When Do Commodities Diversify a Stock Portfolio? Evidence From Every Major Drawdown Since 1970
Eco3min Research · Commodities & Cross-Asset
Commodities rose 136% while U.S. stocks fell 47% in 1973–74 and gained 26% in 2022’s selloff — but dropped 35% alongside stocks in 2008. Whether commodities diversify a stock portfolio depends on one thing: the inflation regime.
The correlation between a broad commodity basket and U.S. equities has not been stable over the past 56 years. Measured on a 36-month rolling basis from January 1970 through April 2026, it ranged from −0.69 (mid-1993) to +0.61 (March 2020). Across the full sample the correlation of monthly returns is +0.05 — close to zero, and uninformative, because the relationship varies systematically with the level of inflation. This page documents that variation, decomposes commodity behaviour across the five largest U.S. equity drawdowns since 1970, and provides the full monthly dataset.
The textbook claim that commodities diversify stocks, and the revisionist claim that financialisation “broke” that diversification after 2008, are both incomplete. The commodity–equity correlation is a function of the inflation regime: across the full record it is +0.27 when annual CPI is below 2%, +0.02 at 2–4%, and −0.06 above 4%. In the five largest equity drawdowns since 1970, commodities rose in the two that were inflation-driven (1973–74, 2022) and fell in the three that were demand-driven (2000–02, 2008, 2020).
Scope: this dataset measures spot-price index behaviour, not the return of an investable commodity fund (which also earns roll and collateral yield), and the basket is roughly two-thirds energy by weight — the “commodity hedge” in inflation shocks is largely an oil hedge (see decomposition and Methodology).
- In 1973–74 a broad commodity basket rose +136% while U.S. equities fell −47%; in the 2022 selloff commodities gained +26% as stocks fell −25%; but in 2008 commodities dropped −35% alongside a −50% equity decline. The diversification was present in the first two and absent in the third.
- The 36-month rolling correlation round-tripped — −0.23 in the 1990s, peaking at +0.61 in March 2020, and back to −0.23 by April 2026. A permanent structural break would not have reversed; the positive era of 2008–2021 was a 12-year regime, not a one-way change.
- Conditioning on inflation makes the pattern explicit: the full-sample correlation is +0.27 below 2% CPI (n=144), +0.02 at 2–4% (n=298), and −0.06 above 4% (n=234) — a monotonic gradient.
- In months when equities fell, commodities returned an average +0.89% when CPI was above 4% (median +0.35%, n=108) versus −0.20% when CPI was below 4% (median −0.10%, n=145). The hedge in selloffs is concentrated in high-inflation periods.
- The hedge is largely an energy effect: in 1973–74 the energy sub-index rose +328% versus +90% for non-energy; in 2022 energy rose +41% while industrial metals fell −16%. A commodity allocation that hedged inflation shocks was, mechanically, mostly long oil.
- As of April 2026 the correlation is negative (−0.23) during an active energy-driven inflation episode — consistent with the regime framework. The dataset covers 676 monthly observations, 1970–2026, and is released under CC BY 4.0.
676 observations · monthly · Jan 1970 – Apr 2026 · Data: CC BY 4.0
The correlation that comes and goes
The chart above plots the rolling 36-month correlation between the monthly returns of the World Bank broad commodity index and a total-return index of the U.S. equity market, from 1972 (the first point with three years of history) through April 2026. A reading of +1 would mean the two move in lockstep; −1 would mean they move exactly opposite; zero would mean no linear relationship.
Three features stand out. Through the 1980s and 1990s the correlation was usually negative — commodities and stocks tended to move in opposite directions, which is the textbook case for diversification. From the mid-2000s it climbed and stayed positive for more than a decade, peaking at +0.61 in March 2020. Then it fell back below zero, reaching −0.23 by April 2026. The shaded band marks the 2008–2021 positive era; the red segment marks the post-2022 return to negative territory.
The single full-sample number — a correlation of +0.05 across all 676 months — is therefore close to meaningless. It averages a −0.69 trough and a +0.61 peak into a figure that describes neither. The useful question is not “what is the correlation” but “under what conditions is it negative.” For the longer-horizon context, see our S&P 500 historical returns dataset and the WTI crude oil price history.
