Best Gold Trading Strategy? We Backtested 14 Popular Strategies on 22 Years of Gold

00The short answer

No gold trading strategy we tested is proven to work. Of 97 versions of 14 popular strategies, one passed all six of our gates: the 50/200 golden cross on 1-hour bars, long only, with no stop and no target, which made +646 R after costs over 387 trades. A later audit found its early data too thin to trust, so its verdict is Inconclusive and it is being tested forward. Four more strategies passed five of the six gates and failed one.

• The closest were long-only trend rules on the 1-hour chart that let each trade run until an opposite signal. The golden cross, the EMA 9/20 crossover and the TTM Squeeze with its patient exit made +646 R, +610 R and +421 R after costs in their best versions (R is the risk of one trade; see below).
• Gold rose about elevenfold in these years, and that helped them. In the one stretch when gold fell, 2013 to 2019, the EMA 9/20 and the TTM Squeeze nearly stopped making money (profit factors 1.05 and 1.02; we require 1.2). Seven versions were strong enough in 2004 to 2012 to pass our development gate; six of them then failed in 2013 to 2019.
• Fast charts lost to costs. None of the 12 EMA 9/20 versions on 15-minute bars and none of the 4 RSI 30/70 versions on 5-minute bars made money after costs. A $0.30 round trip costs about 0.16 R a trade on the 15-minute EMA cross and about 0.07 R on the 1-hour one.
• Buying oversold and selling overbought did worst. The classic RSI 30/70 rule won about two trades in three and lost 4,382 R on 5-minute bars; the classic WaveTrend signal lost money even before costs.
• The short side lost almost everywhere. In 57 of the 61 versions that could sell short, the shorts lost money after costs.
• Beating random entries was not enough. All 12 opening range breakout versions beat their random control, and 9 of them still lost money after costs.

01Who we are, and why you can check these numbers

RIP Backtest is an independent site run by one retail trader. We test the strategies and indicators that traders share, under one fixed protocol: the rules are written down and dated before any data is run, costs are charged on every trade, the years are split into periods that are judged separately, and every version is published, including the ones that fail.

We sell nothing. No strategies, signals, indicators, robots or courses; there are no affiliate links on the site, and this guide names no broker or vendor. Many pages that answer “what is the best gold trading strategy” show one strategy, one chart and a win rate, and some exist to sell an indicator or an expert advisor. This guide runs no new test: every number below comes from the 14 gold tests already published on the site, each with its specification, code and complete results, and each section links to the full write-up. Read more about the site and our testing protocol.

02How every gold strategy was tested

The 14 tests share one setup, so their results can be put side by side.

• Data. Spot gold (XAUUSD), 1-minute bid prices from a retail broker's MT4 demo server, from 11 June 2004 to 7 October 2026, built into 5-minute, 15-minute, 1-hour or daily bars as each strategy needs. This is a spot (CFD-type) price, not a futures contract. One test, the EMA 9/21 crossover, used gold futures (GC) from 2020 to September 2026 instead.
• Costs. $0.30 an ounce for every round trip on spot gold, $30 for a standard lot of 100 ounces; on gold futures, commission plus one tick of slippage on every fill. Overnight financing (swap) is left out of the baseline because the broker's past rates are not known; most tests also show what today's swap would do, and we come back to it below.
• R, the unit of most results. R is the risk of one trade, fixed when the signal fires: in most tests 1.5 times the 14-bar average true range (on 1-hour gold a median of about $5.20 an ounce), in others the strategy's own risk (the Turtle's 2N, the size of the opening range, the height of a head and shoulders, the stop at the engulfing candle or at the day's extreme in the H1 Open Trap). A result of +2 R means twice that distance. R lets a trade at $400 an ounce in 2004 count as much as one at $4,000 in 2026. Because R is not defined the same way in every test, compare it between strategies only roughly. The noise area test reports dollars per ounce, the daily golden cross dollars per ounce and percent, and the night vs day test the sum of its trades' percent returns.
• Three periods. Development, 2004 to 2012, where a result must be statistically convincing; validation, 2013 to 2019, years it was not judged on; and a holdout, 2020 to 2026, as a last check. Earlier tests had already used the gold holdout, so it is flagged as used, not clean. (Gold futures: 2020 to 2023, 2024 to March 2025, April 2025 to September 2026.)

