Screens
Compression iseasy to find.Direction is thehard part
A squeeze tells you volatility has contracted and a move is likely. It says nothing about which way. The screens that work here are the ones that add a direction condition; the ones that did not were killed by a backtest in minutes.
The setup, and what we learned running it
What a squeeze is
Bollinger bands measure volatility with standard deviation; Keltner channels measure it with average true range. When the Bollingers contract inside the Keltners, the two measures agree that range has collapsed. Historically that resolves into an expansion. The condition is mechanical and cheap to screen for.
The version that did not survive
A pure "squeeze fired" screen — compression released, take the break — was backtested and killed within minutes. It finds the setup and cannot tell you the side, so half the firings are the wrong half, and the fees and slippage on the wrong half eat the edge from the right one. That is not a criticism of the indicator; it is the indicator being used for something it does not measure.
The version that shipped
A directional squeeze-breakout on the hourly: compression, then a break in a direction that agrees with the trend regime, with the range that was broken defining the invalidation. It is one of the screens in the lineup because it survived the test that removed two others — and the two that were removed are recorded, with what they scored, so nobody rebuilds them next quarter.
Judge it on the firing history, not the memory
Every firing is written down: which symbol, which condition, which run, at what price. A screen that fires forty times a day is noise no matter how good the one you remember was. This is the number that tells you whether a squeeze screen is working for you, and it is the one most tools do not keep.
Questions about squeeze screens
Which timeframe?
The hourly is where our shipped version runs. Lower timeframes produce many more squeezes and a much worse ratio of real expansions to noise; daily produces few enough that you may as well look at a chart.
How do I pick a direction?
With a separate condition — trend regime, a higher-timeframe bias, or the side of the range that broke. What you should not do is take the first candle’s direction as the answer: it is the noisiest possible estimator of the move you are trying to catch.
What invalidates the trade?
Price back inside the range that broke. That is also the natural place for a stop, which means the setup produces its own risk-per-trade number rather than needing one invented.
Will you tell me if the screen stops working?
The run history will. Degraded runs mean the data stalled; clean runs with no hits mean the market changed. Those get treated as the same thing in most tools, and they should not be.
Take the setup for a rehearsal
Run it on paper, read the firing history, and decide with the record rather than the recollection.