Cross-Commodity Correlation vs Causation

The problem

When oil prices rise, corn often rises too. When copper rallies, industrial metals broadly move in sympathy. These correlations are real and widely documented - and they are the natural raw material of commodity market commentary. The temptation is to turn correlation into narrative: corn rose because oil rose via the biofuel mechanism; silver fell as industrial metals tracked copper’s decline; soybeans moved on crude’s gain.

The question MSCIP asks before publishing that narrative is: is the relationship causal? If oil genuinely and reliably causes corn price movements, a subscriber can watch oil to anticipate corn. If the correlation is spurious - driven by a common third factor like USD strength, global demand cycles, or macro risk-off positioning - then using oil as a corn signal is unreliable and potentially misleading.

What the research shows

Granger causality tests and directed-graph analysis of the relationship between oil and eight agricultural commodities (corn, wheat, soybeans, cotton, rice, soybean meal, soybean oil) over 1996–2008 produced findings that are instructive precisely because they complicate the intuitive narrative:

  • Correlations were strong. The oil-corn correlation reached r = 0.93 in the post-2006 biofuel-mandate period. By correlation alone, oil looks like an excellent corn predictor.
  • Directional causation was mixed. Granger causality ran from oil to corn at some lag horizons but not others. For other commodities (wheat, rice), the oil-to-agricultural causation was weaker or absent even when correlation was high.
  • The relationships varied by time period and commodity. There is no stable, universal cross-commodity causal structure that holds across market regimes.

The bottom line: the presence of strong correlation does not establish predictable causation, and treating correlation as causation produces unreliable signals.

The most common misapplication in commodity commentary - attributing a market-wide price move to a single company’s news, a single adjacent commodity’s move, or a loosely stated biofuel “mechanism” - does not meet the evidentiary standard this research establishes.

How MSCIP applies this discipline

In the OSINT brief and Lane 2 content:

  • Cross-commodity co-movements are described as co-movements: “corn rose alongside oil” rather than “corn rose on oil’s gain.”
  • When a specific causal mechanism is invoked (biofuel spread, freight cost pass-through, shared USD exposure), MSCIP names the mechanism explicitly and qualifies it: “Historical corn-oil correlation via biofuel demand is documented but causation varies by market regime.”
  • Single-company news items are not used as attribution for market-wide commodity price moves. Vizsla Silver’s project milestone does not move the silver market; ExxonMobil’s earnings do not move WTI spot.
  • When no identifiable catalyst explains a move, MSCIP reports the observation without forced attribution: “Silver fell $0.62 to $74.85/oz.” Silence on causation is more accurate than invented causation.

In dashboard features:

  • Any MSCIP panel displaying cross-commodity correlations ships at TREND · v1 maximum maturity, never NOWCAST · v1, until the specific relationship clears the §24.1 two-benchmark gate against an uninformed baseline.
  • Cross-commodity forecast features that do not pass the two-benchmark gate are dropped or caveated at the model-spec level, not buried in fine print.

In source attribution:

  • When cross-commodity content is editorially important, MSCIP cites supporting literature or labels the claim as correlational rather than asserting causation.

See also