How to Read the Long/Short Ratio

Open any perpetual futures dashboard and there it is, usually in green: Long/Short Ratio 1.87. Under it, a second one for the top traders. It looks like the cleanest sentiment number in the market — not a price, not a derivative of a price, but a direct count of which way the traders on the exchange are actually leaning.

And two things get done with it, over and over. The first is to fade it: everybody's long, so the crowd is wrong, so sell. The second is to follow the top-trader version: the big accounts are long, and they know something. Both are stated with confidence and almost never with a number behind them. So this is the number.

What the number actually counts

The exchange looks at every account with an open position in a contract. It counts how many are net long and how many are net short, and divides the first by the second. That is the whole calculation. A $50 account and a $5 million account each count as one.

Binance publishes three versions every five minutes, and they are not the same thing. The first is every account, by headcount — call it the crowd. The second is the top 20% of accounts by margin balance, also by headcount — the "top traders". The third is those same top accounts, but weighted by the SIZE of their positions rather than counting heads. That third one is the only one that measures money, and it matters, because it is the one that behaves differently.

Finding 1: the crowd is always long

Here is the first thing the folk reading gets wrong. A ratio above 1 — more longs than shorts — is not a bullish reading. It is the resting state. Over six years, bitcoin's crowd had more longs than shorts on 79.5% of days. Ether's on 92.9%. Solana's on 97.6%.

So a bitcoin reading of 1.50 is not "the crowd is bullish". It is a Tuesday. Solana's ordinary day is 2.50 — a reading that on bitcoin would sit in its top 10%. The same number means opposite things on two contracts, which is why the ratio is only ever readable against its own history, on its own contract. There is no universal "too long" line.

Finding 2: the "top traders" are the crowd

The second reflex is to skip the crowd and read the top-trader line instead, on the theory that the big accounts know more. So put the two headcount lines side by side and ask how different they are.

They are not different. On bitcoin the two move together with a rank correlation of +0.97, on solana +0.96. When the crowd leans long, the top 20% of accounts by balance lean long by almost exactly the same amount, on the same day. The "smart money" line on the dashboard is the crowd line, drawn a second time in a different colour.

The line that IS different is the third one — the top traders by position size. Its correlation with the crowd is +0.00 on bitcoin, +0.44 on ether. That one really is measuring something else: not how many big accounts are long, but how much of the big money is. Hold that thought, because it comes back at the end and it does not come back well.

Finding 3: the number is mostly yesterday's candle

Here is the mechanical fact that explains most of what the ratio does day to day. Take every day where price fell 5% or worse, and look at what the crowd ratio did over that same day.

It goes UP. After a 5% down day on bitcoin the crowd ratio rises by a median of 15.7%, and it rises 64% of the time. After a 5% up day it falls 17.4%, 71% of the time. On ether it is sharper still — up 18.6% after a fall, down 21.5% after a rise, and the second one happens 88% of the time.

That is the crowd, by headcount, leaning against whatever price just did. Price drops and the shorts take their profit while the dip-buyers arrive, so the count of longs goes up relative to shorts. Price rips and the longs cash out while a fresh wave of shorts tries to call the top. The ratio's day-to-day change has a rank correlation of −0.60 with the day's price change on bitcoin and −0.69 on ether. It is close to a mirror.

Now the money line. The top-trader position ratio moves −0.3% after the same down days on bitcoin and +1.0% after the up days. On solana it is +0.0% and −0.2%. The correlation with the day's move is −0.14, −0.04, −0.01. Big money does not flip its book because of one candle. Small accounts do — or rather, enough of them enter and exit that the headcount flips.

Finding 4: contrarian — but only at one end

Which brings us to the reflex everybody actually trades. Rank each day's reading against that contract's own trailing year — so "extreme" means extreme for THAT contract, THEN, with no peeking at the future — and take the bottom 10% of days and the top 10%. Score each on what price did over the following 30 days, and compare with the middle 80%.