A diversifier, a “broken” diversifier, or neither
The dominant narrative holds that commodities are a portfolio diversifier: because their prices respond to supply shocks and inflation rather than to corporate earnings, they are supposed to zig when equities zag. A widely-cited revision to that view argues the property disappeared after the mid-2000s, when commodity index funds linked raw-material prices to financial flows — the so-called financialisation of commodity markets — pushing commodity and equity returns to move together.
The data supports neither claim in its general form. The diversification did not exist as a constant, and it was not permanently broken. The rolling correlation was negative in the 1990s, rose to +0.39 on a static basis across 2008–2021, and fell back to −0.12 across 2022–2026. The positive era was real and lasted twelve years — but it reversed. What changed between the two negative eras and the positive one in between was the inflation environment: low and stable through 2008–2021, elevated before and after.
This is a spot-price index, not an investable return. The series measures the price level of a commodity basket. An actual commodity allocation — a futures-based fund or a total-return index — earns additional roll yield and collateral interest and is periodically rebalanced; its return can differ materially from the spot index, particularly when futures curves are in backwardation or contango. The patterns here describe how commodity prices behave relative to equities, not the realised return of any specific fund.
The basket is roughly two-thirds energy. By construction the World Bank total index is approximately 67% energy by weight. Commodity behaviour in this dataset is therefore dominated by oil and gas. Industrial metals, precious metals and agricultural goods behave differently and are shown separately in the decomposition below. Any reading of “commodities hedge inflation” should be understood, in practice, largely as a statement about energy.
The commodity–equity correlation round-tripped from −0.23 in the 1990s to +0.61 in 2020 and back to −0.23 by 2026. The diversification property is conditional, not constant — and not dead.
Every major U.S. equity drawdown since 1970
The clearest test of a diversifier is what it does when stocks fall. The table below takes the five largest drawdowns of the U.S. equity total-return index since 1970 and measures, over each peak-to-trough window, what the broad commodity basket and its energy, non-energy and metals components did. Peak and trough are identified algorithmically (the maximum of the equity index before the decline and its minimum after); the exact windows and method are in the Methodology.
| Equity drawdown | Window | Avg CPI | Equities | Commodities | Energy | Non-energy | Metals |
|---|---|---|---|---|---|---|---|
| 1973–74 oil shock | 1972-12 → 1974-09 | 7.9% | −46.5% | +136.0% | +328.2% | +89.6% | +53.6% |
| 2000–02 dot-com | 2000-08 → 2002-09 | 2.4% | −45.0% | −2.9% | −7.1% | +4.7% | −16.0% |
| 2008–09 financial crisis | 2007-10 → 2009-02 | 3.4% | −50.3% | −35.0% | −41.5% | −20.1% | −55.4% |
| 2020 COVID crash | 2020-01 → 2020-03 | 2.1% | −20.2% | −31.1% | −43.9% | −7.5% | −11.6% |
| 2022 inflation selloff | 2021-12 → 2022-09 | 8.2% | −24.8% | +25.6% | +40.8% | −3.9% | −16.2% |
Peak-to-trough returns over each equity drawdown window. “Avg CPI” is the mean year-over-year U.S. CPI across the window. All values computed from the dataset.
The split is clean along one axis: average inflation during the episode. The two windows where commodities rose — 1973–74 (mean CPI 7.9%) and 2022 (mean CPI 8.2%) — are the two high-inflation episodes. The three where commodities fell — 2000–02 (2.4%), 2008–09 (3.4%) and 2020 (2.1%) — are the lower-inflation ones. When the equity selloff was itself driven by an inflation or supply shock, commodities were the source of the problem for stocks and a hedge for the portfolio. When it was driven by collapsing demand, commodities fell with everything else. Related material: the roll-cost view of commodities.
Commodities hedged the two inflation-driven equity drawdowns (1973–74: +136%, 2022: +26%) and amplified the three demand-driven ones (2008: −35%, 2020: −31%). The dividing line is the nature of the shock, proxied by the inflation rate.
