Research · Dataset

Do gold's classic technical rules actually work?

Almost every gold trading course teaches a rule set: buy when the fast moving average crosses the slow one, buy on a 12-month breakout, sell when RSI is overbought. Very few of them ever put the rule on the same data as buy and hold, over the same period, with the same costs. This page re-implements four of the most widely taught rules from scratch and runs them over 706 monthly gold closes — then tests whether the signals say anything about the months that follow.

706 monthly observations, 1968-01 to 2026-10 · Source data updated November 03, 2026

The answer, first

Buy and hold 8.40% a year, 62.1% max drawdown
Best rule 8.25% 12-month Donchian breakout
Rules that beat holding 0 of 4 on annualised return
"Overbought" months +19.0 pp 12-month forward vs average month

Over 706 months, holding gold returned 8.40% a year and 114.0x the money, with a maximum drawdown of 62.1% between 1980-01 and 1999-07. Every rule tested here returned less. The 12-month Donchian breakout came closest at 8.25%, and it did one thing better than holding: it cut the maximum drawdown to 41.5% and produced the best risk-adjusted result of the four, with a Sharpe ratio of 0.61 against 0.56 for buy and hold.

The second finding is sharper and less expected. On gold, the standard "sell when overbought" instruction is not just unproven — it is backwards. Months with RSI(14) above 70 were followed by an average 12-month return of +25.3%, against +6.3% for the average month in the sample. The rule tells you to sell into the months that carry the returns.

What this is not. These are not trading signals with an entry price and a stop. They are descriptions of which market states were followed by better or worse returns, measured on monthly average prices with no spread, no financing, no slippage and no leverage. A monthly-average rule that would be hard to trade exactly is still worth measuring — but nothing here is a backtest of an executable system, and none of it is a return you should expect.

Source and method

Prices
World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet — Gold ($/troy oz), col 70
Headline coverage
706 months, 1968-01 to 2026-10, monthly averages
Full sample
802 months, 1960-01 to 2026-10
Download
gold-technical-rules-signals.csv (month, price, moving averages, RSI, high/low, signal flags, forward returns) · gold-technical-rules-tests.csv (every rule × era × horizon test result)
Reproduction
scripts/gold_technical_rules_analysis.py; every published number independently recomputed by scripts/verify_gold_technical_rules.py

Pink Sheet pins the gold price at $35/oz through 1967-12; those months are excluded from headline statistics and reported separately as a policy artefact. The 96 pegged months are excluded from the headline statistics and reported separately as a control below.

Rules are evaluated at the close of each month and held for the stated forward horizon. Moving averages use the trailing 10, 50 and 200 monthly averages; the breakout rule takes a long position when the monthly close exceeds the highest close of the previous 12 months and goes flat when it falls below the lowest close of the previous 12 months; RSI(14) uses Wilder's smoothing on monthly average prices.

Four rules, one table

Each rule is a long-or-flat system: hold gold when the condition is true, hold cash when it is not. No shorting, no leverage, no compounding of position size. The comparison below is the fairest version of the question, because every rule is measured against the same monthly price series and the same cash alternative.

System Return / yr Total Vol Sharpe Max DD Time in market Longest underwater
Buy and hold 8.40% 114.0x 16.9% 0.56 -62.1% 100% 315 mo
10-month MA above 50-month MA 7.32% 63.6x 14.9% 0.55 -59.2% 65.6% 333 mo
50-month MA above 200-month MA 6.04% 31.3x 14.2% 0.48 -55.7% 64.9% 333 mo
12-month Donchian breakout 8.25% 105.6x 14.8% 0.61 -41.5% 65.6% 309 mo
RSI(14) between 30 and 70 4.18% 11.1x 11.7% 0.41 -46.3% 69.8% 285 mo

The pattern is consistent and not flattering to the rules. Each one spends a substantial part of its time in cash — the moving-average rules around 66%, the RSI rule 70% — and in this sample the months that a long-only rule steps aside from were mostly months gold went up. Sitting in cash is a real position with a real opportunity cost, and it is the reason all four rules end up behind.

The one thing the breakout rule genuinely buys is drawdown control: 41.5% instead of 62.1%, and the same 309-month longest underwater period against 315 months for holding. If the drawdown you can fund is your binding constraint, that is the trade being made — lower return in exchange for a shallower, shorter decline.

Do the signals say anything about the future?

Return comparisons alone can be misleading: a rule that was flat through the 1970s and 1980s and long afterwards might do fine for reasons unrelated to the signal. The stronger test asks a different question. When the rule says "be in gold", what did gold do over the next 3, 6 and 12 months — compared with what it did after the months when the rule said "be in cash"?

