Over 20 years of daily closes, copper is the dollar's same-day mirror: when the dollar drops, copper jumps, almost perfectly in proportion. But that mirror barely predicts tomorrow. Where the dollar does leave a fingerprint on copper is over the next quarter, through how overbought or oversold it has become.
Ask any trader for a rule about copper and the dollar and you will get the same one: they move opposite. Copper is priced in dollars on world markets, so a stronger dollar makes the metal more expensive for everyone outside the United States, demand softens, and the price slips. A weaker dollar does the reverse. It is one of the most repeated relationships in all of commodities. The question we wanted to answer is not whether the textbook is right (it mostly is), but something more useful and more demanding: can the dollar actually predict copper, or does it only ever describe what is already happening?
Those are very different claims, and most write-ups blur them. We pulled front-month COMEX copper futures (HG) and the U.S. Dollar Index (DXY) over 20 years of daily closes (June 2006 to June 2026, 5,029 aligned trading days), measured the live relationship, then tested every honest way we could think of to turn the dollar into a forward signal for copper. The short version: the same-day mirror is real and strong, the next-day prediction is essentially zero, and a genuine but mild forward tilt only shows up at the one-to-three-month horizon, driven not by the dollar's level but by how stretched it has become.
The premise is mechanical. Copper trades globally but settles in dollars, so the dollar sits inside copper's price by construction. That gives us a strong prior that daily moves should be inversely correlated. But a contemporaneous link, however tight, is not a trading edge. If copper and the dollar move opposite at the same instant, by the time you see the dollar's move the copper move has already happened. So we ran the study in two passes. First, the descriptive pass: how strong is the same-day mirror, and how steady is it across years? Second, the harder predictive pass: does anything about the dollar today tell you something about copper tomorrow, next month, or next quarter? The first pass confirms folklore. The second is where it gets interesting, and where most of the folklore quietly falls apart.
Start with the headline number. The same-day correlation between copper's daily return and DXY's daily return is -0.32. That is a strong, dependable inverse relationship for two separate markets: the negative sign is exactly what the textbook predicts, and the magnitude is large enough that you can feel it day to day. The R-squared is 0.10, meaning the dollar accounts for roughly 10% of copper's daily variance. The beta is -1.17, so on average copper falls about 1.17% for every +1% day in the dollar. Copper does not just lean against the dollar, it moves more than one-for-one against it.
The most striking feature is how symmetric it is. If you bucket every trading day by the size of the dollar's move, copper's average response steps cleanly in the opposite direction, in order, every single time.
| Dollar's day | Copper same-day avg | Copper finished up | Days |
|---|---|---|---|
| DXY down more than 0.5% | +0.92% | 71% | 591 |
| DXY down 0.15% to 0.5% | +0.40% | 62% | 1,122 |
| DXY flat (within 0.15%) | 0.00% | 49% | 1,556 |
| DXY up 0.15% to 0.5% | -0.30% | 41% | 1,146 |
| DXY up more than 0.5% | -0.85% | 29% | 613 |
Read it top to bottom and the monotonicity jumps out: the harder the dollar moves, the harder copper moves the opposite way, with no exceptions and a near-perfect mirror around the flat middle. On the dollar's biggest down days copper averaged +0.92% and closed green 71% of the time; on the dollar's biggest up days copper averaged -0.85% and closed green just 29% of the time. The flat-dollar bucket is, fittingly, a coin flip at 49%. This is about as clean as a real-world relationship between two markets gets.
The mirror is strong and remarkably symmetric, but R-squared 0.10 carries an equally important message: about 90% of copper's daily move is not the dollar. Supply, Chinese demand, inventories, risk appetite, and pure noise do the rest. The dollar is one large input among many, not copper's puppet master.
A single 20-year correlation can hide a lot of drift. To see whether the mirror is structural or just an artifact of a few wild years, we recomputed the same-day correlation year by year, alongside each year's copper and dollar returns for context.
