Gold-Copper Ratio: 620 Months Ranked Since 1975, and Why the Historical Average Is Not One Number

Latest observation

Executive summary

The record in six numbers

Where the reading sits

The copper-gold ratio divides the copper price by the gold price. Both legs come from the World Bank Pink Sheet, which has published monthly averages of both since January 1960, so the series needs no splicing across providers and no deflator: it is a relative price, and the currency cancels.

Over the 620 months from January 1975, the ratio ranges from 1.170 to 6.350, a span of 5.43 times, which is why the chart above is drawn on a log scale. The median month is 2.410. August 2026 reads 1.473 and ranks 24th from the bottom.

The interesting part is what sits underneath it. The fourteen lowest months of the whole record are the fourteen consecutive months from April 2025 to May 2026, and the fifteenth is April 2020. Read the other way round, one ounce of gold has bought more than 745 pounds of copper in fourteen separate months since April 2025, and had never done so in the fifty years before.

Why the average moves and the rank does not

The dominant way to read a ratio like this one is against its historical average. It is the natural instinct, it is what a mean-reverting series invites, and it is how the gold-silver ratio is usually presented, with some justification: that pair has spent long stretches oscillating around a level. The copper-gold ratio invites the same treatment, and search traffic shows readers asking for exactly that number.

The number does not exist. Counting from January 1960 the mean is 4.745. From January 1971, the year the dollar left gold, it is 3.057. From January 1975 it is 2.620, from January 1990 it is 2.624, from January 2000 it is 2.534, from January 2010 it is 2.095, and over the trailing ten years it is 1.924. The largest of those is 2.47 times the smallest. Exclude the years when gold was fixed by treaty rather than by trade and the remaining seven still span a factor of 1.59. The median is better behaved and still moves by 1.43.

Eight ways to compute the average, one reading below all of them

Every dot is a defensible historical average. The August 2026 reading of 1.473 sits below all eight.

Source: World Bank Commodity Price Data, the Pink Sheet. Chart: Eco3min Research.

PNG SVG PDF

Reuse permitted with attribution to Eco3min Research under CC BY 4.0.

Now run the same test on the ranking. Take the ten lowest months of the copper-gold ratio counting from 1960. Then from 1971, 1975, 1980, 1990, 2000, 2010, and from September 2016, a window that holds only 120 months. The answer is the same ten months every time, in the same order: February and March 2026, then November, October, September and December 2025, then January and April 2026, then April and August 2025. The agreement extends to the fifteen lowest and breaks at the seventeenth, for a mechanical reason: the trailing ten-year window cannot contain September 1980.

That asymmetry is the finding. A mean summarises a level, and a series with a five-decade drift has no stable level to summarise. A rank summarises a position, and a position survives the drift. It is the same reason the raw copper-gold ratio dataset is more useful than any single statistic drawn from it.

What the ranking does not say

A rank is a position in a distribution, and a distribution is not a mechanism. Three qualifications belong on the page rather than in a footnote.

The first is the near-tie. February 2026 reads 1.1702 and March 2026 reads 1.1703. They are 0.009 percent apart, which is inside the rounding the Pink Sheet applies to its current vintage. Recompute the whole series with LBMA gold instead of Pink Sheet gold and the two swap places: March becomes the record at 1.169. Every other conclusion on this page survives that substitution unchanged, including the identity of the ten lowest months, the rank of April 2020 and the average since 1975, which moves from 2.6204 to 2.6203. The record is a plateau, and the honest sentence is that the ratio has spent a year at levels it never reached before, not that one particular month is special.

The second is that a record low in a ratio can be produced entirely by its denominator. That is the case here, it is documented at length in the macro signal study, and this page takes it as given rather than re-deriving it. The consequence for the measurement is direct: the ranking on this page describes the ratio, and the ratio alone.

The third is that the ratio has behaved as its users expect at the two moments those users cite most. It reached what was then a record low in April 2020, within weeks of the deepest point of the pandemic shock, and it sat in its bottom decile through the 1979 to 1981 episode. A measurement page that only reported the awkward result would be selecting evidence as surely as one that only reported the flattering one.

Explore the record

The argument above rests on a claim that is easier to test than to read: that the average you get depends on the window you pick, and the rank does not. The module below lets you pick the window. Drag either handle to change the start and end of the sample, and the mean, the median and the rank of the highlighted month are recomputed on that window alone.

The module reads the published CSV directly. If it does not load, the decade chart above is the static equivalent of its default view.

The supplementary view, on a linear scale

The hero chart is drawn on a log scale because the ratio spans 5.43 times over the window, above the threshold at which a linear axis compresses the early decades into a single band. The decade view below is the linear complement: it drops the time axis entirely and shows each decade as the range between its lowest and its highest month.

The 2020s median sits below the lowest month of the 1990s

Each bar runs from the decade’s lowest month to its highest. The dot is the decade median. The dashed line is the 1990s floor.

Source: World Bank Commodity Price Data, the Pink Sheet. Chart: Eco3min Research.

