Research · Dataset

The gold/oil ratio, 1960–2026

The gold/oil ratio is quoted as a recession gauge: when the number climbs, oil has fallen and gold has not, and the story goes that this combination means the economy is about to weaken. It is a genuinely popular claim, and as far as we could find it has never actually been measured against monthly data with a proper recession definition. So this page measures it. The answer does not go the way the story does.

802 monthly observations, 1960-01 to 2026-10 · Source data updated November 03, 2026

What we tested

Three claims circulate about gold and oil. All three are checkable against 67 years of monthly data, so we checked them.

  1. Gold and oil move in opposite directions, because both are treated as inflation hedges and only one can be right at a time.
  2. A high gold/oil ratio signals an incoming recession, because cheap oil means weak demand.
  3. Gold hedges against an oil price shock, so an oil collapse should send gold higher.

We used 802 monthly averages of gold and crude oil from the World Bank Pink Sheet, and the NBER recession dating from FRED to define recessions. Every number below is produced by scripts/gold_oil_ratio_analysis.py and independently recomputed from the published CSV by scripts/verify_gold_oil_ratio.py, which uses numpy and a deliberately different code path.

The answer, first

Gold/oil return correlation 0.182 positive, not negative
Recession after top-quintile ratio 18.0% base rate 14.4%
Since 2000, top quintile 5.4% vs 37.5% in bottom quintile
Gold in oil-crash months -0.49% 22 of 49 positive
The short version. Gold and oil returns are positively correlated, not inversely. A high gold/oil ratio does not precede recessions — in the modern era it is associated with fewer of them. And gold does not catch a bid when oil collapses. What the ratio level does carry information about is the shape of forward gold returns, and that shape is a U, not a warning signal.

Claim 1: gold and oil are inversely correlated

This is the easiest one to settle, and it comes out backwards. Monthly gold and oil returns are correlated at 0.1819, which means the two move in the same direction more often than not. The squared correlation is 0.0331, so oil explains about 3.3%% of the variance in gold's monthly return. That is a real relationship, and it is the wrong sign for the claim.

The level correlation is much higher at 0.7498, but that is a trap. Two series that both drift upward over 67 years will look strongly correlated in levels no matter what their monthly moves are doing. Any honest test of "do they move together" has to be run on returns.

Correlation of monthly returns 0.1819 801 monthly observations
Correlation of price levels 0.7498 inflated by shared trend

Because a single correlation estimate can hide regime shifts, we also computed the correlation in every rolling 36-month window. Across 706 windows the correlation was negative in only 24.1% of them. The median window correlation was 0.117, with a minimum of -0.318 in 1978-11 and a maximum of 0.579 in 1993-02.

There is one further check worth making before dismissing the claim entirely: does oil lead gold? We ran a cross-correlation from lag −6 to +6 months. The strongest absolute correlation is at lag 0, i.e. contemporaneous. For comparison, gold's own month-over-month autocorrelation is 0.269, so gold's own momentum explains more of its next-month move than any oil signal does.

Claim 1 is not supported. Gold and oil move together, not in opposition, and there is no lead-lag structure that rescues the inverse-correlation story.

Claim 2: a high gold/oil ratio signals a recession

This is the claim that matters most, and it needs a careful definition before it can be tested at all.

How the recession test is defined

A recession is dated to its peak by the NBER, so "is there a recession in the next twelve months" is ambiguous: a recession that started five years ago is not a fresh signal. We therefore test whether an NBER recession onset falls within the next 12 months, and we drop every anchor month that is already inside a recession. Each anchor is then a genuinely forward-looking observation.

That leaves 695 usable anchors and a base rate of 14.4% — the share of all non-recession months that were followed by an onset within 12 months.

Ratio quintile Ratio range Anchors Recession onset rate 1960–1999 2000–2026
Top quintile (highest) > 25.9 139 18.0% 26.8% 5.4%
Bottom quintile (lowest) < 12.5 139 25.9% 19.5% 37.5%
All anchors (base rate) — 695 14.4% 15.5% 12.8%

A high ratio is associated with a recession rate of 18.0% against a base rate of 14.4% — a gap of -7.9 pp versus the bottom quintile. That is not a signal, and the sign is wrong: the claim says high ratio should mean more recession.