Rule Horizon After signal After no signal Gap 95% bootstrap interval Non-overlapping p
10-month MA above 50-month MA 3 mo +3.3% +1.0% +2.3 pp -0.2 pp to +4.9 pp (crosses 0) 0.105
10-month MA above 50-month MA 6 mo +7.0% +1.8% +5.2 pp +0.3 pp to +10.6 pp 0.125
10-month MA above 50-month MA 12 mo +15.0% +4.1% +10.9 pp +1.1 pp to +21.0 pp 0.104
12-month Donchian breakout 3 mo +3.6% +0.4% +3.2 pp +0.8 pp to +5.7 pp 0.024
12-month Donchian breakout 6 mo +7.4% +1.0% +6.4 pp +1.8 pp to +11.3 pp 0.057
12-month Donchian breakout 12 mo +15.1% +4.0% +11.2 pp +1.9 pp to +21.2 pp 0.207
50-month MA above 200-month MA 6 mo +5.6% +4.4% +1.3 pp -5.1 pp to +6.9 pp (crosses 0) 0.646
50-month MA above 200-month MA 12 mo +11.9% +10.1% +1.8 pp -10.0 pp to +14.7 pp (crosses 0) 0.774
RSI(14) between 30 and 70 6 mo +2.8% +10.6% -7.8 pp -14.9 pp to -1.5 pp 0.129
RSI(14) between 30 and 70 12 mo +7.3% +20.6% -13.3 pp -28.1 pp to -0.6 pp 0.887

The 10-month-above-50-month rule has the cleanest forward signal: after a signal month, gold averaged +15.0% over the next 12 months against +4.1% after a no-signal month, a gap of +10.9 pp with a 95% bootstrap interval of +1.1 pp to +21.0 pp that excludes zero. The 12-month breakout rule shows a similar gap of +11.2 pp at 12 months.

Here is the caveat that has to travel with those numbers. A 12-month forward return computed from every month in the sample heavily overlaps itself — twelve consecutive months share eleven months of price history — so the effective number of independent observations is far smaller than 451. The naive Welch p-value (3.31e-10) is therefore far too small, and it is printed in the dataset for completeness only. The column to read is the last one: on the non-overlapping subsample, the 10/50 gap has p = 0.104 and the breakout gap has p = 0.207 — neither is significant on its own.

The more convincing evidence is the era split. Running the same test on 2000-2026 only — a period with no overlap with the earlier decades in terms of market regime — the 10/50 rule gives a 12-month gap of +12.3 pp with an interval of +3.4 pp to +21.0 pp, and the breakout rule +5.1 pp with an interval of -5.1 pp to +15.4 pp. The moving-average gap holds its direction and its size in the later era. The breakout rule does not: its interval crosses zero after 2000, so the full-sample result is carried by the earlier decades. The 50/200 rule and the RSI-tail rule have no comparable stability — their intervals cross zero in both eras.

Honest summary. The 10/50 moving-average state carries forward information, in the direction of "gold keeps going up in a trend" — which is a description of gold's long bull market, not an edge you can trade cleanly, and not significant once the overlapping observations are removed. The breakout rule's full-sample result does not survive the era split. The 50/200 rule does not survive at all, and the RSI-tail rule's signal points the wrong way.

The overbought problem

This is the most directly actionable part of the study, because it is the rule most retail gold advice is built on. RSI(14) above 70 is supposed to mean "stretched, sell it". Here is what actually happened after those months.

Condition at the signal month Months Avg forward return Average month Gap 95% interval Share positive
RSI(14) above 70 181 +25.3% +6.3% +19.0 pp +5.9 pp to +34.4 pp 76.2%
RSI(14) above 60 307 +21.7% +2.9% +18.8 pp +9.1 pp to +30.2 pp 74.6%
RSI(14) below 30 24 -5.2% +11.8% -17.0 pp -19.5 pp to -14.5 pp 20.8%
RSI(14) below 40 120 +3.6% +12.8% -9.2 pp -14.5 pp to -3.6 pp 53.3%

A month that closed with RSI above 70 was followed by an average 12-month return of +25.3%, and 76.2% of those months were positive. The 95% interval on the gap against the average month runs from +5.9 pp to +34.4 pp and does not cross zero. At the 6-month horizon the same pattern holds: +12.0% against +2.7%.

The mirror image is equally clear. Months with RSI below 30 — the "oversold, buy the dip" signal — were followed by an average 12-month return of -5.2%, with only 20.8% positive months. At 6 months it was -0.9%.

