The Instrument That Enabled Expansion Caused the Collapse
In 1994 JP Morgan released Value at Risk. The VaR was a statistical model that reduced the maximum probable loss of a portfolio over a given time horizon and confidence level to a single number. It was elegant, communicable, and scalable. Banks worldwide adopted it. Regulators made it official. Basel II embedded it into capital requirements. A tool created in a quantitative office on Wall Street had become, within a decade, the cognitive infrastructure of the global financial system.
The problem with the VaR was not technical. It was epistemic. The model assumed that returns followed a normal distribution and that correlations between assets remained stable over time. Both assumptions failed in the conditions that matter, which are crises. In periods of stress markets do not behave as they do in normal times. Correlations converge toward one. Rare events occur far more frequently than a Gaussian curve predicts. The model was precise in measuring ordinary risk and blind to extraordinary risk, the only kind capable of destroying a system.
MacKenzie (2006) showed that financial models do not describe markets from the outside. They transform them from within once adopted at sufficient scale. The VaR followed the same path. When all banks measured risk with the same model, they built similar portfolios, held the same assets, and reacted to the same signals simultaneously. The diversification the model assumed was systematically undermined by its universal adoption.
2008 was the outcome of this structure. It was not an accident, an excess of greed, or a lack of regulation. It was the emergent effect of a system in which the tool used to measure and distribute risk had rendered that risk invisible and concentrated. Structured products, CDOs, and credit default swaps were constructed within the logic of the VaR, calibrated on historical data that excluded systemic crises, and sold to investors who trusted ratings generated by the same models.
When the system collapsed, it did so everywhere simultaneously. Not because everyone had made the same mistakes independently, but because everyone had delegated their decisions to the same instrument, and that instrument had built the same vulnerability on a global scale.
The lesson extends beyond the VaR. It concerns the general relation between models and complex systems. When a measurement tool is adopted universally, it stops measuring the system and begins constructing it. At that point systemic risk is no longer calculable with the tool that created it. An external point of observation becomes necessary. It was missing then and it is missing now.