I tested the opening range breakout 47,555,893 ways. A coin usually did better.

01 The strategy

Draw a box around the open. Trade the break.

The market opens. You wait a bit (five minutes, fifteen, half an hour, depending on who you ask) and mark the highest and lowest price of that first stretch. That's the opening range. If price breaks above it, you buy. If it breaks below, you sell. That's the classic version. I also tested the two ways people flip it: fading the break, betting it fails, and waiting for a break to fail, then trading the reversal.

It isn't a straw man. It goes back at least to the eighties, and in 1990 Toby Crabel wrote Day Trading with Short Term Price Patterns and Opening Range Breakout, built for US stocks and futures, where the open is a real event and the first half hour behaves differently from the rest of the day. And in 2018, a paper in the Journal of Financial Economics looked at twenty years of S&P 500 data and found that the first half hour predicts the last half hour. Something real happens at the open.

But that paper's data stops in 2013, and something that held for twenty years doesn't have to hold now. So: take 12 documented variations, test each one millions of ways on seven very different markets, hold them all to rules fixed before the final run, and see what's left.

02 The test

Twelve variations, every tick, every cost

The opening range breakout isn't one strategy. Everyone has their own. I found 12 documented variations: the classic break, gap-and-go, gap fade, a volume spike, price against VWAP, a volatility range, a trend filter, where today's range sits against yesterday's, order flow read off the ticks, and a few more. Mostly, it's the same idea with different opinions about which days to trade. Then every variation was played three ways: with the break, against it, and on the reversal after a failed break. Of the 1,073 survivors, 745 traded with the break, 112 faded it and 216 took the reversal.

Seven markets: the Nasdaq 100, the S&P 500 and the Dow, where the strategy was born; EURUSD, GBPUSD and USDJPY; and gold. And I didn't choose the settings, because a person choosing them is exactly what's being tested. The machine tried every combination of session, entry and stop type, each at 512 spread-out settings of the rest (range length, break distance, stop size, when to stop trading), with every target and every filter.

Every trade was filled from tick data, every recorded change in price: 3,137,557,989 of them, 37.7 GB of prices and their timestamps. So a stop or a target gets hit at the price that was really there, at the moment it was there. Every trade paid the real spread when it filled, plus commission where the market charges one. It adds up: for the versions that passed round one, about 42 cents of every dollar they made before costs went on costs.

The prices are Dukascopy's tick quotes, bid and ask. "The open" means different things on different markets, so the opening range was tried at three of them: Tokyo, London and New York. And on top of spread and commission, trades paid slippage, between 0.0012 and 0.0193 R a trade on average, depending on the market and the round.

Three rounds, and the last one never chooses

Round one, in-sample: every version gets scored on the first stretch of years, and the ones that clear the bar go on. Round two, validation: the next four years, no changes allowed. Round three, out-of-sample: the last two years, never used to optimise, rank or pick anything. One thing to know: earlier runs of this study had bugs, and they'd scored round three too. The fixes were written down before the final run, which is the one published here.

The bar. Making money wasn't enough to pass a round. After every cost, a version's trades had to average at least one-twelfth of what they risked (about $8.33 per $100). For EURUSD, that's one round trip's cost at the tightest stop the test allowed. It also had to trade at least 12 times a year. It's a real bar: in round two, 15,380 versions made money and still didn't clear it.

R is one unit of risk. Risk $100 and make $150, and that's +1.5R. Hit the stop, and it's −1R. Dollar figures here assume every trade risks $100 on a $10,000 account, whatever the stop, so versions with tight and wide stops compare fairly.

04 Walk it forward

Think you'd have picked better?

It's the end of round one. You're holding 128,892 versions that all cleared the bar, and you can't see the years ahead. Pick them however you like, then walk them forward.

Picks are ranked across all seven markets together, using each version's own round-one numbers, with ties broken in a fixed order. Every dot is a real version from the study, and what happens to it is what really happened.

Ranking by profit seems to help: 1.38% to 1.60% of the top 5,000 to 25,000 survive, against 0.83% overall. But the list leans on EURUSD, which survives at 1.87% alone, and near the top, random same-market picks do as well. It's market mix. The top ten of each market-and-variation pair, like the video, does worse.

The ones you'd frame

Everyone wants the equity curve that climbs without a wobble. 437 versions made money in at least nine out of ten half-years of round one. The kind you'd frame. 0 of them survived to the end. And it wasn't only them: the steadier a version looked in round one, the less likely it was to last.

05 The coin

Is 1,073 a lot?

1,073 versions that cleared the bar for years, then for four more, then for two more on top. Sounds like a lot. But a count of survivors means nothing on its own. It only means something next to what luck would have produced.

