Research · Dataset

Gold vs. real yields and the dollar

285 months of gold returns paired month-by-month with US real yields, nominal yields and the broad dollar index. Every number on this page is produced by our published script from the raw sources.

Rates sample 2003-02 → 2026-10 · Dollar sample 2006-02 → 2026-10 · Gold data updated November 03, 2026 · FRED accessed 2026-11-08

Source and method

Gold price
World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet, column 70 (Gold, $/troy oz), monthly average, nominal USD
Rates and dollar
FRED (Federal Reserve Bank of St. Louis), accessed 2026-11-08:
DFII10 — 10-year TIPS real yield
DGS10 — 10-year nominal Treasury yield
T10YIE — 10-year breakeven inflation
DTWEXBGS — nominal broad dollar index
CPIAUCSL — CPI, used for year-over-year inflation
Alignment
Daily FRED series are averaged into calendar-month means to match the Pink Sheet monthly average gold price. Mixing a monthly average with a month-end value would manufacture a lead or lag that is not in the data.
Download
gold-monthly-macro.csv (merged monthly series, 285 rows)
Reproduction
scripts/gold_macro_correlation.py — recomputes every table below from the workbook and the FRED CSVs

Daily FRED series are averaged into calendar-month means to match the Pink Sheet monthly average gold price. Correlations are computed on monthly CHANGES, never on price levels. Gold here is a price series with no carry, financing or roll cost.

The three relationships, measured on changes

Gold monthly return against the monthly change in each macro series. Correlating levels would be meaningless here: both series trend, and a correlation between two trending series measures their shared drift, not any relationship a trader could act on.

Relationship Correlation Months R² p-value Sample
vs. change in 10y real yield -0.459 285 21% 2.92e-16 2003-02 → 2026-10
vs. change in 10y nominal yield -0.268 285 7% 4.35e-06 2003-02 → 2026-10
vs. change in broad dollar index -0.413 249 17% 1.11e-11 2006-02 → 2026-10

All three correlations are negative and statistically significant, and the real yield is the strongest of the three by a clear margin (1.7× the nominal yield, 1.1× the dollar). But note the R² column: the real yield explains 21% of the variance of monthly gold returns. Even the best of these three channels leaves the majority of monthly gold movement unexplained.

The level correlations are near zero — and one has the wrong sign

Gold returns vs. real yield LEVEL 0.124 changes give -0.46
Gold returns vs. dollar LEVEL 0.042 changes give -0.41
Variance explained by real yield 21% of monthly gold returns

These are correlations of gold MONTHLY RETURNS against the LEVEL of the macro series. Both are close to zero, and the real-yield one is POSITIVE -- the opposite sign to the change-based relationship. Reading gold off a real-yield chart in levels tells you almost nothing about the next month, and the chart is the form most commentary uses.

Regime split: high real yield vs. low real yield

Months split by whether the 10-year real yield was above or below its sample median of 1.06%.

Real yield above median -0.412 142 months
Real yield below median -0.571 143 months
Difference 0.159 95% CI -0.05 to 0.38

The relationship is stronger in the low-real-yield half, and the bootstrap says the gap leans that way in 7.1% of resamples — but the confidence interval includes zero, so this is a hint, not a demonstrated effect. The split variable is derived from the same data and the two halves are not independent samples, so a standard significance test on the difference would be invalid; the interval here is a moving-block bootstrap (12-month blocks, 2000 resamples, fixed seed) for exactly that reason.

Rolling 36-month correlation

The relationship is not a constant. Over 250 overlapping 36-month windows, the correlation ranged from -0.77 (window ending 2021-12) to -0.07 (window ending 2006-09), with a median of -0.56.

Window ending Correlation 10y real yield
2026-10 -0.277 2.78%
2026-09 -0.261 2.64%
2026-08 -0.243 2.40%
2026-07 -0.284 2.35%
2026-06 -0.257 2.18%
2026-05 -0.254 2.04%
2026-04 -0.256 1.94%
2026-03 -0.273 1.91%
2026-02 -0.263 1.82%
2026-01 -0.271 1.91%
2025-12 -0.296 1.90%
2025-11 -0.319 1.83%

The most recent 12 windows are shown. In 0% of all 250 windows the correlation was positive — it was negative in every single one. The sign is remarkably durable; the strength is not. The deepest reading came when real yields were negative (-0.99% in 2021-12), and the shallowest when real yields were high (2.32% in 2006-09).

Lead-lag test: does anything move first?

Cross-correlation of gold monthly returns against the macro change at lags of −6 to +6 months. A negative lag means the macro variable moved first and could in principle act as a signal; a positive lag means gold led.