The era split changes the sign

The ratio's structural centre has drifted enormously — the median was 24.2 in the 1960s and the current reading is 37.8 — so a raw level cut-off does not compare like with like across time. Splitting the same test at the year 2000 is revealing.

1960–1999 · top quintile 26.8% bottom quintile 19.5%
2000–2026 · top quintile 5.4% bottom quintile 37.5%

Before 2000 the relationship at least pointed the way the story says: the top quintile ran 7.3 pp above the bottom quintile. Since 2000 it has flipped hard — the bottom quintile was followed by a recession 32.1 pp more often than the top quintile, and the top quintile's 5.4% rate is well below the 12.8% base rate.

Normalising the ratio by its own trailing five-year median, so that each era is measured against its own centre, does not rescue the claim either. On that measure the top quintile ran 12.2% against a 14.4% base rate, and since 2000 the top quintile produced 0.0% recession onsets across 56 anchors.

The full decile table

Decile Gold/oil ratio Anchors Onset rate 1960–1999 2000–2026
1 7.1 – 10.9 69 42.0% 50.0% 39.6%
2 11.0 – 12.5 69 10.1% 10.5% 9.7%
3 12.5 – 14.5 69 10.1% 17.9% 0.0%
4 14.5 – 17.0 69 1.4% 3.7% 0.0%
5 17.0 – 20.5 69 13.0% 12.8% 13.6%
6 20.5 – 22.7 69 18.8% 25.0% 0.0%
7 22.7 – 24.1 69 7.2% 3.4% 30.0%
8 24.1 – 25.6 69 5.8% 2.0% 15.0%
9 25.6 – 29.8 69 5.8% 4.7% 7.7%
10 29.9 – 64.4 74 28.4% 46.5% 3.2%

Rates above the 14.4% base rate are highlighted. The pattern is not a clean monotone signal in either era — which is itself the finding.

Claim 2 is not supported. A high gold/oil ratio does not predict recessions. In the era that most people actually trade gold in, it is associated with roughly a quarter as many recession onsets as the base rate.

Claim 3: gold hedges against an oil price shock

The third claim is that when oil collapses — the classic demand-shock recession signal — gold catches a bid as the flight-to-safety trade. The data says no.

Months oil fell 10%+ 49 over the full sample
Gold's average move -0.49% median -0.31%
Months gold rose 22 / 49 44.9% of the time
Bootstrap 95% CI -1.44 to +0.61% straddles zero

In the 49 months where oil fell 10% or more, gold's average monthly move was -0.49% and gold was positive in only 22 of them. The bootstrap interval (-1.44% to +0.61%) contains zero, so there is no evidence gold does anything unusual in these months — it behaves roughly like a normal month.

The bigger oil moves tell a similar story. Across the 50 months where oil moved 15% or more in a single month, gold's average move was +4.35% during the ten largest oil advances and +1.22% during the ten largest oil declines. Gold outperformed oil in 10 of the ten largest oil declines — the one place the hedge story shows up, and it shows up because oil fell further than gold, not because gold rose.

Claim 3 is not supported. Gold does not respond systematically to oil crashes. It behaves like an ordinary month with a small negative drift.

What the ratio level does tell you

None of the three claims survived, which would be a fine place to stop. But the ratio does carry one piece of information, and it is about forward gold returns rather than about recessions.

Sort every month with a complete forward window by its gold/oil ratio and look at gold's return over the next 12 months. The shape is a U.

Decile Gold/oil ratio Months Mean forward 12m gold Bootstrap 95% CI 1960–1999 2000–2026
1 6.8 – 10.7 79 +25.44% +20.45 to +31.12% +61.50% +15.55%
2 10.8 – 12.2 79 +16.76% +11.08 to +22.29% +19.56% +12.42%
3 12.2 – 14.1 79 +3.32% -2.55 to +8.80% -0.06% +8.03%
4 14.1 – 16.5 79 +5.22% -0.99 to +12.13% +0.73% +9.38%
5 16.5 – 19.6 79 -0.31% -4.57 to +4.13% -4.92% +7.63%
6 19.7 – 21.9 79 +0.84% -1.15 to +2.82% -2.04% +10.60%
7 22.0 – 23.3 79 +3.86% +1.16 to +7.12% +1.49% +13.20%
8 23.4 – 25.6 79 +10.98% +5.06 to +17.53% +10.11% +13.53%
9 25.6 – 30.0 79 +12.69% +7.33 to +18.48% +12.33% +13.33%
10 30.0 – 80.1 79 +21.51% +7.90 to +36.36% +22.10% +20.59%

Gold's mean forward return was +25.4% in the lowest ratio decile and +21.5% in the highest — but only -0.3% in the middle of the distribution. Both tails of the ratio are associated with stronger forward gold returns, and the middle of the range is where gold has historically gone nowhere.

