Insight, Research & Film

To reach that answer, Synterra ran a series of statistical tests on 5,060 trading days of market data, from January 2006 to September 2026. This article explains what we did and what we found, and it explains every technical term along the way. No prior knowledge of finance is assumed. Where a term needs more than a sentence, it is explained in a Plain English box.

1. Why should the dollar matter to gold at all?

Gold is a lump of metal. It pays no interest, issues no dividends and has no management team. So why should its price care what happens to one particular currency? There are two reasons, and both follow from the fact that gold is priced, traded and settled almost entirely in US dollars.

The first reason is arithmetic. Imagine a jeweller in Frankfurt who buys gold at $4,000 an ounce. If the dollar strengthens by 10% against the euro, that same ounce now costs her 10% more in her own currency, even though the dollar price has not moved. She buys less. So do buyers in Mumbai, Shanghai and Dubai. Lower demand pushes the dollar price of gold down until it finds a new balance. A stronger dollar therefore tends to mean cheaper gold in dollar terms, and a weaker dollar tends to mean dearer gold.

The second reason is competition. In a crisis, frightened money looks for somewhere safe to hide. The two most popular hiding places are US dollars, usually held as short-term US government debt, and gold. When the dollar becomes more attractive, typically because American interest rates rise, gold loses customers to its rival.

Put the two together and you get what traders loosely call a see-saw: when one end goes up, the other tends to go down. The question this article asks is how strong that see-saw really is, whether it has changed, and whether it still explains where gold trades.

2. Rounding up the usual suspects

The dollar does not act alone. If we simply compared gold with the dollar, we would risk crediting the dollar with work done by other forces that happen to move at the same time. It is rather like concluding that umbrellas cause traffic jams because both appear on rainy days. To isolate the dollar's own effect, we must also include the other main suspects, and hold them constant while we look at the dollar. Economists call these control variables.

Why not CPI? The Consumer Price Index (CPI) is the official measure of inflation, and an obvious candidate. The difficulty is that it is published once a month. In a daily model, CPI would appear as a flat line that steps up once every few weeks, which would tell us nothing about why gold moved on a Tuesday afternoon. It would be like checking a monthly bank statement to discover what you spent this morning. For the daily analysis we therefore use breakeven inflation, which the bond market updates every second of the trading day. CPI comes into its own in the long-run monthly analysis in Section 8.

3. How we tested it, without the jargon

Our data run from 3 January 2006 to 18 September 2026, the last trading day before our 20 September cut-off. That gives 5,060 trading days, 176 of them in 2026. The start date is set by the Federal Reserve's broad dollar index, which begins in 2006. The period is long enough to include the global financial crisis of 2008, the gold boom of 2011, the crash of 2013, the pandemic and the extraordinary rally of 2022 to 2026. The gold price is the New York close of COMEX gold futures; the other series come from the Federal Reserve Bank of St. Louis. The method had five steps.

Step 1: compare daily changes, not prices

Gold has risen from about $530 to over $4,400 since 2006. So have house prices, CPI and the number of people who own a smartphone. Anything that trends upwards will appear related to anything else that trends upwards, whether or not there is any connection. Statisticians call this a spurious regression. To avoid the trap, we test whether each series "wanders" (it does) and then work with daily percentage changes, which do not.

Step 2: run a regression

A regression is a statistical machine for separating ingredients. Feed it gold's daily moves alongside the moves in the dollar, real rates, inflation expectations, fear and oil, and it estimates how much of gold's movement belongs to each, holding the others fixed. The number attached to the dollar is called its beta: the percentage change in gold associated with a 1% change in the dollar.

Step 3: allow for calm and storms

Markets have weather. Calm days follow calm days, and storms come in clusters. An ordinary regression assumes the weather is always the same, which makes it overconfident during storms. A GARCH model fixes this by letting the expected size of gold's daily moves rise and fall over time. We ran every important test both ways.

Step 4: look for change

We measured the dollar effect in rolling one-year windows to see how it evolved, and we ran a structural break test that searches the data for the dates on which the relationship changed, without being told where to look.

Step 5: test the long run

Finally, we switched to monthly data, added CPI, and asked a different question: not "does gold move when the dollar moves?" but "do gold, the dollar, real rates and prices stay bound together over the years?" This uses a technique called ARDL and a concept called cointegration, both explained in Section 8.

4. Finding one: the see-saw is real

Across twenty years, the dollar beta is −1.35, with a t-statistic of −15.7. In plain terms: on a day when the dollar rises by 1%, gold falls by about 1.35%, and the probability that this pattern is a statistical accident is vanishingly small. The GARCH version, which allows for market storms, gives almost the same answer: −1.33.

Figure 1 shows the see-saw in its simplest form. On days when the dollar rose by more than 0.3%, gold fell by an average of 0.69%, and it fell on 73% of those days. On days when the dollar fell by more than 0.3%, gold rose by an average of 0.78%, and it rose on 80% of them. Across all days, gold moved in the opposite direction to the dollar two times out of three.

A word of humility is in order. The model's R² is 0.19, so the dollar and the four other suspects together explain about a fifth of gold's daily movement. The dollar alone explains 16%. The other four-fifths is everything else: news, speculative flows, physical buying in Asia, central-bank purchases and the general mood of the market. The dollar is the single most important daily influence we can measure, but it is very far from the whole story.

The other suspects behave broadly as theory predicts. A 25 basis point rise in real interest rates is associated with a fall in gold of a little under 1%. Fear and oil both push gold up. The one surprise is inflation expectations, which carry a small negative sign. The reason is that, with real rates held constant, a rise in breakeven inflation is mechanically a rise in the headline interest rate, and gold dislikes higher interest rates of any kind. The effect is small and disappears after 2022.

Crucially, the dollar effect is stable. Split the twenty years into periods and the beta barely changes, until 2026.

5. Finding two: 2026 was louder, not different

At first sight, 2026 is extraordinary. The dollar beta jumps to −3.83, nearly three times its long-run average, and a formal test confirms that the 2026 figure is statistically different from 2022–2025. A casual reader might conclude that the dollar has seized control of gold. That conclusion would be wrong, and the reason is instructive.

A beta measures the size of gold's moves relative to the dollar's. It can rise for two quite different reasons: because gold follows the dollar more faithfully, or because gold simply moves more violently in general. In 2026, it was mostly the second. Gold's annualised volatility, a measure of how wildly a price swings, reached about 31%, the highest in our sample and edging past even 2008. Its average daily move was 1.4%, against 0.8% over the previous twenty years. Gold moved by more than 3% in a single day on 15 occasions in 2026, compared with an average of about six a year before. On 30 January, the day after gold's record close of about $5,318, COMEX gold fell almost 11% in a single session.

Meanwhile, the dollar was, if anything, calmer than usual. The correlation between the two, which measures how reliably they move in opposite directions regardless of size, was −0.54 in 2026. That is a little tighter than the twenty-year average of −0.40, but well within the range seen in earlier years: 2007 and 2018 were both tighter still. When we break the 2026 increase in beta into its two parts, roughly 70% comes from gold's higher volatility and only 30% from a tighter link. The song has not changed. Someone has turned up the volume.