Six gates. A version must pass all six to survive:

1. A profit factor of at least 1.2 in validation.
2. At least 50 trades in each period.
3. A development t-statistic of at least 2.0, or 2.5 when the test tries more than four versions or the idea was tested before.
4. A result above the 95th percentile of a random control: the same kind of trades placed at random or shifted times, many times over.
5. A positive result without its 10 best trades.
6. A positive result in the holdout.

A version that fails any gate is Buried; one with fewer than 30 trades, or whose verdict changes when doubtful data is removed, is Inconclusive. The gates are strict for a reason: 97 versions were tried on the same 22 years of gold, and with that many tries some will look good by luck alone.

03The ranking: 14 gold strategies, from closest to furthest

Each strategy is ranked by its best version on gold: first by the number of gates it passed, then, for ties, by its development t-statistic. A higher rank means it got closer, not that it works: none of the 14 has a proven edge.

Gates passed by each strategy's best version. * Passed all six in the run; the verdict became Inconclusive after the data check.

After costs: the whole test, 2004 to 2026 (gold futures: 2020 to 2026), before any financing; R is defined in each test, $/oz is dollars per ounce, % (sum) is the sum of trade returns in percent. The gates are numbered as in the list above. Each strategy's full write-up explains its version codes.

04Golden cross on 1-hour gold: passed every gate, then the data check

The rule: buy when the 50-bar simple moving average of 1-hour closes crosses above the 200-bar one, and close the trade (or, in the long-and-short versions, reverse it) when it crosses back below. All 24 hours count, and positions are held over weekends. We ran 12 versions on gold: long only and long and short, each with no stop, with a 1 R stop, and with targets at 1 R, 2 R, 3 R or a third at each.

One version passed all six gates: long only, with no stop and no target. Over 387 trades it made +646 R after costs (+$2,864 an ounce), with a profit factor of 1.91 and a development t-statistic of 2.65, above the 2.5 required. Only 41% of its trades won; the money came from a few long trends.

On 8 October 2026, an audit of the gold data found the development years, 2004 to 2012, thin, with 2,099 suspect gaps. Without the 26 trades whose signal or fill touched one of them, the development t-statistic falls to 2.29, below the bar. Our data rule says a verdict that changes this way becomes Inconclusive, and that is where it stands: not buried, not proven.

The other versions were not close. Every version with a stop or a target failed at least three gates; the best long-only one made +88 R. The long-and-short version with no stop made +551 R, but its shorts lost 95 R and it failed the development gate (t 1.36).

Two cautions. It did not beat holding gold: one ounce held from the first signal in June 2004 gained +$3,716, against the rule's +$2,864, both before financing. And financing at 3% a year, charged on every day held, cuts the result to +542 R. The rule is now in forward tests, frozen, each with a weekly journal: the first one, whose demo series was stopped on 15 October 2026 by trades unrelated to the rule, and a second, separate test. Read the full 1-hour golden cross test.

05EMA 9/20 crossover on gold: five gates on the 1-hour chart, none on the 15-minute

The rule: buy when the 9-bar exponential moving average crosses above the 20-bar one, sell when it crosses below. 24 versions: 15-minute and 1-hour bars, long only and long and short, each with no stop, a 1 R stop, and targets at 1 R, 2 R, 3 R or a third at each.

On 15-minute bars, none of the 12 versions made money after costs: they ended between −697 R and −4,038 R. The crosses come often (12,030 trades for the long-only version with no stop), and a 15-minute stop is small, so the same $0.30 round trip costs about 0.16 R a trade, more than twice the 0.07 R it costs on 1-hour bars.

On 1-hour bars, the long-only version with no stop made +610 R over 2,865 trades, beat the random control and reached a development t-statistic of 3.80. It failed the validation gate: from 2013 to 2019 its profit factor was 1.05, against the 1.2 required. Two other long-only versions, one with a 1 R stop and one with a 3 R target, also passed five gates and failed the same one.

The full post adds two warnings about the best version: with today's swap charged on every night held it falls to +29 R, and without the trades that crossed gaps in the thin early data its development t-statistic falls below 2.5. Read the full EMA 9/20 test.