PNG SVG PDF

Two decades overlap the 2020s from above. The 1980s reach down to 1.387 and the 2010s to 1.608, so the current decade is not alone at the bottom of the range. What separates it is the concentration: its median is lower than the floor of the 1990s, and 54 of its 80 months are.

Forward distribution

The relation most often attached to this ratio is with the US 10-year Treasury yield, so the file carries that yield and its forward changes at six and twelve months, alongside the ratio’s own forward return. The table below is conditional on the expanding percentile, which uses only information available at the date, never the ex-post rank.

The caveat is larger than the table. The 79 months that fall in the lowest bucket are not 79 independent observations: they come from 13 contiguous episodes, and the largest single episode, August 2024 to August 2026, is 31.6 percent of the bucket. Counted by episode rather than by month, 11 of the 13 had a positive median 12-month move in the 10-year yield, which is a weaker statement resting on a much smaller sample than the month count suggests. The windows also overlap, so consecutive rows share most of their forward period.

Eight of the 20 months since January 2025 have a complete 12-month forward window, so the current episode contributes to the table only in part. Past distributions are not predictive of future outcomes. Regime-conditional statistics describe historical patterns, not expected returns. Whether the relation between this ratio and the 10-year yield still holds is examined in the macro signal study, and this page takes no position on it.

Levels to watch

Decade and episode tables

Ranks 1 to 10 are identical under all eight start dates tested. Ranks 1 to 15 are too. The set of ten is also unchanged when gold is taken from the LBMA benchmark instead of the Pink Sheet, though ranks 1 and 2 swap.

Historical turning points

August 1976, the high. The ratio reads 6.350, with copper at 1,539 US dollars a tonne and gold at 110 US dollars a troy ounce. One ounce of gold bought 157 pounds of copper, the least of the whole window. Gold had been freely traded for five years and had not yet begun the run that would take it above 600 dollars by 1980.

September 1980, the first low that stood. The ratio reads 1.387, with copper at 2,060 dollars a tonne and gold at 674 dollars an ounce. It held the record for the lowest month for nearly forty years, and it is the reason the trailing ten-year window disagrees with the longer ones beyond rank sixteen: a ten-year window ending in 2026 cannot see 1980.

April 2020, the record the search traffic remembers. The ratio reads 1.363, with copper at 5,058 dollars a tonne and gold at 1,683 dollars an ounce, and the 10-year yield at 0.66 percent. It beat September 1980 and became the lowest month since 1975, at the 0.18th percentile of the data available at the time. It stood for five years.

April 2025, when it fell. The ratio reads 1.294, the first month to print below April 2020. Copper is at 9,177 dollars and gold at 3,218 dollars. Note the direction of both legs: copper is 81 percent above its April 2020 level and gold is 91 percent above its own. The record did not come from a collapse in either price.

February 2026, the current low. The ratio reads 1.170, with copper at 12,951 dollars a tonne, gold at 5,020 dollars an ounce and the 10-year yield at 4.13 percent. Gold’s highest monthly average of the record and a copper price 156 percent above April 2020 produce the lowest ratio of the record.

August 2026, the latest observation. The ratio reads 1.473, with copper at 14,326 dollars a tonne, its highest monthly average in the file, and gold at 4,411 dollars an ounce. The reading is 26 percent above the February low and 44 percent below the average since 1975, and it ranks 24th of 620. Two months separate it from the last lower reading.

Methodology

The ratio. The copper price in US dollars a tonne is converted to dollars a pound at 2,204.622621848776 pounds a tonne, then divided by the gold price in US dollars a troy ounce and multiplied by 1000. The inverse column, gold_copper_ratio_lb, is the gold price divided by the copper price in dollars a pound, and reads directly as pounds of copper per ounce of gold. The two are reciprocal by construction and the file asserts it.

Frequency, and why it is monthly. Monthly, because no daily copper series with fifty years of depth exists from a source that can be redistributed under CC BY 4.0. This is disclosed in the direction that hurts: averaging within the month compresses extremes, so a daily series would probably make both the 2026 low and the 1980 low deeper, not shallower. The ranking of monthly averages is a ranking of monthly averages.

Why the analysis window starts in January 1975. The published file starts in January 1960 and carries a gold_regime column, so the choice can be reversed by the reader. Gold was an administered price at 35 dollars an ounce until the London Gold Pool collapsed in March 1968, and only fully market-determined after the United States closed the gold window in August 1971. Private ownership of gold became legal for US residents on 31 December 1974. The file marks 98 months as pegged, 41 as two_tier and 661 as floating. The analysis window takes the 620 months from January 1975. It makes no difference to the ranking: the ten lowest months are the same from 1960.

Percentiles. Every percentile column is computed on information available at the date. pct_expanding_1975, pct_expanding_1990 and pct_expanding_2000 rank a month within the months elapsed since their start. pct_trailing_10y and pct_trailing_20y rank it within the trailing 120 and 240 months. The single ex-post column is named rank_asc_1975_expost and is never used to form a bucket in the forward table.