Two caveats apply. First, 176 trading days is a short sample, and the 2026 estimate is far less precise than the twenty-year figure. Second, when we use two-day rather than one-day changes, the 2026 beta falls to about −2.8, which suggests that some of gold's same-day reactions in 2026 were overreactions that partly reversed the following day.

6. Finding three: no grudges and no delays

Two further questions are worth asking. Does gold respond more strongly to a rising dollar than to a falling one? And does gold react with a delay, giving an alert investor time to act?

The answer to both is no. Splitting dollar moves into rises and falls produces almost identical betas over the full sample, −1.35 and −1.36, and the difference is statistically meaningless. In 2026 the gap is wider, −4.3 against −3.4, but it is still well within the range of chance. Gold holds no grudges: it responds to the dollar symmetrically. Nor does yesterday's dollar move help explain today's gold price. The reaction is complete within the same trading day.

This matters because it tells us that the market is doing its job. Information about the dollar is absorbed into the gold price within hours. There is no simple trading rule here of the "buy gold the morning after the dollar falls" variety. Anyone who claims otherwise should be asked for their data.

7. Finding four: when the see-saw changed

A popular story holds that the gold–dollar relationship "broke" in 2022, when central banks began buying gold on an unprecedented scale. We tested that story directly. A structural break test, specifically the Bai–Perron method, searches every possible combination of dates for the points at which the relationship shifted, and it is not told where to look.

It found two breaks, and neither is in 2022. The first falls in late summer 2008, on the eve of the Lehman Brothers collapse; the second in July 2011, as gold raced towards what was then a record high. Before August 2008, the dollar beta was a steep −2.7. Through the crisis years it weakened to about −1.0. Since mid-2011 it has settled at about −1.3, and it held there through 2022, 2023, 2024 and 2025. The test works on a monthly grid, so the dates are accurate to within a few weeks.

So at the daily level, the see-saw did not break in 2022. Day to day, the dollar went on pushing gold around exactly as before. Something did break in 2022, but to see it we need a different lens.

8. Finding five: the anchor that snapped

Everything so far concerns daily movements. A quite different question is whether gold, the dollar, real interest rates and the price level are tied together over the long run, so that if gold drifts too far from where they say it should be, it is eventually pulled back.

We built a monthly model covering 248 months from January 2006 to September 2026, linking the gold price to the dollar, real interest rates and CPI. Over the full period, the lead appears to have been cut. The formal test is borderline, but the error-correction term is close to zero (about 1% a month, implying it would take around five years to close even half of any gap) and is statistically indistinguishable from zero. A second standard test, the Engle–Granger method, also fails to find a stable long-run relationship.

The story becomes vivid when we estimate the long-run rulebook using only the years before 2022. That rulebook says that a 1% stronger dollar means about 1.2% cheaper gold; a one percentage point rise in real interest rates means gold about 18% cheaper; and a 1% higher price level means gold about 2.3% dearer. For sixteen years it worked remarkably well. In 98% of the months from 2006 to 2021, gold stayed within 20% of the rulebook value, and on average it was within 8%.

Then it stopped working. Between 2021 and 2026, average real interest rates rose from about −0.9% to about +2.1%, and the broad dollar strengthened by about 5%. By the old rulebook, that is a near-perfect recipe for falling gold. Instead, gold climbed from about $1,790 in December 2021 to about $4,420 in September 2026. The rulebook value for September 2026 is about $1,680. The gap is roughly $2,740 an ounce: gold trades at about 2.6 times what the old relationship says it should.

Our model cannot tell us what filled that gap, because whatever it is does not appear in our data. The most plausible candidate, which Synterra discussed in its October 2026 Investor Note, is official-sector demand. Central banks bought more than 1,000 tonnes of gold in each of 2022, 2023 and 2024, roughly double their 2010–2021 average, after the freezing of Russia's foreign-exchange reserves in 2022 reminded every finance ministry that a dollar reserve is only as safe as America's goodwill. These buyers do not care about the yield on Treasury bonds. They care about who can freeze their savings. A model built on yields and exchange rates is blind to them by construction.

One caution is essential. The gap is not a forecast. It does not mean gold "should" fall to $1,680, and the statistical tests are clear that the old rulebook no longer binds. It is at least as likely that the rulebook is obsolete as that gold is overvalued. What the gap does measure is how much of today's gold price rests on forces other than the dollar, interest rates and inflation, and therefore how exposed the price would be if those forces faded.

9. Why this matters

For students of economics

There are two lessons here that extend well beyond gold. The first is that a relationship in changes and a relationship in levels are different questions. The dollar and gold can move against each other faithfully every day while drifting apart over years, much as two passengers on the same train can jostle each other at every bend while the train carries both somewhere neither expected. The second is that every statistical model is a rulebook written from the past. When the identity of the marginal buyer changes, the rulebook can fail without warning, even while each individual day still looks normal.

For investors

The dollar remains a useful guide to gold's short-term risk. An investor hedging a gold position over days or weeks can rely on the see-saw. As a guide to value, however, the dollar is now close to useless. The level of gold depends on whether sovereign buyers keep buying. If they slow, the forces our model can see would reassert themselves, and those forces currently point a long way down.

For Synterra

• Price decks, not spot. Synterra values tailings and brownfield projects on conservative gold price assumptions rather than the spot price. This analysis reinforces that discipline. The rulebook value is a useful extreme downside case for stress-testing project economics and debt sizing.

• Confidence in in-situ gold. For SYNTGOLD and any instrument offering economic rights to unmined gold, the question investors will ask is what supports the price. If the answer is sovereign demand rather than rates, then independent verification, proof of reserves and conservative reserve haircuts matter more, not less.

• Signposts. The variables worth watching are central-bank purchase data, ETF flows and real interest rates. A slowdown in official buying combined with high real rates would be the scenario in which the gap begins to close.

10. Health warnings

Good analysis states its own limits. Five apply here.

1. Correlation is not causation. The dollar and gold often react to the same piece of news, such as a strong US jobs report. Our betas measure how they move together, not proof that one drives the other.

2. 2026 is nine months. Conclusions about 2026 rest on 176 trading days and are far less certain than the twenty-year results.

3. Timing. The Federal Reserve's dollar index is measured at noon in New York, while gold and bond prices close later in the day. We checked with two-day changes, which confirm the main results.

4. The missing suspect. Central-bank and ETF demand is not available daily, so it sits outside the model. That is precisely why the long-run gap appears.

5. Rulebooks change. The relationship has shifted before, in 2008 and 2011, and will shift again.

11. Conclusion

The US dollar has not lost its influence over gold. Day by day, it remains the single most powerful force we can measure, and in 2026 its effect, amplified by extraordinary volatility, was the strongest in two decades. But influence over the next move is not the same as authority over the price. Since 2022, the dollar, interest rates and inflation have set the rhythm of gold while something else has set the level. For those who value gold, and for those of us who dig it out of old mine dumps for a living, the most important number in this study is not the beta of −1.35. It is the gap of $2,740.

