63.2% Win Rate on Gold, From 11,520 Tests and Zero Lines of Code

Gold reverts to its London open. I have the numbers on it now: 87 trades over twelve months, 63.2% of them winners, profit factor 2.60.

It took me about a minute to find, and I did not write a line of code to do it.

The setup nobody will give you a number for

Gold pulls away from its London open, then comes back. Traders have described that behaviour for years. Almost nobody has measured it.

Look up opening range breaks and you will have a hit rate inside five minutes. Inside days, gap fills, the New York open, all measured, all quoted endlessly. Time based range reversion gets drawn on a chart, described in three paragraphs, and then the article ends before anyone tells you how often it actually works.

So here it is stated properly, in a form you could hand to a stranger and have them take the identical trade.

Take the price London opens at. Build a volatility band around it, measured in standard deviations of that session's own movement. The first time price stretches out and touches that band, you fade it. You are betting it comes back towards the open. Your stop sits further out, another sigma beyond the band. And the trade only counts inside a set window after the open. Miss the window and the setup is dead for that day, however good it looks at three in the afternoon.

Three dials, then. How far price has to stretch. How far behind that your stop sits. How long the entry stays live.

Every dial changes the answer, and none of them has an obvious right value. Which is exactly why nobody publishes a number for this setup: to get one, somebody has to test all of them.

So I tested all 11,520 of them

Every combination of those three dials, across chart timeframes, direction assignments, and a full grid of stops and targets. 11,520 configurations, run against 257 real London sessions of gold.

Then the part that makes the number worth anything.

Search 11,520 versions of anything and you will find one with a gorgeous equity curve. On pure noise, with no edge present at all, a grid that size hands you a beautiful winner essentially every time. This is why most backtest screenshots are worthless, and why the people posting them never mention how many variations they tried before they found the one in the picture.

So the winner gets put through two more tests before it is allowed to mean anything.

A walk forward. The year is cut into five windows. In each one the entire search re-runs on the earlier portion only, and whatever it lands on is measured on the dates that came next, which it has never seen. A real edge keeps getting found and keeps working on the unseen half. A curve fit collapses the moment you hide data from it.

A permutation test. The direction of every trading day is flipped at random, a thousand times over, building a picture of exactly what luck looks like on this data. The real result is measured against that, and adjusted for the size of the search, because beating luck once is not impressive when you looked at eleven thousand candidates to do it.

Most configurations fail one or both. This one passed both, which is the only reason it is in this article.

The configuration that survived

A 15 minute chart. Band at 0.75 standard deviations from the session open. Stop a full sigma beyond it, stop multiple 0.5, target multiple 0.75. An entry window of 240 minutes from the London open.

Then the detail I would never have guessed in a hundred years of screen time.

It trades a different direction depending on the day of the week. Short Monday. Long Tuesday. Short Wednesday. Long Thursday. Long Friday.

Read that again, because it is the whole argument for testing over theorising. No amount of chart time hands you a weekday direction map. It either falls out of the data or it never occurs to you at all.

What it paid

87 trades across the twelve months, roughly one every three sessions. 63.2% closed green. Profit factor 2.60, so it made two pounds sixty for every pound it handed back.

Expectancy came in at 0.50R per trade. Half your risk, banked, on average, every time you take it. Across 87 trades that is about 44R for the year, and the worst drawdown along the way was about 3.4R.

Those two numbers together are the ones that matter, and they are the only ones that do not change when you change your size. Forty four R gained against a worst stretch of three and a half. Everything below is just that result expressed in somebody's account.

What that looks like on micro gold

The full COMEX contract is 100 ounces at $100 a point, which with gold near $4,400 is around $440,000 of notional. Very few people are trading that, and I am not going to pretend otherwise.

The micro is a tenth of it: 10 ounces, $10 a point. On the micro, this strategy risked roughly $220 per trade and returned about $9,700 across the year, with a worst drawdown near $760.

Now put that on an account. Risking $220 a trade on a $25,000 account is just under 1% of it per trade, which is a normal, unremarkable amount of risk. At that sizing the year is worth roughly 39%, and the worst drawdown you would have sat through is around 3%.

That is the honest way to read any strategy result, including this one. Not the headline dollars, which only tell you what size the tester happened to run, but the R and the drawdown, converted into whatever account and risk level you actually use. Two contracts doubles both halves of it. Half the risk halves both. The 44R does not move.

If 39% a year sounds underwhelming, that is the problem

I know exactly how that number lands right now. Trading Twitter is wall to wall with people posting six figure weeks, and it has quietly rewired what traders believe a good strategy is supposed to look like.

It is body dysmorphia for traders. See a distorted picture often enough and your own perfectly healthy result starts to look pathetic to you.

So let me put it plainly. A strategy returning 44R a year against a three and a half R worst drawdown is not a small result. It is a very good one. The reason it does not print a six figure week is not the strategy. It is the capital behind it.

That is the thing people have backwards. They go hunting for a bigger edge when what they actually need is a bigger balance. I have just shown you the identical signals, unchanged, returning $97,417 instead of $9,700, purely because the position was ten times the size. Nothing about the strategy got better. The money behind it did.

Which is how this works as a business rather than a hobby. You find something that holds up, you prove it holds up out of sample, and then you put capital behind it. In practice that means trading one account and mirroring it out with a copier across as many funded accounts as you can pass and keep, within each firm's own rules. The edge does not change at all. The income multiplies with the number of accounts standing behind it.

And the same move repeats at the next size up. A track record built on a strategy that survived proper testing is the thing you take to other people's money, whether that is a handful of backers or your own fund.

Find the edge. Prove it. Then feed it capital. In that order. A validated 44R strategy running across ten funded accounts is a serious income. The trader searching instead for a strategy that makes 44R a week will be searching forever, because it does not exist, and the people implying otherwise are selling you something.

What answering this used to take

Five years ago, getting the paragraph above in front of you meant a project.

You needed clean intraday data for gold going back a year, which meant paying for a feed and then discovering it had gaps in it. You needed a backtester, so either you learned enough Python to write one or you fought with somebody else's. You needed a cost model, because a strategy tested without commissions and slippage is fiction, and you needed the right contract specification for every instrument you touched, because a tick value entered wrongly quietly turns a losing system into a winner on paper.

Then you needed the statistics. Not a backtest, which anyone can run, but a walk forward across rolling windows and a permutation null corrected for the size of your search. That is where almost everyone stops, which is why so much of what circulates as a proven edge is a curve fit nobody stress tested.

Days of work, assuming you already knew how. And at the end of it you had the answer to one question, about one setup, on one market.

What it takes now

You choose an instrument, a session and a strategy. You set the ranges you want swept, or leave the defaults. You press run.

Seconds later you get a ranked table of what worked, a response surface showing whether your winner sits on a broad plateau or a lucky spike, and a verdict on every row telling you whether the checks backed it or threw it out.

The data is already there. The commissions, the slippage and the correct contract specification for every instrument are already there. The walk forward and the permutation test run on every result without you asking. That is Edge Lab, and it is built for traders rather than for quants.

Start with the setup you already trade. Genuinely, that is the best first use of it. Most traders have never seen their own strategy measured across a full grid of stops and targets, and the usual result is finding out that the version they trade sits near a good one, but not on the good one. That gap is free money inside a strategy you already know how to execute.

Or point it at gold and the London session, run time based range reversion, and see what it says on today's dates. The window rolls forward, so the answer moves as new sessions land, and you should be checking it rather than trusting mine.