ORIGINAL RESEARCH · 26 PAGES
Gold market research
Studies run against our own bar archive — the same data our backtest engine replays. Every figure is reproducible from a published CSV, and where a study contradicted something we had already written, the correction is in the study.
Start here1
Published studies3
- XAUUSD spread and volatility by hourCSV70,546 five-minute bars across 312 sessions. Median daily range $76.12, spread over $1.00 in 3 bars, and a $2 stop hit by noise 81% of the time within an hour.
- How long gold losing streaks actually lastCSV257 sessions and 61 separate losing streaks. The longest ran nine days, but the worst three-day run cost almost exactly as much as it — so a drawdown limit is met by size, not duration.
- Do gold session breakouts follow through?CSV234 overnight-range breakouts, of which 51.7% followed through. The confidence interval contains 50%, so direction is a coin flip and only the asymmetry in size is real.
Reference data1
Studies published as articles21
- What gold actually does at NFP releases11 Non-Farm Payrolls releases. Median release minute 2.7x a normal one, largest 22.6x, elevated volatility over in a median of 2 minutes. Direction continued at 81.8% against a 30,000-minute control of 54.4%.
- Do round numbers act as support and resistance?35,002 approaches to price levels. Round numbers turned price back 56.6% of the time; levels chosen because nobody watches them, 56.1%. The difference contains zero.
- Daily high and low timing by sessionAsia sets the daily low 48.1% of the time and London sets either extreme about one day in ten — the opposite of what most session guides claim.
- The stochastic 80/20 rule tested772 signals. The oversold buy beats doing nothing for about an hour and then loses to it; the overbought sell was wrong at every horizon measured.
- Weekend gaps on XAUUSD52 gaps, median $7.60, and 48% exceeding $10 — larger than a sensible intraday stop.
- How many trades it takes to judge a strategyA strategy with no edge finished profitable 44.1% of the time after ten trades. Distinguishing a real edge at 95% confidence took about 416.
- What trailing drawdown really looks likeGold closed below its own running high on 84.5% of sessions, with a median retracement of 9.47% — comfortably past the 5% limit many funded programmes use.
- What moves the gold priceGold against real yields, the dollar and the Fed since 1971. Real yields and the dollar still move it in the same month but predict nothing about the next. A rule fitted on 2006-2021 puts August 2026 at $1,655; gold averaged $4,411, a miss none of 2,000 resampled histories matched. After the six first Fed cuts since 1994, gold was higher a year later four times.
- Is gold a good investment?Every month since August 1971 treated as a purchase. Up after one year 61% of the time and after twenty years 88%, but ahead of inflation only 55% and 63%. The longest wait to break even was 26 years (January 1980). Over the following year gold’s trend tended to continue more often than random price paths explain; over five years no buying condition did.
- Gold price history and every crashEvery year since 1925, before and after inflation, and every fall of 20% or more since August 1971: 16 in all. Falls that stopped short of 30% were back above their peak within 3.4 years; deeper ones took 3.6 to 28 years. Buying the first close 20% down was higher a year later 8 times in 15.
- Gold vs the S&P 500Every month since August 1971, with dividends counted. $10,000 became $3.48 million in the S&P 500 and $1.03 million in gold; without dividends, stocks trailed gold. Gold beat stocks in 27% of ten-year periods and rose in 5 of 7 big stock declines. The S&P 500 to gold ratio looks predictive, but random price paths match it too often to count.
- The gold-silver ratio, testedEvery month since August 1971. Above 80, silver beat gold over the next three years 90% of the time, but the ratio has crossed 80 only five times and kept rising first each time but one. The 80/50 swap rule roughly doubled a gold holding, on three trades. Random price paths match both results often enough that neither is proof.
- Is gold a good hedge against inflation?Every month since August 1971 against the US consumer price index. Gold beat inflation in only about 55% of one- to ten-year holding periods, and lost ground in every high-inflation year since 1981. Adjusted for inflation, its 1980 peak took 45 years to regain.
- Gold vs the US dollar3,905 trading days against a Dollar Index rebuilt from its six-currency formula. Gold moves 1.04% against every 1% of dollar, and the link held every year since 2014 — including 2024, when both rose. The dollar cannot predict gold.
- Does RSI work on gold?Every RSI level mapped across three timeframes. 31 of 75 tests beat the base rate where 3.8 were expected, and 10 survive correction — but 7 of those 10 point to momentum, the opposite of what overbought is taught to mean.
- Fixed or volatility-scaled stops?Both rules risking the same average distance. Identical through eight of ten volatility deciles; in the fastest tenth the fixed stop won 44.5% of 1:1 trades against 50.0%. One configuration contradicts it, and that is in the study.
- Does the golden cross work?Nine 50/200 daily crosses since 2014, each measured against a randomly chosen day. Negative edge at every horizon; 2 of 36 tests beat chance where 1.8 were expected.
- Do Fibonacci retracements work?32,592 mechanically identified retracements. None of the five Fibonacci levels was more popular than the levels beside it; the golden ratio scored exactly 1.00.
- Is gold seasonality real?151 months. January averages +3.61% but fails once twelve months are tested at once. Volatility seasonality is real: March runs 42% wider than September.
- Does the risk-reward ratio create edge?66,572 trades per ratio with edge-free entries. Win rate tracked the break-even line to within 0.021 points, and expectancy settled at almost exactly the spread.
- Does scalping work on gold?Spread is flat at $0.16 in 23 of 24 hours, so spread timing is not the lever. A $2 stop is hit 81% of the time within an hour by ordinary movement.
How these studies are run
Every study replays the same archive of five-minute XAUUSD bars taken from a live MetaTrader 5 feed on a raw-spread account. Nothing is simulated except where a page says so explicitly — the sample-size study is a Monte Carlo model and labels itself one. Method, sample size and date range are stated on each page, and the figures on the page are rendered from the published data file rather than typed in, so a page cannot drift from the numbers behind it.
The studies are run by the same people who build the bot — who we are and why this section exists. That is worth stating plainly rather than hiding: we have an obvious interest in gold trading being tractable, which is exactly why the measurements that went against us are published here too.
Findings that went against us are published alongside the ones that did not. The session breakout study found a coin flip. The stochastic study disproved a rule we had partly relied on, and the strategy page that relied on it was corrected. That is the point of the section: a measurement that only ever flattered the product would not be worth citing.