Lag (months) vs. dollar vs. real yield
-6 -0.034 0.009
-5 -0.026 -0.019
-4 0.021 -0.022
-3 0.076 0.016
-2 -0.003 0.023
-1 -0.111 -0.115
0 (same month) -0.413 -0.459
+1 -0.125 -0.125
+2 -0.052 -0.076
+3 -0.081 0.003
+4 -0.137 -0.045
+5 -0.130 -0.042
+6 -0.029 -0.012

Both relationships are strongest at lag zero — dollar -0.41, real yield -0.46 — and decay in every direction away from it. The strongest absolute reading at any lag from −6 to +6 is the same-month one in both cases. At monthly frequency, the macro variable and gold move together; neither leads. A strategy that waits for the rates or dollar move to "confirm" a gold move is adding a month of latency to a signal that was already fully present.

The largest real-yield shocks and what gold did

The 15 biggest one-month moves in the 10-year real yield across 285 months, and gold's return in the same month.

Month Δ real yield Gold return 10y real yield CPI y/y
2008-10 +0.90 pp -2.77% 2.75% 3.73%
2022-09 +0.75 pp -4.76% 1.14% 8.19%
2008-12 -0.72 pp +7.23% 2.17% -0.02%
2013-06 +0.61 pp -5.02% 0.25% 1.72%
2022-04 +0.58 pp -0.56% -0.14% 8.23%
2011-08 -0.49 pp +11.82% 0.14% 3.75%
2022-10 +0.45 pp -1.01% 1.59% 7.76%
2004-04 +0.44 pp -0.98% 1.90% 2.29%
2007-11 -0.43 pp +6.75% 1.77% 4.37%
2003-07 +0.39 pp -1.40% 2.11% 2.06%
2010-10 -0.38 pp +5.59% 0.53% 1.17%
2010-12 +0.38 pp +1.53% 1.04% 1.44%
2023-10 +0.37 pp +0.00% 2.41% 3.25%
2023-12 -0.36 pp +2.12% 1.84% 3.32%
2022-05 +0.35 pp -4.54% 0.21% 8.54%

Of the 10 largest real-yield rises, gold fell in 8 of them, with an average decline of 2.63% in those months. All 5 of the largest real-yield falls were accompanied by a gold gain, averaging 6.70%. The relationship holds sharply at the extremes — but "sharp" is not "certain": two of the ten biggest rises still produced a rising gold price.

All months, split by the direction of the real-yield change

Months real yield rose 136 avg gold -0.56%
Months real yield fell 149 avg gold +2.30%
Spread of monthly averages +2.87% per month, one-way
Months in sample 285 2003-02 → 2026-10

These are conditional averages of monthly gold returns, not a trading rule. A +2.87% monthly gap is a real effect but a small one — and it is dwarfed by the dispersion within each group. Knowing which way the real yield moved tells you something about the average month; it does not tell you what this month will do.

By period

Period Months corr. real yield corr. dollar Gold over period Avg real yield Avg CPI y/y
2003-2009 83 -0.412 -0.422 +216.2% 1.96% 2.56%
2010-2015 72 -0.552 -0.313 -3.8% 0.37% 1.68%
2016-2020 60 -0.698 -0.285 +69.2% 0.27% 1.78%
2021-2023 36 -0.589 -0.742 +8.5% 0.40% 5.61%
2024-2026 34 -0.312 -0.475 +101.8% 2.02% 2.97%

The most recent block, 2024-2026, shows the correlation at -0.31 — less than half the strength of the 2016–2020 block (-0.70), and in a period where real yields averaged 2.02%, close to the 2003–2009 average. The dollar correlation is missing for the first block because the broad dollar index only begins in 2006.

Where the data stands now

Latest gold month2026-10
Monthly average price$4,105 / troy oz
Gold return that month-4.95%
10y real yield (TIPS)2.78%
10y nominal yield5.13%
10y breakeven inflation2.35%
Broad dollar index121.48
CPI year-over-year—%
Gold, trailing 12 months+1.2%
Gold, trailing 36 months+114.2%
Average real yield, last 12 months2.14%

The rate environment is at its most hostile to the gold thesis in this sample: the 10-year real yield is at 2.78% against a sample median of 1.06%, and the rolling 36-month correlation is well below its own average. Gold rose +1.2% over the last twelve months anyway. That is the pattern in the regime table above — high real yields weaken the relationship without reversing it.

How to use this — and how not to

  • These are correlations between monthly series. They describe a tendency across 20+ years, not a signal for any single month.
  • Real yields explain 21% of monthly gold variance. Treat the rest as unexplained rather than trying to find a macro story for every candle.
  • The lead-lag test covers monthly data only. It says nothing about intraday or weekly behaviour, and it does not rule out a short-lived intraday reaction.
  • Gold here is a price series with no carry, financing, storage or roll cost. A real position pays for those, and the correlations would not change but the returns would.
  • Nothing on this page is a forecast or a performance claim. Past relationships do not establish a forward edge.

