The Same Strategy, Four Hours Later, Loses Money

Here is the cleanest result I've gotten out of any backtest, and it took me a while to understand what it was telling me.

I ran an opening range breakout on the 09:30 open. It scored a t-statistic of +4.06 across 265 trades.

Then I ran the identical rules on the 14:00 bar. Same range logic, same breakout condition, same risk model, same instrument, same 1,254 sessions. The only difference was the clock.

It lost money.

strategy	n	PF	t
morning ORB 09:30–09:45	265	1.72	+4.06
afternoon ORB 14:00–14:15	500	0.91	−0.99
power hour VWAP 15:00	561	0.95	−0.56

Note the afternoon version had nearly twice as many trades. It wasn't starved of data. It had more opportunities and did worse with them.

The pattern isn't the edge

This is the thing I'd been getting wrong, and I think most people learning this stuff get it wrong the same way.

I thought I had a breakout strategy. I did not. I had a volume strategy that happens to express itself as a breakout.

Look at what actually differs between 09:30 and 14:00. Not the price structure — ranges form and break at both times, and you can find textbook-looking setups all afternoon. What differs is who's in the market. Median volume per 5-minute NQ bar runs about 3,708 during regular hours; the open is the densest part of that, and it thins out sharply as the day goes on. Evening bars run around 554.

A breakout is only meaningful if there's enough participation behind it to keep going. In the afternoon the same shape appears on the chart with nobody behind it, and it fails.

The chart pattern was never doing the work. The pattern is just the shape the volume makes on its way past you.

Why this generalizes

Once you see it, you start asking a different question about every strategy you test. Not "what does the setup look like?" but "what is the mechanism, and when is the mechanism present?"

This also gives you a real test. If you think you have an edge, run it at a time or on an instrument where your stated mechanism should be absent. If it still works, your mechanism story is wrong — you found something else, and you don't know what. If it stops working exactly where the mechanism disappears, that's genuine confirmation, and it's much harder to fake than a good backtest score.

My afternoon result is a negative control. I'd rather have it than another winner.

If you can only trade one session a week

A practical note, since the open is a narrow window and plenty of people can't be at a screen for it every day.

I looked at whether any weekday was better. Friday came out at 62 trades and an average of +0.353R — the best-looking day. But all five weekday confidence intervals overlapped heavily, which means the ranking is noise. There's no good day. There's just the open.

That turned out to be fine, because the question isn't whether Friday is better. It's whether Friday is worse. It isn't. So if one session a week is what you have, you're trading the same edge at a lower frequency, not a degraded version of it.

The math is sobering, though. One session a week is about 12 trades a year, roughly +4.4R, which at a single contract is on the order of a thousand dollars. That's not an income. That's a training budget — enough reps to learn something, not enough to matter financially.

Which is the honest answer to "can I trade around a schedule": yes, but treat the first year as tuition rather than earnings.