Gold and Real Yields: Test the Relationship Across Regimes and Rolling Windows
Macro Research Workbench · MR06
Gold and Real Yields: Test the Relationship Across Regimes and Rolling Windows
The simple claim that gold must fall whenever real yields rise often misses. A rise in real yields can raise the opportunity cost of holding gold, which pays no interest, so an inverse relationship is sometimes observed, yet it is not a fixed law. This article separates the gold price and real yields into levels and changes, then tests the stability of the relationship with 12-, 26- and 52-week rolling correlations and by regime, using one consistent set of illustrative educational data and figures.
- The theoretical link between real yields as opportunity cost and gold
- Spurious level correlation versus correlation of changes
- Reading regime shifts through rolling correlation
- Confounders: the dollar, inflation uncertainty and liquidity
Key takeaways
- A rise in real yields can raise the opportunity cost of gold, so an inverse relationship is sometimes observed, but it is not a permanently fixed inverse correlation.
- A correlation of levels tends to look strong simply because of trends and is prone to spurious correlation. Using gold returns and real-yield changes changes the assessment.
- In this article’s fictional data, the level correlation is a strongly inverse -0.86, yet the full-sample correlation of changes is roughly 0.00, and the sign flips across regimes.
- The dollar, inflation uncertainty, liquidity and physical, investment and central-bank demand act as confounders that strengthen or weaken the link. Correlation does not prove causation.
- Every number, table and figure here is illustrative educational data (fictional), not actual market values, forecasts or trade recommendations.
Open the table of contents
The answer: the gold-real-yield relationship is a tendency, not a law
Here is the answer up front. The gold real yields relationship is sometimes observed as an inverse tendency, where gold softens as real yields rise, but it is not a fixed inverse correlation that always holds. The reasoning is simple: gold pays no interest or dividend, so when the real yield (the yield net of expected inflation) rises, the opportunity cost of holding gold can rise relatively. In practice, however, the measured correlation strengthens, weakens or even turns same-direction depending on how the sample period is chosen, whether you look at levels or changes, and confounders such as the dollar and inflation uncertainty.
So in practice, rather than concluding in one step that “real yields rose, therefore gold falls,” you need to test the stability of the relationship itself. The aim of this article is to help gold, FX and macro researchers avoid oversimplifications like “it’s inverse, so sell,” and instead check the data with rolling correlations and by regime. It helps to first get an overview of the whole research process in the Macro Analysis Guide that connects COT, rates, real yields and EIA data, which makes it easier to place this lesson in context.
This article explains, for educational purposes, the idea of mechanically organizing and visualizing public macro data. It does not recommend buying, selling, holding, entering or exiting any specific metal, currency or Treasury, nor any price target or expected return. All numbers, tables and figures below are illustrative educational data (fictional), not actual market values, forecasts or trade recommendations.
Separate nominal rates, expected inflation and real yields
Before testing the link with gold, it helps to sort out the types of yield. Confusing them leads to misreading the correlation. The rough relationship is captured by the following identity.
Real yield ≈ nominal yield − expected inflation. The nominal yield is the headline rate, expected inflation is the market’s priced-in future price rise (approximated by the breakeven inflation rate), and the real yield is their difference, a purchasing-power yield. In discussing gold’s opportunity cost, the theoretical link is considered stronger with the real yield, which removes the inflation component.
To show it with a single example (a fictional single week), when the nominal 10-year yield is 4.10% and the breakeven inflation rate is 2.30%, the real yield is roughly 4.10 − 2.30 = 1.80%. What matters here is that even if the nominal yield rises, the real yield may not move if expected inflation rises by the same amount at the same time. In other words, do not lump everything together as “higher rates”; you need to separate which of nominal, expected inflation and real actually moved. In gold research, the 10-year real yield (such as the TIPS real yield), which is easy to compare against maturity-free gold, is often the starting reference. Yields and real yields themselves are covered in detail in How to Read Treasury Yields and the Yield Curve (2s10s and real yields).
Fictional dataset, dates, units and transforms
Every analysis below is based on the following consistent set of illustrative educational data (fictional dataset W). The same values are used across the prose, tables, figures and interactive tool, and levels are distinguished from changes.
| Series | Unit | Frequency | Level transform | Change definition |
|---|---|---|---|---|
| Gold price (XAUUSD-equivalent) | Index (start 100) | Weekly | Index level | Weekly return (%) |
| 10-year real yield (TIPS-equivalent) | % | Weekly | Yield level (%) | Week-over-week change (bp) |
| U.S. dollar index | Index | Weekly | Index level | Week-over-week change (index pts) |
The period is a fictional run of 52 consecutive weeks (W1-W52), with observations assumed aligned to the same time each week’s end. Levels and changes (difference/return) are clearly separated: real-yield change is expressed in bp (basis points; 0.01% = 1bp), gold in weekly return (%) and the dollar in week-over-week index points. All correlations are Pearson correlation coefficients, with the sample size n stated alongside. Because weekly series are joined to weekly series, any week with a holiday or missing value is excluded, with no forward fill. Because forward-filling economic data does not mean the value was knowable at the time, missing data is treated plainly as missing.
