Correlated Exposure
Several risks that look independent resolving to the same underlying variable, so they move together at exactly the moment separation would have helped.
Reviewed July 16, 2026Stable
When not to use it
- Where the correlation is known, priced and accepted, which is a position rather than a blind spot.
- As a prediction. Identifying correlated exposure says what would happen together, not whether it will happen.
- Where exposures genuinely are independent, which requires testing rather than assuming in either direction.
Reach for something else instead
- Dependency tracing — follow each exposure to its origin rather than counting exposures, which is the whole method.
- Stress testing against a common shock — model the single event that moves everything, which reveals correlation that per-component analysis conceals.
- -
Further reading
- Bouthillier et al. (2021), Accounting for Variance in Machine Learning Benchmarks — why sources of variation must be separated rather than counted.
- Kleinberg et al. (2016), Inherent Trade-Offs in the Fair Determination of Risk Scores — a case where apparently separate criteria prove jointly unsatisfiable.
Primary sources, listed so you can check the claims on this page rather than take them on trust.
Where people go wrong
- Treating the number of suppliers, sources or models as a measure of resilience.
- Diversifying in the dimension that is easy to measure rather than the one that binds.
- Counting corroborating sources without checking whether they share an ancestor.
At a glance
FieldFoundations
Diagnostic questionwhat single event moves all of these
Three formsshared input, shared counterparty, shared chokepoint
Formal basiscovariance, not count
Related failurecorroboration from sources with one ancestor
DifficultyIntermediate
Where this sits
A starting point. Nothing needs to come before it.
0Levelstarting point
0Needs firstnothing
0Opens upnothing further
1Areastays here
Computed from the prerequisite graph, not assigned. How this works