Educational content only. This is not financial advice or an investment recommendation. The author is not a registered financial adviser. A correlation describes what two series did together in the past; it is not a forecast and it is not a trading rule.
In short: Between January 2003 and August 2026, the correlation between monthly changes in the gold price and monthly changes in the 10-year real yield was −0.46. The same calculation against the nominal yield gives −0.26. The real yield is roughly twice as closely related to gold as the number everyone quotes — and if you run the comparison the way most people run it, you get almost zero and conclude, wrongly, that neither matters.
The two numbers
Take the monthly average gold price and the monthly average 10-year yield. For each month, work out how much gold moved in per cent and how much the yield moved in percentage points. Then correlate the two columns. Over 283 months:
| Gold, monthly change, against | Correlation | Share of gold’s movement explained |
|---|---|---|
| The real 10-year yield | −0.46 | 21% |
| The nominal 10-year yield | −0.26 | 7% |
| The breakeven inflation rate | +0.20 | 4% |
The sign is the one the mechanism predicts: real yields up, gold down. Gold pays no interest, so when a safe bond pays more after inflation, holding metal costs more. That part is covered in How Bond Yields Move Gold.
What the figures add is the size. The real yield is about twice as closely related to gold as the nominal yield, and that gap is the practical case for doing the extra two minutes of work. The headline yield on the front of every financial site is the weaker of the two measures.
And then the third column, which is the one to sit with. A correlation of −0.46 means real yields account for about a fifth of gold’s month-to-month movement. Four fifths of it is something else: the dollar, central bank buying, physical demand, fear. Anyone presenting real yields as the driver of gold is overstating a relationship that is real but partial.
Why the usual comparison gives the wrong answer
Almost every version of this analysis you will find correlates the levels: the gold price against the real yield, month by month. Run that and you get this:
| Method | Real yield | Nominal yield |
|---|---|---|
| Correlating changes | −0.46 | −0.26 |
| Correlating levels | −0.09 | −0.04 |
Essentially nothing. Someone who does it that way concludes that real yields have no bearing on gold, and publishes a chart to prove it.
The reason is not subtle once you see it. Over this period gold went from $357 to $4,411 — a long climb driven by everything from central bank buying to the money supply. The real yield did no such thing: it wandered up and down inside a band and ended near where it started. Correlating a trending series against a cycling one measures the mismatch between a trend and a cycle, and tells you nothing about whether the two move together day to day.
Differencing removes the trend from both and asks the question people actually mean: when the real yield moves, does gold move the other way? The answer is yes, about a fifth of the time — and the level comparison cannot see it at all.
This is worth being explicit about because it cuts both ways. A correlation of levels can just as easily produce a spuriously high figure when two series happen to trend together for unrelated reasons. Either way the number is an artefact of the method.
One number for twenty years is still too few
−0.46 is an average across 283 months, and averages hide things. Recomputing the same correlation over a rolling three-year window shows what it hides:
Two things in that line, and they point in opposite directions.
It never changes sign. Across 248 overlapping three-year windows, from January 2006 to August 2026, the correlation is negative in every single one. Not one window out of 248 shows gold and real yields moving the same way. For a relationship in financial markets, that is unusually durable.
Its strength varies enormously. The deepest reading is −0.77 in December 2021 — a period when real yields were the dominant story and gold tracked them closely. The shallowest is −0.07 in September 2006, which is effectively no relationship at all. The average across all windows is −0.51.
So the honest statement is not “the correlation is −0.46”. It is that the relationship has been consistently negative for twenty years, and has ranged from negligible to strong depending on what else was happening. Anyone quoting a single figure without a window is offering a precision that the data does not contain.
By period
| Period | Months | vs real yield | vs nominal yield |
|---|---|---|---|
| 2003–2007 | 59 | −0.26 | −0.12 |
| 2008–2012 | 60 | −0.57 | −0.23 |
| 2013–2018 | 72 | −0.51 | −0.46 |
| 2019–2022 | 48 | −0.62 | −0.38 |
| 2023–2026 | 44 | −0.33 | −0.25 |
The real yield beats the nominal yield in every period. The gap is widest in 2008–2012, when inflation expectations were moving violently and the nominal yield was therefore at its least informative, and narrowest in 2013–2018, when they were not.