The hedge is mostly oil
The decomposition columns in the table expose what is doing the work. In 1973–74 the energy sub-index rose +328% against +90% for non-energy and +54% for metals. In 2022 the divergence is starker still: energy gained +41% while industrial metals fell −16% and non-energy was roughly flat at −4%. The “commodity hedge” against an inflation-driven equity selloff was, mechanically, an energy position. A basket of industrial metals — which respond to the same global growth cycle that drives corporate earnings — behaved much more like equities, falling in four of the five drawdowns.
This matters for interpretation. Because the World Bank total index is about two-thirds energy, the dataset’s “commodities hedge inflation” finding rests heavily on oil. For metals the relationship is weaker and at times inverts. An investor reading this as a general property of “commodities” would be over-generalising from what is principally an energy effect. For the underlying series, see the WTI and copper dataset pages.
A legitimate analytical qualification is that the inflation regime may not be the cause of the correlation pattern but a proxy for something deeper. Two alternatives deserve weight. First, the 2008–2021 positive correlation coincides with the zero-interest-rate and quantitative-easing era, when most risk assets — equities, credit, commodities — moved together on shifts in central-bank liquidity. On that reading, commodities co-moved with stocks because both tracked the same monetary impulse, not because of index-fund financialisation; the reversion after 2022 then reflects the return of positive real rates rather than a change in commodities themselves. The dataset cannot cleanly separate the financialisation channel from the liquidity-regime channel, because both predict the same thing in the same years. A related perspective: Our note on commodities as macroeconomic regime signals.
Second, inflation is correlated with, but not identical to, the type of shock. The 2008 episode began with elevated inflation and an oil price near its all-time high in mid-2008, yet commodities ultimately fell −35% because the demand collapse dominated. The cleaner causal statement is that commodities hedge supply-driven equity selloffs and amplify demand-driven ones; the inflation rate is the observable proxy for that distinction, and it is an imperfect one. For monetary-regime context, see our net liquidity index and real interest rate history.
The correlation is a function of inflation
Across the full 1970–2026 record, sorting every month by the contemporaneous year-over-year CPI produces a monotonic gradient in the commodity–equity correlation.
| Inflation regime | n (months) | Commodity–equity correlation | Mean CPI |
|---|---|---|---|
| CPI < 2% (low) | 144 | +0.27 | 1.2% |
| CPI 2–4% (moderate) | 298 | +0.02 | 2.9% |
| CPI ≥ 4% (high) | 234 | −0.06 | 7.1% |
Correlation of contemporaneous monthly commodity and equity returns within each inflation regime. Full sample 1970–2026.
The correlation falls steadily as inflation rises — from +0.27 in low-inflation months to roughly zero in the middle band to −0.06 in high-inflation months. Because the move is monotonic across three buckets rather than a jump at one threshold, it does not depend on the precise definition of “high” inflation; shifting the cut from 4% to 3% or 5% preserves the ordering (see sensitivity).
Restricting attention to the months that matter most for a diversifier — those when equities fell — sharpens the result. In down-equity months with CPI above 4%, the broad commodity basket returned an average +0.89% (median +0.35%), rising in 55% of them. In down-equity months with CPI below 4%, it returned an average −0.20% (median −0.10%), rising in only 46%. The diversification benefit in selloffs is concentrated almost entirely in the high-inflation state.
When equities fell and inflation was above 4%, commodities rose +0.89% on average (n=108). When equities fell and inflation was below 4%, they fell −0.20% (n=145). The hedge appears only in the high-inflation regime.
One observation in the sub-4% bucket illustrates how fuzzy the threshold is: in March 2026, equities fell −4.9% while commodities rose +26% on the Iran-war energy shock — a textbook inflation hedge occurring at 3.3% CPI, just below the cut. The median absorbs the single outlier, but it is a reminder that the mechanism is continuous, not a switch that flips at exactly 4%.
