Model risk
What happens when the assumptions a risk model rests on are wrong, and how do the resulting losses show up?
In plain words
Model risk is the chance of loss caused by a model whose assumptions, inputs or limits do not match how markets actually behave. The model can be internally consistent and still describe a world that never arrives, leaving positions sized for risks that differ from the ones that occur.
Why it matters
Risk limits, margin buffers and position sizes are usually derived from models of volatility, correlation and liquidity. When those estimates are too small or too rigid, the real market delivers moves larger or differently shaped than the model allowed for, so losses come from the sizing rule itself rather than from the market direction alone.
In a real week: BTC, week of 1 June 2020
In the week of 1 June 2020, BTC's realised volatility was 65.7% a year, yet its largest single hourly move was -5.66% at 14:00 UTC on Tuesday 2 June, inside a week whose range spanned 12.02% from a high of 10380.0 to a low of 9266.0. A model using the 65.7% figure to set hourly stop distances would have been overtaken by that hour. The weekly close of 9746.99, up 3.16%, hides the intra-week path that the model failed to capture.
A common mistake
Treating a model's output as a fact about the market rather than an estimate built on assumptions that can break, so that a calm model reading is read as evidence that calm conditions are guaranteed.