The Gap-and-Go Made Money Every Year for Five Years. Then It Stopped.

The idea: trade the big gaps, not the small ones

A gap happens when the market opens at a very different price from the last close. Most gap-trading lessons teach the same habit: gaps fill (the price goes back to the old close), so trade against them. This is called fading the gap.

The numbers behind that habit are real. On Nasdaq futures, gaps smaller than a third of the daily range fill by the close 78% of the time. But the lessons skip the other end of the same table. Gaps bigger than 1.2 times the daily range fill only 8% of the time. So the people who fade gaps are right on small gaps. On big gaps they lose, and the big gaps are where the money is.

That difference is the whole idea. Here are the rules:

• When the S&P 500 CFD opens at least 0.5% above the last session's close, buy at the open.
• Put the stop loss (the price where we close a losing trade) at that last close. If the gap fills completely, our idea was wrong.
• Sell at 15:55 ET.
• Buy only, never sell short.

The traders who fade the gap are the ones on the other side of our trade.

We measured it before we built anything

Our rule is simple. A strategy must first show that its basic effect exists. Only then do we write an EA (an automatic trading program for MT5).

We wrote two separate scripts and used five years of 1-minute data, from 2017 to 2021. We did not touch the later years. Both scripts gave exactly the same answer:

• 182 sessions with a gap up of 0.5% or more
• +0.15R per trade, with the stop at the last close (R is the amount we risk on one trade)
• A 95% confidence range of +0.03R to +0.28R
• All five years positive

The check we cared about most is the one that failed our opening-range-breakout test in July. The question: is this just being long the market on good days? We compared:

• Gap-up sessions returned +0.172% from open to close.
• An average session in the same years returned +0.023%.

That is seven times more, with a t-value of 1.9 (a measure of how likely a result is not chance). The statistics were borderline. But the difference was real, and not just a result of a rising market.

Then we wrote down the conditions that would make us drop the idea. We also wrote down the market risk: a falling market where early strength gets sold. After that, we built the EA.

On old data, it did exactly what the measurement said

Index CFDs at our test broker have no commission, only a spread (the cost to trade). So the only cost is a 0.75-point spread on every trade. Every number here includes it.

The optimizer tested 36 combinations of the gap size, the exit time and an optional profit target. Two stable groups of settings came out, where all the nearby settings also made money:

PF means profit factor (money won divided by money lost; above 1 is profit). Max DD means maximum drawdown (the biggest drop from a high point). In-sample is the data used to choose the settings. Out-of-sample is new data the strategy never saw.

The first version had 181 in-sample trades. That is the 182 events from our measurement, minus one. In the first short test, the EA traded exactly the twelve sessions the raw data said it should. The stops sat exactly on the last close. The yearly returns grew from +0.9% in 2017 to +8.7% in 2021. We did not fit this to the data. It was a pattern that really existed.

Net return by year. Five green years in sample, then 2022 and nothing.

On new data, one good year and three flat ones

We fixed both versions and gave each one single run on 2022 to 2025.

The first version made 5.4% in 2022. That surprised us. 2022 was a falling market, the exact risk we had written down, and the strategy did fine. Then it stopped working:

• 2023: −0.7%
• 2024: −3.2%
• 2025: −0.2%

Over four years and 169 trades, it made $111 net. The profit factor was 1.02, and our minimum is 1.15. The second version did the same thing with more trades: PF 1.04, $397.

Out-of-sample equity of the 0.5% version, after the test spread. It went up and came back to where it started.

We also test every strategy with higher costs, $5 per side, even if the broker does not charge this. With those costs, both versions lose several thousand dollars. The first version failed six of our tests. The second failed four. We rejected both.

We had planned two extra checks in advance: the Dow and the Nasdaq, with the same fixed settings, on old data only.

• The Dow did best: PF 1.52 on 208 trades.
• The Nasdaq did worst: PF 1.17 on 259 trades.

This is the same order our first measurement showed. We did not use an out-of-sample run on either one. The S&P 500 had already answered the question.

What we think happened

The traders who fade gaps did not go away. What changed is that fading a big gap stopped costing them money.

In 2020 and 2021, a big gap up in the index pulled in many traders betting against it at the open. By lunch, they were forced to buy back, and the price went up more. Since 2023, big gap-ups have opened and then moved sideways.

Before the spread, the edge on new data is about 0.7 index points per trade. The middle gap size is 26 points. So the edge is just random movement that happens to be slightly positive.

This is the second index-open idea to fail here in three months. The two failures are different:

• The opening-range breakout never had an effect. It was just the normal rise of the market, and it only looked like a breakout strategy.
• The gap-and-go did have an effect. We measured it with a confidence range. We repeated it in the tester. And then we saw it end.

We prefer to publish this than to pretend the first five years did not happen.

Two parts of our process are worth keeping:

• The measurement we planned in advance. It lets us say the in-sample result was not fitted. The tester matched a script we wrote before the EA existed.
• The one-time out-of-sample rule. It lets us say the effect is gone. We did not just find worse settings.

Neither one cost us anything, except the temptation to run the test again.