06TTM Squeeze on gold: five gates, stopped by the lean years

The squeeze is on when the Bollinger Bands (20 bars, 2.0 standard deviations) sit inside the Keltner Channels (20 bars, 1.5 average true ranges). It fires when they move back out, and you trade in the direction of the momentum histogram. We tested it on 1-hour gold, long and short or long only, with two exits: the quick one usually taught with it, out after two bars of fading momentum, and a patient one, out only when the momentum crosses zero.

The exit decided the result. With the quick exit both versions ended near zero (−61 R and +63 R). With the patient exit, long and short made +392 R and long only +421 R over 1,670 trades. All four versions beat a control that traded the same direction at other times.

The best version, long only with the patient exit, passed five gates and failed validation: from 2013 to 2019 it made only +6.9 R (profit factor 1.02). It has no stop, so one trade went 11.3 R against it while open, and today's swap, charged on every night held, would cut the result from +421 R to +121 R. Its forward test, a paper series on data that came after the backtest, has its own frozen protocol and journal. Read the full TTM Squeeze test.

07Head and shoulders on 15-minute gold: five gates, weak first years

Drawn by eye, everyone sees a different head and shoulders, so we used a published computer definition (Osler and Chang, Federal Reserve Bank of New York, 1995) on 15-minute bars, at two pattern sizes: sell on a close below the neckline of a top, buy on a close above the neckline of an inverse pattern, with the stop beyond the right shoulder and the target the height of the pattern.

The smaller patterns made money in every period: +46 R over 279 trades, +0.16 R a trade after costs, with the tops (+21 R) and the inverse patterns (+25 R) both positive, and above all 500 runs of a control that placed the same trades at other times. They failed one gate: the 2004 to 2012 evidence alone was too weak (development t-statistic 1.26; 2.0 needed). The larger patterns did no better than random: −6 R over 135 trades.

Two things set it apart from the rest of this list. Its trades are measured against the whole height of the pattern, so the $0.30 round trip costs only about 0.04 R a trade, and its trades last hours, not weeks, so today's swap barely changes the result (+46 R to +45 R). And its tops made money as well as its inverse patterns, which is rare here: in 57 of the 61 gold versions that could sell short, the short side lost. Read the full head and shoulders test.

08EMA 9/21 crossover on gold futures: five gates, most of the profit late

This test ran on gold futures (GC), 1-hour bars from 2020 to September 2026, as part of a test on three futures markets. The rule: buy when the 9-bar exponential average crosses above the 21-bar one, sell when it crosses below. Four gold versions: long only and long and short, each with no stop, or with a 1 R stop and a 2 R target.

The two versions with no stop passed five of six gates, including the random control (percentiles 95.6 and 95.2): long only made +209 R over 860 trades, long and short +160 R. Both failed the development gate, with t-statistics of 1.08 and 0.71 in 2020 to 2023 (2.5 needed). Both versions with a stop and a target lost money.

About two thirds of each result (68% and 64%) came in 2024 to 2026, when one gold contract rose $177,560, against $29,810 in 2020 to 2023 (buying and holding, before costs). The years it was judged on did not support it, and its shorts lost money (−49 R). Read the full EMA 9/21 test.

09WaveTrend on gold: the classic signal lost before costs

WaveTrend is LazyBear's free TradingView oscillator, close to a smoothed CCI. With its default settings (10 and 21, zones at ±53) on 1-hour gold: buy when the fast line crosses up through the slow one in the oversold zone, sell when it crosses down in the overbought zone, and get out at the next opposite cross or hold until the next opposite signal.

The classic signal lost money even before costs. Long and short, out at the next cross, it lost 533 R over 3,441 trades, −0.09 R a trade before costs, in 21 of 23 calendar years, and it did worse than 99% of the runs of a control that placed the same trades on other days.

One version made money: long only, held until the next sell signal, +169 R over 493 trades, with four gates passed. It made +3.6 R from 2013 to 2019 (profit factor 1.01), was no better than random long entries held the same way (percentile 63.4), and with today's swap it would have ended at −283 R. Read the full WaveTrend test.

