Research · Dataset
The gold/silver ratio, 1960–2026
The ratio is quoted as a risk gauge almost every day: high ratio means risk-off, low ratio means risk-on. This page computes it from 802 monthly averages and then tests what actually happened next — because the conventional reading is not what the numbers say.
Last updated: 2026-10 (data through 2026-10) · Source data updated November 03, 2026
Source and method
Both legs are monthly average prices, so this is a ratio of averages, not of closing prices. It is not a tradable instrument: no storage, financing or roll cost is included in either leg. A monthly average also smooths intramonth moves, so a ratio built from daily closes would be noisier and would not reproduce these figures.
Where the ratio is now
The ratio sits at the 73rd percentile of its own history — above the 72.7 75th percentile line, well below the 83.2 90th. On the conventional reading that is not a low-risk reading: a high ratio means gold has been bid relative to silver.
Over the last 60 months the ratio has moved from 75.3 to 72.0 (-4.3%). Gold is +1.2% over twelve months and silver +15.4%, with gold at $4,105/oz and silver at $57.00/oz.
Decade by decade
| Period | GSR mean | GSR range | Gold | Silver | Gold vol | Silver vol | Months silver beat gold |
|---|---|---|---|---|---|---|---|
| 1960-1969 | 28.2 | 16.4–38.9 | +0.0% | +100.0% | 5.3% | 14.2% | 17 / 119 |
| 1970-1979 | 31.6 | 18.4–46.2 | +1,200.0% | +1,084.2% | 21.2% | 31.2% | 50 / 120 |
| 1980-1989 | 52.7 | 17.4–75.7 | -39.4% | -85.6% | 24.2% | 40.7% | 51 / 120 |
| 1990-1999 | 73.6 | 43.7–98.4 | -31.0% | +0.0% | 10.5% | 18.1% | 56 / 120 |
| 2000-2009 | 61.5 | 47.4–79.2 | +299.6% | +238.5% | 13.9% | 23.4% | 60 / 120 |
| 2010-2019 | 67.5 | 34.7–90.6 | +32.3% | -3.4% | 11.6% | 21.9% | 49 / 120 |
| 2020-2029 (to 2026-10) | 80.8 | 51.6–111.5 | +163.0% | +216.7% | 13.4% | 32.7% | 40 / 82 |
Price totals, not annualised, and gross of any cost. Volatility is the annualised standard deviation of monthly changes within the decade.
The ratio's own history is the story here: it averaged 28.2 in the 1960s and 73.6 in the 1990s. Any threshold used to define "high" or "low" changes meaning every decade. A ratio of 60 meant something very different in 1960-1969 than it does today.
The test: what happened after an extreme ratio
This is the part the conventional reading gets wrong. We took the 20 months with the highest ratio and the 20 with the lowest, and looked at what each metal did over the following 12 months.
The ratio's own direction of travel was a decent signal for silver: after the highest readings silver averaged +40.30% over the next year, and after the lowest readings it averaged -18.36%. That is the "mean reversion" story the ratio is usually sold on.
But gold, which is the metal most people trade the ratio for, moved in the same direction in both cases: +9.73% after the highest readings and +7.73% after the lowest. Gold was positive in 70% of the high-ratio months and 85% of the low-ratio months. The ratio level said almost nothing about gold's next year.
Caveat on sample size: 20 months per tail, drawn from a 66-year series. The tails are not evenly spread: the highest-ratio months cluster in 1990–1993, 2020 and 2025, while the lowest cluster in 1967–1970 and 1980. So these averages describe a handful of very different environments, and the low tail is dominated by the era when silver was still priced against a pegged gold rather than as a precious metal in its own right.
Conditioning the whole sample by ratio level
A cleaner version of the same question: split all 790 months with a full 12-month forward window at the median ratio (56.4), and compare the two halves.
| Group | Months | Forward gold | Forward silver | Forward ratio |
|---|---|---|---|---|
| Ratio above median | 395 | +8.42% | +13.75% | -1.42% |
| Ratio below median | 395 | +11.64% | +10.57% | +6.65% |
Is the difference real? No.
The gold difference is -3.22 percentage points (high-ratio months +8.42% vs low-ratio months +11.64%). Three checks, in increasing order of honesty:
- Naive t-test: t = -1.79, p = 0.0729. Looks borderline.
- Non-overlapping subsample (every 12th month, so no two 12-month windows share a month): 66 months, difference -2.64pp, t = -0.35, p = 0.729. The effect is gone.
- Moving-block bootstrap (12-month blocks, 2000 resamples, fixed seed 20261108): 95% CI -14.57 to 6.03 pp. It straddles zero.
The naive test is inflated because 12-month forward windows overlap — a bad month is counted in eleven different windows. That is why the non-overlapping subsample and the bootstrap are the numbers to quote.
Does the ratio move first?
