A big candle, a small one inside it, then a close outside the big one. One of the most taught gold setups, tested over ten years.
One of the 593 trades in the test: Tuesday 21 January 2025, XAU/USD on H1, times in New York. Each picture is the chart as you would see it replaying bar by bar.
| SPLIT | TRADES | WIN RATE | RESULT |
|---|---|---|---|
| All trades | 593 | 37.8% | −16.6R |
| Buys | 309 | – | −12.8R |
| Sells | 284 | – | −3.9R |
| In-sample, 2016–2020 | 279 | 36.9% | −10.8R |
| Out-of-sample, 2021–2026 | 314 | 38.5% | −5.8R |
The win rate looks fine on its own: 37.8% against the 33.3% a 2:1 trade needs. But that break-even only holds when every win pays the full 2R. Trades still open at 16:00 New York are closed wherever price is, so part of the wins are small ones, and the average trade came out at −0.028R.
The same test was run twice more on the same days and hours, with the same stop and 2:1 target, but without the pattern choosing:
The inside bar did worse than the average random entry; only 28.5% of random runs finished below it. Its −16.6R also sits well inside the range random entries produce by chance, so the test cannot tell this pattern apart from no pattern at all. See how much a backtest can be off by chance.
Verdict: on gold’s one-hour chart, the inside bar breakout is a nice-looking pattern, not an edge. It lost in both halves of the test, in 7 of 11 years, and a random candle did better.
Untested ideas — each one a single change you can run through the replay in an afternoon:
The first minute of the New York stock open sets the range; trade the first break of it. Same rule as our Nasdaq test, fixed before the run.
One of the 2,773 trades in the test: Thursday 4 June 2026, XAU/USD on M1, times in New York. Each picture is the chart as you would see it replaying minute by minute.
| STRATEGY | COIN FLIP | RANDOM MINUTE | |
|---|---|---|---|
| Trades | 2,773 | same trades | about 2,771 |
| Win rate | 36.5% | 32.3% | 36.1% |
| Total | +258.9R | −87.2R | +215.1R |
| 2016–2020 (in-sample) | +162.2R | −32.8R | – |
| 2021–2026 (out-of-sample) | +96.7R | −54.4R | – |
Of all trades, 35.0% reached the target, 63.5% were stopped out and 1.5% were still open at 16:00. The median trade lasted 3 minutes. Both sides made money: 1,417 buys returned +96.6R and 1,356 sells +162.3R.
The same entries with a nearer and a further target, as a robustness check (not in the video):
| TARGET | WIN RATE | BREAKS EVEN | TOTAL | COIN FLIP | 2021–2026 |
|---|---|---|---|---|---|
| 1:1 | 52.8% | 50% | +152.3R | −50.3R | 52.5%, +74.2R |
| 2:1 | 36.5% | 33.3% | +258.9R | −87.2R | 35.5%, +96.7R |
| 3:1 | 27.6% | 25% | +277.9R | −74.2R | 26.6%, +87.9R |
Every target is positive in both halves, and every one beats the coin flip in all 400 runs. A result that holds across targets is harder to explain as luck than one that only works at a single setting.
Two comparisons, on the same rule:
The honest reading: trading the break of a one-minute range on gold made money from almost any minute of the session, and 09:30 did a little better than average — but +258.9R sits inside the random-minute range, so the opening minute itself is not proven special. What the test does show is that following the first break beats guessing its direction, every time it was checked.
This is a one-minute strategy, and that is where costs bite. 1R is the 09:30 candle's range: a median $1.13 over the whole test, and $5.07 over the last twelve months as gold’s price climbed. Every trade pays the spread once.
The $0.61 is gold’s median spread around the New York open, measured by FxBacktest from Dukascopy bid and ask data over the past year — the same figure the replay charges. Charged on each of the last 258 trades, it takes +41.9R down to +9.1R: still positive, but about +0.035R a trade, before any commission or slippage on the breakout fill. Earlier years cannot be costed the same way, because their spreads were not measured. If your broker’s gold spread is wider than $0.61 at the open, this rule may not survive at all.
