Does a trading filter have a cause, or is it just a statistic?

Most filters traders use, like “only trade with the trend” or “fade moves away from VWAP”, come with a story. Usually the story is told after the numbers looked good. We did it the other way round: we wrote the reason down first, said which markets it should work on and which it should not, and only then ran the test.

• NQ, ES, gold, oil and Bitcoin futures, 22 Sept 2025 to 26 Oct 2026, 276 sessions.
• A long and a short on every 5-minute bar of the regular session. Stop and target are each half a day's average range, held at most 60 minutes, after costs. Results in R, where 1 R is the stop.
• Eleven predictions about data points we already show in the app, each with its mechanism, written before the test.
• Five checks per market: the effect shows on the markets the reason names and not on the others; same sign in both halves; a bigger dose gives a bigger effect; it beats most of 200 placebos where the filter's values are reshuffled between whole sessions; and it keeps its sign in the last three months.
• Every data point is fixed and simple. Nothing was tuned: no moving-average lengths or thresholds chosen after the fact.

The effect is the result when the trade agrees with the reason minus when it disagrees, in R per trade.

Not predicted, so only a lead: NQ and ES longs did better when interest rates were rising since the open, in both halves. Rates tend to rise on strong-growth days, and stocks like growth. But we found it rather than predicted it, so it has to hold on new months before it counts.

A reason that holds names its market. Oil and stocks share one driver, the outlook for growth, so they move together during the day. Gold against the dollar, the classic story, flipped sign within the year.

Doses matter. Gold's pull back to VWAP is invisible until price is far away. Averaging over every distance hides it.

Some stories are backwards. Bitcoin did worse when traded in the Nasdaq's direction. That is why a filter needs both a reason and a test.

Most good-sounding reasons don't show up in intraday results. Before you trust a filter, write down why it should work and where, then check that the market you trade is the one it names, that it holds in both halves and that a bigger reading gives a bigger effect.

Simple, fixed data points keep the test honest. With adjustable settings, any of these could have been made to “work” on the past.

About 14 months. The 60-minute bracket is one exit among many. Trades on overlapping bars are not independent, which is why the placebo shuffles whole sessions. Results allow one tick of slippage and no more. This is the past, not a promise it repeats.

Simulated results on past market data from 2025 and 2026. Research only, not trading signals or advice. Past results do not promise future ones; see the hypothetical performance disclosure at the bottom of the page.