Methodology
Most trading ideas don't hold up. Here's how we check.
Before anything becomes a signal in Holdout, it runs through the same process. Most ideas fail it. That's the point.
The process
- Years of one-minute data. Over 12 years of NQ and ES futures (2014 onward), at real contract prices, including the 2022 bear market.
- Realistic costs and fills. Every trade pays commission and slippage; entries fill at the next bar, not at the signal price; stops are assumed hit first when a bar touches both stop and target.
- No peeking. Every level and indicator value uses only data available at that moment. We rebuild calculations on truncated history to prove nothing from the future leaks in.
- Out-of-sample checks. A rule designed on one period has to work on another it never saw — and in walk-forward tests, on the next year after each training window.
- Controls. A level effect is compared against random prices nearby; an entry rule against random entries with the same exits. If the control does as well, there is no edge.
- Clear guidance. Each indicator page explains how it works and when it
- Honest limits. Each indicator page says what didn't work and when the rule fails. apos;s at its best.
What made it into the suite
✓ 15-minute opening range breakout
Held up out of sample with a wide stop at the other side of the range; improved by requiring cumulative volume delta to agree. Worked from 2021 on, not before.
✓ Noise-area intraday momentum
A published rule (Zarattini, Aziz & Barbon, 2024) tested without changing its logic. Positive in both our older and newer data, with no fitting.
✓ Value-area and round-number confluence
Not a signal by itself, but supply/demand zones containing these levels did clearly better than zones without them.
What we tested and rejected
Popular ideas — many from books, papers and social media — that didn't survive costs or out-of-sample data:
✕ Fading every key-level touch
Price was no more likely to reverse at real levels than at random prices nearby.
✕ Value-area rotation
Fading the value-area edges toward the POC: zero of 36 configurations stayed positive out of sample.
✕ Range fades
A mechanical detector can't tell a range that will hold from one about to break; every break is a stop-out.
✕ Absorption and order-flow fades on bars
Real in R-multiples, but trading costs ate the entire edge.
✕ Fading failed opening-range breaks
Looked good in some years, failed in others; combining it with the breakout made results worse.
✕ ICT-style ERL → IRL sweeps
At arbitrary intraday swings the sweep-and-reclaim fade lost money; it only works at prior-day extremes.
✕ Historical analog matching
Comparing today's chart shape to the past had essentially zero correlation with what happened next.
✕ Cross-market lead signals
Crude oil, bonds, VIX and ES moves are simultaneous with NQ, not ahead of it.
✕ Levels as profit targets
Capping winners at the next level hurt every strategy tested — big winners run through levels.
What this means for you
The context indicators (levels, profile, structure, zones) show you where the market is likely to react, and your own read decides which way. Like any trading rule, the momentum rules have losing streaks and depend on market conditions. Use them as part of your own plan, size small, and test on a simulated account first.
Past and simulated results don't guarantee future results. See the risk disclosure, including CFTC Rule 4.41 on hypothetical performance.