Verdict: the first gold rule on this page with a real result. Positive in both halves, at every target, in 10 of 11 years, and far ahead of a coin flip. But the edge is thin once the spread is paid, and it is not unique to 09:30 — test it with your own broker’s costs before you trade it.
On a free account you can replay the last six months of gold’s one-minute chart; minute history back to 2003 is part of Ultra.
Results in R, where 1R is the amount risked on a trade. Every strategy is tested the same way and stays on this list whatever the result. Data from Dukascopy.
Most strategy pages show the trades that worked. These are the rules every strategy on this page goes through instead, so the results can be compared with each other.
Entry, stop, target, time exit and filters are written down before the first trade is counted, and nothing is tuned after seeing the result.
Dukascopy XAU/USD data from 2016 to the present — the same candles you replay in the app. When stop and target fall inside one candle, the stop counts.
2016–2020 and 2021 onwards are reported separately. A rule that only works in one half is not a rule you can lean on.
Every rule is run against random entries or a coin-flip direction with the same stop and target. If random does as well, the pattern is not what made the money.
Results are shown before spread and commission and labelled that way, and where costs decide the verdict we charge gold’s measured spread too.
A strategy is added with its result, good or bad, and stays on the page. You see the whole record, not the highlights.
No download, no data to buy. A browser and a free account.
Pick gold, the timeframe your rule uses and a start date.
Keep gold’s measured spread or type your broker’s, add commission, set risk per trade.
Step candle by candle. When the setup appears, enter with the stop and target your rules give.
Win rate, expectancy, profit factor and drawdown — with the spread already paid.
Short answers about the strategies, the results and testing them yourself.
There is no best one until it has been tested on your rules, your timeframe and your costs, which is why every strategy here comes with its full result. Of the two tested so far, the 1-minute opening range breakout at the New York open is the only one with a positive result: +258.9R over 2,773 trades before costs, and thin after the spread. The inside bar breakout on H1 lost −16.6R.
Before costs, yes: trading the first break of the 09:30 New York one-minute candle made +258.9R over 2,773 trades from 2016 to 2026, was positive in both halves of the test and in 10 of 11 years, and beat a coin-flip direction in every one of 400 runs. After gold's measured spread, the last twelve months kept +9.1R of +41.9R, so the edge is thin and depends on your broker's costs.
Not as tested here. On XAU/USD H1 from 2016 to 2026 it took 593 trades, won 37.8% at a 2:1 target and finished at −16.6R before costs, with 7 of 11 years losing. Entries on a random candle with the same stop and target averaged +5.7R.
Because 33.3% is the break-even win rate only when every win pays the full 2R and every loss costs exactly 1R. Trades still open at 16:00 New York are closed at the market, so part of the wins are smaller than 2R. The inside bar breakout's average trade came out at −0.028R.
No. The main results are before costs and labelled that way. For the one-minute opening range breakout we also charged gold's spread measured around the New York open: it cut the last twelve months from +41.9R to +9.1R. When you replay a strategy in FxBacktest, the measured spread for each hour is charged on every trade, or you can type your broker's spread and commission.
It depends on the strategy, so each one states its own. The inside bar breakout was tested on H1 with signals from 03:00 to 11:00 New York; the opening range breakout on M1 at the 09:30 New York open. You can replay gold from M1 to monthly, so the same rule can be tried on another timeframe in minutes.
Yes. Open XAU/USD on the strategy's timeframe, step through the chart candle by candle and place each trade the way the rules say, with the stop and target set before you enter. The session report gives win rate, expectancy, profit factor and drawdown. A free account is enough to start, no card; one-minute gold history beyond the last six months is part of Ultra.
Yes. Each new strategy is tested with the same protocol before it is added: rules fixed before the run, 2016 to the present on real XAU/USD data, an in-sample and out-of-sample split, and a random-entry comparison. It goes on this page whatever the result.
More on gold: gold backtesting, how to backtest gold and gold tick data. Trading faster timeframes? See forex scalping strategies.
Create a free account, open XAU/USD and run any strategy on this page — or your own rule — through real gold charts.