Where retail crowding sits across the 20 most-traded Bybit perpetuals, refreshed every 30 minutes. An event fires when a coin's long/short ratio reaches an extreme — above 80% long, or below 40%. The label says which way the crowd is leaning; it is not a buy or sell.
| # | Coin | Retail positioning | 24h | Turnover |
|---|---|---|---|---|
| 1 | BTC | 65.6% |
-0.8% | $4707M |
| 2 | ETH | 72.9% |
-1.4% | $2480M |
| 3 | SOL | 77.4% |
-1.7% | $510M |
| 4 | SOXL | 67.3% |
-3.1% | $475M |
| 5 | ZEC | 63.5% |
-4.9% | $333M |
| 6 | XRP | 80.8% |
-2.3% | $271M |
| 7 | 1000PEPE | 73.5% |
+4.7% | $221M |
| 8 | NEAR | 66.2% |
-4.6% | $205M |
| 9 | KORU | 66.0% |
-0.9% | $176M |
| 10 | XAU | 76.4% |
+0.1% | $158M |
| 11 | HYPE | 74.6% |
-2.7% | $149M |
| 12 | CL | 37.3% |
-0.1% | $134M |
| 13 | ENA | 74.2% |
-3.0% | $126M |
| 14 | SKHYNIX | 66.1% |
+5.0% | $119M |
| 15 | SNDK | 74.6% |
+0.1% | $91M |
| 16 | Q | 64.5% |
+30.2% | $86M |
| 17 | GOOGL | 85.0% |
-9.7% | $71M |
| 18 | EVAA | 26.9% |
-7.9% | $70M |
| 19 | DOGE | 79.2% |
-1.6% | $69M |
| 20 | SKHY | 53.7% |
+4.5% | $67M |
The trade levels are marked unvalidated, and they mean it. Direction fades the crowd. Stop sits 1.5× ATR against the entry, targets at 2× and 3× ATR, and size comes from risking 1% of equity across the stop — a consistent structure, not a fitted one. What none of it carries is evidence of profitability: testing this signal across 194,213 symbol-days found no edge (t = −0.04 over 400 days). The levels tell you how to structure a position if you take one. They do not tell you the position is worth taking.
Why they were removed. A crowding effect measured over 14 days looked strong — crowded coins underperformed the basket by 1.16% over 24h, t = −2.88. Re-run over 400 days and 12,937 symbol-days with day-clustered errors, the same effect measured t = −0.04. It was noise. Twenty-two alternative rules were then tested — breakout, momentum, reversal, crowding in both directions — and every one either flipped sign between the first and second halves of the year, or went flat once a realistic ATR stop was applied.
On win rates. A 78% win rate is easy to manufacture: take profit at 2% and stop at 10%. Tested on this data that configuration wins 77.5% of the time and loses 0.809% per trade. Win rate says nothing about whether a strategy makes money, because it says nothing about the size of the losses. It is the number that made the old martingale bot look safe.
Positioning is still real information. Crowding is measurable, it is published by the exchange, and it tells you where the fragility sits. It is a useful input to your own judgement. It is not, on this evidence, a signal you can trade mechanically.