10Opening range breakout on gold: paid before costs, not after

Mark the high and low of the first 30 minutes of the Asian, London or New York session and trade the break, with the stop at the other side of the range and no target, with and without an RSI filter: six versions on 1-minute gold, then six more with a 1 R target, added after the first results and reported as Round 2. R here is the size of the day's range.

Every original version made money before costs, about 0.07 to 0.18 R a trade, and costs of 0.09 to 0.17 R a trade took most of it. The best net result was London with the RSI filter: +109 R over 3,646 trades, +0.03 R a trade. Three versions ended slightly positive and three negative, and all six Round 2 versions lost money.

All 12 versions sat at the top of their random control: the breakouts were better than random entries, by less than it costs to trade them. On a small range, a fixed cost is a big part of every trade. Read the full gold opening range breakout test.

11Golden cross on daily gold: too few trades to judge

The classic version: buy when the 50-day moving average crosses above the 200-day (the golden cross), sell when it crosses back below (the death cross). Four versions on daily gold: simple and exponential averages, long only and long and short.

The long-only version with simple averages made +$3,015 an ounce over 15 trades, against +$3,685 for buying in March 2005 and holding. Seven of the 15 trades won, and a few long trends carried the total. Its timing beat 97.4% of a control with the same pattern of episodes.

Too few trades to judge. The four gold versions made 11 to 31 trades in 21.5 years, so three of them are Inconclusive. The protocol asks for 50 trades in each period; at this rule's pace, even 50 trades in total would take about 72 years of daily data. The one gold version with enough trades to be judged, long and short with simple averages (31 trades), was buried, and the death cross lost money as a short signal in both long-and-short versions. Read the full daily golden cross test.

12Turtle trading on gold: positive in every period, still fragile

The rules Richard Dennis taught his trainees in 1983, unchanged, on daily gold: System 1 buys a 20-day high and sells short a 20-day low, System 2 uses 55 days; the stop is 2N (twice the 20-day average true range), and the pyramiding versions add a unit every half N, up to four. Four versions, long and short.

All four made money after costs in every period, from +33 R to +225 R over the whole test, and System 2 beat the time-shifted control (percentiles 98.6 and 97.2).

It is still fragile. Without its 10 best trades every version ends below zero; the long side made all the money, and the shorts lost in all four versions; no development t-statistic reached 2.0; and System 2 traded too rarely for the protocol (26 to 44 trades a period, 50 needed). With today's swap, System 2 with four units falls from +225 R to +75 R. Read the full Turtle trading test.

13Noise area breakout on gold: about the size of the costs

The noise area comes from a 2024 paper by Zarattini, Aziz and Barbon: for each minute of the session, average how far the price has usually been from the open at that time of day over the last 14 days. Buy when the price closes above that band, sell short when it closes below, exit at the band or the session's VWAP, and close everything at the end of the session. Six versions: the Asian, London and New York sessions, long and short or long only, one ounce at a time.

Every version ended close to zero after costs, between −$0.34 and +$0.12 an ounce a trade. Before costs they made −$0.04 to +$0.42 a trade, against a $0.30 round trip. The Asia and New York versions ranked above a control that moved the same entries to other days (percentiles 98.6 to 98.8), but beating moved entries is not the same as making money.

The best, New York long only, made +$254 an ounce over 2,114 trades and failed validation (profit factor 1.00). The same rule passed every gate on Nasdaq-100 futures in the New York session; on gold it did not carry over. Read the full noise area test on gold.

14RSI 30/70 on gold: wins two trades in three, loses the money

The most searched indicator rule of all: buy when RSI(14) turns back up through 30, sell when it turns back down through 70, and exit at the other level. Eight versions: 5-minute and 1-hour bars, long and short or long only, classic (no stop) or protected (a 200-bar EMA trend filter and a 1 R stop).

The classic rule won about two trades in three and lost money. Its versions won 61% to 65% of their trades and ended at −4,382 R long and short on 5-minute bars and −573 R on 1-hour bars. Without a stop the losers were huge: the worst open loss inside one trade reached 101.8 R. Selling at 70 lost money even before costs, on both charts.

None of the four 5-minute versions made money after costs. The best results came from long only on 1-hour bars, +45 R classic and +36 R protected, but neither reached a development t-statistic of 1.1 (2.5 needed), from 2013 to 2019 one lost money and the other broke even (profit factor 1.00), and neither beat the control. Read the full RSI 30/70 test.