For the mechanism behind these numbers, see Gold, the Dollar, and Real Yields. For the long-run price context, see the gold price history dataset. For the same history net of inflation, see the inflation-adjusted gold dataset. For the same series split by calendar month — and the tests that decide whether the calendar carries any real information — see Gold and the calendar. For the policy-event version of the same question — whether gold behaves differently in the months the Fed actually cuts — see Gold and the Fed cycle. For the policy-balance-sheet version of the same question — whether gold and Fed total assets move together at all — see Gold and the Fed balance sheet. For the recession-forecast version of the same question — whether an inverted yield curve says anything about gold over the next year — see Gold and the yield curve inversion. For the volatility side of the same question — what the options market is charging for gold volatility, measured against what gold actually did — see Gold implied volatility. For the physical-demand side of the same question — whether official central bank buying leads the price — see Central bank gold reserves and the gold price.

Frequently asked questions

How strongly do real yields and the dollar actually move gold?

Gold's monthly return correlates with the CHANGE in the 10-year real yield at -0.4591 (n = 285, R² = 0.2107, p = 2.92e-16), with the change in the nominal 10-year yield at -0.2683 (R² = 0.072), and with the change in the broad dollar index at -0.4131 (n = 249, R² = 0.1707). Real yields are the strongest of the three and about 1.7 times the nominal relationship. But note what R² means here: real yields explain roughly 21% of monthly gold variance, so most of the monthly movement has nothing to do with them.

Why does the same real-yield series look like it has the opposite sign in charts?

Because charts show levels and this page computes changes. Correlating gold's monthly return with the LEVEL of the real yield gives 0.1237 — near zero and POSITIVE, the opposite sign to the change-based figure. The same is true of the dollar: 0.0422 in levels. Reading gold off a real-yield chart in levels tells you almost nothing about the next month, and levels are the form most commentary uses. All the correlations on this page are computed on monthly changes.

Is the real-yield relationship stronger in some periods than others?

Yes, and that is the honest caveat. Splitting the 285 months at the sample median real yield of 1.06%, the correlation is -0.4118 in the high half and -0.571 in the low half, a difference of 0.1592. The block bootstrap interval for that difference runs from -0.0452 to 0.3832 and CROSSES ZERO, so the strength difference is a hint rather than a demonstrated effect — the split variable is derived from the data, which rules out a Fisher z-difference test. The sign of the relationship is the stable part; the strength is not.

Does the real-yield relationship hold across every period since 2003?

No, and it has been weakening. The block correlation was -0.6982 in 2016-2020 and is -0.312 in 2024-2026 — the weakest block in the table, with an average real yield of 2.02% against 0.27%. Gold still rose +101.8% over that block. A rolling 36-month correlation series of 250 windows is negative in 0% of them with a median of -0.5639, ranging from -0.7724 to -0.0669, so the SIGN is stable while the STRENGTH is not.

Do real yields or the dollar lead gold?

No. Scanning lags from -6 to +6 months, the strongest correlation for both is at lag 0: -0.4591 for real yields and -0.4131 for the dollar, with every other lag in the scan weaker. There is no calendar position at which the macro series has consistently forecast gold. On a monthly frequency, gold and the macro backdrop move together rather than one leading the other.

What happens in the extreme months?

The tails are where the relationship is sharpest. Of the 10 largest monthly rises in the 10-year real yield, gold fell in 8 of them, averaging 2.6% — but 2 of those shock months still saw gold rise, so even the extremes are not deterministic. All 5 of the largest real-yield falls saw gold rise, averaging 6.7%. Across all months, real-yield-up months averaged -0.563% and real-yield-down months 2.302% — a real difference, but only 2.865 percentage points on a monthly basis.

Where do gold and the macro backdrop stand right now?

As of 2026-10 the 10-year real yield is 2.78%, the nominal 10-year is 5.13%, the 10-year breakeven inflation rate is 2.35% and the broad dollar index is 121.48. Gold closed the month at $4105 with a -5% monthly return, and is up 1.2% over 12 months and 114.2% over 36 months. That is the regime where the real-yield relationship has historically been at its weakest, and gold rose anyway.

What are the limits of these correlations?

Three. They are monthly correlations across 285 months for the rate pair (from 2003-02) and 249 months for the dollar pair (from 2006-02) — each pair uses its own longest sample rather than one shared window, which is why the sample sizes differ. They describe a tendency, not a signal for any single month. And gold here is a price series with no carry, financing, storage or roll cost, so the correlations would not change for a real position but the returns would. Nothing on this page is a forecast.

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