Strategy Research

The commodity edges that actually hold up

Most "winning strategies" you find online are overfit backtests. These are different: fundamental edges (positioning, inventories, the futures curve, weather) drawn from peer-reviewed research, each adversarially fact-checked — with the hype stripped out and the caveats left in.

24 sources reviewed25 claims verified · 6 killedNBER · AQR · J. Futures Markets
1

Carry / Term-Structure

Strongest, most buildable edge

Buildable on your data now
The rule
Favor commodities in backwardation (a downward-sloping futures curve); avoid or short those in contango. The slope of the curve — the roll yield — is the signal.
Why it works
Backwardation reflects scarcity: buyers pay a premium for prompt delivery, and you collect the roll as each contract converges to spot. It's model-free and observable in advance.
The evidence
High-basis minus low-basis ≈ +10.2%/yr (t=3.73). Over 139 years, commodities returned 7.9% in backwardation vs 1.5% in contango.
Both figures survived adversarial fact-checking.
The catch
Carry, momentum and inventory are ~0.87 correlated — they're the same underlying bet. Don't stack them as if they diversify.
On your data
Computed from your 40 years of futures term-structure data.
2

Hedging-Pressure Positioning (COT)

Genuine diversifier — and it's not 'fade the speculators'

Buildable on your data now
The rule
Go long where commercial hedgers are net short, and short where they're net long. The signal lives in the commercial (producer/user) category — not managed money.
Why it works
Hedgers pay a premium to offload price risk; speculators earn it for taking the other side. This 'hedging pressure' is grounded in Keynes/Cootner/Hirshleifer theory.
The evidence
Significant abnormal returns with a Sharpe that exceeds long-only commodities — and statistically distinct from carry and momentum, so it adds real diversification.
Direction and significance verified. Specific performance numbers were refuted in fact-checking, so we don't quote them.
The catch
The popular 'fade the crowd' / speculator-extreme contrarian rule is weak — the financialization-distortion premise was tested and largely unsupported (Sanders-Irwin).
On your data
Directly usable from your weekly CFTC COT z-scores per commodity.
3

Inventory / Theory of Storage

Scarcity premium

Buildable with a proxy
The rule
Favor commodities with low inventories (relative to their own history); avoid those flush with supply. Low stocks → high convenience yield → higher expected return.
Why it works
When inventories are scarce, holders of the physical good earn a convenience yield, which shows up as backwardation and a higher futures risk premium.
The evidence
Low-inventory minus high-inventory ≈ +8.1%/yr (t=3.19), positive in 56% of months (1969–2006).
Figure verified. Note it partly overlaps the carry/momentum factor.
The catch
It's the same family as carry — and a clean cross-commodity inventory panel (days-of-supply) needs harmonizing across the full set.
On your data
EIA crude (since 1982) + nat-gas storage; use a 5-year-average baseline since you don't hold analyst consensus.
4

Nat-Gas Storage + Weather

Fundamentals that move the tape

Buildable on your data now
The rule
Trade natural gas off storage deviations from the 5-year average and temperature shocks — heating/cooling degree days (HDD/CDD) drive demand.
Why it works
Nat-gas returns and volatility are explained by storage announcements and temperature surprises, not price-only signals — a rare clean fundamental link.
The evidence
Storage and HDD/CDD significantly explain weekly storage changes and daily futures returns and variance (Chen, 2023).
Verified. Caveat: your storage history starts in 2010, limiting sample length.
The catch
Short backtest window (2010+); weather is noisy and forecasts decay fast.
On your data
EIA Lower-48 storage (2010+) + your US population-weighted degree-day series.
5

Crude Inventory Surprise

Event reaction

Buildable with a proxy
The rule
React to the EIA weekly crude report relative to expectations: larger-than-expected builds push price down, draws push it up. It's the surprise, not the level.
Why it works
Unexpected inventory changes hit returns inversely and raise volatility immediately — and the EIA report carries more signal than the API number.
The evidence
Significant immediate inverse price reaction to the surprise; EIA shocks are larger and longer-lived than API (Ye & Karali, 2016).
Verified. Part of the move is pre-empted by the prior-day API leak, which complicates timing.
The catch
Needs an expectations baseline you don't have — we'd proxy consensus with a 5-year-average or model.
On your data
EIA crude inventory (since 1982); surprise computed against a 5-year-average proxy.

What the research is honest about

  • The flashy numbers don't survive. We threw out the specific Sharpe/return stats from several papers when fact-checking refuted them — only direction and statistical significance held up. We quote a figure only where it was verified verbatim.
  • Carry, momentum and inventory are the same bet (≈0.87 correlated). Combining carry with hedging-pressure diversifies; stacking carry + momentum + inventory does not.
  • "Fade the crowd" is mostly a myth. The idea that speculator/index positioning distorts prices was tested and largely unsupported. Commercial hedging-pressure is the part with theory and evidence behind it.
  • All of this is in-sample and gross of costs, and these premia have visibly decayed since 2010 with financialization. Historical returns overstate what's live today — which is exactly why we re-test everything on our own data.

See these run on real data

Tara can pull the live readings, or build and backtest a rule yourself.

Educational research, not investment advice. Past performance does not guarantee future results.