On gold, overbought has meant strength and oversold has meant continued weakness. That is the opposite of the mean-reversion story the indicator is usually sold with, and it is the single most important thing on this page: the classic RSI rule on this asset would have been a persistent loser, not a marginal one.

It also explains the RSI-tail system's return. Holding gold only when RSI sat between 30 and 70 returned 4.18% a year — roughly half of buy and hold. The forward test shows exactly why: after a neutral-RSI month, gold averaged +7.3% over the next 12 months, and after a month outside the band it averaged +20.6%. The band you avoid is where the money is made.

Why the sample starts in 1968

The official gold price was legally fixed at $35 per ounce from 1960-01 to 1967-12 in the World Bank Pink Sheet; the free London market did not open until 1968. Testing a technical rule against an administratively frozen price measures the price controls, not the market. The 96 pegged months are excluded from every headline number on this page.

The full 802-month series is published alongside so the effect of that choice is visible rather than hidden. Including the peg era lowers every return, because it adds eight years of flat $35 gold and the flat months dilute the CAGR.

Rule 1968-2026 (headline) Full sample 1960-2026
Buy and hold 8.40% 8.40%
10-month MA above 50-month MA 7.32% 6.42%
50-month MA above 200-month MA 6.04% 5.30%
12-month Donchian breakout 8.25% 7.23%
RSI(14) between 30 and 70 4.18% 3.67%

The ranking of the rules does not change when the peg era is added back. Only the levels do.

Statistical hygiene: why the obvious p-values are wrong

This page runs 24 rule-by-horizon comparisons, plus twelve RSI threshold tests. Reading a 0.05 threshold across that many tests would produce roughly one false positive by construction, before any of the overlap problems are considered. The Bonferroni-adjusted level for the headline set is 0.002, and the dataset publishes it so nobody has to assume the wrong one.

Three tests are reported for every comparison, and they disagree on purpose:

  • Welch t-test. Reported for completeness. It assumes the observations are independent, which overlapping forward windows are not, so its p-values are systematically too small.
  • Moving-block bootstrap. Resamples contiguous blocks of months, preserving the serial correlation that a month-by-month resample would destroy. The confidence interval is the honest read on the size of the effect.
  • Non-overlapping subsample. Keeps only every h-th starting month, so the observations genuinely do not share price history. This is the most conservative test, and usually the most instructive.

Read the interval and the non-overlapping p-value. The t-test is there so you can see how much it overstates.

Under these criteria, one result survives: the 10/50 moving-average state, in the 1968-2026 and the 2000-2026 eras alike, in the same direction and of a similar size. The 50/200 rule's intervals cross zero at every horizon. The 12-month breakout rule's full-sample interval excludes zero, but its 2000-2026 interval does not, so it does not survive the era split. The RSI-tail rule is a consistent loss rather than an uncertain one.

Where gold stands against these rules right now

Gold (2026-10) $4,105 monthly average per ounce
10-month MA $4,507 above the 50-month
RSI(14) 57.8 inside the 30-70 band
RSI range, full sample 15.4–100 24 months <30, 270 months >70

The monthly average closed 2026-10 at $4,105 per ounce, with the 10-month average at $4,507, the 50-month at $2,904 and the 200-month at $1,811. The 10-month sits above the 50-month — the state this study's forward test associates with the better forward returns — and well above the 200-month.

RSI(14) was 57.8, inside the band this study finds has the weakest forward returns, and far from the overbought readings that have historically preceded the strongest 12-month outcomes. Over the full sample only 24 months printed RSI below 30 and 270 printed above 70 — the extremes are rare on monthly data, which is worth remembering before treating either as a signal.

What this page is not

It measures monthly average prices, so every rule is evaluated as if it could be entered exactly at a monthly average and held to the next one. A real order fills at some intraday price and pays a spread, and a rule that trades often — the breakout rule especially — pays more of that than a rule that does not. No transaction cost, financing, storage or tax is included anywhere, so all four rules are if anything flattered. There is no leverage, no position sizing, no compounding of returns across accounts, and no shorting.

It also does not test whether these rules work on intraday gold, on other instruments, or on today's price series. A rule that works on monthly averages may be unusable on a tick chart, and the opposite is also possible. The only claim being made here is narrow and specific: on 706 monthly gold closes from 1968-01 to 2026-10, holding beat all four rules on return, the moving-average and breakout states carried forward information, the 50/200 rule did not, and the RSI overbought rule pointed the wrong way.

Nothing on this page is investment advice, and past results do not predict future returns. Gold is volatile, a monthly rule can be underwater for years, and a rule that reduces a drawdown also reduces the return that pays for it.

Frequently asked questions

Do moving-average and breakout rules actually beat holding gold?

