Case Study 1: Value at Risk — The Number That Missed the Catastrophe
The Number
Value at Risk (VaR) was the financial industry's primary risk measure from the mid-1990s until the 2008 crisis — and, in modified form, remains in use today. The concept: estimate the maximum expected loss at a given confidence level over a given time period. A VaR of $100 million at 99% confidence over one day means: "There is a 99% chance our daily loss will not exceed $100 million."
The Precision
VaR was calculated to the dollar. It was updated daily. It was reported to boards of directors, regulators, investors, and counterparties. It was used to set trading limits, calculate capital requirements, and determine executive compensation. The number appeared in regulatory filings, annual reports, and risk management presentations with the visual trappings of scientific measurement: charts, tables, time series, confidence bands.
The Failure
On September 15, 2008, Lehman Brothers' VaR was approximately $113 million. The actual loss was the firm's entire equity — approximately $26 billion. The VaR was off by a factor of roughly 230.
But the failure was not a single-day anomaly. Throughout 2007-2008, VaR numbers at major banks consistently understated the losses that actually materialized. JPMorgan, Goldman Sachs, Citigroup, Bear Stearns, and others all experienced losses that their VaR models said were virtually impossible.
The Structural Analysis
The VaR failure was not a failure of mathematics. The math was correct, given its assumptions. It was a failure of precision without accuracy — a case where the precision of the number disguised the inaccuracy of the assumptions.
| Feature | What VaR Implied | What Was Actually True |
|---|---|---|
| Distribution | Returns are normally distributed | Returns have fat tails |
| Correlations | Asset correlations are stable | Correlations spike during crises |
| Historical basis | Past data predicts future risk | Past data can miss unprecedented events |
| Confidence level | 99% means only 1% tail risk | The 1% tail contains most of the catastrophic risk |
| Precision | Risk is known to the dollar | Risk is uncertain by orders of magnitude |
Discussion Questions
- If VaR had been reported as a range ("Our maximum daily loss is somewhere between $50 million and $5 billion") rather than a point estimate, would the crisis have been prevented or mitigated?
- The fat-tail problem was known to specialists before 2008. Why didn't this knowledge prevent the VaR failure? (Hint: think about incentive structures, institutional demand for precise numbers, and the Archer C problem.)
- VaR is still used (in modified form) by regulators and banks. Has the precision problem been adequately addressed?
- Design a risk communication system for a bank board that honestly conveys the uncertainty of financial risk without producing decision paralysis.
References
- Taleb, N. N. (2007). The Black Swan. Random House. (Tier 1 — the definitive critique of normal-distribution risk models)
- Mandelbrot, B. & Hudson, R. L. (2004). The (Mis)Behavior of Markets. Basic Books. (Tier 1 — fat tails in financial data)
- The Basel Committee on Banking Supervision has published extensive documentation on VaR requirements and their evolution. (Tier 2)