What happened next? Forward returns by inflation regime
Classifying each month by its inflation regime and measuring the subsequent 12-month return of both assets shows how the regimes have historically resolved. The forward window is calendar-month based; recent months for which 12 months of subsequent data do not yet exist are excluded.
| Regime at month t | n | Median 12m commodity | Median 12m equity | Both fell (12m) |
|---|---|---|---|---|
| CPI < 2% (low) | 144 | +12.0% | +17.3% | 4.2% |
| CPI 2–4% (moderate) | 286 | +2.6% | +13.4% | 7.3% |
| CPI ≥ 4% (high) | 234 | +0.9% | +14.1% | 12.4% |
Median forward 12-month total returns and the share of months in which both assets fell over the subsequent year, by inflation regime. 1970–2026, calendar-month forward windows.
Median forward commodity returns are highest from low-inflation months and weakest from high-inflation months — consistent with mean reversion, since high-inflation months tend to follow large commodity rallies. The share of 12-month windows in which both assets fell rises with inflation, from 4.2% in the low regime to 12.4% in the high regime: the periods when commodities most reliably diversify within a selloff are also the periods when joint drawdowns are most frequent. The two facts are not contradictory — they describe co-movement at different horizons.
Past distributions are not predictive of future outcomes. Regime-conditional statistics describe historical patterns, not expected returns.
▸ 36-month correlation at −0.23 (Apr 2026): a sustained move back above zero would mark a return to the 2008–2021 positive-correlation regime. Historically, such moves have coincided with falling and stable inflation.
▸ U.S. CPI at 3.8% (Apr 2026), 4.2% (May 2026): a sustained reading above 4% places the data in the regime where, historically, the commodity–equity correlation has been negative. See our U.S. CPI inflation history.
▸ Next World Bank Pink Sheet and BLS CPI releases: the energy sub-index and headline CPI are the two series that, in combination, have historically determined which correlation regime is in force.
How to read the correlation regimes
Commodities and equities move sharply opposite. Observed in the early-1990s trough (−0.69) and in disinflationary stretches. The textbook diversification case.
The most common state historically. Commodities provide a modest offset to equity moves. Where the data sits as of April 2026 (−0.23).
Commodities and equities drift together. Characteristic of the 2008–2021 low-inflation, abundant-liquidity era (+0.39 static).
The two assets move as one, eliminating the diversification benefit. The 36-month peak of +0.61 in March 2020 is the only sustained instance.
Historical turning points
1973–74 — the original commodity hedge
From the December 1972 equity peak to the September 1974 trough, U.S. equities lost −46.5% while the broad commodity basket rose +136.0%, led by energy (+328.2%) during the first oil embargo. Average CPI over the window was 7.9%, rising from 3.4% to 11.9%. This is the canonical episode behind the “commodities diversify” intuition — and it was overwhelmingly an energy event.
2000–02 — diversification by absence of decline
Across the dot-com unwind (August 2000 to September 2002), equities fell −45.0% while commodities were roughly flat at −2.9% (average CPI 2.4%). Commodities did not rise, but by holding their value while equities collapsed they provided diversification through non-participation rather than through a hedge.
2008–09 — the diversifier that failed
The financial crisis (October 2007 to February 2009) is the counter-example. Equities fell −50.3% and commodities fell with them, −35.0%, with metals down −55.4%. The episode began with elevated inflation — oil reached its record high in mid-2008 — but the collapse in global demand dominated, and the average CPI of 3.4% masks a swing from 5.5% to outright deflation by the trough. When the shock is a demand collapse, commodities offer no protection.
2020 — a fast demand shock
The COVID crash (January to March 2020) was brief and demand-driven. Equities fell −20.2% and commodities fell harder, −31.1%, with energy down −43.9% as oil demand evaporated. Average CPI was 2.1%. The 36-month correlation reached its all-time peak of +0.61 in this month.
2022 — the hedge returns
From the December 2021 equity peak through the September 2022 trough, equities fell −24.8% while commodities rose +25.6%, energy up +40.8%, against the highest inflation in four decades (average CPI 8.2%, peaking at 9.0%). Industrial metals, however, fell −16.2% as recession fears mounted — the same split between energy and metals seen in 1973–74.
April 2026 — current observation
As of the latest paired data, the 36-month correlation is −0.23 and the 24-month is −0.11, during an active energy-driven inflation episode. The World Bank energy sub-index rose +41% in March 2026 and U.S. CPI reached 3.8% in April and 4.2% in May , driven by an oil supply shock; Brent crude briefly touched roughly $126 per barrel in late April before a mid-June ceasefire pulled it back near $80 . Unlike the 1973–74 and 2022 episodes, U.S. equities did not enter a major drawdown over this period, so 2026 is presented as a confirmation of the negative-correlation regime rather than as a completed crisis-hedge episode.