Not on this sample, once you measure the whole period rather than a favourable window. Buy and hold returned 8.4% a year with a 62.1% maximum drawdown over the 706 months from 1968-01. The 10-month-above-50-month rule returned 7.32% a year, the 50-month-above-200-month rule 6.04% and the 12-month Donchian breakout 8.25%. All three lost to simply holding. The breakout rule did cut the maximum drawdown to 41.5% and produced the best risk-adjusted result of the three, with a Sharpe of 0.61 against 0.56 for buy and hold.

Do these rules tell you something about the next 12 months?

Yes, but not as a trading signal — as a description of which market states pay. When the 10-month moving average was above the 50-month, the following 12 months averaged 15% against 4.1% when it was not, a gap of +10.92 pp. The 95% bootstrap interval on that gap is +1.1 pp to +21 pp, which excludes zero, but the gap is not significant on the non-overlapping subsample (0.104). The same rule in 2000-2026 shows the same direction with a similar size, which is the more convincing evidence. The 50-month-above-200-month rule produced no reliable gap at any horizon, and the 12-month breakout rule's full-sample gap does not survive the era split.

Is "sell when RSI is overbought" supported by the data?

No — on gold it is contradicted. Months with RSI(14) above 70 were followed by an average 12-month return of 25.3%, against 6.3% for the average month, and 76.2% of them were positive. The 95% interval on the gap is +5.9 pp to +34.4 pp. Being "overbought" on gold has historically been a sign of strength, not of a pending reversal. The mirror image holds too: months with RSI below 30 were followed by an average 12-month return of -5.2%, and only 20.8% of them were positive.

What does "avoiding the RSI tails" cost you?

It costs a measurable amount. Holding gold only in months when RSI(14) was between 30 and 70 returned 4.18% a year, well below buy and hold's 8.4%. The forward-return test explains why: in the 12 months after a neutral-RSI month, gold averaged 7.3%, and in the 12 months after a month outside the 30-70 band it averaged 20.6% — a gap of -13.34 pp in favour of staying invested through the extremes. Avoiding the tails means avoiding the months that carry the returns.

How much of these results is statistical artefact?

The page runs 24 rule-by-horizon comparisons, so a plain 0.05 threshold would be far too generous — the Bonferroni-adjusted level here is 0.0021. That is why every result carries three tests: a Welch t-test (which ignores the overlap between adjacent forward windows and is therefore too small), a moving-block bootstrap on the confidence interval, and a non-overlapping subsample test that treats the observations as independent. Read the interval and the non-overlapping p-value, not the t-test. Under those criteria the moving-average and breakout signals hold up directionally; the 50/200 rule and the RSI-tail rule do not.

Why does the sample start in 1968?

Because the official gold price was legally fixed at $35 per ounce from 1960-01 to 1967-12 in the World Bank Pink Sheet — the free London market did not open until 1968. A rule tested against a frozen price measures policy, not markets. The 96 pegged months are therefore excluded from the headline statistics and reported separately as a control, and the full 802-month series is shown alongside so the effect of the exclusion is visible rather than hidden.

Where is gold relative to its own indicators right now?

At the end of 2026-10 the monthly average was $4,105 per ounce, with the 10-month average at $4,507.3, the 50-month at $2,904.06 and the 200-month at $1,811.03 — the 10-month is above the 50-month, which is the state this page's forward-return test associates with better forward returns. RSI(14) was 57.8, inside the 30-70 band that this study finds has the weakest forward returns. Over the full sample RSI(14) ranged from 15.4 to 100, with 24 months below 30 and 270 months above 70.

Can I reproduce these numbers?

Yes. The monthly price series, every indicator value, every signal flag and every forward return are published as CSV files alongside the dataset, and the production script and an independent numpy verification script are both referenced on the page. The verification script recomputes the headline statistics from the published CSVs through a different code path and checks them against the published values.

Sources

  • World Bank Commodity Price Data (The Pink Sheet) — Monthly Prices sheet, Gold ($/troy oz), col 70, monthly averages in USD per troy ounce. Data updated November 03, 2026.
  • scripts/gold_technical_rules_analysis.py — production script; writes src/data/research-gold-technical-rules.json and both CSVs.
  • scripts/verify_gold_technical_rules.py — independent second implementation; recomputes every published statistic from the published CSVs with numpy through a different code path.
  • Wilder, J. (1978), New Concepts in Technical Trading Systems — the RSI smoothing method used here, and the original Donchian channel definition.

Last updated: 2026-11-11

Educational content only. Not investment advice. Trading forex and CFDs carries substantial risk of loss.