Jargon buster

Methodology and sources

Daily data: COMEX gold front-month futures, New York close (Yahoo Finance, GC=F); Federal Reserve broad trade-weighted dollar index (DTWEXBGS); 10-year TIPS yield (DFII10); 10-year breakeven (T10YIE); VIX (VIXCLS); Brent crude (DCOILBRENTEU); all from FRED, Federal Reserve Bank of St. Louis. Monthly: CPI-U, seasonally adjusted (CPIAUCSL). Daily model: OLS on log changes with Newey–West standard errors, and GJR-GARCH(1,1) with Student-t errors. Robustness: DXY, 5y5y forward inflation, two-day returns, lags, asymmetric dollar terms, 2010–2026 sample. Breaks: Bai–Perron global search, minimum segment 10%, BIC selection. Long run: ARDL bounds test (Pesaran, Shin and Smith, case III), NARDL, Engle–Granger and dynamic OLS. Central-bank purchase figures are from the Synterra Investor Note of 19 October 2026 and were not re-estimated here. The full results workbook and code are available on request.

We tested both views against forty-five years of data. The answer is more interesting than either camp would like. Debt matters, but only one kind of debt, and only up to a point. Beyond that point lies something the debt statistics cannot see, and it currently accounts for a large part of what you pay for an ounce of gold. As in our previous article, every technical term is explained along the way, and longer explanations sit in Plain English boxes.

1. A price without a cash flow

Most assets can be valued by asking what they will pay you. A share is worth the present value of the dividends a company will distribute. A bond is worth the present value of its interest payments and the return of the sum lent. Property is worth the rent it will earn. In each case the answer comes from discounting a stream of future cash, which is the bread and butter of financial analysis.

Gold pays nothing. It earns no interest, issues no dividends and collects no rent. Strictly, it costs money to hold, because it must be stored and insured. And yet it has served as money for several thousand years and still sits in the vaults of almost every central bank on earth. Its value cannot come from cash flows, so it must come from something else: from what gold is not.

A government bond is a promise. Its value depends on the government's ability and willingness to pay you back, in money whose value the government also controls. A bank deposit is a promise by the bank. Even a banknote is, in a sense, a liability of the central bank that issued it. Gold is the only widely held financial asset that is no one else's promise. When an ounce of gold sits in your vault, there is no one who can default on it, inflate it away or freeze it.

That is why debt enters the valuation question. If gold's value comes from being the alternative to promises, then the quantity and quality of promises outstanding ought to matter. The more the world owes, and the less credible the promises look, the more valuable the one asset that promises nothing should become. That is the hypothesis we set out to test.

2. The thesis: interest rates and the dollar decide

The conventional view, taught in finance courses and repeated by most bank research desks, is that gold is priced off two forces. The first is the real interest rate: the return you could earn on safe government bonds after inflation. Because gold pays nothing, a higher real return elsewhere makes it less attractive. The second is the US dollar: gold is priced in dollars, so a stronger dollar makes it dearer for everyone else and pushes its dollar price down. Our previous article, The See-Saw and the Snapped Anchor, showed how reliably the dollar still moves gold from one day to the next.

On this view, global debt is noise. Debt may rise or fall, but unless it changes interest rates or the dollar, it should leave gold untouched. Formally, this is our null hypothesis: once real rates, the dollar and inflation are allowed for, debt adds no further explanation of gold's price.

The competing view, our alternative hypothesis, is that rising government debt, especially when it looks hard to repay without inflation, raises the demand for an asset with no issuer. If that is right, gold should command a higher value than interest rates and the dollar alone would justify, and the extra should grow with the debt.

3. How much gold stands behind each dollar of debt?

Before running any regressions, it is worth asking a simple accounting question. If we add up the market value of all the gold ever mined and divide it by the world's debts, what share of those debts could gold notionally cover? We call this the coverage ratio.

The World Gold Council estimates that 222,600 tonnes of gold had been mined by mid-2026, about 7.2 billion troy ounces, of which roughly 39,000 tonnes sit in central bank vaults. At $4,010 an ounce, the whole stock is worth about $28.7 trillion. Global non-financial debt, meaning what governments, companies and households owe excluding banks' borrowing from each other, was about $251 trillion in 2024 according to the IMF, and we estimate roughly $276 trillion by late 2026. Gold therefore covers about 10.4% of the world's debts.

On its own, that number means little. Its power comes from history. Using IMF data for countries producing about 90% of world output, we rebuilt the coverage ratio for every year since 1980. Figure 1 shows the result.

The pattern is striking. In 1980, at the height of the last great crisis of confidence in paper money, when US inflation ran above 13% and gold briefly touched $850, the world's gold was worth 16.4% of its debts. Over the following two decades, as central banks restored their credibility, that ratio fell to a low of just 2.0% in 2001. Between 1980 and 2024 the median was 4.3%. Today it is 10.4%, about two and a half times its normal level and higher than at any time since 1981.

We can turn the question round and ask what gold would be worth at different coverage ratios, given today's debt and today's stock of gold:

This is the honest version of the "gold should be worth $10,000" arguments that circulate online. Those arguments pick a coverage ratio, usually without saying why, and divide. The table shows why that is unsatisfactory: the answer depends entirely on the ratio you choose, and history has offered ratios ranging from 2% to 16%. The useful point is not any single number. It is that gold is currently priced at a coverage ratio seen only once before in modern history.

What moves the ratio? A simple regression shows that two factors, the dollar and the level of government debt, explain about 82% of its year-to-year variation. When government debt is high and the dollar weak, gold covers more of the world's debt. That is our first hint that government debt matters. To test it properly, we need to go further.

4. How we tested it

We assembled two datasets. The first is global and annual: the IMF's Global Debt Database, which records government, household and company debt country by country (our sample covers about 90% of world output), combined with world GDP, the gold price, US real interest rates and the dollar, from 1980 to 2024. That gives us 45 years. The second is American and quarterly: US federal debt, the government's interest bill, the term premium on Treasury bonds and the same market variables, from 1973 to mid-2026, giving us 214 quarters. Using both lets us check that a result found in one place also appears in the other.

Following the approach of our first article, we asked two different questions. The first is about changes: when debt rises in a given year, does gold rise by more than interest rates and the dollar would predict? The second is about levels: are gold and debt bound together over the long run, so that if gold drifts too far from what debt justifies, it gets pulled back? The first question tells us whether debt matters. The second tells us whether there is such a thing as a stable "true value".

5. The antithesis: debt matters, but only one kind

The first result is clear. Debt does help explain gold, even after interest rates, the dollar and inflation have been accounted for. But which debt matters turns out to be the most revealing finding of the study.

Over 1980–2024, a one-percentage-point rise in world government debt as a share of GDP has been accompanied by a rise in gold of about 1.3–1.4% in the same year, with t-statistics between 2.4 and 3.3. For US federal debt the figure is 1.1%, with a t-statistic of 6.7. Adding government debt to the conventional model raises the share of gold's annual movements it explains from 53% to 61%. By contrast, private debt, which is what households and companies owe, has no measurable effect at all. Its estimated impact is slightly negative and statistically indistinguishable from zero.

This makes economic sense. A mortgage or a corporate loan used to build a factory is a private contract, usually backed by an asset, and it does not threaten the value of money. A government debt is different. Governments that borrow heavily have the means, and sometimes the motive, to reduce what they owe by allowing inflation to erode it. Investors who fear that outcome reach for the one asset no government can print. The data say they do so in measurable amounts.