Methodology
The dataset merges three primary sources at monthly frequency, January 1970 to April 2026: the World Bank Commodity Price Data (“Pink Sheet”) monthly index series (2010 = 100), the Kenneth R. French Data Library U.S. market total-return series, and the U.S. Bureau of Labor Statistics Consumer Price Index. Commodity and equity returns are monthly log changes; the correlation is the Pearson correlation of those returns.
equity return = ln( US_market_total_return_index[t] / [t−1] )
rolling corr(W) = Corr( commodity return, equity return ) over trailing W months
inflation regime = low if CPI_YoY < 2% · moderate if 2% ≤ CPI_YoY < 4% · high if CPI_YoY ≥ 4%
Equity proxy
The U.S. equity series is the Kenneth R. French total U.S. market return (market excess return plus the risk-free rate, compounded to an index). It is a total-return measure, includes dividends, extends consistently to 1970, and correlates approximately 0.99 with the S&P 500 total-return index. The correlation results are materially unchanged if the S&P 500 price index is substituted.
Crisis-window selection
The five episodes are the largest drawdowns of the equity total-return index since 1970 (consensus bear markets). For each, the peak is the maximum of the equity index in the run-up and the trough is its minimum after the peak; component returns are measured over that same peak-to-trough window. The selection is mechanical, not discretionary.
trough = argmin( equity_index ) after the peak
episode return(series) = series[trough] / series[peak] − 1
Filter definitions
All conditional statistics use these explicit definitions, applied identically throughout:
“moderate inflation” = 2% ≤ CPI_YoY < 4%
“high inflation” = CPI_YoY ≥ 4%
“down-equity month” = equity monthly return < 0
“2008–2021 era” = date BETWEEN 2008-01 AND 2021-12
“post-2022” = date ≥ 2022-01
Sensitivity
The monotonic gradient (low +0.27, moderate +0.02, high −0.06) does not depend on the 4% cut. Moving the high-inflation threshold to 3% or 5% leaves the ordering intact: the highest-inflation bucket remains the only one with a negative correlation, and the lowest remains the most positive. The down-equity-month contrast is similarly robust to the threshold, and the median is reported alongside the mean precisely because the high-inflation bucket contains large positive outliers (1970s, 2022, the March 2026 energy shock).
Dataset design
| Variable | Type | Unit | Source | Calculation |
|---|---|---|---|---|
| total, energy, nonenergy, metals, … | float | index (2010=100) | World Bank | direct |
| eq_tr_index | float | index | French Library | compounded total return |
| total_ret, eq_ret, … | float | log return | derived | ln(x[t]/x[t−1]) |
| cpi_yoy | float | % | BLS | 12-month % change of CPI-U |
| corr24m / corr36m / corr60m | float | — | derived | rolling Pearson corr |
| infl_regime | str | — | derived | low / moderate / high on cpi_yoy |
| comm_fwd12_pct / eq_fwd12_pct | float | % | derived | forward 12-month % change |
Reproduction
# Reproduce the core conditional finding from primary sources import pandas as pd, numpy as np df = pd.read_csv("commodity-equity-correlation-1970-2026.csv") d = df.dropna(subset=["cpi_yoy"]) for lo, hi in [(None,2),(2,4),(4,None)]: m = (d.cpi_yoy >= (lo or -9)) & (d.cpi_yoy < (hi or 99)) s = d[m] print(lo, hi, round(s.total_ret.corr(s.eq_ret),3), len(s)) # CPI < 2% -> +0.27 (n=144); 2-4% -> +0.02 (n=298); >=4% -> -0.06 (n=234)
Data sources & references
- PrimaryWorld Bank, Commodity Markets “Pink Sheet”, monthly price indices (energy, non-energy, metals, agriculture, precious metals), retrieved June 2026.
- PrimaryKenneth R. French Data Library, U.S. market return factors (Mkt-RF, RF), monthly, retrieved June 2026.