With real data, keep the observation date, period end, release date, retrieval date and revision date managed separately, and never mix values published after the analysis date into a historical analysis. TIPS real yields and gold closing prices can be revised, delayed or missing. For handling point-in-time data, see the article on historical regime comparison, data revisions and look-ahead bias.
Levels vs changes: beware spurious correlation
The first pitfall is the “correlation of levels.” If you correlate the level of the gold price directly with the level of the real yield, a strong correlation can appear simply because both trend heavily over the same period. In fictional dataset W, the correlation between the gold level and the real-yield level is r = -0.86 (n = 52), which looks like a rock-solid inverse relationship. Yet this is a strong candidate for spurious correlation, largely reflecting the fact that each series moved in one direction across the period.
So when you re-take the correlation using changes – gold’s weekly return and the real yield’s week-over-week change (bp) – it drops to r ≈ 0.00 (n = 51) over the full sample. In other words, viewed by “which way each moved week to week,” their co-movement essentially averages to nothing, sharply at odds with the strong inverse impression from the levels. The danger of correlating trending series in levels is plainly on display here.
So which is right? The answer is “look at both separately.” A level correlation reflects long-run direction, while a change correlation reflects short-term co-movement. If your goal is to know the day-to-day or week-to-week co-movement, changes are appropriate, and it is dangerous to judge “it’s inverse” from a strong level correlation alone. From the next section, we look at how this relationship between changes evolves over time.
Confounder hypothesis map: the dollar, inflation uncertainty, liquidity and demand
The gold price is not determined by real yields alone. Part of the apparent relationship between real yields and gold may be a third factor moving both. Below are the main confounders organized as hypotheses. The arrows are hypotheses and do not prove causation.
The dollar is especially important. Because gold is priced mainly in U.S. dollars, a stronger dollar can be a downward factor for gold through a channel separate from real yields. And because real yields and the dollar often move together, part of the gold-versus-real-yield correlation may run through the dollar. That is exactly why we check real-yield changes and dollar changes side by side. The relationship between the dollar and rate differentials is explored in Interest Rate Differentials and FX (comparing nominal, real and expected rates). In fictional dataset W, the full-sample correlation of gold returns and dollar changes is r = -0.24 (n = 51), a weaker inverse link than with real-yield changes.
What is free and what Pro adds
The free basic view lets you review Treasury yields, real yields and the Gold x Real Yield overlay. When you need longer history, percentiles and z-scores across multiple tenors, a rate-differential builder, local saving and PDF or CSV exports, Pro covers those.
Rolling correlation and regime shifts
Compute one correlation over the whole sample and the changes along the way cancel out on average and disappear. That is where the rolling correlation comes in. It recomputes the correlation of gold returns and real-yield changes over a fixed window (say 12 weeks), sliding one week at a time, and lines the results up as a time series. A shorter window captures regimes more responsively but has a smaller sample and more noise; a longer window is smoother but slower to react.
The figure below shows the 12-week and 26-week rolling correlations for fictional dataset W. The first part (roughly W11-W21) is a strong inverse -0.5 to -0.84, the middle (W22-W34) weakens to -0.2 to about 0, and the latter part (W35-W52) turns same-direction (positive) at +0.2 to +0.36. The near-0.00 full-sample correlation turns out to be the result of these opposite-signed regimes canceling each other out. The 26-week window traces the same arc more smoothly and with a lag.
Strong inverse, weak and same-direction: three regimes
The movement of the rolling correlation is summarized in a table of three pre-defined regimes. To avoid post-hoc storytelling like “this is where it was inverse,” the regimes are mechanically pre-defined by week number (W1-W17 / W18-W34 / W35-W52). For each regime, the Pearson correlation of gold returns with real-yield changes and with dollar changes is placed side by side.
| Regime (pre-defined) | Gold x real-yield change | Gold x dollar change | Interpretation (hypothesis) |
|---|---|---|---|
| Regime A (W1-W17) strong inverse | -0.65 | -0.62 | A period where rising real yields and a stronger dollar aligned the downward pull |
| Regime B (W18-W34) weak / unstable | -0.08 | -0.19 | A period where other factors mixed in and co-movement became ambiguous |
| Regime C (W35-W52) same-direction | +0.35 | +0.01 | A period where yields and gold moved together and the inverse link broke down |
How to read the table: a minus sign means inverse (as one rises the other tends to fall), a plus sign means same-direction. In Regime A, both real-yield change and dollar change are strongly inverse, consistent with the intuition that “higher rates push gold down,” but in Regime C the gold x real-yield change flips to +0.35. Because the relationship changes this much even within a single year, it is important not to treat a single correlation value as a permanent law. All numbers are illustrative educational data.