A sceptic would raise an obvious objection. Government debt jumps in crises, as it did in 2008 and 2020, and gold also rises in crises because frightened investors buy it. Perhaps both are simply reacting to fear. We tested this in three ways. Removing the worst crisis quarters leaves the effect intact, at about 0.7% of gold per point of debt. Debt growth in one quarter helps predict gold in the next, which a purely simultaneous panic would not produce. And controlling for corporate credit stress, a direct measure of financial fear, does not change the result. The debt effect is not simply crisis behaviour in disguise.

6. The snowball test

If the credibility argument is right, debt should matter most when it is hardest to repay. Economists have a precise way of describing that condition, and it is worth understanding because it lies at the heart of every modern debate about public finances.

We tested whether debt has a bigger effect on gold during the periods, about half of all quarters since 1973, when the US government paid more in interest than its economy grew. It does not. The extra effect is essentially zero, with a t-statistic of 0.1. Gold responds to the quantity of government debt, but not, it seems, to whether the arithmetic of that debt is becoming dangerous.

Worse for the credibility thesis, rising interest costs have recently pushed gold the other way. Between 2022 and 2026, as the US government's interest bill climbed to about 4% of GDP, quarters in which that burden rose fastest were quarters in which gold did worse. The explanation is the opportunity-cost channel from Section 2: higher interest payments mean higher yields on Treasuries, and higher yields make a non-yielding asset less attractive. The two forces the credibility thesis describes, fear of default and the lure of high yields, operate at the same time and partly cancel out.

There is one piece of supporting evidence, and it is suggestive rather than conclusive. The term premium is the extra return investors demand for lending to the US government for ten years rather than rolling over short-term loans. It is a direct, if imperfect, measure of how nervous bond investors are about long-term US finances. Before 2022, a rising term premium went with a slightly weaker gold price. Since 2022, it has gone with a stronger one: about 11% more gold for each one-point rise. The change is in the direction the credibility thesis predicts, but with only eighteen quarters of post-2022 data it is not quite statistically secure, with a p-value of 0.09.

7. The synthesis: a floor and a premium

We now have the ingredients to answer the question posed in the title. If debt matters, how much of today's gold price can it explain?

The first step is to ask whether gold, debt and the other usual suspects are bound together over the long run, the condition economists call cointegration and which we described in our first article using the image of a man walking a dog on a lead. Across every combination we tried, with different start dates, US and global data, and with and without debt, the formal tests fail to find a lasting lead. Debt improves the fit of the models considerably, but gold and its fundamentals are perfectly capable of drifting apart for a decade or more. There is no fixed "true value" towards which gold is inexorably pulled.

That does not make fundamental value meaningless. It makes it a range. Figure 4 shows gold's estimated fundamental value from a model that includes world GDP, real interest rates, the dollar and world government debt, together with the band within which the actual price has historically fallen 95% of the time.

For most of the past forty-five years, gold strayed above and below its central estimate but rarely left the band, and it was still inside it in 2024. Then, in 2025 and 2026, it broke clean through the top. The model puts gold's fundamental value for 2026 at about $2,430 an ounce, with a 95% range of roughly $1,740 to $3,400. Gold is trading at $4,010.

Now comes the decisive comparison. Run the same model without debt and the 2026 fundamental value is about $2,180. Adding the whole of the world's post-2020 government borrowing lifts it by only about $250, or roughly 12%. The debt effect is real, but it is small next to the gap it is being asked to explain. Government debt has raised gold's floor, but it cannot carry the price to where it now trades.

8. So what is gold actually worth?

Three independent methods, built on different data and different logic, agree that the measurable forces behind gold justify a price of roughly $1,700 to $2,500 an ounce. Interest rates, the dollar, economic growth and government debt together account for something like half of today's market price. The remainder, somewhere between 65% and 140% on top of fundamental value depending on the yardstick, is what we call the credibility premium.

It would be a mistake to call this premium irrational, and an equal mistake to call it permanent. It reflects something real that our statistics do not capture: a change in who wants to hold gold and why. As our previous article noted, central banks bought more than 1,000 tonnes a year in 2022, 2023 and 2024, roughly double their earlier average, after Russia's foreign-exchange reserves were frozen. Those buyers were not responding to the level of global debt or to interest rates. They were responding to the discovery that a reserve held as another country's promise can be switched off by that country. That is a judgement about the credibility of money itself, and it is exactly the kind of judgement the debt statistics cannot see.

History suggests both the power and the fragility of such premiums. In 1980, gold was priced at a coverage ratio of 16.4%. When Paul Volcker's Federal Reserve raised interest rates far enough to convince the world that the dollar would hold its value, the premium did not collapse overnight. It eroded over twenty years, with gold falling in real terms for most of that period. A credibility premium is built on a belief, and beliefs change slowly, but they do change.

The honest answer to the question "what is gold worth?" is therefore conditional. If the world's trust in sovereign money remains as damaged as it appears today, gold at $4,000 is not obviously expensive, and a further loss of confidence could carry it towards the coverage ratios of 1980. If that trust is gradually restored, the evidence says gold has a long way to fall before it meets its fundamentals, and government debt, on its own, will not stop it.

9. Why this matters

For students of economics

This study illustrates a general truth about financial markets: prices reflect both measurable fundamentals and unmeasurable beliefs, and the balance between them shifts over time. It also shows the value of separating a hypothesis into its parts. "Debt drives gold" turned out to be partly true (government debt), partly false (private debt), and partly untestable with current data (sovereign trust). A careful researcher reports all three.

For investors

Anyone buying gold today should understand that roughly half of the price rests on a premium for monetary distrust. That premium may well persist or grow, but it is a different kind of risk from interest-rate or currency risk. Investors who hold gold as a hedge against fiscal excess are, on this evidence, buying a modest structural benefit at a substantial premium.

For Synterra

• Conservative price decks. Synterra values tailings projects on gold price assumptions well below spot. These results support that discipline. A price deck anchored near the upper end of the fundamental range, around $2,500 to $3,000, rather than at $4,000, is defensible to lenders and investors.

• Credibility as a product. The premium in today's gold price is, in essence, a payment for an asset that is no one else's promise. For SYNTGOLD and any instrument offering rights to in-situ gold, that is a warning as well as an opportunity: the product must earn the same trust, through independent verification, proof of reserves and conservative haircuts, or it will not share in the premium.

• Signposts. The factors to watch are central bank purchases, the US term premium and real interest rates. A slowdown in official buying alongside high real yields would be the clearest sign that the premium is beginning to erode.

10. Health warnings

1. Correlation is not causation. Government debt and gold may both respond to deeper forces, such as wars or pandemics, that our controls only partly capture.

2. Short samples. The global data cover 45 years, and the post-2022 findings rest on 18 quarters. The term-premium result in particular should be treated as suggestive.

3. Estimated inputs. The IMF does not publish a single global debt total for every year; we rebuilt it from country data, and our 2024 figure of $255 trillion is close to the IMF's published $251 trillion. Gold stocks before 2026 are back-calculated from approximate mine production and may be up to about 3% out.

4. No stable equilibrium. Because the long-run tests fail, the fundamental value is a statistical range, not a price the market must return to.

