held and rallied. And on 5 August 2024, the day the yen-carry unwind flushed every risk asset, the ratio sat at 2.55 — the crowd had been long into the drop, and paid for it. Each of those is a tail. Between them, the line wandered without telling you much.

Six anchor windows since the start of the daily series make the indicator’s character visible. Spot prices are nightly closes from our own pipeline; ratios are pulled from the same daily snapshot that powers the chart above. Note that the ratio led correctly at the November 2024 leg (short-crowded at 0.61 right before the Trump rally squeezed shorts) and misled at the yen-carry flush (long-crowded into the drop). The signal is in the extreme, not the direction it happens to sit.

Three regimes resolve cleanly. Below 0.7, short-account share has pushed past roughly 59% and the book is shorts-crowded — historically a setup for a squeeze, since the under-capitalised side covers fast on any rally. Above 1.3, long-account share has pushed past 56% and the book is longs-crowded — a setup for a leverage flush, since stops cascade when the trend stalls. The middle band is descriptive only: it names who is positioned, not where price is going.

The discipline is to treat extremes as setups, not triggers. A reading below 0.7 is the kind of book that squeezes when spot breaks higher; a reading above 1.3 is the kind that flushes when spot stalls. Neither tells you the timing. Pair it with funding rate to see the price the crowd is paying to hold its lean, and the liquidation heatmap to see where the cascade actually lives. A single-venue account-count read, on its own, is not a trade.

The head count ignores size, so it can point the wrong way. This is the metric’s oldest and fairest criticism. A hundred small retail longs stacked against five large institutional shorts will print long-crowded even though the dollar imbalance runs the other way. The same venue’s position-weighted surface regularly disagrees on direction, not just degree. If a directional decision rests on this call, cross-read the notional version before you act — the two can sit on opposite sides of 1.0 on the same day.

The dominant venue rotates, and the ratio rides the migration. The sharpest case is the FTX collapse. On 2 November 2022 reporting exposed the Alameda balance sheet; by 11 November FTX — then the second-largest perp venue — had filed Chapter 11 and its long/short feed went from live to dark over a weekend. The surviving venue’s reading became the de-facto “global” number with no change in methodology, even though the user base feeding it had just been swelled by a forced migration that was nothing like organic positioning. The post-FTX liquidity-gap analysis — the “Alameda Gap” — found global crypto liquidity halved in the following week and was still at half its pre-FTX depth eleven months later. A ratio can move that much without anyone trading anything. Our series starts after that window for exactly this reason.

It is the most gameable metric on the site. Funding rates reflect a real cash payment and open interest reflects a settled book, so both are hard to fake without leaving evidence. A long/short ratio is a derived classification of accounts: new sub-accounts opening, dormant accounts closing, or a regional ban culling part of the user base can shift the readout without one trader changing their view. The account-vs-position fork compounds it — a dashboard can flip the sign just by pulling the other surface. Treat extremes as suggestive and absolute levels with suspicion; the level is venue-specific and population-specific, and neither is stable.

The body is long-skewed, which dulls the “above 1 is bullish” read. Across the 1,000 daily prints in this series, the ratio sits above 1.0 far more often than below it — perp accounts lean long as a baseline. That means a reading of 1.2 is not crowding; it is Tuesday. Only the genuine tails carry information, which is why the rails sit at 0.7 and 1.3 rather than at the neutral 1.0 a symmetric reading would assume.