A post-hoc explanation such as “Regime C was positive because of X” can be manufactured endlessly. To avoid distorting the analysis, define the regime boundaries, window width and target tenor before looking at the data, and set the break conditions in advance too (for example, the correlation flipping sign for two consecutive weeks). Correlation does not prove causation.
Rolling-correlation explorer (educational)
You can check the correlations covered so far by hand, using the same fictional dataset W. Choose the variable (real yield / dollar index), the comparison type (levels / changes) and the window (12 weeks / 26 weeks / 52 weeks = full sample), and it computes the Pearson correlation, sample size and rolling range in the browser only. Neither inputs nor results are sent over the network, and nothing is saved. It does not convert into bullish/bearish or buy/sell judgments.
Static worked example when JavaScript is disabled (illustrative educational data): with variable = real yield, comparison = changes and the full sample, the Pearson correlation of gold returns and real-yield changes is r ≈ 0.00 (n = 51). With the same variable in levels over the full sample it is r = -0.86 (n = 52). The 12-week rolling range (changes) is about -0.84 to +0.36, and the latest 12-week window is about +0.24. Switching the variable to the dollar index gives r = -0.24 (n = 51) for changes over the full sample. These match Table 2 and Figure 3. The interactive tool below starts from this static example and recomputes as you change the selection.
Rolling-correlation mini tool (fixed, fictional data)
This tool is a teaching aid for getting a feel for correlation, not a full reproduction of the formal workbench. For testing with real data, arbitrary periods and multiple tenors, plus saving and exports, use the formal Gold x Real Yield feature from the workflow below.
Interpretation limits and lead-lag caution
Correlation analysis has inherent limits. To avoid misreading them, they are stated explicitly.
- Correlation is not causation: even if gold and real yields are inversely correlated, neither is necessarily the cause. Always leave open that a third factor (the dollar, liquidity, etc.) is moving both.
- Spurious correlation: trending series in levels show inflated correlation. Convert to changes or returns and state the sample size and period.
- Regime dependence: the correlation can even change sign across periods. Do not treat a single full-sample value as a permanent law.
- Lag search invites multiple testing: brute-forcing “how many weeks of shift maximizes the correlation” tends to pick up chance high correlations. You need a procedure that accounts for publication lags and holdouts. For the correct approach to time-shifted correlation, see Lead-Lag Analysis (testing lead and lag with time-shifted correlation).
Rather than “real yields rose, therefore gold falls,” keep it to a conditional statement: “in this period and this window an inverse link was observed, but in another regime it weakened and was at times same-direction.” Do not conclude a price direction from one correlation value; pair it with break conditions (a sign flip or loss of correlation) and additional data to watch (the dollar, expected inflation). When reading positioning extremes together with COT data, COT Percentile and Z-Score is also useful.
Practical checklist and the workbench workflow
Below are the items worth checking before and after testing gold x real yields, and the confirmation workflow in the SG Group Macro Research Workbench.
Pre-testing checklist
- Have you decided which of nominal, expected inflation or real yield to look at? Have you fixed the tenor (such as 10-year)?
- Have you separated levels from changes? Have you suspected spurious correlation from trends?
- Have you defined the window (12 / 26 / 52 weeks) and the regime boundaries before looking at the data?
- Have you reviewed confounders such as the dollar and expected inflation side by side?
- Have you checked the sample size n, missing data, revisions and point-in-time alignment (release date versus retrieval date)?
Workflow in the workbench
Basic view
Open the basic Treasury-yield and real-yield view and the basic Gold x Real Yield template (free).
Levels and changes
Switch between the level overlay and the comparison of changes to avoid spurious correlation.
History and saving
Use Pro for longer history, percentiles, z-scores and a rate-differential builder, plus local saving and PDF or CSV exports.
Regimes and lead-lag
Use Premium to test historical regime comparison, lead-lag and point-in-time, avoiding look-ahead bias.
A comfortable progression is to first confirm the levels and changes of Gold x Real Yield with the free basic template, then consider Pro once you need longer history or saving and exports, and Premium when you want to go as far as historical regime comparison and lead-lag. Because the scope can change, check the latest on the plan comparison page.