5. The unmeasured factor. We attribute the residual premium to monetary credibility and official buying. That is an interpretation consistent with the evidence, not a result the models prove.

11. Conclusion

The claim that gold rises because the world is drowning in debt is half right. Government debt, but not private debt, does lift gold's value, by about one per cent for every point of GDP borrowed. That effect has raised gold's floor. But it cannot explain why the world's gold is now worth a tenth of the world's debts when, for most of modern history, it was worth a twentieth. The difference is not a quantity of debt. It is a quality of trust. Gold today is priced not as a hedge against how much governments owe, but against the possibility that their promises may not be kept. That is the price of monetary credibility, and at somewhere between $1,600 and $2,300 an ounce, it is now one of the largest components of every bar's price.

Jargon buster

Methodology and sources

Global annual data, 1980–2024: IMF Global Debt Database (general government debt; private debt, loans and debt securities), aggregated by Synterra using IMF World Economic Outlook GDP in US dollars; IMF WEO world government debt projections for 2025–2026. US quarterly data, 1973Q1–2026Q2: FRED series GFDEGDQ188S, GFDEBTN, GDP, A091RC1Q027SBEA, THREEFYTP10 (Kim–Wright term premium), REAINTRATREARAT10Y (Cleveland Fed real rate, from 1982; earlier years 10-year yield less CPI inflation), CPIAUCSL, TWEXBMTH spliced with TWEXBGSMTH, BAA10YM. Gold: World Bank monthly prices via DataHub; COMEX close of 23 October 2026 for the current price. Above-ground stock: World Gold Council, 222,600 tonnes at end-Q2 2026, back-calculated using approximate annual mine production. Models: OLS with Newey–West standard errors on annual and quarterly changes; dynamic OLS and ARDL bounds tests (Pesaran, Shin and Smith) for long-run relationships; Engle–Granger cointegration tests; prediction intervals from the 1980–2024 levels model. Central-bank purchase figures are from Synterra's Investor Note of 19 October 2026 and were not re-estimated here. The full workbook and code are available on request.

The assumption is plausible. It may even be partly true. But it has rarely been tested against the obvious alternative: that ETF investors are not leading the market but following it, buying after prices have risen and selling after they have fallen. This article tests both stories against twenty-two years of daily data. As with the previous articles in this series, every technical term is explained, and longer explanations sit in Plain English boxes.

1. What a gold ETF actually is

An exchange-traded fund, or ETF, is a fund whose shares trade on a stock exchange like any company's. A physically backed gold ETF uses investors' money to buy real gold bars, which it stores in a vault, usually in London. Each share represents a fixed amount of that gold. The largest, SPDR Gold Shares, known by its ticker GLD, was launched in New York in November 2004. Before then, an ordinary investor who wanted gold had to buy coins or bars, pay for storage and worry about authenticity. GLD let them buy gold with a click, as easily as buying a share in Apple.

That convenience transformed the market. GLD held 58 tonnes on its first day. Eight years later, at its peak in December 2012, GLD held over 1,350 tonnes, more than the official gold reserves of all but a handful of countries. Worldwide, gold ETFs now hold around 4,000 tonnes.

2. The thesis: flows drive the price

The conventional story runs as follows. When investors buy ETF shares, authorised participants must buy physical gold to create new shares. That is genuine new demand for a scarce metal, and it should push the price up. When investors sell, gold flows out of the vaults and back onto the market, pushing the price down. On this view, ETF flows are an independent force that moves the gold price, and watching them gives investors an early signal of where the price is heading.

There is a powerful and intuitive case for this story. ETFs made gold accessible to pension funds, wealth managers and private investors who had never owned it. It is widely argued that this new pool of demand helped to drive gold's rise from $440 in 2004 to over $1,900 in 2011. And the sight of hundreds of tonnes leaving the vaults during gold's collapse in 2013 is etched into the memory of every gold investor.

3. The antithesis: the price drives the flows

The opposing story is rooted in behavioural finance. Many investors are momentum traders, whether they admit it or not. They notice that gold has risen, read headlines about record prices and buy. When gold falls, they lose confidence and sell. If that is how ETF investors behave, then flows will look as though they cause price movements, because the two move together, when in fact the causation runs the other way.

A third possibility is that both are driven by something else entirely. A fall in interest rates, a weaker dollar or a geopolitical shock could prompt investors to buy ETFs and simultaneously push up the gold price through other channels. In that case flows and prices would move together without either causing the other.

Separating these stories is a classic problem in economics, known as endogeneity: when two variables influence each other, a simple correlation cannot tell you which way the causation runs. Our approach is to exploit timing. If flows drive prices, today's flows should help predict tomorrow's price. If prices drive flows, today's price should help predict tomorrow's flows.

4. How we tested it

We used the funds' own daily records of the gold held in SPDR Gold Shares (GLD) and SPDR Gold MiniShares (GLDM), from GLD's launch in November 2004 to 22 October 2026: 5,464 trading days. Together these two funds hold about 1,280 tonnes, around a third of the world's ETF gold. We measured flows as the daily percentage change in tonnes held, so that a flow of 1% means the funds' holdings grew by 1% that day. We matched these to the New York gold futures price and, as controls, to the US dollar, real interest rates and the VIX index of stock market fear, because each could move both flows and prices at the same time.

Figure 1 shows the raw material. The two lines rise together until 2012. After that they go their separate ways.

5. Finding one: investors chase the price

The clearest result comes at the monthly horizon, the one at which investment commentary usually discusses flows. A 1% rise in gold in one month is followed by an increase in ETF holdings of about 0.13% the following month, with a t-statistic of 2.5. Investors buy after the price has risen. Running the test the other way, last month's flows tell us essentially nothing about this month's price: the statistical evidence is so weak (p = 0.30) that we cannot rule out pure chance.

At the weekly horizon the result is starker still. Flows have no detectable power to predict next week's gold price (p = 0.77). The price has some power to predict next week's flows, though more weakly (p = 0.09). Whichever way you slice the data, the arrow of prediction points from the price to the flows, not the reverse.

6. Finding two: a ripple, not a current

Look at daily data and a more nuanced picture emerges. Here the influence runs in both directions. Gold rising today predicts more ETF buying tomorrow, as we would expect. But ETF buying today also predicts a small rise in gold tomorrow: about 0.17% for every 1% increase in holdings. This effect is statistically robust. It appears in every sub-period we examined, from 2004–2012, 2013–2019 and 2020–2026, and it applies to both inflows and outflows.

So the conventional story is not entirely wrong. When authorised participants buy physical gold to create new shares, they do nudge the price. But the nudge is small and brief. Adding past ETF flows to a model of daily gold movements improves its explanatory power by less than one percentage point, and the effect has disappeared entirely by the time we look at weekly data. It is a ripple on the surface, not a current that carries the price.

One technical caution applies. GLD records new gold in its holdings with a delay of up to a day after investors place their orders. Some of the apparent "price leads flows" effect at the daily horizon may therefore reflect reporting timing rather than investor behaviour. This caveat does not affect the weekly and monthly results, which are the ones that matter for the argument.

7. Finding three: the limits of the data

There remains the strongest relationship of all: gold and ETF holdings move together within the same month. A 1% rise in gold goes with roughly 0.4% more ETF gold in the same month, and the correlation between the two is 0.35. Who is causing whom within those weeks?