There is an obvious mechanical explanation for the right-hand tail, and it is worth stating plainly rather than dressing it up as a signal. A very high gold/oil ratio usually happens because gold has just risen a lot. In 1973-06, the ratio hit 50.0 with gold at $120/oz; in April 2020 it hit 80.1 with gold at $1,683/oz. High-momentum gold followed by more gold momentum is a continuation pattern, not an early-warning pattern. It is also the reason the recession reading above is weak: high-ratio months are disproportionately months in which gold was rallying for its own reasons.

The left-hand tail is a different story and is more interesting. The lowest decile covers ratios of 6.8 to 10.7 — a mix of the oil-shock years (1974, 1976, 1977 and 2008), when oil ranged $12–$133 while gold averaged $471/oz. Gold's forward 12-month return from that decile averaged +25.4%, with a bootstrap interval of +20.5% to +31.1%. It is the single strongest cell in the table, and unlike the high end it is not a momentum artefact — it is gold catching up after being beaten down while oil spiked.

Careful. This is a description of what followed, not a rule. Two tails of a historical distribution are not a trading signal, the decile sample sizes are small, and the split between 1960–1999 and 2000–2026 shows the relationship is not stable across time. Nothing here is a recommendation to position.

Where the ratio is now

Current ratio 37.8 2026-10
Percentile since 1960 95.3 of 802 months
5-year median, 60m 34.6 93.8th percentile
vs its 5-year median 1.363 1.000 = at the median

The ratio closed 2026-10 at 37.8, with gold at $4,105/oz and crude at $108.50/bbl. That places it in the 95.3th percentile of the full history and above its own trailing five-year median.

For scale, the ratio has spent 15 months at or above 50 since 1960. Every one of them except 1973-06 falls in 2020 or later:

Month Ratio Gold $/oz Oil $/bbl Forward 12m gold Recession onset
1973-06 50.0 120 2.40 +28.33% yes
2020-04 80.1 1,683 21.00 +4.58% no
2020-05 56.4 1,716 30.40 +7.81% no
2025-05 52.8 3,309 62.70 +38.62% no
2025-08 50.5 3,368 66.70 +30.97% no
2025-09 55.2 3,668 66.50 +17.75% no
2025-10 64.4 4,058 63.00 +1.16% no
2025-11 65.6 4,087 62.30 not yet observable no
2025-12 70.8 4,309 60.90 not yet observable no
2026-01 74.6 4,753 63.70 not yet observable no
2026-02 73.8 5,020 68.00 not yet observable no
2026-03 50.8 4,856 95.60 not yet observable no
2026-06 51.8 4,228 81.70 not yet observable no
2026-07 51.0 4,073 79.80 not yet observable no
2026-08 52.3 4,411 84.40 not yet observable no

The most recent of these is 2026-08 at 52.3. On the test above, that is a reading associated with a low recession-onset rate and a U-shaped forward-return profile — but the recession sample behind it is small and the forward returns for the most recent months are not yet observable.

How far the ratio can travel

The ratio is not a bounded instrument. Its drawdown episodes, measured peak to trough, are larger than almost anything in gold itself.

Peak Ratio at peak Trough Ratio at trough Depth Months to trough Recovered
1973-06 50.0 2008-06 6.8 -86.5% 420 yes, 2020-04 (142m)
2020-04 80.1 2022-06 15.7 -80.4% 26 not yet
1969-05 33.1 1971-01 23.8 -28.2% 20 yes, 1972-06 (17m)

The longest drawdown took 420 months to bottom, from 1973-06 to 2008-06, and took a further 142 months to recover. Anyone using the ratio as a timing tool has to be willing to sit through moves of that size for years at a time.

