Source: https://forexiapro.com/research/gold-strategies-tested/

# Gold Trading Strategies Tested: What Failed and Why

Dylan Forexia11 November 20267 min read

Gold trading strategies tested on our data mostly failed: every intraday rule we ran on XAUUSD lost money after costs, in every period. What stayed above zero was trend following in a market that rose. Gold's long rise did not rescue the rule: buys lost as well as sells.

R is the amount risked on one trade, so -0.10R means losing a tenth of your risk on average. The chart below splits the setup we teach, run as a fixed rule, into its buys and sells on gold alone. Neither side made money in any period.

[![Gold trading strategies tested: bars of the loss per trade on XAUUSD for buys and sells in four periods from 2012 to 2026, all below zero](/seo/gold-strategies-tested/gold-trading-strategies-tested-buys-and-sells.webp)](/seo/gold-strategies-tested/gold-trading-strategies-tested-buys-and-sells.webp)

One fixed version of the setup on XAUUSD, after costs, March 2012 to 2 October 2026. Buys and sells both lost in all four periods, in a market that ended far above where it began.

## How were these gold trading strategies tested?

They were written as fixed rules and run on real bid and ask prices, with no judgement, as part of three larger studies. Gold was read apart from each for this page.

|  |  |
| --- | --- |
| Data | Dukascopy 5-minute bid and ask for XAUUSD, 2012 to 2 October 2026 |
| Costs | The real spread on every bar, plus $0.15 a round trip for commission and slippage |
| Typical spread | $0.39, against a median hunt of $7 past the Asian range (2021 to 2026) |
| Hour-based tests | Buy at the ask, sell at the bid, plus 0.7 basis points a round trip |
| Fills | When a 5-minute candle touched both the stop and the target, the stop came first |

The setup is described here as it was tested, not as it is taught. A program flagged [the stop hunt](/glossary/the-stop-hunt/), what most traders call a liquidity sweep, as it happened, and traded the reversal the same way each time. The settings are not on this page.

## Which gold trading strategies failed, and by how much?

All of the short-hold rules did. Here they are side by side.

| Rule on gold | Period | Trades or days | Result |
| --- | --- | --- | --- |
| The setup, out the same day | 2012-14 | 757 | -0.107R a trade |
|  | 2015-20 | 1,622 | -0.115R a trade |
|  | 2021-24 | 1,147 | -0.146R a trade |
|  | 2025-26 | 358 | -0.089R a trade |
| The setup, held to Friday | 2012-14 | 297 | -0.110R a trade |
|  | 2015-20 | 609 | -0.043R a trade |
|  | 2021-24 | 433 | -0.201R a trade |
|  | 2025-26 | 158 | -0.076R a trade |
| Short between the London auctions | 2015-2026 | 3,033 days | -7.1% a year net |
| Long overnight, short between the auctions | 2015-2026 | 2,981 days | -13.2% a year net |

The auction tests have no stop, so they are stated in percent a year. The same-day rows come from a rerun on 30 October 2026, and the held-to-Friday rows from the study's earlier results file, which has the same-day 2025-26 row at 356 trades and -0.094R. The first set of rows is [the same test as our main backtest](/research/mechanical-setup-backtest-2015-2026/), on gold alone. Gold was about in line with the group: all eight markets lost, in every period.

## Why did the setup fail on gold?

It failed for the same reason it failed everywhere: the moment itself has no lean. Across all eight markets in the fitting years, price after the trigger moved about zero before any costs, and the whole rule made about -0.03R a trade with no costs at all. We did not run that no-cost ladder on gold alone, so we say this of the group.

**The bull run is not the excuse.** It is tempting to say a market in a long rise runs over anyone selling its highs. But the worst sells, -0.159R a trade, came in 2012-14, when gold mostly fell, and the buys lost in every period as well, from -0.046R to -0.152R a trade. If a rising market were the whole story, the buys would have made money.

**Good-looking slices do not last.** Gold on New York Tuesdays was the best slice of 2012 to 2014 when every market was split by session and day. In the next period it lost -0.070R a trade. Held to Friday, the same slice looked best in 2015 to 2020, and then lost -0.426R a trade on 37 trades. Choosing by past profit selects luck.

The [M and W patterns](/price-action/m-pattern-forex/) you see on a finished gold chart are real shapes, and across all eight markets the M or W slice lost too.

## Why did the London auction trade fail?

Because what is there is smaller than the cost of collecting it. The rule published by Speck and others, as the study names it and which we could not find at its source, is that gold rose overnight and fell between the two [London auctions](https://www.lbma.org.uk/prices-and-data/precious-metal-prices) for decades. Our data agrees on where the rise came from: gold's rise was outside those hours. Between the auctions it was about flat before costs, not falling.

