Research · Dataset
Gold implied volatility against realised volatility, 2008–2026
Every desk has an opinion about whether gold is "expensive" right now. The trouble is that gold's volatility is quoted in two different numbers at once — what the options market is charging, and what gold has actually been doing — and they rarely agree. This page measures the gap between them on 4521 daily observations, and tests the three claims that usually get made about it.
Data through 2026-11-10 · implied volatility through 2026-11-11 · updated 2026-11-13
Source and method
Every statistic uses only dates on which BOTH the LBMA gold fix and the GVZ print were published, so no implied-vol observation is carried forward across a market holiday or a data gap.
A moving-block bootstrap (21-day blocks, 2,000 resamples) that keeps the implied/realised pairing intact, plus a non-overlapping subsample t-test, because overlapping volatility windows otherwise inflate the effective sample size.
No transaction cost, spread, financing, custody, margin or tax is modelled anywhere on this page. The sizing figures are illustrative arithmetic on a single historical path.
Where the market is right now
Gold implied volatility was 24.43 on 2026-11-10, against 22.54 realised over the trailing 21 days — a premium of +1.89 points. That implied reading sits at the 85th percentile of its own history since 2008-06-03: elevated against the last few years, well short of the extremes.
Percentile is the share of daily readings on record at or below the current one. It is a rank, not a forecast.
Is the premium real, or does it average away?
The implied-minus-realised premium is the number this page is really about. If it were noise, the whole idea of reading implied volatility would be worthless — and if it is persistent, it is the reason option sellers earn a spread and the reason anyone who sizes gold positions by trailing volatility alone is sizing on a number that is always stale.
| Window | Implied mean | Realised mean | Premium mean | Premium positive | 95% block bootstrap |
|---|---|---|---|---|---|
| 10 days | 19.02 | 16.16 | +2.86 | 78.1% | +2.46 to +3.27 |
| 21 days | 19.01 | 16.51 | +2.50 | 79.8% | +2.09 to +2.89 |
| 63 days | 18.92 | 16.81 | +2.11 | 77.3% | +1.67 to +2.55 |
The premium is persistent, and it survives the harder tests.
On the 21-day window the premium averaged +2.50 points, with a median of +2.64 and a positive reading on 79.8% of observations. The 95% block-bootstrap interval runs from +2.09 to +2.89 and excludes zero.
Overlapping windows would quietly inflate the sample size, so the headline is also measured on a non-overlapping subsample of 215 observations: mean +2.61, p = 1.46e-20.
Every calendar year on the sample carried a positive average premium: 19 of 19. The largest was +5.69 points in 2009; the smallest was +0.36 in 2013.
A premium this consistent is not a free lunch for the seller. It is compensation for the tail risk that comes with the other side of the trade, and the sample of years where it was smallest is instructive: 2013 is the year implied volatility was least useful as a risk signal.
Does implied volatility lead realised volatility?
For a premium to be worth acting on, implied volatility has to carry information about volatility that has not happened yet. If it merely trailed realised volatility, it would be a lagging echo of the same number.
Correlation with the next window is 0.7525 — lower than the 0.8573 same-window figure, but far from zero, and higher than the 0.8977 that the same test produces against the 63-day window. Implied volatility does carry forward-looking information about how volatile gold is about to be.
The VIX is used only as a control: if gold implied volatility merely tracked equity implied volatility, it would carry no gold-specific information. The VIX and gold implied volatility correlate 0.7473 over 4522 common observations, and gold implied volatility has a lag-1 autocorrelation of 0.9836 — it moves slowly, so it tracks the general mood of the market as much as gold itself.
Does a volatility spike mark a top?
The claim worth testing is the one people repeat after every selloff: implied volatility has spiked, therefore the top is in, therefore sell.
It does not. Across the 40 highest implied-volatility observations on record — mean implied vol of 53.3 at those events — the forward 21-day return averaged -1.48% with a median of +1.49%, and was negative in 50% of cases. That is a coin flip, not a top signal. At 63 days the sign flips and the mean forward return is +11.75%, negative in only 22.5% of cases.
Read the sample size before the average.
The 40 highest-implied-volatility observations on record are not spread across two decades. They are concentrated in one period — the peaks listed below run from 2008-10-14 to 2008-10-10. A statistic computed on one episode, with a 21-day forward return ranging from -20.35% to +17.84%, cannot separate a tradable signal from a memory of one crisis.
This page reports the number because it is genuinely interesting. It does not offer it as a rule.
Highest implied-volatility readings on record: 2008-10-10, 2008-10-13, 2008-10-28, 2008-10-29, 2008-10-27, 2008-10-30.
Does implied volatility forecast the gold price?
Split into thirds by implied volatility, here is what gold did over the following 21 days.
| Implied vol tercile | Mean implied | Mean forward return | Median | Months that fell |
|---|---|---|---|---|
| Lowest (8.9–15.8) | 13.17 | +0.69% | +0.23% | 46.8% |
| Middle (15.8–19.7) | 17.57 | +1.41% | +1.31% | 38.8% |
| Highest (19.7–64.5) | 26.32 | +0.43% | +0.51% | 46.3% |
The ordering is not monotonic: the middle tercile produced the best forward return (+1.41%), ahead of the cheapest third (+0.69%) and the most expensive (+0.43%). The spread across all three is small, and the share of falling months is close to 47% in every tercile. Implied volatility is a price that people pay, not a forecast of direction.