Cross-correlating gold's monthly return against the change in the ratio at lags −6 to +6 months, a negative lag meaning the ratio moved first:
| Lag (months) | -6 | -5 | -4 | -3 | -2 | -1 | 0 | +1 | +2 | +3 | +4 | +5 | +6 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Ratio change vs gold | -0.0367 | 0.0028 | -0.0237 | 0.0215 | 0.0688 | -0.0339 | -0.1325 | -0.0478 | 0.0339 | 0.0608 | -0.0125 | -0.0182 | 0.0511 |
| Gold change vs itself (control) | -0.0300 | 0.0150 | 0.0747 | 0.0290 | -0.0359 | 0.2691 | 1.0000 | 0.2691 | -0.0359 | 0.0290 | 0.0747 | 0.0150 | -0.0300 |
The strongest reading is at lag 0 (-0.1325), so the ratio and gold move together, not one after the other. Every lagged value is smaller in absolute terms.
The control row matters more than the first row. Correlating gold's return against itself shifted by one month produces 0.269 at the strongest non-zero lag (lag -1) — so any apparent lead-lag structure with a correlation near 0.27 is indistinguishable from noise. The ratio's best lag of -0.1325 is well inside that noise band.
Correlation traps
Two numbers that look meaningful and are not:
- Ratio change vs gold return: -0.1325. The trap here is arithmetic, not statistics. Because the ratio is gold divided by silver, its change is (to first order) the gold return minus the silver return. Correlating that identity with the gold return gives -0.1650, which sits within 0.03 of the number above. So it is a restatement of how often silver moves more than gold — not a discovered relationship.
- Ratio level vs gold level: 0.5174. This one is commonly dismissed as "near zero", which is wrong: gold and silver correlate 0.9472 in levels as two trending series, and that alone produces a substantial ratio-level correlation. It is a trend artefact, not a relationship. The same failure mode is covered in the real-yield and dollar study.
- For reference, gold and silver monthly returns correlate 0.7008 over 801 months. They are related, not independent — which is why sizing both as if they were separate risks is wrong.
Silver's volatility premium
The trailing 60-month ratio of annualised volatility (silver ÷ gold) currently sits at 2.19, with gold at 14.0% and silver at 30.6% annualised.
The widest reading is an artefact of the 1960s: gold was pegged at $35/oz, so its measured volatility was tiny while silver's was not. The tightest reading (2003-05) is the modern case where silver was momentarily calmer than gold. Median across the whole sample: 1.74×.
Practically: silver carries roughly 2.2× gold's volatility. If you size a silver position the same notional size as a gold position, you are running about 2.2 times the risk. The position size calculator exists for exactly this.
Which metal won each year
Over the 66 calendar years in the sample, silver beat gold in 29 of them (43.9%). By decade, silver won the 1960s (gold was pegged at $35 while silver doubled) and the 1990s; gold won the 1970s, the 1980s, the 2000s and the 2010s. The 1960s "win" is an artefact of the peg: gold could not go anywhere.
| Year | Gold | Silver | GSR start | GSR end |
|---|---|---|---|---|
| 2025 | +59.0% | +104.9% | 89.1 | 69.2 |
| 2024 | +30.2% | +34.5% | 88.8 | 86.0 |
| 2023 | +6.7% | +0.8% | 80.1 | 84.8 |
| 2022 | -1.0% | +0.4% | 78.3 | 77.2 |
| 2021 | -4.1% | -13.1% | 72.1 | 79.6 |
| 2020 | +19.0% | +38.9% | 86.7 | 74.3 |
| 2019 | +14.5% | +9.6% | 82.8 | 86.5 |
| 2018 | -6.1% | -13.5% | 77.8 | 84.5 |
| 2017 | +6.0% | -4.1% | 70.5 | 78.0 |
| 2016 | +5.4% | +16.3% | 77.9 | 70.5 |
| 2015 | -14.0% | -18.0% | 72.7 | 76.3 |
| 2014 | -3.5% | -18.1% | 62.5 | 73.7 |
| 2013 | -26.9% | -36.7% | 53.8 | 62.0 |
| 2012 | +1.9% | +3.9% | 53.9 | 52.8 |
| 2011 | +20.6% | +6.3% | 47.7 | 54.1 |
| 2010 | +24.4% | +65.5% | 63.2 | 47.5 |
| 2009 | +32.1% | +55.8% | 76.0 | 64.5 |
| 2008 | -8.3% | -35.2% | 56.0 | 79.2 |
| 2007 | +27.3% | +11.7% | 49.3 | 56.2 |
| 2006 | +14.5% | +46.2% | 60.4 | 47.4 |
| 2005 | +20.3% | +30.3% | 64.2 | 59.3 |
| 2004 | +6.8% | +12.7% | 65.7 | 62.3 |
| 2003 | +14.0% | +16.7% | 74.4 | 72.7 |
| 2002 | +17.7% | +2.2% | 62.7 | 72.2 |
Last 24 calendar years shown; the full series is in the downloadable CSV. Year figures are January-to-December changes in monthly average prices.