Then a trade every day pays the spread every day. Shorting between the auctions lost 7.1% a year net, with one year up of twelve and a t-statistic of -2.8. Doing both legs pays it twice, and lost 13.2% a year, with one year up of twelve and a t-statistic of -3.6. Holding gold overnight by itself was not judged, because gold rose through the sample and that would only test the rise.

The rest of those tests, on forex, are on [the page for the twenty tests of published strategies](/research/published-trading-strategies-backtest/).

## Did Gann levels or the COT report work on gold?

Gann's Square of Nine levels and time counts lost on gold, and his trend rules did what trend rules do. The typical variation on XAUUSD alone was -0.158R a trade for Square of Nine levels, -0.118R for time counts and -0.067R for squaring price and time. Swing and overbalance rules stayed above zero, and so did a plain channel breakout with no Gann in it and, by less, his retracement levels, in a market that rose. [The full Gann test](/research/gann-levels-75-markets/) has the controls.

The COT report ran against its general finding here. Across all eight markets, joining the crowd lost and fading it held up best. On gold alone the two swapped places: fading the crowd lost, and joining it did not. It is one market with about 150 trades each way, so we would not build a gold rule on that. It does mean you should not assume the headline COT result covers gold. [Does the COT report work](/research/does-the-cot-report-work/) and [fading the crowd](/research/fading-the-crowd-cot-level-hunt/) carry those figures with their caveats, and [our gold COT page](/cot-report/cot-report-gold/) shows how to read the report on gold.

## Did anything on gold make money?

Trend following stayed above zero, and that is not a claim we would make to you. Trend rules on gold finished ahead, and gold rose a long way. A rule that buys a rising market earns the rise. We have no test showing it would have done the same in a market that went nowhere.

## What does this test not prove?

It does not prove that gold cannot be traded, or that a person reading a day would lose. Here is what it leaves open.

- **Judgement is not tested.** These are rules with no opinion about the day. Nothing here says whether a trader can pick the good ones.
- **Gold's news is not modelled as a reason.** Real moves on gold have real causes, as [one weak jobs report](/why-is-gold-falling-after-weak-jobs-report/) shows. The tests read candles.
- **It is one data source.** Spreads on gold differ widely between brokers, and real fills around a release are worse than 5-minute candles show.
- **The samples are uneven.** The 2025-26 rows are short, and the sides split one market's trades further.
- **It is not investment advice about gold.** Holding gold is a different question, and it is not tested here.

The rest is on [the research page](/research/). Course 3 of ForexiaPRO, Reverse Price Action, teaches the read behind the setup step by step, and lesson 3.3.4 covers the hunt that keeps going.

## What changed when this page came back?

This page was taken down on 31 October 2026 and put back on 11 November 2026. Its figures were rerun on 11 November 2026 with scripts/seo-research-figures.mjs and cherrypick.mjs. The same-day gold rows already came from a rerun and still match: 358 trades at -0.089R in 2025-26. Nothing changed. No finding was reworded to read better than the numbers.

## Frequently asked questions

### Is there a gold strategy that works?

We did not find one we can stand behind. Every intraday rule we ran on XAUUSD lost after costs. Trend following made money in a market that rose, and that is a different statement.

### Does buying gold dips beat selling gold rallies?

Not in the setup we tested. Run as a rule, both sides lost in every period from 2012 to 2026, though gold ended the sample far above where it began. The buys ranged from -0.046R to -0.152R a trade and the sells from -0.093R to -0.159R.

### Does gold really rise overnight and fall in London hours?

Half of it showed up before costs: gold's rise came outside the hours between the two London auctions, and between them gold was about flat, not falling. It did not survive costs. Shorting between the auctions lost 7.1% a year net, and trading both legs lost 13.2% a year.

### Do these tests mean gold cannot be traded?

No. They mean these rules, run with no judgement on these prices, did not pay. They say nothing about a person's read of a given day, and nothing about holding gold as an investment.

## More from the blog

### [Copy Trading vs Forex Signals: What You Actually Get](/learn-forex/copy-trading-vs-forex-signals/)

Copy trading places the trades in your account; a forex signal leaves them to you. What each gives you and costs you, on one real dated EUR/USD signal.

### [Double Top Failure Rate: How Often Double Tops and Bottoms Fail, Counted](/price-action/double-top-failure-rate/)

The double top failure rate, counted: 983 tops and bottoms on 15-minute forex and gold candles, what the neckline did, and two real days that failed.

### [COT Report Release Date: When It Comes Out, With the 2026 Schedule](/cot-report/cot-report-release-schedule/)

The COT report release date is Friday 15:30 New York, positions counted the Tuesday before. The CFTC's 2026 dates, six holiday Mondays and the next four.