15Night vs day: gold rose overnight, but not by enough

Published studies of gold futures found that gold tends to rise overnight and fall during the COMEX day session. We tested the plain timing rule on spot gold: long from 13:30 to 08:20 New York (the night), short from 08:20 to 13:30 (the day), or both, every trading day.

The night was positive; shorting the day was not. Long only overnight made +145.2% (sum) over 5,704 nights after costs and passed three gates, but it failed validation (profit factor 0.96), its development t-statistic was 1.27, and it did not beat a control that held gold for the same length of time at other hours of the same days (percentile 90.6): most of the night's gain is that the night is long, and gold rose. Shorting the day session made money before costs and lost it after (−93.5% (sum)). Read the full night vs day test.

16Engulfing candle on 1-hour gold: beat chance, lost to costs

Buy after a bullish engulfing candle, sell after a bearish one, stop at the engulfing candle's low or high, target 1 R or 2 R, with and without a reversal-context or a 200 EMA trend filter: four versions on 1-hour gold.

All four made money before costs, beat a control that moved the same trades to other days, and lost after costs. The stop at the candle was only about $2.98 away (median), so the $0.30 round trip cost 0.14 to 0.15 R a trade, more than the edge. The best, with the trend filter, ended at −549 R over 6,254 trades. Read the full engulfing candle test.

17H1 Open Trap: a retail indicator, half winners, costs took the rest

A trap signal from a retail MetaTrader indicator, tested with its default settings on 1-minute gold: a reversal entry at the hour's opening level, the stop at the day's extreme, targets of 1, 2 or 3 R. A follow-up widened every stop to at least $5.

The 1 R version won about half its trades, and every version lost heavily. The stops were tiny (median $1.37 an ounce), so the cost was about 0.29 R a trade: the 1 R version made +57 R before costs and −2,583 R after, over 9,026 trades. With a $5 minimum stop the cost fell to 0.06 R a trade, and the signal turned out to lose money even before costs (−37 R). Read the full H1 Open Trap test.

18Why gold makes long-only strategies look good

Gold rose about elevenfold in our data, from $384 an ounce in June 2004 to $4,110 on 7 October 2026. That rise sits inside every long trade, so a rule that is long most of the time looks good on gold whether or not its signals add anything. This is why our controls trade the same direction at random times, and why most of these tests show long-only and long-and-short versions side by side.

Gold (XAUUSD) at the end of each year, from the June 2004 open to 7 October 2026, in the data used by the tests. Shaded: development (2004 to 2012), validation (2013 to 2019, in red) and holdout (2020 to 2026).

The rise was not even. In the development years, 2004 to 2012, gold went from $384 to $1,675. In the validation years, 2013 to 2019, it fell: −28% in 2013 alone, to a low of $1,050 in December 2015, and it ended 2019 at $1,520, below where the period began. From 2020 it rose again, to $4,110.

That falling stretch is where the best-looking results broke. Seven of the 97 gold versions were strong enough in 2004 to 2012 to pass the development gate. Six of those seven then failed validation in 2013 to 2019. The seventh is the 1-hour golden cross, whose development years were later found too thin to trust.

The same lean shows in the totals: 23 of the 36 long-only versions ended positive after costs, against 19 of the 61 versions that could sell short, and in 57 of those 61 the short side lost money. In every one of the 28 long-and-short versions of the trend rules (the moving-average crosses, the Turtle rules and the TTM Squeeze), the shorts lost.

None of this makes long-only rules useless. It means a long-only backtest on gold has to beat something harder than zero: random long entries held the same way, and simply holding gold. The two best 1-hour rules did not beat holding: the golden cross made +$2,864 an ounce against +$3,716 for one ounce held over the same years, and the EMA 9/20 made +$3,626 against +$3,727, all before financing.

19What is the best timeframe for trading gold? What our tests say

No test on this site compares timeframes with everything else held equal, so this is not proof that one chart is best for gold. What the 97 versions do show is the arithmetic of costs: the same $0.30 round trip is a bigger share of a small trade.