What this dataset does and does not support
Supported by the data: the ratio's long-run level has risen structurally since the 1960s, so any fixed threshold is era-dependent; silver's volatility premium over gold currently runs about 2.2×; and the extreme tails of the ratio do show clear mean reversion in the ratio itself.
Not supported: the ratio as a timing signal for gold. On the full sample the forward-return difference between high- and low-ratio months is -3.22pp, which disappears on a non-overlapping subsample (p = 0.729) and has a bootstrap interval of -14.57 to 6.03pp. The lead-lag test finds nothing outside the noise floor set by the control series.
A ratio can be a useful descriptive label for what precious metals are doing. That is not the same as it being a tradeable signal, and this dataset is offered as the evidence for the difference.
Frequently asked questions
What is the gold/silver ratio and where is it now?
The gold/silver ratio is the number of ounces of silver that buy one ounce of gold. As of 2026-10 it stands at 72, against a sample mean of 55.4 and a median of 56.7, which places it at the 73.4th percentile of the 802-month sample from 1960-01 to 2026-10. The sample range runs from 16.4 in 1968-06 to 111.5 in 2020-04.
Does the gold/silver ratio predict gold or silver returns?
Not in any way that survives scrutiny. Splitting the sample at the median ratio of 56.4, the high-ratio half's 12-month forward gold return averaged 8.4% against 11.6% in the low half, a difference of -3.22 pp. The moving-block bootstrap interval runs from -14.6 to 6 percentage points and includes zero. On the non-overlapping subsample of 66 windows the difference shrinks to -2.64 pp with p = 0.729.
Is the ratio just a restatement of gold and silver moving together?
Largely yes, and this is worth knowing before using it. The correlation of the ratio's monthly change with the gold return is -0.1325, while the correlation with the silver return is -0.783: the ratio responds mostly to silver, which is the smaller and more volatile leg. The arithmetic identity gold return minus silver return correlates with the gold return at -0.165, within 0.05 of the direct figure. The level correlation between the ratio and gold is 0.5174, positive rather than negative, because two trending series correlate in levels; reading that as a signal is reading a trend artefact.
What happened after the most extreme ratio readings?
The extreme readings produced large moves in silver, not in gold. Across the 20 months with the highest ratios (90.3 to 111.5), gold averaged 9.7% forward while silver averaged 40.3% and the ratio itself fell 17.8%. At the low end (16.4 to 19.4), silver averaged -18.4% and the ratio rose 37.8%. Gold rose in 85% of the low-ratio months and 70% of the high-ratio months, which is to say it went up most of the time in both.
Has silver outperformed gold more or less often?
Less often than gold, over the 66 complete years compared. Silver beat gold in 29 of those years, or 43.9%. The ratio has averaged down over the long run but not with a reliable cadence. Over the last 60 months the ratio moved from 75.3 to 72, a change of -4.3%, while gold returned 1.2% and silver 15.4%.
How volatile is silver relative to gold?
Roughly three times as volatile, and this is the constant behind every ratio move. In the ratio's most volatile month silver's annualised volatility reached 9 times gold's. That asymmetry is why the ratio is usually framed as a risk-appetite gauge: silver carries more industrial demand and more downside, so it needs a larger move in either direction to keep the ratio roughly unchanged.
Does the ratio lead gold or silver at any lag?
No usable lead or lag exists. Scanning from -6 to +6 months, the strongest contemporaneous correlation is -0.1325 at lag 0 and every other lag in the scan stays below 0.07 in absolute value. There is no point on the calendar where the ratio has consistently forecast either metal.
Sources
- World Bank Commodity Price Data (The Pink Sheet),
CMO-Historical-Data-Monthly.xlsx— Monthly Prices sheet, columns 70 (Gold) and 72 (Silver), updated November 03, 2026. https://www.worldbank.org/en/research/commodity-markets -
Analysis script:
scripts/gold_silver_ratio_analysis.py(in this project's source tree). Independent recomputation:scripts/verify_gold_silver.py. - For the correlation caveats discussed here, see Gold vs. real yields and the dollar and Gold price history.
- The other cross-metal comparison: Gold versus Bitcoin, 2014-2026 — the same method applied to the pair that is neither precious metal nor diversifier.
- The monetary-versus-industrial comparison: Gold versus platinum, 1960-2026 — the same method applied to the pair that separates the two stories people tell about gold, with an industrial-cycle test and the same regime-dependence caveat.
Data accessed 2026-11-08. Prices are nominal US dollars and include no carry, financing or storage cost. Nothing on this page is investment advice or a forecast.