Trading the opening-range breakout

We wrote each configuration down, committed it, and only then ran it. That order matters: if you fix the rules after seeing the results, you are no longer testing an idea, you are describing one. Here it matters twice, because five tests give you five chances to fool yourself.

The base setup, identical in all five: the opening range is 09:30–10:00 ET. Entry is the first close beyond the range high or low, long and short, filled on the next bar. The stop is one opening-range width. Flat at the end of the day. Costs — one point round turn plus one tick of adverse slippage — are charged on every fill.

The five variants:

1. Volatility regime. Only take the break when the day is expansive: yesterday's ATR above its 14-day average, or VIX above 18, or both. Four reward-to-risk targets each.
2. Order flow. Only take the break when cumulative bar delta makes a new extreme at the same moment — the "confirmed breakout". Plus the mirror version: fade the break when confirmation is missing.
3. All of it at once. Regime and a clear path to target (no prior-day point of control in the way) and order-flow confirmation. This is the version most traders actually believe in — the one where the edge is supposed to come from the combination.
4. Let the winner run. On days when the dealer-gamma proxy says moves extend: take half off at +2R and trail the rest, instead of a fixed target. Three trailing methods, three caps.
5. Volume and trend. Break of the session high or low, but only with volume above twice its 20-bar average and ADX above 25.

What ties them together: every one of the 63 cells was written down before the run, and every one is reported — including the ten that came out positive. The instrumented runs we exported these numbers from reproduce the original engines trade for trade, maximum deviation zero, in all five families.

Hypothetical backtest results. Read the full disclaimer at the bottom of this page.

Ten of the 63 cells made money on average. None of them passed.

That sounds like a contradiction and it is the most useful thing on this page. A cell passes only if it also beats its own control — the same days, the same costs, entries placed at random. The volatility family's headline cell — VIX above 18, a 1.5R target, 959 trades and +0.0135 R — landed at the 63rd percentile of that control distribution. We require the 95th. So the filter did something real: it picked days that moved. It just did not pick entries that were better than chance on those days. The volatility filter is exposure selection, not a trigger edge — it changes when you are in the market, not whether your entry is worth anything.

Three findings we did not expect and are not going to bury:

The confirmation hurt. Waiting for order flow to confirm the break did not improve anything — it made it worse. At a 2R target the confirmed breakouts averaged −0.0975 R against −0.0587 R for taking every break. Roughly four hundredths of an R worse, per trade, for being patient. The confirmation arrives when the move is already paid for.

Stacking made it look better and hold worse. The version with everything switched on — regime and clear path and confirmation — produced the second-best average in the whole study, +0.0119 R. It also produced 345 trades where taking every break in the same family produced 2,288, and it sat at the 18th percentile of its null distribution and the 10th of the random-entry control. That is what stacking does: each filter removes trades, the survivors look handpicked because they were, and the result does not survive contact with a control. We tested the thing everyone means by "confluence", and it is the weakest result we have.

The runner never showed up. The "let your winners run" version takes half off at +2R and trails the rest, which costs a third fill on every trade. Of 1,572 breakouts on those days, 168 — under 11 % — ever reached the +2R scale-out at all, and 55 ran far enough to hit the 3R cap. The remaining 1,404 paid for a runner they never got. All nine runner cells came out below the plain 2R baseline. The fat tail is real; it is just rarer than the cost of fishing for it.

And the one from the volume-and-ADX version, which we put in words rather than in a picture because it needs its window named: on IS 2015–21, a random entry on the same volume-and-ADX days averaged +0.30 R, while the filtered breakout averaged −0.02 R. The filter found the days. The trigger then gave the day back.

This is what a single chart can never show you. One session shows the mechanics. Sixty-three versions across nine years show the outcome.

Here is the thing worth taking away: the trigger was never the interesting part.

Look at what actually carried information in all five tests. Not the entry — the day. Whether volatility was expanding. Whether the path to your target was clear. Whether the session had any range to give in the first place. Every filter that helped was a statement about the environment, and every attempt to turn that statement into an entry signal failed.

That is a useful split, because the environment is knowable before you trade and the entry is not. Our Regime Lab reads it off the market and shows the distribution of realised daily ranges in the current regime, with the sample size next to it — so you can see how thin the evidence is when it is thin. It is free, it is published on TradingView with the same measurement we use ourselves, and it deliberately does not tell you to enter anything.

Hypothetical backtest results. Read the full disclaimer at the bottom of this page.

Everything above can be checked right here, cell by cell — no signup. If you want to run the numbers yourself, the trade-by-trade logs of all three studies and their frozen configurations are available as a data bundle on the Free Tools page.