- PrimaryU.S. Bureau of Labor Statistics, Consumer Price Index for All Urban Consumers (CPI-U, series CUUR0000SA0), retrieved June 2026.
- ResearchTang, K. & Xiong, W. (2012), “Index Investment and the Financialization of Commodities”, Financial Analysts Journal.
- ResearchGorton, G. & Rouwenhorst, K. G. (2006), “Facts and Fantasies about Commodity Futures”, Financial Analysts Journal.
- ResearchErb, C. & Harvey, C. (2006), “The Strategic and Tactical Value of Commodity Futures”, Financial Analysts Journal.
- ReferenceNational Bureau of Economic Research, U.S. business cycle reference dates.
Methodological limitations
- The commodity series is a spot-price index, not an investable total return; roll yield and collateral return are not captured, so realised fund returns differ.
- The total index is approximately 67% energy by weight; results are dominated by oil and gas and do not generalise cleanly to all commodities.
- Rolling 36-month correlations use overlapping windows, which induce autocorrelation; they are descriptive of the trajectory, not a significance test. Static non-overlapping period correlations are reported as corroboration.
- The inflation regime is a proxy for the supply-versus-demand nature of a shock, not a direct measure of it; the two diverge (notably in 2008).
- The most recent months of CPI reflect BLS releases that remain subject to revision; the equity-paired series ends April 2026.
- The financialisation and liquidity-regime explanations for the 2008–2021 positive correlation cannot be separated within this dataset.
Frequently asked questions
Do commodities diversify a stock portfolio?
It depends on the inflation regime. Across 1970–2026 the correlation of monthly commodity and equity returns was +0.27 when annual CPI was below 2%, +0.02 at 2–4%, and −0.06 above 4%. In the five largest equity drawdowns since 1970, commodities rose in the two inflation-driven episodes (1973–74: +136%, 2022: +26%) and fell in the three demand-driven ones (2000–02, 2008, 2020). They diversify reliably only in high-inflation conditions.
Did financialisation break the commodity–equity diversification after 2008?
Not permanently. The correlation did rise to +0.39 across 2008–2021, consistent with the financialisation hypothesis, but it fell back to −0.12 across 2022–2026 once inflation returned. A structural break would not reverse. The positive era coincided with the zero-rate, quantitative-easing period, and the data cannot distinguish whether financialisation or the common liquidity regime drove the co-movement.
Why did commodities fall in 2008 if they are supposed to be an inflation hedge?
Because 2008 was fundamentally a demand and credit collapse, not an inflation shock. Although the episode began with high oil prices, the broad basket fell −35% and industrial metals fell −55% as global demand contracted. Commodities hedge supply-driven selloffs and amplify demand-driven ones; the average CPI of 3.4% over the window masks a swing from 5.5% to deflation by the trough.
Is this “commodity hedge” just an oil bet?
Largely, yes. The World Bank total index is about two-thirds energy, and the hedge in inflation shocks comes overwhelmingly from energy: +328% in 1973–74 and +41% in 2022, versus much weaker or negative moves in industrial metals (which fell −16% in 2022). A diversification claim about “commodities” based on this dataset is, in practice, mostly a claim about oil and gas.
What is the current commodity–equity correlation?
As of April 2026 the trailing 36-month correlation is −0.23 and the 24-month is −0.11. This is occurring during an energy-driven inflation episode, with U.S. CPI at 3.8% in April and 4.2% in May 2026 and the World Bank energy sub-index up 41% in March — consistent with the negative-correlation, high-inflation regime documented here.
Does this dataset measure the return of a commodity fund?
No. It measures a spot-price index. A futures-based commodity fund additionally earns roll yield and collateral interest and rebalances periodically, so its realised return can differ substantially from the price index, especially when futures curves are in backwardation or contango. The dataset describes price behaviour relative to equities, not fund performance.
Last updated — 12 July 2026
Disclaimer – Financial Information: The analyses, commentary, and content published on eco3min.fr are provided for informational and educational purposes only. They do not constitute investment advice or a solicitation to buy or sell financial instruments. Past performance is not indicative of future results. All investment decisions involve risk and are the sole responsibility of the reader.