Here the honest answer is that our data cannot say. To decompose same-month movements, an analyst must make an assumption about which variable reacts first. Assume flows come first and they explain about 13% of gold's monthly price swings. Assume the price comes first and flows explain less than 0.5%. Neither assumption can be tested with this data. Figure 2 makes the point.

This is a valuable lesson for anyone reading financial research. Many confident claims about what "drives" a market rest on an assumption buried in the method. When the result reverses under the opposite, equally plausible assumption, the honest conclusion is not that one side is right but that the question cannot be answered this way. What we can say is that even under the assumption most favourable to the flows-drive-prices story, ETFs explain only around one-eighth of gold's monthly movements.

8. Finding four: the rally that ETFs missed

Statistics can feel abstract. Four episodes make the argument concrete.

2013: the stampede out. Gold fell 28% and the funds lost 41% of their gold, 553 tonnes. This is the episode most often cited as proof that ETFs move the price, and there is something to it: outflows began in the first quarter of 2013, before gold's dramatic crash in April. But holdings then kept falling for almost three years, to 642 tonnes by the end of 2015, long after the price had stabilised. For most of that period, investors were retreating from a market that had already turned, not pushing it down.

2023–2024: the rally without them. Gold rose 45%, from about $1,830 to $2,640 an ounce, while the gold held by the SPDR funds actually fell by 2%. If ETFs drove the gold price, this rally should not have happened. It did happen because the marginal buyer had changed. As our first two articles showed, central banks were buying more than 1,000 tonnes a year, roughly double their historical average. Western investors were sellers into the rally.

2025: the latecomers arrive. Only after gold had broken through record after record did ETF investors return in force. Holdings rose 28% as the price climbed 64%. Momentum had done its work: Western investors, net sellers in 2023 and 2024, were buying back at far higher prices.

September 2026: buying the dip. Gold fell 7% during the month, yet the SPDR funds' holdings rose by about 19 tonnes. Here, ETF investors bought into weakness, and their buying did not stop the fall. Even substantial inflows proved no match for the forces pushing gold down: during the month the 10-year real interest rate rose by half a percentage point and the dollar strengthened by nearly 2%.

9. The synthesis: a mirror, not an engine

Put the evidence together and a coherent picture emerges. Gold ETFs are not the engine of the gold price. They are better understood as a mirror of Western investor sentiment. They reflect what has already happened to the price, they amplify it slightly at the margin over a day or two, and they occasionally lean against it. But the forces that set the level of gold, namely real interest rates, the dollar, central bank reserve decisions and confidence in sovereign money, operate largely independently of them.

That does not make ETF flows useless. They are an excellent real-time measure of how Western investors feel about gold, which is valuable information in its own right. And because they tend to buy after rises and sell after falls, ETF investors can deepen both rallies and slumps, much as a crowd at a football match makes a winning team feel invincible and a losing one feel doomed. But a crowd does not score goals.

There is also an irony worth noting. The great appeal of a gold ETF is that it makes gold as easy to trade as a share. That ease is precisely what allows investors to behave like share traders, chasing performance and selling in fear. The financial innovation that democratised gold also turned part of it into a momentum asset.

10. Why this matters

For students of economics

This study is a case study in the gap between correlation and causation, and in the hidden assumptions behind statistical claims. Timing tests such as Granger causality can tell us a great deal, but within a single period the direction of influence may be impossible to identify without additional information, such as an event that changes one variable but not the other.

For investors

Monthly ETF flow reports make good headlines but poor forecasts. On this evidence, they tell you where Western sentiment has been, not where the price is going. An investor who bought gold in response to strong inflows would, on average, have been buying after the move rather than before it.

For Synterra

• Tokenised gold will face the same behaviour. SYNTGOLD and any other digital gold instrument will make gold even easier to trade than an ETF. On this evidence, demand is likely to be pro-cyclical: strong when gold is rising, weak when it is falling. Issuance plans, liquidity provision and market-making must be designed for that pattern, not for steady demand.

• Distribution decides success. Our long-held view that distribution, not technology, is the hard part of tokenisation is reinforced. GLD succeeded because it plugged gold into existing brokerage accounts and institutional mandates. A token must earn the same access and the same trust.

• Do not mistake flows for value. When valuing gold assets, Synterra will continue to rely on fundamental drivers and conservative price decks rather than on investment flow data, which tends to follow the price rather than anticipate it.

11. Health warnings

1. Two funds, one market. Our data cover GLD and GLDM, about a third of global ETF gold. Funds listed in Europe and Asia, and other US funds such as the iShares Gold Trust, may behave differently.

2. Reporting timing. Holdings are recorded with up to a one-day lag, which affects the interpretation of the daily results but not the weekly or monthly ones.

3. Prediction is not causation. Granger tests show which variable carries information about the other's future, not the mechanism behind it.

4. The identification problem. Same-month causation cannot be resolved with this data, as Section 7 explains.

5. Global flow figures not verified. Industry estimates of worldwide ETF inflows in September 2026 have not been independently checked; the US data in Figure 3 support the same point.

12. Conclusion

Gold ETFs were one of the great financial innovations of the century, and they genuinely changed who owns gold. But the evidence does not support the popular belief that their flows drive the gold price. Over weeks and months, the price leads and the flows follow. The daily nudge that ETF buying gives the price is real but tiny, and it fades within a week. The most dramatic proof is the rally of 2023 and 2024, when gold rose by nearly half while Western investors sold. ETFs hold up a mirror to the market's mood. To understand what moves the price of gold, you have to look past the mirror, to interest rates, to the dollar and, above all, to the central banks that are now buying the metal on a scale ETFs never matched.

Jargon buster

Methodology and sources

Daily holdings in tonnes: SPDR Gold Shares (GLD), 19 November 2004 – 22 October 2026, and SPDR Gold MiniShares (GLDM), from 27 June 2018, from the World Gold Trust Services historical archives. Gold: COMEX front-month futures close (Yahoo Finance). Controls: US Dollar Index (ICE, Yahoo Finance), 10-year TIPS yield (DFII10) and VIX (VIXCLS) from FRED. Flows are daily log changes in combined tonnes. Methods: OLS with Newey–West standard errors; predictive regressions with five lags; Granger causality tests in bivariate VARs at daily, weekly (Friday close) and monthly (month-end) frequency, with lag length chosen by AIC; forecast-error variance decompositions at 12 periods under both Cholesky orderings; asymmetry tests splitting inflows and outflows and bull and bear regimes (price above or below its 200-day average). Central bank purchase figures are from Synterra's Investor Note of 19 October 2026. The full workbook and code are available on request.

That arrangement is now under strain in ways that have no close historical precedent. US federal debt stands at more than 120% of GDP. The share of it owned by foreigners has fallen by almost a third since 2013. Sanctions have shown that dollar reserves can be frozen. China is building an alternative financial plumbing and its currency has strengthened sharply. Share prices are near their most expensive in history, credit spreads near their tightest, and geopolitical risk near its highest in a generation. This article asks what role gold plays in that system. As with the earlier articles in this series, we use historical data as a contrast, explain every technical term, and keep longer explanations in Plain English boxes.

