Has Gold Decoupled From Real Yields? What the Data Actually Show
For most of the last two decades, gold and the real 10-year Treasury yield have moved with an inverse relationship. This is consistent with our understanding of the relationship, as gold pays no yield, so when the inflation-adjusted return on safe government debt rises, holding gold becomes more expensive in opportunity-cost terms, and the price tends to fall. When real yields sink, that cost disappears, and gold tends to rise. This relationship has been a reliable assumption for macro investing over time.
However, since 2022, this relationship has looked broken. Real yields rose sharply as the Fed hiked into the worst inflation in forty years, and yet gold hit record highs. The purpose of this analysis is to assess if this break shows up in the data as more than a chart pattern, and to use the findings to inform current portfolio decisions.
Methodology
Using monthly data from January 2003 (when TIPS-based real yield data begins) to mid-2026, we tested whether gold's price and the real 10-year yield share a cointegrating relationship. A cointegrated relationship is a statistical equilibrium where, even though both series wander over time, their first difference (value in period t - value in period t - 1) stays bounded and tends to close around a consistent mean. This is the standard econometric test to assess if two variables are statistically linked, rather than just trending over time.
Over the full sample, there is no evidence that the pair are cointegrated. The Engle-Granger cointegration test comes back firmly negative, suggesting there is no stable long-run equilibrium linking gold's level to the real yield's level across 2003 - 2026. This result is discouraging, but by exploring it further, we can create a more nuanced view.
Structural Breaks
The E-G test assumes a consistent relationship across time, but as discussed, post 2022, we have seen a reversal of the traditional path. Using the Gregory-Hansen test allows for a single break in the structure. This is the same idea as the Engle-Granger test, but it allows the data to search for the point where the relationship might have shifted, rather than assuming a fixed one throughout. Although this test still does not produce formal cointegration, the best-fitting break date the search converges on is January 2023. This point is independently arrived at by scanning every possible month in the sample, and has no information from external factors.
This finding gives the most important results of this exercise. Before 2023, the textbook inverse relationship existed. Although not processing a cointegrated relationship, running a simple OLS regression from 2003 - 2022 finds a beta of -0.41 (Regressing gold price on real yield finds a negative relationship). After 2023, the coefficient flips positive, with a value of +0.39. Gold and the real yield have been moving in the same direction, not inversely. While this is a much smaller sample, it represents a material shift in how gold is treated as an asset, and is a valuable insight even if variables are spurious.
The scatter plot above plots the real 10yr yield of a given month against the log gold price of that same month. An OLS regression is then plotted to highlight the slope of the relationship, and the blue slope highlights observations from 2003 - 2022, while the red line highlights observations from 2023 - 2026.
The blue line's downward slope (−0.41) is the inverse relationship we assume. The red line's upward slope (+0.39) is the same relationship, estimated the same way, in the 43 months since the estimated break, and instead points the opposite way. Note the red cluster sits at both higher real yields and higher gold prices than the blue cluster. This is a visual separation post 2022, and rather than the relationship having weakened, it has in fact shifted.
As neither half of the sample achieves formal cointegration on its own, conclusions must not be overstated. Rather than moving from one equilibrium relationship to another, results indicate that the loose, correctly-signed association investors have long relied on gave way, right around the start of 2023, to a loose, oppositely-signed one.
What Could Cause This Shift?
The leading real-world candidate for what changed is not a monetary story but rather a shift in demand. Central banks bought a record 1,136 tonnes of gold in 2022, the most in any year since 1967, and kept buying above 1,000 tonnes annually through 2023 and 2024 — more than double the 2010 - 2021 average of roughly 473 tonnes a year. That surge is widely tied to reserve managers, concentrated in emerging markets, diversifying away from dollar assets after Russia's dollar and euro reserves were frozen in 2022, and a de-dollarisation narrative took over.
The sequence of timing seems implicit. While buying surged, the estimated break data and the sign flip all land within the next few months. While this is not a causal test, and persistent fiscal deficits, a weaker dollar, and elevated geopolitical risk premia could also contribute to this change, it serves as a clean, qualitative coincidence between a documented demand shock, an independently estimated statistical break point, and a measured reversal in sign.
What Stayed The Same
While this longer-term, levels shift occurred, it is worth exploring the shorter-term change in levels. The diagram below shows the shorter-term change in values remained negative, even after the structural break.
When regressing on the change in levels, we discover an important distinction. The original regression we ran estimates the gold price as an independent variable. In other words, across the entire 13-year history, what absolute price level of gold is associated with a given absolute level of real yields, on average?
Instead, switching to changes, every point is a regression coefficient estimated using only the trailing 36 months of data, re-run for every month. In other words, if you only had the last three years, how much did a change in the real yield move gold's monthly return, on average? A negative value points towards the inverse relationship in the short term, and even after the break, the line has stayed below zero.
Understanding this difference is important. It is the level of gold, or where the price sits relative to what the real yield traditionally implied, that shifted. The short-run relationship has stayed consistent in sign, meaning a change in real yield in a given month may still cause an inverse change in gold. What has changed is the relative levels they sit at, not how they react month to month.
Understanding This Idea
Put together, the evidence supports a specific claim: the traditional inverse relationship between gold and real yields is not absolute. Inferring recent dynamics, this should be via a large enough and persistent enough shift in demand. Central-bank buying looks like exactly that kind of shock, as it has provided a sustained, multi-year elevated flow of purchases, rather than a one-off purchase.
While this can not support any claim of future gold price movements, if official-sector buying tapers back toward its old baseline or the narrative around dedollarisation and currency debasement shrinks, the older inverse relationship could plausibly reassert itself. Inversely to this, if sovereign demand persists as debt positions and currency credibility stutters, it is feasible to infer this relationship could break out further. Nevertheless, the post-break sample is short (43 months), and no formal cointegration was ever established in either regime, so results should be considered as suggestive, not conclusive.
So understanding this dynamic should be done as follows: the old macro assumption of real yields increasing leading to gold prices falling has worked reasonably well for twenty years. It has stopped describing it so well for the last four. The best available explanation for why is a persistent, well-documented demand shock large enough to invert the old relationship. Under this assumption, it is reasonable to assume that real yields have become a signal of weaker government credibility, as debt repayments become more unsustainable, driving demand into gold irrespective of the opportunity cost.
Applying It To Markets
The levels outlook:
If the yield climb currently underway is being read by the market at least partly as a debt-sustainability signal rather than purely a growth/inflation one, the post-2023 regime, where rising yields no longer force gold's price level down, has a real mechanism to keep holding. Under that lens, Gold’s consolidation at $4,300 - 4,440 while G7 yields hit multi-decade highs reinforces this shift in the level relationship. Real yields are climbing without dragging the price level back down with them. If the fiscal-pressure component of this yield move persists or intensifies, this relationship could strengthen.
The change outlook:
The mechanical, month-to-month relationship never went away; as we discussed, the rolling beta stayed negative right through the sample. As yields spike this week, we have seen a slight dip in gold. So even inside a regime where the level is structurally elevated, sharp, fast increases in yields may still place pressure on the current gold price as a function of shorter-term movements. This may be especially relevant for more like conventional rate-shock moves, such as the Iran/oil-driven component of the current spike, rather than slow-building fiscal-credibility repricing. The current consolidation range itself may partly be that short-run drag temporarily capping upside, even while the structural floor underneath stays higher than the old real-yield relationship alone would justify.
In other words: don't expect the elevated level to insulate gold from near-term dips when yields move abruptly; instead, expect the dips, and expect the floor to hold anyway.