The four competing averages. mean_since_1975, mean_since_1990, mean_since_2000 and mean_trailing_10y record, at every date, what a reader would obtain as the historical average given the start date they happened to pick. mean_dispersion_x is the largest of the four divided by the smallest, and it reads 1.364 in August 2026. The eight-way comparison quoted on this page extends the same idea to eight start dates and is computed on the full file.

Sensitivity, four tests. First, the start date: the mean moves by a factor of 2.47 across the eight windows, the median by 1.43, and the ten lowest months not at all. Second, the gold benchmark: recomputing the ratio with the LBMA afternoon price instead of Pink Sheet gold leaves the set of ten lowest months unchanged, swaps ranks 1 and 2, leaves April 2020 at rank 15 and moves the mean since 1975 from 2.6204 to 2.6203. Third, the unit basis: computing the ratio directly as dollars a tonne over dollars an ounce, without the pound conversion, gives February 2026 as the lowest month, as the published basis does. Fourth, rounding: the September 2026 Pink Sheet vintage rounds prices to the unit, and the largest combined effect that can have on any month of the analysis window is 0.4873 percent, in August 1976.

The two vintages. The January 2025 vintage of the Pink Sheet carries full decimal precision through December 2025; the September 2026 vintage rounds every price to the unit but extends to August 2026. The file uses the first through December 2025 and the second thereafter. The 792 overlapping months are asserted to agree within 0.5 units on both legs before the build proceeds, which is exactly the rounding and no more. Pink Sheet monthly prices are historical averages of quoted prices and are not revised.

Cross-check against the issuers. Pink Sheet gold against the LBMA afternoon benchmark, resampled to monthly means over 701 common months: median relative gap 0.0011 percent, 95th percentile 0.182 percent, and four months above 1 percent, all between December 1980 and November 1985, when the Pink Sheet drew on legacy sources. Pink Sheet copper against the IMF global copper price over 415 common months from January 1992: median relative gap 0.0515 percent, 95th percentile 0.419 percent, two months above 1 percent. The IMF series is used as a control and is never redistributed, because it carries the marker “Copyrighted: Citation Required” on FRED.

Reproduction. The two Pink Sheet workbooks, the LBMA series and the 10-year yield are all public. The ranking reduces to four lines.

Data sources and references

• World Bank, Commodity Price Data (the Pink Sheet), monthly prices workbook, vintages of January 2025 and September 2026. Copper is LME grade A settlement, gold is the London price, spot average of daily rates from June 2025 and the London afternoon fixing before that. Both series begin in January 1960. CC BY 4.0.
• London Bullion Market Association, LBMA Gold Price PM, daily, from 1 April 1968. Used as the independent second gold benchmark in the robustness test, resampled to monthly means.
• Board of Governors of the Federal Reserve System, H.15 selected interest rates, 10-year Treasury constant maturity, monthly, via FRED series GS10. Public domain.
• International Monetary Fund, global price of copper, via FRED series PCOPPUSDM. Used only as a cross-check on the copper leg and not redistributed, because the series carries the FRED marker “Copyrighted: Citation Required”.
• Erb, C. B., and Harvey, C. R. (2013), “The Golden Dilemma”, Financial Analysts Journal 69(4), 10 to 42. On the instability of any long-run anchor for the real gold price, which is the argument this page reaches from the ratio side.
• Harvey, D. I., Kellard, N. M., Madsen, J. B., and Wohar, M. E. (2010), “The Prebisch-Singer Hypothesis: Four Centuries of Evidence”, Review of Economics and Statistics 92(2), 367 to 377, together with the authors’ subsequent erratum correcting data construction errors. On secular trends in primary commodity prices, and on why a trending series resists summary by a mean.
• National Bureau of Economic Research, business cycle expansions and contractions, used only to date the episodes named in the turning points section.

Limitations

• Monthly averaging compresses extremes. Every reading is a monthly average of daily quotes. A daily series would very likely show deeper lows in both 2026 and 1980. The ranking is a ranking of monthly averages and nothing more.
• Neither leg is a constant-quality good. Copper’s end uses and gold’s marginal buyer have both changed across fifty years. The ratio is a relative price, not an index of like for like, and the file makes no adjustment for it.
• The record is a plateau, not a point. February and March 2026 are 0.009 percent apart and swap order under a different gold benchmark. Treating any single month as the record overstates the precision of the data.
• Two vintages, one series. The published file splices the January 2025 and September 2026 vintages of the same World Bank workbook because the newer one rounds to the unit. The overlap is asserted to agree within rounding, but it is a splice, and it is the one place the file departs from a single download.
• The 1980s gold gap. The Pink Sheet and the LBMA benchmark disagree by more than 1 percent in four months between December 1980 and November 1985. Those months sit at ranks 17 and beyond, so they do not touch the headline, and early-1980s work is best done with both sources side by side.
• The forward table rests on 13 episodes. The lowest percentile bucket holds 79 months drawn from 13 contiguous episodes with overlapping forward windows. The month count overstates the evidence by roughly a factor of five.
• The current decade is incomplete. The 2020s row covers 80 months to August 2026, not 120, and its statistics will change as the decade finishes.

Frequently asked questions