1. A financial system unlike any before

Every financial system rests on a safe asset: something that holds its value when everything else is falling, and which everyone accepts as collateral. Under the classical gold standard, that asset was gold itself. Under the Bretton Woods system from 1944 to 1971, it was the US dollar, convertible into gold at $35 an ounce. Since President Nixon ended that convertibility in 1971, it has been the US Treasury bond, backed by nothing more tangible than the taxing power and the credibility of the US government.

That last arrangement has worked remarkably well for half a century. But several features of today's system would have been unrecognisable to the investors of the 1980s or 1990s:

• Record sovereign debt. US federal debt is above 120% of GDP, and interest payments absorb around 4% of national income, as our second article showed.

• A shrinking foreign buyer base. The world's appetite for US debt has peaked. China's holdings of Treasuries are reported to be at their lowest since 2008.

• A weaponised dollar. The freezing of Russia's reserves in 2022 showed that a reserve held as another country's promise can be switched off.

• A rising alternative. The yuan has strengthened from about 7.30 to 6.69 per dollar since the end of 2024, and China is building payment systems that bypass the dollar.

• Extraordinary financialisation. Share valuations are near their highest on record, corporate credit spreads near their lowest, and a large private credit market has grown up largely outside the banking system and its regulators. The IMF has warned of the risks.

• Heightened geopolitical risk. The Caldara–Iacoviello index of geopolitical risk, built by counting newspaper coverage of wars, terrorism and international tensions, has averaged 188 in 2026, almost double its 1985–2019 average of 100.

2. The thesis: Treasuries are the safe haven

Start with the conventional view and test it on history. We examined every trading day from August 2000 to October 2026, more than 6,500 days, and asked how gold and 10-year US Treasuries behaved on the worst days for US shares.

The results support the textbook. On the worst 10%, 5% and 1% of days for shares, Treasuries rose, and the relationship is statistically strong at every threshold. Gold, by contrast, was essentially a coin toss. On the worst 1% of days, when shares fell by 5% on average, gold rose only 52% of the time. Statistically, its behaviour on crash days is indistinguishable from its behaviour on ordinary days. Over the past quarter-century, gold has been an excellent diversifier but not a reliable safe haven in the strict sense.

The pattern becomes clearer when we separate slow bear markets from fast panics.

In slow, recession-driven bear markets, when central banks cut interest rates, gold protected investors handsomely. In fast panics, when investors sold whatever they could to raise cash, gold was sold too, precisely because it was liquid and usually in profit. It then recovered: in the three months after each stock market low, gold rose by between 4% and 13%. The long monthly record since 1973 tells the same story. Gold rose in only five of the twelve worst months for US shares.

If the article stopped here, the conclusion would be simple: Treasuries are the fire extinguisher, gold is the foundation. But the evidence from the past four years suggests the fire extinguisher itself has developed a fault.

3. The safe harbour moved

A Treasury bond protects a portfolio only if it rises when shares fall. That depends on why shares are falling. In a recession, investors expect interest rates to be cut, so bond prices rise while shares fall. But when the threat is inflation or a government's ability to finance itself, interest rates rise, and both shares and bonds fall together.

Figure 1 shows the transformation. From 2001 to 2021 the correlation between US shares and Treasuries was firmly negative, averaging about −0.4: when one fell, the other tended to rise. In 2022 that relationship broke. Since 2023 it has mostly been positive. The share of trading days on which both shares and bonds fell rose from 15% before 2022 to 26% since. For the classic 60/40 portfolio of shares and bonds, the cushion has gone.

Gold has not filled that gap day by day. On the worst days when shares and bonds fell together, gold also slipped, by about 0.4% on average, because gold is itself sensitive to interest rates. But over longer periods the picture is different. In 2022, the year the cushion failed, shares fell 19%, Treasuries 15%, and gold was unchanged. Gold did not rescue investors on the day; it preserved their capital over the year.

4. Geopolitics and the weaponised dollar

The most consequential financial event of the decade may not have been a crash at all. In February 2022, Western governments froze around half of Russia's foreign-exchange reserves. Overnight, every central bank in the world learned that dollar and euro reserves are only as safe as the political goodwill of the countries that issue them. Gold held in one's own vaults carries no such risk.

We tested whether this changed how markets price geopolitical risk, using the daily Caldara–Iacoviello index. The answer is that it did. Between 2000 and 2021, gold showed no response to geopolitical shocks: on the days when the index spiked to its most extreme levels, gold actually fell slightly, by 0.18% on average. Since 2022, gold has risen on those days, by 0.26% on average, and its general sensitivity to rising geopolitical risk has turned positive. At the same time, Treasuries have shown a tendency to fall as geopolitical risk rises, the opposite of what a safe haven should do. These post-2022 results rest on fewer than five years of data and are suggestive rather than conclusive, but the direction is unmistakable.

5. Credit: the quiet risk

If share prices look stretched, the corporate debt market looks positively relaxed. The credit spread is the extra interest that investors demand to lend to companies rather than to the US government. It is a measure of how much risk lenders believe they are taking. The spread on BAA-rated corporate bonds stands at about 1.46 percentage points, against an average of 2.59 since 2000 and close to its tightest level in a quarter of a century. Lenders, in other words, are asking for very little reward for the risk of default.

Much of the borrowing growth of the past decade has also moved into private credit: loans made by investment funds directly to companies, outside the public bond market and the regulated banking system. These loans are rarely traded, so their prices do not move every day. That makes the market look calm, but it also means that losses may only become visible when they are already large.

Here too, gold's behaviour changed after 2022. Before then, gold showed no response when credit spreads widened. Since 2022, gold has risen by about 0.4% for every 0.1 percentage point of widening, a statistically strong result, and on the most stressful credit days it rose by almost 1% on average. Treasuries also rose on those days, so this is not yet a story of gold replacing them. But it is a sign that investors now treat gold as protection against financial stress in a way they did not before.

6. Dollars, yuan and the fading petrodollar

For fifty years, the dollar's dominance rested on a virtuous circle. America ran trade deficits, the rest of the world earned dollars, and it invested those dollars in Treasuries. Oil exporters, who priced their oil in dollars from the 1970s onwards, were central to this petrodollar system: when oil prices rose, their surpluses flowed back into US government debt.

Figure 3 shows that the circle is weakening. Foreign investors held 34% of US federal debt at the peak in 2013. Today they hold 24%. The dollar still accounts for around 57% of the world's official foreign-exchange reserves, according to recent IMF data, but the marginal buyer of Treasuries is increasingly domestic, and the marginal reserve asset of choice for many central banks has been gold.

Our data also show the petrodollar link fading. Before 2014, a 1% rise in the oil price was associated with a 0.07% rise in gold on the same day, a strong and consistent relationship. Since 2014, that link is no longer statistically detectable. And while an ounce of gold bought about 17 barrels of oil on average between 2000 and 2021, it bought around 50 barrels in 2025, and still buys about 35 today even with Brent crude above $115. Gold has become far more valuable relative to the world's most important commodity, consistent with it being priced increasingly as money rather than as a commodity.

