Everyone Using the Same Model
Different Institutions, Correlated Behaviour
Single-firm risk management assumes your decisions are yours. Increasingly they are not. Institutions buy models from the same small set of providers, license the same market and alternative data, reference the same benchmark indices, and now build on the same handful of general-purpose models. Two firms running genuinely independent processes over shared inputs will make correlated decisions with no coordination at all, and neither can see it from the inside. The concentration is not only in models: it runs through data vendors, hosting, ratings inputs and the pricing sources that feed valuation. A common dependency is a common failure mode, and it does not appear in a risk report that stops at the institution's boundary.
- Shared providers, shared data and shared benchmarks produce correlated decisions without coordination
- The dependency map that matters is your providers' providers, and few firms have ever drawn it
- A common dependency is a common failure mode, invisible from inside any single firm
- A single-firm risk view has no vantage point from which the correlation is observable
Crowding, Herding and the Unwind
The market expression of this is crowding: many participants holding similar positions arrived at by similar reasoning. While conditions are calm, crowding is invisible and feels like confirmation that the signal works. Under stress it becomes the mechanism of the loss, because the same participants try to reduce the same exposures at the same time and the exit is narrower than the entrance. Quantitative equity strategies have already demonstrated the pattern: deleveraging in similar portfolios produced losses that forced further deleveraging, across strategies that were not correlated by construction and were badly correlated in practice. Which assumption fails first is usually liquidity, because it was measured in conditions that no longer apply.
- Crowding is invisible in calm conditions and reads as evidence that the signal is real
- Correlated exits are the loss mechanism — the door is narrower than the entrance was
- Strategies uncorrelated by construction can be tightly correlated in practice through shared positioning
- Liquidity assumptions measured in normal conditions are the first thing to break
The Systemic Dimension
Supervisors and international bodies watch this for reasons no single firm's risk report can capture: procyclical behaviour amplified by automation, herding from common models and data, concentration in a small number of critical third parties, and the speed at which automated systems act relative to human intervention. The implication for one institution is modest but real. Find out which external dependencies you share with your peers, because that answer is not on your own risk register. Ask whether your models would behave like everybody else's under stress. Keep a fallback that does not rest on the same provider. And carry third-party concentration as a named risk rather than as a procurement matter.
- Automation can amplify procyclicality and compress the time available for human intervention
- Concentration in a few critical third parties is a supervisory concern, not only a vendor question
- Which dependencies you share with peers is decision-relevant and is not on your own risk register
- Keep a fallback that does not rest on the same provider, and name concentration as a risk
Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.