Tercile boundaries at implied vol 15.78 and 19.71. Forward returns are the cumulative simple return over the 21 aligned trading days after the observation.
How big is the implied move, really?
Implied volatility converts into a statement about how far gold should move. Here is that statement, checked against what happened.
| Horizon | Implied 1-sigma move | Realised average move | Realised 1-sigma | Move exceeded the band |
|---|---|---|---|---|
| 21 days | 5.49% | 3.84% | 4.76% | 26.3% |
| 30 days | 6.56% | 4.61% | 5.69% | 27.1% |
| 63 days | 9.48% | 6.78% | 8.22% | 28.1% |
Over 21 days the implied one-standard-deviation move averaged 5.49%, while the realised average absolute move was 3.84%. The realised move exceeded the implied one-sigma band on only 26.3% of occasions, against 31.7% for a normal distribution — the implied band is wider than gold usually moves.
Implied one-sigma move is implied volatility scaled by the square root of the horizon over 252 trading days. The normal-distribution benchmark is the share of observations outside a one-sigma band under a Gaussian assumption, and is shown as a reference point, not a fitted value.
What it does to position sizing
The practical use of implied volatility is not predicting direction. It is telling you how much size to carry now, because it knows about volatility that trailing estimates have not caught up with yet. Here is that tested, illustratively.
| Sizing rule | Annualised return | Max drawdown | Volatility | Sharpe |
|---|---|---|---|---|
| Trailing realised volatility | 8.18% | -43.46% | 14.13% | 0.345 |
| Implied volatility | 8.10% | -36.55% | 12.65% | 0.363 |
Both rules target 15% annualised volatility, capped at no leverage, on gold alone. The implied-volatility version produced the shallower drawdown (-36.55% against -43.46%) at a similar annualised return (8.10% against 8.18%) — because implied volatility rises before realised volatility does, so it de-risks earlier.
Illustrative, cost-free, unlevered volatility-targeting on gold alone. Weights are capped at 1.0, so this is de-risking only. No transaction cost, spread, financing, margin, custody or tax is modelled. A single historical path; not a forecast.
If you size by volatility in practice, the position size calculator on this site does the arithmetic directly, and the volatility-targeting position sizing covers the same method on a much longer monthly series, and gold tail risk and stop-loss evidence measures what the worst 1% of days actually cost.
Year by year
| Year | Implied mean | Realised mean | Premium | Premium positive |
|---|---|---|---|---|
| 2008 | 40.97 | 36.64 | +4.34 | 80.5% |
| 2009 | 27.21 | 21.52 | +5.69 | 91.9% |
| 2010 | 20.93 | 16.26 | +4.67 | 90.7% |
| 2011 | 22.25 | 19.17 | +3.08 | 83.2% |
| 2012 | 18.17 | 17.42 | +0.75 | 63.6% |
| 2013 | 20.72 | 20.36 | +0.36 | 58.1% |
| 2014 | 16.56 | 12.99 | +3.56 | 93.1% |
| 2015 | 17.06 | 14.26 | +2.80 | 87.8% |
| 2016 | 17.81 | 16.09 | +1.72 | 68.3% |
| 2017 | 12.44 | 11.23 | +1.21 | 70.2% |
| 2018 | 11.81 | 9.00 | +2.81 | 99.6% |
| 2019 | 12.39 | 11.09 | +1.30 | 79.7% |
| 2020 | 21.51 | 18.72 | +2.80 | 72.5% |
| 2021 | 16.74 | 14.71 | +2.03 | 81.7% |
| 2022 | 18.35 | 14.32 | +4.03 | 90.5% |
| 2023 | 14.86 | 12.66 | +2.21 | 88.5% |
| 2024 | 16.05 | 14.10 | +1.95 | 73.2% |
| 2025 | 19.41 | 17.23 | +2.18 | 75.8% |
| 2026 | 27.97 | 27.29 | +0.68 | 66.7% |
2026 covers 213 aligned days to date, so its figures are partial-year. 2008 begins in June, when the GVZ series starts.
What this dataset does and does not support
Supported by the data: a persistent implied-minus-realised premium on gold. It averaged +2.50 points on the 21-day window, was positive on 79.8% of observations, was positive in every calendar year on the sample, and survives both a block bootstrap and a non-overlapping subsample test. Implied volatility also carries forward-looking information: it correlates 0.7525 with the realised volatility of the following 21 days.
Not supported: any claim that a volatility spike marks a top. Across the 40 highest readings on record, the forward 21-day return was negative in only 50% of cases — and that sample comes almost entirely from one episode in one quarter. Nor is implied volatility a directional forecast: the three terciles produced forward returns of +0.69%, +1.41% and +0.43%, which is not a pattern.
Nothing on this page forecasts the gold price or the future level of implied volatility. The sizing figures are illustrative arithmetic on a single historical path with no transaction cost, spread, financing, margin, custody or tax modelled.