| Year | Same-day corr | Copper return | DXY return |
|---|---|---|---|
| 2006 | -0.21 | -20% | -1% |
| 2007 | -0.12 | +6% | -8% |
| 2008 | -0.47 | -54% | +7% |
| 2009 | -0.35 | +130% | -5% |
| 2010 | -0.38 | +31% | +2% |
| 2011 | -0.44 | -23% | +1% |
| 2012 | -0.55 | +3% | 0% |
| 2013 | -0.26 | -8% | 0% |
| 2014 | +0.01 | -17% | +12% |
| 2015 | -0.16 | -25% | +8% |
| 2016 | -0.13 | +21% | +4% |
| 2017 | -0.08 | +32% | -11% |
| 2018 | -0.34 | -19% | +5% |
| 2019 | -0.13 | +6% | 0% |
| 2020 | -0.30 | +24% | -7% |
| 2021 | -0.38 | +25% | +6% |
| 2022 | -0.45 | -14% | +8% |
| 2023 | -0.41 | +3% | -3% |
| 2024 | -0.38 | +3% | +6% |
| 2025 | -0.16 | +41% | -10% |
| 2026 | -0.42 | +11% | +2% |
The relationship is negative in 20 of 21 years, which is about as durable as a market regularity gets. But its strength swings with the macro regime. The tightest inverse link came in 2012 at -0.55, the heart of the post-crisis, central-bank-driven era when nearly everything traded as one big dollar-versus-risk macro bet. The crisis and tightening years run consistently strong: 2008 (-0.47), 2011 (-0.44), 2022 (-0.45), 2023 (-0.41). When the world is trading on liquidity and the dollar, copper and the dollar lock together. In calmer, more idiosyncratic stretches the link loosens, and in 2017 (-0.08) it nearly faded.
The lone exception is 2014, the one year the daily link basically vanished (+0.01). It is not a data glitch, it is a story. The dollar staged a powerful post-taper rally (DXY +12% on the year) while copper fell on its own (-17%), but the two were marching to different drummers: the dollar was responding to a Federal Reserve pivot, while copper was sliding on a China-driven demand slump. Both went down-for-copper and up-for-dollar on the year, so the annual numbers look inverse, yet day to day they barely moved together because different forces were driving each. It is the cleanest reminder in the whole table that an annual co-move and a daily correlation are not the same thing.
Here is the result that should change how you use the dollar. The same-day mirror is strong, but it has almost no memory. The correlation between the dollar's move today and copper's move the next day is +0.001, which is to say nothing at all. Run it the other way, copper today versus the dollar tomorrow, and it is -0.011, also nothing. Neither market leads the other by a day.
| Relationship | Correlation | Read |
|---|---|---|
| Same day (DXY vs copper, same session) | -0.32 | Strong inverse mirror |
| DXY today vs copper next day | +0.001 | Essentially zero |
| Copper today vs DXY next day | -0.011 | Essentially zero |
This is the single most important distinction in the entire study, and it is the one most often glossed over: the difference between contemporaneous and predictive. A contemporaneous relationship is already happening and, by the time you can observe it, already priced. A predictive relationship is one you could act on in advance. The dollar and copper share a powerful contemporaneous link and essentially no predictive one at the daily level. You cannot trade copper off yesterday's dollar tick. Knowing the dollar jumped today tells you copper probably fell today, which you already knew by looking at copper. It tells you nothing actionable about tomorrow.
A -0.32 same-day correlation and a +0.001 next-day correlation describe the same two markets. The first is a description of the present; the second is the (failed) attempt to turn it into a forecast. Confusing the two is how a real relationship gets sold as a fake edge.
If the daily move does not predict copper, maybe the dollar's level does. The intuitive story is tidy: a cheap dollar should mean a strong copper backdrop, an expensive dollar a weak one. We split the sample into quintiles by DXY level and measured forward copper from each. It does not work, and it fails in an instructive way.
| DXY level | 5-day fwd (avg, up%) | 20-day fwd (avg, up%) |
|---|---|---|
| Below 80.2 | -0.08% (50%) | -0.63% (49%) |
| 80.2 to 86.0 | +0.13% (53%) | +1.03% (55%) |
| 86.0 to 95.6 | +0.17% (51%) | +0.61% (50%) |
| 95.6 to 99.3 | +0.26% (53%) | +0.90% (58%) |
| Above 99.3 | +0.18% (53%) | +0.78% (56%) |
The pattern is non-monotonic and, worse, era-confounded. If the naive story were right, the lowest-dollar quintile should have the best forward copper. Instead it has the worst (-0.63% over 20 days, the only negative cell). That is not because a weak dollar is bad for copper. It is because the dollar spent its lowest stretch in roughly 2011 to 2014, which happened to be exactly when copper was rolling over into a multi-year bear market of its own. The level column is really a disguised calendar: it sorts history into eras and then attributes each era's copper performance to the dollar's price. DXY is non-stationary (it wanders across decades), so its raw level carries almost no clean information about what copper does next. The level is a red herring.