Decade by decade

Decade Months Mean Median Min Max
1960s 120 25.2 24.2 21.9 33.1
1970s 120 20.0 15.8 7.7 50.0
1980s 120 18.2 14.7 9.6 36.3
1990s 120 19.9 20.1 11.0 28.2
2000s 120 11.2 10.5 6.8 22.6
2010s 120 19.2 17.3 11.6 38.7
2020s (to 2026-10) 82 35.3 28.8 15.7 80.1

The decade medians climb from 24.2 in the 1960s to 28.8 in the 2020s. A level that meant one thing in 1975 means something entirely different today, which is why the era splits above matter more than the headline numbers.

What this page is not

These are price-only series. No financing, storage, roll, refining or dividend cost is included on either leg. Oil carries far larger carry costs than gold — futures roll contango or backwardation depending on the regime, and the cost of rolling a crude position is not trivial — so any total-return comparison would flatter gold considerably more than it flatters it here.

Monthly averages smooth intramonth moves, so a ratio built from daily closes would be noisier and would not reproduce these figures. The ratio is not itself a tradable instrument; you would need to express it through two separate positions, each with its own financing and transaction cost.

The recession test uses one recession definition (NBER onset) on one country (the United States) and does not test other economies. The 12-month horizon was chosen in advance and repeated for the normalised variant, but other horizons were not explored, so the specific split of the sample is not searched over. Finally, the extreme-ratio table includes months whose forward returns are not yet observable, and those are marked as such rather than excluded.

Nothing on this page is investment advice, and past results do not predict future returns. A ratio that has made a new high in the last year can make a new low in the next one, as the table above shows.

Related reading

Frequently asked questions

What is the gold/oil ratio?

It is the price of one troy ounce of gold divided by the price of one barrel of crude oil, usually expressed as a barrel-per-ounce figure. A higher number means gold is expensive relative to oil; a lower number means oil is expensive relative to gold. On this dataset it ranges from 6.8 in 2008-06 to 80.1 in 2020-04.

Does a high gold/oil ratio predict a recession?

Not on this data. Anchors in the top quintile of the ratio were followed by an NBER recession onset within 12 months 18.0% of the time, against a base rate of 14.4%. The relationship is not only weak, it points the opposite way in the modern era: since 2000 the top quintile was followed by a recession 5.4% of the time and the bottom quintile 37.5% of the time.

Are gold and oil inversely correlated?

The claim is the opposite of what the data shows. Monthly gold and oil returns are positively correlated at 0.1819, which explains 3.3% of the variance in gold's monthly return. The correlation was negative in only 24.1% of rolling 36-month windows.

Does gold protect against an oil price crash?

No. In the 49 months where oil fell 10% or more, gold's average monthly move was -0.49%, and gold was positive in only 22 of 49 of them. The bootstrap 95% interval runs from -1.44% to +0.61%, which straddles zero.

What does the ratio's level actually predict?

The shape of forward gold returns is U-shaped, not monotonic. Gold's mean forward 12-month return was +25.4% in the lowest ratio decile and +21.5% in the highest, but only -0.3% in the middle. Very high ratios follow large gold advances, so the high end is not a measure of the gold being expensive.

What are the important caveats?

These are price-only series with no financing, storage, roll, refining or dividend cost on either leg. Oil carries far larger carry costs than gold, so any total-return comparison would favour gold even more. The ratio's structural centre moved from about 24.2 in the 1960s to 37.8 now, so a raw level is not comparable across eras without normalising; this page reports every headline split separately for 1960-1999 and 2000-2026.

Sources

  • World Bank Commodity Price Data (The Pink Sheet) — Monthly Prices sheet, gold column and crude oil average column, monthly averages in USD. Data updated November 03, 2026.
  • FRED USREC — NBER US recession indicator, monthly. Used to date recession onsets.
  • scripts/gold_oil_ratio_analysis.py — production script, writes src/data/research-gold-oil-ratio.json and both CSVs.
  • scripts/verify_gold_oil_ratio.py — independent second implementation; recomputes every published statistic from the published CSVs with numpy and a different code path.
  • gold-oil-ratio-monthly.csv — every month, both prices, the ratio, the ratio against its trailing five-year median, monthly returns and the recession flag.
  • gold-oil-ratio-events.csv — the 50 months where oil moved 15% or more.

Last updated: 2026-11-10

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