Frequently asked questions
What does implied volatility on gold actually measure?
It is the market price of insurance on gold, quoted as an annualised volatility figure. This page uses the CBOE Gold ETF Volatility Index (GVZCLS), which tracks the expected volatility of gold ETF options over the next 30 days. Every statistic here uses only the 4522 dates on which both the LBMA gold fix and the GVZ print were published — 2008-06-03 to 2026-11-10 — so no implied-vol reading is carried forward across a gap. Realised volatility is computed from the same aligned gold fixes, not from the longer gold history.
Is gold implied volatility unusually high right now?
As of 2026-11-10, gold implied volatility was 24.43, against trailing realised volatility of 22.54 over 21 days and 21.19 over 63 days. That puts the implied-minus-realised premium at +1.89 pp against the 21-day window and +3.24 pp against the 63-day window. The implied reading itself sits at the 84.6th percentile of its own history since 2008-06-03, so it is elevated against recent years but far from the extremes.
Is there a persistent premium between implied and realised volatility?
Yes, and it survives the checks that usually kill a result like this. Over the full sample the mean premium on a 21-day window is +2.5 pp, the median is +2.64 pp, and it was positive on 79.8% of observations. A moving-block bootstrap that keeps the implied and realised legs paired gives a 95% interval of +2.09 pp to +2.89 pp over 2000 resamples. Because overlapping windows would inflate the effective sample size, the premium is also measured on a non-overlapping subsample: 215 observations, mean +2.61 pp, p = 1.46e-20. The premium was positive in every one of the 19 calendar years on the sample, and the largest of them was +5.69 pp.
Does a spike in gold implied volatility mark a top or a bottom?
Not reliably. Taking the 40 highest implied-volatility observations on record — mean implied vol of 53.3 at those events — the forward 21-day return averaged -1.48 pp with a median of +1.49 pp, and it was negative in 50% of them. At 63 days the picture inverts: the mean forward return was +11.75 pp and negative in only 22.5% of cases. A sample of 40 events, drawn almost entirely from one quarter of one year, cannot separate a signal from an episode. The page reports it as the sample it is.
Can implied volatility tell me where the gold price goes next?
No, and the data says so plainly. Split into thirds by implied volatility, the 21-day forward return averaged +0.69 pp in the cheapest third, +1.41 pp in the middle third and +0.43 pp in the most expensive third — with negative months at 46.8%, 38.8% and 46.3% respectively. The ordering is not monotonic and the spread is small. Implied volatility is a price, not a forecast of direction.
Is gold implied volatility just the VIX in disguise?
No, though the two are related. Gold implied volatility correlates 0.7473 with the VIX across 4522 common observations, and it has a lag-1 autocorrelation of 0.9836 — it is a slow-moving series that reflects the market's mood rather than gold-specific stress. It is not a copy: gold implied volatility averages 19.04 against the VIX's 19.79 over the same period, and the relationship between implied and realised gold volatility holds after the equity link is acknowledged.
Is the implied move bigger than the realised move?
Yes, consistently, and that is what a premium means. Over 21 days the implied one-standard-deviation move averages 5.49% while the realised average absolute move was 3.84%. The realised move exceeded the implied one-standard-deviation band on only 26.3% of occasions, against 31.7% for a normal distribution. At 63 days the figures are 9.48% implied against 6.78% realised.
Does this change how I should size a gold position?
It offers a defensible alternative to sizing on trailing volatility alone. The page runs an illustrative volatility-targeting comparison on gold alone at a 15% annualised volatility target, capped at no leverage. Sizing on trailing realised volatility produced a 8.18% annualised return with a -43.46% maximum drawdown; sizing on implied volatility produced 8.1% with a -36.55% drawdown. The implied-volatility version produced the shallower drawdown because it de-risks before realised volatility catches up. These are cost-free, unlevered, single-path figures with no financing, spread or margin modelled, and the position size calculator on this site does the arithmetic directly.
Sources
-
Gold: LBMA Gold Price PM benchmark fix, USD per troy ounce —
scripts/daily_gold_lbma_pm.csv(1968-04-01 to 2026-11-10). -
Implied volatility: CBOE Gold ETF Volatility Index (GVZCLS), daily
close, carried by FRED —
scripts/fred_GVZCLS.csv(2008-06-03 to 2026-11-11). Source: Chicago Board Options Exchange; copyright 2016 Chicago Board Options Exchange, Inc., reprinted with permission. -
Control: CBOE S&P 500 Volatility Index (VIXCLS), daily close, carried
by FRED —
scripts/fred_VIXCLS.csv. -
Analysis script:
scripts/gold_implied_vol_analysis.py. Independent recomputation with numpy:scripts/verify_gold_implied_vol.py(222 checks over the published figures). - For the sizing method on a longer series, see volatility-targeting position sizing; for the macro drivers of gold volatility, see gold vs. real yields and the dollar.
Realised volatility is the trailing sample standard deviation of daily returns, annualised by the square root of 252. Nothing on this page is investment advice, a recommendation or a forecast.