If the level fails because it is non-stationary, the fix is to measure the dollar against itself: not how high it is, but how stretched it has become. We computed the dollar's 14-day RSI, a standard momentum oscillator, and measured forward copper from each RSI band. This is where a real, sign-consistent forward edge finally appears, and it points exactly the way the inverse relationship predicts.
| DXY RSI | 5-day | 20-day | 60-day | Days |
|---|---|---|---|---|
| Below 30 (oversold dollar) | +0.10% (51%) | -0.13% (57%) | +4.83% (60%) | 231 |
| 30 to 45 | +0.32% (53%) | +1.08% (55%) | +2.14% (54%) | 1,457 |
| 45 to 55 | +0.27% (54%) | +0.78% (56%) | +2.04% (54%) | 1,521 |
| 55 to 70 | -0.16% (50%) | +0.05% (52%) | +1.20% (52%) | 1,533 |
| Above 70 (overbought dollar) | +0.11% (49%) | -0.88% (40%) | -4.04% (44%) | 273 |
Look at the two extreme rows, especially the 60-day column. When the dollar is overbought (RSI above 70), copper went on to lose 4.04% over the next quarter and finished up only 44% of the time, well below the +1.62% baseline. When the dollar is oversold (RSI below 30), copper gained 4.83% over the next quarter and finished up 60% of the time. That is a spread of nearly 9 percentage points over 60 days based purely on the dollar's RSI, and the direction is exactly the inverse logic stretched out over months: a dollar that has run too hot tends to mean-revert, and copper benefits on the way back down, while a dollar that has been beaten down tends to bounce, and copper pays for it. The middle bands sit close to baseline, as they should.
This is the real forward edge in the study, but it deserves an honest caveat that we will press harder in the reality-check section. The RSI extremes are small, overlapping samples. There are only 231 oversold days and 273 overbought days, and because RSI moves slowly, those days arrive in clusters of mostly-consecutive sessions. So while the cell counts look like a few hundred observations each, the number of truly independent overbought or oversold episodes is far smaller, perhaps a few dozen. The effect is real and it is sign-consistent, but it rests on fewer distinct events than the raw counts suggest.
A single oscillator could be a fluke, so we checked the result against a completely different, simpler measure of dollar stretch: where DXY sits relative to its own 5-day moving average. If the RSI story is real, a falling dollar (below its short average) should be a copper tailwind and a surging dollar (well above it) a headwind. It is, and in the same direction, just milder.
| DXY vs 5-day MA | 5-day | 20-day | 60-day | Days |
|---|---|---|---|---|
| Well below (more than 0.4% under) | +0.15% (52%) | +0.85% (58%) | +2.49% (56%) | 908 |
| Below (0 to 0.4% under) | +0.34% (54%) | +0.75% (54%) | +1.85% (54%) | 1,589 |
| Above (0 to 0.4% over) | +0.08% (52%) | +0.48% (53%) | +1.30% (51%) | 1,565 |
| Well above (more than 0.4% over) | -0.10% (50%) | -0.03% (50%) | +0.92% (53%) | 963 |
The gradient is the same shape as the RSI table. A dollar trading below its 5-day average has been a mild copper tailwind (the well-below bucket returned +2.49% over 60 days), and a dollar surging well above its average has been a mild headwind (the well-above bucket managed only +0.92% over 60 days and was slightly negative at the shorter horizons). The spread is narrower than the RSI version because the 5-day average is a blunter, faster measure of stretch than a 14-day oscillator, but two independent gauges pointing the same way is more convincing than either alone.
If two different stretch measures agree, the natural move is to combine them and require both to flash before calling the dollar broadly strong or broadly weak. We defined a strong dollar as RSI 55 or higher and trading above its 5-day average, and a weak dollar as RSI 45 or lower and trading below its 5-day average, then measured forward copper from each state.
| Dollar state | 5-day | 20-day | 60-day | Days |
|---|---|---|---|---|
| STRONG dollar (RSI 55+ and above 5-day MA) | -0.14% (50%) | -0.10% (50%) | +0.37% (50%) | 1,417 |
| WEAK dollar (RSI 45 or under and below 5-day MA) | +0.27% (53%) | +0.96% (56%) | +2.32% (55%) | 1,328 |
The combined gauge produces the cleanest summary of the whole forward story. When the dollar is broadly weak, copper returned +2.32% over the next 60 days and finished up 56% of the time. When the dollar is broadly strong, copper managed just +0.37% and a 50% coin flip. That is a consistent tilt of roughly 2 percentage points per quarter, in the inverse direction, and it holds at every horizon: the weak-dollar bucket beats the strong-dollar bucket at 5, 20, and 60 days. Crucially, requiring both conditions also gives each bucket more than 1,300 days, so this combined view rests on a far broader base than the RSI tails alone.
