Overbought" sounds like a sell signal. On the 4 hour chart it usually wasn't.
We measured every time Harmonic Oscillator printed Exit Overbought on the 4 hour chart: 76 perpetual futures on Bybit (63 crypto coins and 13 stock, ETF and commodity contracts), 22 April 2025 to 29 September 2026, 633 first fires (a repeat within three bars counts once).
What followed, 12 hours later
The coin was higher: 52.4% of the time Other coins on the same bar, with a similar run-up: 46.7% Median move: +0.16%
Comparing with other coins on the same bar matters. It takes out the market's own move that day, so what's left is the coin and its signal. Those other coins were higher only 46.7% of the time, so the period leaned down, and overbought coins still did better.
The crowd decides which kind of overbought it is
We split the fires by the long/short account ratio at the bar's close, compared with that coin's own last 30 days.
Longs not crowded (468 fires): higher 53.6% of the time, against 46.2% for other coins. Median move +0.26%. Longs crowded (149 fires): higher 47.0%, against 46.7%. Median move -0.47%.
When the crowd wasn't already leaning long, overbought more often meant the move kept going. When it was, that edge disappeared and the typical 12 hours ended lower.
On the chart
ENA, 4 hour, 22 August 2026. Harmonic printed Exit Overbought while 61.9% of accounts were long, below ENA's own 30-day normal of 71.8%. Twelve hours later ENA was 3.4% higher. It's one example, picked to show the case; the numbers above cover all 633.
How to do it
Step 1: On the 4 hour chart, note when Harmonic prints Exit Overbought. Step 2: Look up the long/short account ratio for that coin and compare it with its last 30 days. Step 3: Below its usual level: overbought has more often been a pause inside a move that keeps going. Above it: treat the print as a warning worth respecting. Step 4: Read it quickly or not at all. The difference was mostly gone two bars (8 hours) after the print.
Extreme Overbought behaved the same way overall: 623 fires, higher 50.7% of the time against 46.0%.
How fast it fades
The gap against other coins was 5.8 points if you read it at the print, 5.0 if you came in one bar later, and 1.3 two bars later. Most of the difference is gone within 8 hours, which is why step 4 says read it quickly or not at all.
How we checked it wasn't noise
Every test was declared before it ran. We compared each fire with coins that had made the same move on the same bar (a price twin), then with coins that also matched on volume (a volume twin), and we reran the whole measurement on 10 versions of shuffled history to see how often luck alone produces a gap this size. It held up in all three.
What this does not tell you
Higher more often is not a trade. The gap is a few points of probability, the median move is small, and on short holds fees eat a lot of it. One timeframe, one 17-month stretch, measured in sample. It held up against a price-only comparison and against shuffled history, and we've started counting it forward from 3 October on bars that didn't exist when we measured. We'll post what that count shows, either way
.$ENA