It has weakened recently
The most recent three-year window reads −0.24, against a twenty-year average of −0.51. The relationship is less than half its usual strength, and 2023–2026 is the weakest stretch since the mid-2000s.
What that means is a separate question from what the number says, and the number does not answer it. One reading is that something else has been doing more of the work — official sector buying is the candidate usually named, and it is demand that does not respond to yields at all. Another is that three years is a short window and this is noise. The data cannot tell you which, and a weakening correlation has reverted to form several times already in this series.
What it does mean, practically, is that a framework built entirely on real yields is explaining less of the gold price now than it did five years ago.
Why none of this goes back before 2003
Every figure on this page starts in January 2003, and that is not a choice. The 10-year Treasury real yield series begins on 2 January 2003. Before that there is no market-observed real yield to correlate anything with, because inflation-protected Treasuries at that maturity were not being quoted in a usable series.
You will nonetheless find confident statements about gold and real yields in the 1970s, or across the 1980 peak. Those are not using this data, because it does not exist. They are using a proxy: the nominal yield minus some measure of inflation, usually the past twelve months of CPI.
That proxy is a different quantity. A market real yield is what investors are willing to accept in advance, looking forward; nominal minus trailing CPI is an arithmetic operation on what already happened. In calm periods the two track each other. In exactly the periods people want to analyse — 1974, 1980, 2022 — inflation was moving fast enough that backward-looking and forward-looking measures diverge sharply.
So a correlation computed on the proxy is a real result about a different series, and it should be labelled that way. This page does not extend backwards, because doing so would mean quietly swapping the input halfway through and presenting one number as though it measured one thing.
What this does not say
It is not causation. The correlation is consistent with the opportunity-cost mechanism, which is a reason to take it seriously, but correlation alone cannot establish direction. Both series also respond to the same third things — a Federal Reserve decision moves real yields and gold at the same moment, through different routes.
It is not a rule for anything. A correlation of −0.46 over 283 months says nothing about next month. Four fifths of gold’s monthly variation is outside this relationship entirely, and the relationship itself has spent long stretches at −0.1.
And monthly averaging hides what happens inside a month. Both series are monthly means here, to match the gold data. A daily analysis would catch reactions to individual releases that this one averages away, and would also pick up far more noise.
How to check it yourself
Everything here is reproducible in a spreadsheet in about twenty minutes, from two free files.
- Download the daily 10-year real yield, series
DFII10, and the daily nominal 10-year,DGS10, from the Federal Reserve Bank of St Louis. Both are the US Treasury’s own published figures, republished unchanged. - Average each one by calendar month. Blank rows are market holidays; leave them out rather than carrying the previous value forward.
- Get the monthly gold price from the World Bank’s CMO-Historical-Data-Monthly.xlsx, Monthly Prices sheet. It is a monthly average, which is why the yields are averaged to match.
- Compute gold’s percentage change and the yield’s change in percentage points, month on month, then correlate the two columns.
If your figure differs from −0.46, the usual reasons are correlating levels instead of changes, using month-end values instead of monthly averages, or a different gold series.
Related reading
This page measures the relationship; How Bond Yields Move Gold explains why it exists and how to split a headline yield into its parts. The Treasury yields tool shows both components updating every business day, and The Four Releases That Move Gold covers what moves them in the first place.
Sources
- Real and nominal yields: Federal Reserve Bank of St Louis, FRED series DFII10 and DGS10, daily, downloaded 4 October 2026. Both republish the US Department of the Treasury’s daily yield curve rates; the September 2026 values were checked line by line against the Treasury’s own published figures and are identical.
- Gold prices: The World Bank, Commodity Prices — History and Projections, monthly averages. All 800 months used here were compared against the file published on 2 October 2026 and match exactly. Licensed under CC BY 4.0; the World Bank does not endorse this site or its analysis.
- Method: monthly averages throughout. Gold as percentage change, yields as change in percentage points. Pearson correlation over 283 monthly observations, January 2003 to August 2026. Rolling figures use a 36-month window, giving 248 overlapping windows from January 2006.
- Every figure on this page was computed twice, with two separate implementations, and the results agree to four decimal places.