The bottom end pays. When bitcoin's crowd was in the least-long tenth of its year, the next month averaged +6.9% and was up 67% of the time, against +1.8% and 51% in the middle. Ether: +13.7% against +2.5%. Solana: +25.1% against +7.4%. It reproduces on all three.

The top end mostly does not. When the crowd was in its most-long tenth, the next month came in below the middle on all three contracts — but at −2.1%, +0.1% and +3.6%, with an up-month 48%, 38% and 50% of the time. That is a headwind, not a sell signal. "Everybody's long, so short it" is a trade that was, on average, roughly flat.

The asymmetry has a plain reason. Because the crowd is structurally long, "unusually long" is only a little more of the normal thing, and it happens at every stage of a rally including the early ones. "Unusually NOT long" is the rare state — it takes a drawn-out fall to empty the room of dip-buyers — and that is where the month after has tended to be good. The most extreme low reading on solana, 0.55 on 20 October 2023, was followed by +126.2% in thirty days. The most extreme high, 7.17 on 3 December 2021, by −16.7%.

Finding 5: the smart-money line says nothing

The one ratio that genuinely measures money rather than heads is the top-trader POSITION ratio. If "follow the big accounts" works anywhere, it works here. So run the same test on it.

There is no shape here. On bitcoin BOTH extremes did worse than the middle. On solana both extremes did better — +20.3% when big money was least long and +15.0% when it was most long, against +4.3% in between. A line that pays off at both ends and not in the middle is not telling you a direction. It is telling you it moved.

So the honest summary of the two reflexes is this. Fading the crowd works only when the crowd has already been beaten out of the room, on a month's clock, as a tilt. Following the top traders does not work, because the top-trader line most people look at is the crowd, and the one that is not the crowd has no direction in it.

How to read it yourself

• Read it against its own history on its own contract. Bitcoin at 1.5 and solana at 2.5 are both ordinary. There is no number that means 'too long' everywhere.
• Ignore the day-to-day change. After a big red candle the ratio rises and after a big green one it falls, almost mechanically. That is the candle, not new information.
• Know which of the three lines you are looking at. The two headcount lines are the same crowd. Only the position-size line measures money, and it is the least informative of the three.
• The reading that has paid is the LOW one — the crowd unusually not-long, for that contract, relative to its year. Treat a high one as a mild headwind, not a short.
• Give it a month. The tilt is invisible on a day and weak on a week. It is a positioning gauge, not a timing tool.
• Put it next to open interest and funding. Those count money and cost; this counts people. Three gauges that disagree are more useful than one that is read three ways.

Method, and what this cannot tell you

Each UTC day's reading is the last five-minute snapshot of the day. Three ratios are read side by side: every account by headcount, the top 20% of accounts by margin balance by headcount, and that same top 20% by position size. Each ratio is ranked against its own trailing 365 days, so a reading is 'extreme' only relative to what that contract had been doing over the previous year — never against the whole sample, which would use the future to define the past. The bottom and top 10% of days by that rank are then scored on price over the following 30 days, and compared with the middle 80%.

• One venue. Binance publishes the ratio for its own accounts; the number on Bybit or OKX is a different crowd and can disagree.
• The two top-trader ratios are blank in the archive for 315 days, from 31 December 2021 to 13 December 2022 — nearly the whole of the 2022 bear market. Every top-trader figure here is measured on that shorter window, and the crowd figure next to it is not.
• Headcount is not money. The all-account ratio counts a $50 account and a $5 million account as one each. That is why the position-sized ratio behaves so differently, and why neither one on its own is the whole picture.
• Thirty-day windows overlap. Two thousand days of readings give a few dozen independent extreme episodes per contract, not hundreds, and the extremes cluster in particular years — the crowd was least long through 2023 and most long in late 2021. The result is a tilt that shows up in both halves of the sample, not a law.
• In-sample and gross. No fees, no funding, no slippage, and no attempt to trade any of it.