Gold x real yields, with real data and any period
The tool in this article is for building intuition. To choose a tenor on the actual series and check levels, changes, regimes and rolling correlations, go to the free basic view. If you need saving and exports, Pro covers it; historical regime comparison and lead-lag are handled by Premium.
Frequently asked questions (FAQ)
Does gold fall whenever real yields rise?
It often tends to soften, but it does not always fall. A rise in real yields can raise the opportunity cost of holding gold, which pays no interest, so an inverse relationship appears in some periods. However, this link strengthens, weakens or even turns same-direction depending on the dollar, inflation uncertainty, liquidity and physical or central-bank demand. Rather than asserting a single rule, it is important to split the sample by period and regime and check whether the correlation is stable.
Which real-yield maturity is commonly used in gold research?
The 10-year real yield is a common reference. Gold has no maturity and is often compared with a longer-dated real yield, so a longer tenor such as the 10-year TIPS real yield is a natural starting point. That said, using short and medium tenors as well shows that explanatory power changes by regime. Fixing one tenor keeps comparisons consistent, while checking other tenors confirms the relationship does not break down.
How do nominal and real yields differ?
The nominal yield is the headline rate, while the real yield subtracts expected inflation to give a purchasing-power yield. Roughly, real yield equals nominal yield minus expected inflation. When reasoning about the opportunity cost of gold, the real yield, which removes the inflation component, is considered the theoretically stronger link. Note that even if the nominal yield rises, the real yield may not move if expected inflation rises at the same time.
Should gold and real yields be compared in levels or changes?
It is safer to look at both separately. A correlation of levels can appear strong simply because both series trend in the same direction over time, which makes it prone to spurious correlation. Using gold returns and real-yield changes (week-over-week) evaluates shorter-term co-movement. In this article’s fictional data, the level correlation looks strongly inverse, yet the change correlation is close to zero and its sign flips across regimes.
What is a rolling correlation?
It recomputes the correlation coefficient over a fixed window that slides through time, showing how the relationship changes. For example, a 12-week window computes the correlation over the most recent 12 weeks, then shifts one week at a time to form a series. Compared with a single full-sample correlation, it distinguishes periods of strong inverse, weak and positive correlation. A shorter window is more responsive but noisier; a longer one is smoother but lags.
Should the U.S. dollar be reviewed at the same time?
Reviewing it alongside is recommended. Because gold is priced mainly in U.S. dollars, dollar strength can affect the gold price through a channel separate from real yields. Real yields and the dollar often move together, so looking at only one tends to overstate the relationship. Placing real-yield changes and dollar changes side by side helps separate which factor is at work in a given period. Correlation, however, does not prove causation.
Why can the relationship break down?
Because the gold price moves for reasons beyond real yields. When inflation uncertainty rises, geopolitical safe-haven demand increases, central banks buy, or the liquidity environment shifts, the link with real yields can temporarily weaken or reverse. A correlation is a descriptive statistic that depends on assumptions and the sample period, not a permanent law. That is precisely why it helps to define regimes in advance and decide the conditions under which the link breaks and the additional data to watch.
What does the workbench’s gold-versus-real-yield view show?
The free basic template lets you review basic Treasury-yield and real-yield displays and a basic Gold and Real Yield overlay view. Pro centers on longer history, percentiles and z-scores across multiple tenors, a rate-differential builder, local saving and exports such as PDF and CSV. Historical regime comparison, lead-lag and point-in-time testing sit within Premium. Because the scope can change, check the plans page for the latest details.
References (primary sources)
The following primary sources provide background for the definitions and methodology in this article. Check each institution’s official page for usage terms, updates and revisions.
- U.S. Department of the Treasury — Interest Rate Statistics (nominal and real yield statistics): home.treasury.gov
- Federal Reserve Bank of St. Louis — FRED (time series for real yields, breakevens and more): fred.stlouisfed.org
- World Gold Council — Goldhub Data (gold supply, demand and price-related data): gold.org/goldhub/data
Disclaimer
This article explains, for educational purposes, the idea of mechanically organizing and visualizing public macro data and data loaded locally on your device. It is not investment advice, a trade signal, a forecast or a guarantee of returns. All numbers, figures, tables and tool defaults shown are illustrative educational data (fictional) and do not represent real market values, performance or user counts. Correlation does not prove causation, and the relationship between real yields and gold changes with assumptions, period and regime, at times turning same-direction. Data can be delayed, revised or missing. Buying or holding specific metals, currencies or Treasuries, and calculating lot size, margin or trading costs, are outside the core scope of this article and this service. For usage terms of primary sources and the latest on SG Group’s features, free scope and pricing, please check the plans page.