China matters in its own right. Even after allowing for the dollar's overall strength, a 1% strengthening of the yuan against the dollar is associated with a rise in gold of between 0.4% and 0.65%, a result that holds in every period since 2006. Chinese buyers are among the largest in the world, and a stronger yuan increases their purchasing power. It also explains a puzzle of 2026: gold has fallen 7.6% this year in dollar terms but almost 12% in yuan, because the yuan has strengthened so sharply.

7. Inflated markets and the price of gold

A different way to see the current system is to price the stock market in gold rather than in dollars. Figure 4 shows how many ounces of gold it would take to buy one unit of the S&P 500 index since 1900.

The chart describes a series of great swings between financial and real assets. Shares became extremely expensive in gold terms in 1929, 1966 and above all in 2000, when one unit of the index cost 5.4 ounces. Each peak was followed by a long period in which gold outperformed shares, culminating in 1980, when the index cost just 0.16 ounces. Today the ratio stands at about 1.8, down from 2.6 in 2021: gold has already outperformed shares sharply over the past five years, even as share prices have hit records in dollar terms.

Does an expensive stock market predict that gold will outperform? Using Robert Shiller's cyclically adjusted price–earnings ratio, or CAPE, we compared each month since 1971 with the returns that followed over the next ten years.

The pattern is striking: the more expensive shares were, the better gold did relative to them over the following decade. But this is where intellectual honesty demands a heavy caveat. The months with CAPE above 30 come almost entirely from the late 1990s, and the decade that followed was dominated by the dot-com bust, the financial crisis and gold's great bull market. In effect, the result rests on one historical episode. Today's CAPE, reported at around 40, is in territory reached only once before, in 1999–2000. History rhymes; it does not promise.

8. What gold still does not do

It would be easy to conclude from all this that gold has become the perfect safe haven. The evidence does not support that. In the 2020s, gold has risen on only 44% of the worst 5% of days for shares, a weaker record than before 2020, and its day-to-day correlation with shares has turned slightly positive. In March 2020 and April 2025, gold fell alongside shares in the first days of the panic. On the days when shares and bonds fell together, gold usually fell too.

The reason is the very financialisation this article has described. Gold is now held through futures, ETFs and leveraged accounts by investors who use it as a source of cash in a crisis. When margin calls arrive, they sell what they can, not what they want to. Gold's liquidity, which makes it attractive, is also what makes it vulnerable in the first hours of a panic.

9. The synthesis: the last asset standing

Put the evidence together and a coherent, if uncomfortable, picture emerges. The Treasury bond remains the world's main safe asset, and in recessions it still does its job. But the system in which it sits has changed. When the threat is inflation, fiscal stress, sanctions or geopolitical conflict, rather than a simple recession, Treasuries are part of the risk rather than the protection against it. Those are precisely the threats that dominate today's system.

Gold has not become a crash hedge. It has become something more subtle: the residual safe asset of a system in which every other safe asset is someone else's promise. That explains why its behaviour on geopolitical and credit-stress days changed after 2022, why central banks have bought it on a historic scale, and why, as our second article showed, it now carries a large premium over the fundamentals of the old system.

The practical value of that role shows up clearly in portfolios. We simulated a traditional 60% shares and 40% bonds portfolio from 2000 to 2026, rebalanced monthly, and replaced part of it with gold.

Every measure improves as gold is added: higher returns, lower volatility, shallower losses and faster recoveries. Part of the return improvement reflects gold's exceptional performance since 2000 and may not be repeated. The reduction in risk is the more durable lesson. Gold is not a fire extinguisher. It is a foundation.

10. Why this matters

For students of economics

Financial relationships are not laws of nature. The negative correlation between shares and bonds, the safe-haven status of Treasuries and the link between oil and gold were all products of a particular system, built in the 1970s and 1980s. When the underlying system changes, as it appears to be doing now, the relationships change with it. Historical data are indispensable, but only if we ask whether the conditions that produced them still hold.

For investors

A portfolio built on the assumption that bonds will always cushion falling shares is exposed to exactly the risks that dominate today: inflation, fiscal stress and geopolitics. Gold's role is not to protect against tomorrow's crash, but against the slow erosion of confidence in the promises on which other assets rest. Investors who hold it for the first purpose will be disappointed; those who hold it for the second are, on this evidence, well served.

For Synterra

• Market gold as a foundation, not insurance. For SYNTGOLD and other gold-linked instruments, promising crash protection would be misleading and would invite redemptions at the worst moment. The honest proposition is long-horizon protection against monetary and geopolitical risk.

• Non-dollar demand matters. The yuan's independent influence and the shift in central bank reserves point to growing demand from outside the traditional Western investor base. Distribution strategies should reflect that.

• Stress-test for liquidity panics. Because gold can fall sharply in the first days of a crisis, project financing and token liquidity arrangements should be stress-tested against short, sharp falls of 10–15%, even in a long-term bull market.

11. Health warnings

1. A short new regime. The post-2022 findings on geopolitical risk and credit stress rest on fewer than five years of data.

2. One historical episode. The valuation result is driven largely by the period after the dot-com bubble.

3. Approximate returns. Share returns exclude dividends, and bond returns are estimated from changes in yields. The 150-year comparison of 2022 uses similar approximations.

4. Reported figures. China's Treasury holdings, the dollar's reserve share, today's CAPE and the risks in private credit are drawn from press reports of official data and should be checked against the original sources.

5. Correlation is not causation. Gold, Treasuries and shares may all respond to common forces that our measures capture only in part.

12. Conclusion

For half a century, the world has had a simple answer to the question of where to hide: in the debt of the US government. That answer still works in an ordinary recession. But today's financial system is not ordinary. It combines record sovereign debt with a shrinking foreign buyer base, a dollar that can be used as a weapon, a rising Chinese alternative, stock markets near record valuations and credit markets that price almost no risk at all. In such a system, the traditional safe asset is itself exposed. The evidence since 2022 suggests that markets have begun to recognise this, and that gold, the one financial asset that is no one else's liability, has quietly taken on a new role. It will not save you on the first day of the next crash. It may well be the last asset standing at the end of it.

Jargon buster

Methodology and sources

Daily data, 30 August 2000 – 22 October 2026: S&P 500 index and COMEX gold futures (Yahoo Finance); 10-year Treasury yield (FRED DGS10), with daily bond returns approximated as carry less eight times the change in yield; US Dollar Index (ICE); BAA–Treasury credit spread (FRED BAA10Y); USD/CNY (FRED DEXCHUS); Brent crude (FRED DCOILBRENTEU); daily Geopolitical Risk Index (Caldara and Iacoviello). Monthly data: World Bank gold prices from 1973; Shiller S&P 500, dividends and CAPE from 1871 (CAPE to 2023); monthly Geopolitical Risk Index. Quarterly: foreign holdings of US federal debt (FDHBFIN) and total public debt (GFDEBTN). Methods: OLS with Newey–West standard errors; safe-haven regressions with interaction terms for the worst 10%, 5% and 1% of equity days; rolling 250-day correlations; event averages on the most extreme 1% of geopolitical-risk and credit-spread days; ten-year forward returns by CAPE quartile; monthly-rebalanced portfolio simulations. Central bank purchase figures are from Synterra's Investor Note of 19 October 2026. The full workbook and code are available on request.