The 1-hour chart did best on our data: 24 of its 44 gold versions ended positive after costs, and it holds four of the five strategies that passed five or more gates. The faster the chart, the more of each trade's risk the cost takes: about 0.06 to 0.07 R on 1-hour spot gold, 0.16 R on the 15-minute EMA cross and 0.25 to 0.30 R on the 5-minute RSI. The daily chart made money in all 8 of its versions, but gave too few trades to judge most of them.

Two caveats. The 1-hour strategies here were mostly trend rules that hold until an opposite signal, the kind gold's rise helped most. And the cost in R depends on the stop as much as on the chart: the 15-minute head and shoulders, whose R is the height of a whole pattern, paid only 0.02 to 0.04 R a trade, while the 1-hour engulfing candle, whose stop sits at the candle, paid 0.14 to 0.15 R. If you scalp gold in seconds, our study of 40-second gold scalps on 23 million real ticks found that you had to be right about 60% of the time just to break even after the spread.

20What made a difference across the 97 versions

The exit mattered as much as the entry

Five tests compared exits on the same entries. In their long-only versions, the most patient exit made the most every time:

• Golden cross, 1-hour: no stop and no target +646 R; the best version with a stop or a target +88 R.
• EMA 9/20, 1-hour: no stop and no target +610 R; the best version with a stop or a target +529 R. On 15-minute bars the version with no stop also lost the least.
• TTM Squeeze: out when the momentum crosses zero +421 R; out after two bars of fading momentum +63 R.
• WaveTrend: held to the next sell signal +169 R; out at the next cross −109 R.
• EMA 9/21 on gold futures: no stop +209 R; a 1 R stop and a 2 R target −7.1 R.

Exits that let a trade run until the opposite signal suited a market that trended up for most of 22 years. That is a pattern across our tests, not a rule we tested on its own, and it has a price: without a stop, one TTM Squeeze trade went 11.3 R against the position while it was open, and one RSI trade 101.8 R.

A high win rate meant nothing

The classic RSI 30/70 won 61% to 65% of its trades and lost money in three of its four versions. WaveTrend's long-only version held to the next signal won 64% of its trades and was no better than random. The best result in this whole list, the 1-hour golden cross, won 41% of its trades. What decides a strategy is how much the winners make against the losers after costs, the profit factor and the average trade in R, not the share of winners.

Beating random entries was not the same as making money

Our control asks whether a strategy's timing beats chance. 51 of the 97 gold versions passed it, and 26 of those still lost money after costs, among them nine of the 12 opening range breakouts. A signal can be better than random and still not be worth its costs.

Overnight financing can erase a long-only result

Spot gold positions held overnight pay or receive a swap, and long positions usually pay. None of our baselines charges it, because the broker's past rates are not known, but most tests show what today's rate would do if it had applied on every night held:

Today's swap may not match the rates of past years, so these are sensitivity checks, not results. The direction is clear, though: the longer a strategy holds, the more financing matters. The 1-hour golden cross held a trade for about 12 days on average; the head and shoulders, for hours.

Costs decided 42 versions

42 of the 97 versions made money before costs and lost it after; 25 of them traded on charts of 15 minutes or less. Another 13 lost money even before costs. At $0.30 an ounce, a strategy needs an average trade well above the cost to survive, and on fast charts few did.

21Before you trade any gold strategy: a checklist

1. Write the rules down before you test them, exits included. An exit chosen after seeing the results is a result you fitted.
2. Charge costs in R. $0.30 an ounce is about 0.07 R on a typical 1-hour stop and 0.16 R on a 15-minute one. Our trading cost calculator does the arithmetic for your own spread and stop.
3. Look hard at 2013 to 2019. A gold strategy that only made money while gold rose is a bet on gold rising, with extra steps.
4. Compare with random entries in the same direction and with holding gold, not with zero.
5. Count the trades. Under 50 in a period, luck and edge look the same.
6. Remove the 10 best trades. If the result turns negative, it lived on a few trends.
7. Add the swap if you hold overnight, at your broker's current rates.
8. Count every version you tried. We tried 97 on gold; the more versions, the higher the bar should be.
9. Test it forward on data it has never seen, on paper or on a demo account, before you risk money.

The full method is in our guide how to backtest a trading strategy without fooling yourself, and the gates are explained in the protocol. Every result on the site, gold and other markets, is summed up in what the backtests found so far.

