Value at Risk (VaR)

VaR compresses a portfolio’s risk into one number a committee can approve, which is why it conquered the industry. The compression is also the flaw. A model that promises “you will not lose more than X, 99 days in 100” is silent about the hundredth day — and the hundredth day is the only one that ends firms.

The deeper problem is not the tail; it is the feedback loop. VaR is measured from recent market behaviour. Calm markets produce low measured risk, low measured risk frees up risk budget, and freed-up budget becomes leverage — across every institution running the same class of model at once. When a shock finally lands, the loop runs backwards: volatility jumps, every model demands de-risking simultaneously, and the selling itself produces the correlated, gap-down market that the bell-curve assumptions said should almost never occur. The model does not merely miss the crash; it helps synchronise it. This mechanism — forced deleveraging — is why markets sometimes fall with no news attached, and why the professionals can look strangely passive near the bottom: their models are the ones doing the selling.

None of this is an argument against measuring risk. It is an argument against outsourcing judgement to a single number — and it is directly relevant to retail investors, because the margin engine in a broker’s app is a cousin of the same model. The LTCM article below is the canonical case study, from the mathematics to the bailout.