Think of a quiz: ten true-or-false questions. Guessing all ten right is about one in a thousand, so give it to a hundred thousand people who never studied, and about a hundred get a perfect score anyway. It gets worse when they copy. Lots of these versions share the same trades, like a classroom copying one kid. If he's right, they all ace it at once. Luck doesn't spread out. It comes in clumps.

Enter the coin

Take every real trade, exactly where it happened: same days, same markets, same spread, same commission. Then flip a coin for each day. Heads, the day stays as it was. Tails, it's reversed: a day that made $200 before costs loses $200 instead. Costs are paid either way. And every version that traded that day gets the same coin, so the copying stays in.

What's left trades exactly like the strategy and pays exactly what it pays, but has no idea which way the market is going. I took the 89,896 versions that passed round one and traded often enough to be judged, sent the coin through rounds two and three, counted the survivors, and did it 2,000 times. If the strategy has real skill, it should beat almost all of them.

Four coins, same verdict

The first version of the coin test was written down before it ran. Then I built three more, to check it wasn't unfair to the strategy: every market gets its own coin; the coin pays slippage too; or both. The video and the chart up top use the slippage one. The headline number moves between versions. The verdict doesn't.

In every version, a typical coin run finds more survivors than the strategy's 1,073, and no market-and-variation pair passes once you correct for testing 78 of them. The closest is EURUSD · ORB-1.26. It supplies 225 of the survivors on its own, gets to p = 0.0035–0.0060, and still falls short.

One result worth flagging. Put all 35,681 EURUSD versions that reached the coin test into one pile, and measure profit per trade instead of counting survivors. The real trades lost less than their coin twins: −0.029 R a trade, against a typical −0.080 R, with p = 0.0005 to 0.0020 depending on the blocks of days. That pooled measure wasn't one of the tests written down in advance, and even there the versions still lost money after costs. It's the one place this data hints at something beyond luck, and it doesn't make any version worth trading.

06 Market by market

Most survivors came from a market it wasn't built for

Two-thirds of all the survivors came from a single forex pair, EURUSD. The Nasdaq, right on the strategy's home turf of US index markets, gave 13. A typical coin run there found between 92 and 117, depending on the version of the test.

What holds in every version: in five of the seven markets, the strategy found no more survivors than a typical coin run. In the other two, EURUSD and USDJPY, it found more, but not more than luck regularly produces.

07 Every survivor

All 1,073 survivors, and how each one did

Here's every survivor, market by market and variation by variation, with its results in all three rounds: round one, validation, and the final two years. Click a square in the grid to list that group. Sort any column. Search for anything.

08 Long only

"You should've only let it buy." Fair. So I did.

I ran it again on the three US indices, buying only. Ten of the twelve variations can run that way. For the other two, buying only would change what the rule is, and the files explain why. 465 survived, about twice as many as when it could trade both ways. Then the coin took the same test.

All three indices together: 465 survivors. A typical coin run found 1,196 in the pre-registered version, and 934 when the coin pays slippage too. In every version, the coin usually found more, and no market-and-variation pair passes. Buying only didn't save it.

For scale: over the same final two years, buying the index and doing nothing made $2,915 to $4,777 on $10,000. The typical long-only survivor made $601.67 to $724.77. It's not a fair race. Buy-and-hold puts the whole account in the market (its worst fall was 18.9% to 26.4% of the account), while each strategy risks $100 a trade. But it shows how much of the long-only result was just the market going up.

09 What this can't tell you

What this can't tell you

1. Whether a different entry works. This is one family of strategy. The versions that trade with the break or against it get in the moment price breaks out of the range. Only the reversal waits for a candle to close back inside. So a close outside the range before entering, a retest, and a second break are untested. That's part two. And the documented failed-break version ran at one fixed setting of its own parameters, not a sweep.
2. Other markets, other years. Round one starts in 2009, 2011 or 2014 depending on the market, and round three ends in August 2026. Other markets, other years and other sessions could behave differently.
3. How much of long-only is the market. The S&P 500 rose 36.4% in the final two years. A strategy that only buys should beat one that trades both ways on that alone.
4. Whether any single version is worthless. The coin test is about the crowd. It shows that this many survivors is what luck produces. A real edge could still be hiding in there among the lucky ones. The test just can't tell it apart from them.
5. Other data, other costs. Every price is Dukascopy's, and every cost comes from one model of spread, commission and slippage. A different broker would move individual results. The coin paid the same costs as the strategy in every comparison, though, so the comparison stays like for like.
6. Exact dollars. Dollar figures assume $100 of risk per trade on a $10,000 account. Buy-and-hold is price only: no dividends, no financing costs.

10 Check my work

Go find what I missed

Don't trust a backtest because it's on a nice page. That includes this one. Every number here, and every number in the video, comes from these files: the study's 70 CSVs plus 13 made for the page and the video from the same stored results, 83 in all. The READMEs define every column.