Tested the 'big squeeze before takeoff' idea: 516 symbols, 5 years. Tight ranges predict more quiet

Everyone has heard the folk model: when a name goes dead quiet and the range compresses, it is coiling before a big move. We wanted to know if the tightest compressions actually launch, so we tested it.

**Design**

- Universe: 516 optionable symbols, 5 years of dailies (~650k symbol-days)
- Compression: 10-day average true range %, ranked against the symbol's OWN trailing year, split into deciles. Decile 0 = this name is in the tightest 10% of its own recent history (not vs other symbols)
- Forward measure: biggest absolute close-to-close move over the next 5 sessions
- Split-half: everything before June 2024 vs after. A result has to hold in both halves or it is noise.

**Result: monotone in both halves, the wrong way for the launch thesis**

|compression decile|median 5d max move (h1 / h2)|P(move >=5%)|
|:-|:-|:-|
|0 (tightest)|3.0% / 3.1%|25% / 29%|
|5|3.8% / 3.6%|36% / 36%|
|9 (loudest)|4.4% / 4.6%|44% / 46%|

The tightest decile has the SMALLEST forward moves, perfectly monotone across all ten deciles, in both halves. Quiet names stay quiet, loud names stay loud. Volatility clusters - that is the real phenomenon, and it is one of the oldest results out there. Direction is flat too (P(up) runs 52-56% across deciles), so compression also tells you nothing about which way.

**Before you flip it and buy options on the loud names:** implied vol already knows. We measured implied vs realized separately and IV overshoots what actually happens in both tails, so "loud stays loud" is largely priced.

What actually preceded launches in our record was not shape, it was flow - repeated large premium hitting the same name across days, and catalysts. Breakout systems that use compression only as a shortlist and trigger on the actual break are a different claim than "tight = about to launch at a fixed horizon" - we only tested the latter.

Happy to share methodology details. Poke holes.