22What this guide cannot tell you

• One price history. One broker's spot gold data; its early years are thin and have gaps, and another feed would give somewhat different numbers.
• Rules as written. We test each strategy as it is published or usually taught, without discretion. A trader who skips trades by judgment is not what we tested.
• 14 strategies, not all of them. Others may do better on gold; we add them as we test them.
• Many tries on one market. 97 versions on the same 22 years: some results look better than they are by luck, some worse, and the holdout was already used.
• The past. Gold's next 22 years will not repeat its last 22.

This is educational content, not financial advice, and not a recommendation to trade gold or any strategy.

FAQCommon questions

What is the best gold trading strategy?

None we tested is proven. The closest was the 50/200 golden cross on 1-hour bars, long only, with no stop: +646 R after costs over 387 trades from 2004 to 2026 and all six gates passed in the run, until a data check found its early years too thin and made it Inconclusive. The EMA 9/20 crossover on 1-hour bars (+610 R) and the TTM Squeeze with a patient exit (+421 R) passed five gates and failed in 2013 to 2019, when gold fell.

What is the best indicator for gold trading?

In our tests no indicator gave a proven edge on gold. Moving-average crossovers on 1-hour bars and the TTM Squeeze came closest. Overbought and oversold oscillators did worst: the classic RSI 30/70 won about two trades in three and still lost 4,382 R on 5-minute bars, and the classic WaveTrend signal lost money even before costs.

What is the best time frame for trading gold?

In our tests, the 1-hour chart: 24 of its 44 gold versions ended positive after costs, against none of the 12 EMA 9/20 versions on 15-minute bars and none of the 4 RSI 30/70 versions on 5-minute bars. The main reason is costs: a $0.30 round trip is about 0.07 R a trade on a 1-hour stop and 0.16 R on a 15-minute one. Daily rules made money but traded too rarely to judge.

Does the golden cross work on gold?

On daily bars we cannot tell: the 50/200 cross gave only 11 to 31 trades per version since 2005. On 1-hour bars, the long-only version with no stop passed all six of our gates (+646 R over 387 trades), but a later data check made its verdict Inconclusive, and it did not beat holding gold (+$2,864 an ounce against +$3,716). It is in a forward test on new data.

Is scalping gold profitable?

Not in our tests of fast rules: none of the 5-minute RSI 30/70 versions and none of the 15-minute EMA 9/20 versions made money after costs. In a separate study of 40-second gold scalps on 23 million real ticks, you had to be right about 60% of the time just to break even after the spread.

Why do gold backtests look so good?

Because gold rose about elevenfold from 2004 to 2026 and every long trade carries that rise. In our tests 23 of 36 long-only versions ended positive after costs against 19 of 61 that also traded short, and six of the seven versions that looked strongest in 2004 to 2012 failed in 2013 to 2019, when gold fell. Compare a gold backtest with random long entries and with holding gold, not with zero.

Do gold trading strategies beat buy and hold?

The two best 1-hour rules we tested did not: the golden cross made +$2,864 an ounce and the EMA 9/20 crossover +$3,626, against +$3,716 and +$3,727 for one ounce held over the same years, all before financing.

How were these gold strategies tested?

On spot gold (XAUUSD) 1-minute data from 2004 to 2026 (one test on gold futures from 2020), $0.30 an ounce of costs on every round trip, with rules written and dated before the data was run, costs included on every trade, results split into development, validation and holdout periods, a random-entry control as the benchmark, and every variant reported. Read the protocol.

23Check our arithmetic

Code provenance: our own code. No new test was run for this guide.

24Corrections

22 Oct 2026, update: added the three gold tests published after this guide (night vs day, the engulfing candle and the H1 Open Trap, 11 versions), sections 15 to 17; the counts across all versions were recomputed from the results files. No earlier result changed; no strategy moved above the five that passed five gates.

+More tests and guides

• Golden cross 1 hour backtest
• Ttm squeeze strategy
• Ema 9 20 crossover backtest
• Head and shoulders pattern
• Free prop firm challenge calculator
• What all the backtests found so far
• The backtesting protocol
• How to backtest a trading strategy without fooling yourself
• Backtesting glossary