This is the section that matters most, because the forward results above are real but easy to over-sell. Here is the honest accounting of what this study does and does not establish.
The same-day link explains only 10% of copper. An R-squared of 0.10 is meaningful for two separate markets, but it is still small in absolute terms. About 90% of copper's daily variance comes from somewhere other than the dollar. The dollar is an important coincident input, not a controlling one, and any mental model that treats copper as the dollar's shadow is overstating the case by a wide margin.
The forward tilts are real but mild, and live at the 1-to-3-month horizon, not day to day. The tradeable signal is the dollar's stretch (RSI and the 5-day average), and it shows up over quarters, not sessions. At the daily level the dollar predicts copper essentially not at all (+0.001). Do not read the RSI and gauge tables as a day-trading tool; they are slow, context-setting tilts measured over weeks and months.
Correlation is not causation. The cleanest interpretation is not that the dollar drives copper, but that both copper and the dollar respond to the same global forces: world growth, risk appetite, and Federal Reserve policy. When growth expectations rise, copper tends to climb and the dollar often softens, and vice versa. That makes the dollar frequently a coincident thermometer of the same macro weather copper is reacting to, rather than copper's cause. 2014 is the proof: when the dollar and copper were driven by different things (a Fed pivot versus a China slump), the daily link evaporated even though both moved on the year.
The relationship's strength is regime-dependent. The same-day correlation ranged from -0.55 to +0.01 across individual years. It is tightest in crisis and tightening regimes, when everything trades as one macro bet, and loosest in calm, idiosyncratic stretches. A single full-sample number papers over the fact that the link you can lean on in 2012 is much weaker in 2017.
The forward edge rests on smaller samples in the RSI tails. The oversold and overbought dollar buckets hold only a few hundred mostly-consecutive days each, which translates to far fewer independent episodes. The combined gauge is broader (1,300-plus days per side), but the most dramatic single numbers, the +4.83% and -4.04% 60-day extremes, come from the thinnest, most overlapping slices in the study and should be weighted accordingly.
There are no costs here, and no walk-forward. Every forward figure is a frictionless average with perfect execution and no transaction costs, slippage, or financing. And all of it is measured in-sample over a single 20-year window; we did not run it as a live, out-of-sample strategy. Treat these as descriptive tendencies, not a backtested system.
So the defensible takeaway is layered. Copper and the dollar are a strong same-day mirror, beta about -1.17, that you should expect to see almost every day. That mirror does not predict tomorrow, so the dollar is context, not a daily trigger. And when the dollar reaches a stretched extreme, overbought or oversold, it does leave a mild, inverse fingerprint on copper over the following quarter. Use it as a thumb on the scale, never as the scale itself. This is research and education, not investment advice.
One pair, over 20 years of daily closes: front-month COMEX copper futures and the U.S. Dollar Index, measured live and tested forward.
Daily correlation -0.32, beta -1.17. On the dollar's biggest down days copper averaged +0.92% and rose 71% of the time; on its biggest up days, -0.85% and 29%. A clean, monotonic mirror.
Same-day correlation -0.32, but the dollar today versus copper next day is +0.001, essentially zero. The link is contemporaneous, not predictive. You cannot trade copper off yesterday's dollar tick.
The dollar's raw level is a red herring. Its 14-day RSI is not: an overbought dollar preceded copper -4.04% over 60 days, an oversold dollar +4.83%, a near 9-point spread, all in the inverse direction.
Negative in 20 of 21 years, but the strength swings with the macro regime: tightest in 2012 (-0.55) and the crisis years, and in 2014 the daily link vanished entirely (+0.01).
What is real here, what is only context, and how to tell them apart.
The same-day mirror is strong, but it has no next-day memory (+0.001). Knowing the dollar rose today tells you copper likely fell today, which you already saw on the copper chart. Treat the dollar as backdrop that frames copper, never as a same-day or next-day entry signal.
The dollar's price level carries almost no clean forward information (it is non-stationary and era-confounded). How stretched it is does: an overbought dollar has preceded copper weakness over the next quarter and an oversold dollar copper strength, a mild but consistent inverse tilt of roughly 2 points per quarter when the dollar is broadly weak versus strong.
The whole study turns on one distinction. A -0.32 same-day correlation describes what is already happening and already priced; a +0.001 next-day correlation is the failed attempt to forecast from it. Both markets also answer to the same macro forces, so the dollar is often a coincident thermometer, not copper's cause. Demand evidence a relationship is predictive before you trade it.
We test popular market ideas the honest way: every instance counted, every result measured against a plain baseline. See what else held up, and what did not.
Browse the Research
Comments
Loading comments…