Module 20 · Advanced
Backtesting & VaR limits
You've built VaR (Module 6). But how do you know it's any good? You backtest it: over the last 250 trading days, count how often the actual daily loss exceeded the 99% VaR. At 99%, you expect a breach roughly 1 day in 100 — so about 2–3 exceptions a year is healthy.
Regulators formalize this with a traffic light: green (0–4 exceptions — the model is fine), yellow (5–9 — questionable), red (10+ — broken). And it has teeth: a worse zone raises the capital multiplier, so a model that understates risk literally makes the bank hold more capital. This is the discipline that keeps risk numbers honest — and it's the last piece: from pricing a single bond all the way to proving the whole risk framework works. Slide “actual risk” above 1× to see an under-calibrated model drift into the red.
🎛 Backtest simulator
Exceptions / 250
2
Basel zone
green
Capital multiplier
3.00×
A risk model is only trusted if it's backtested: over the last 250 days, how often did the actual loss blow through the 99% VaR? You'd expect ~2–3 breaches a year. Basel's traffic light grades it: green (0–4, fine), yellow (5–9, questionable) or red (10+, broken) — and a worse zone raises the capital multiplier, so a bad model literally costs more. Slide “actual risk” above 1× to simulate a model that underestimates, and watch the breaches and zone deteriorate. Educational tool — not investment advice.
Things to try
- • Keep “actual risk” at 1.0× and re-run — a good model mostly stays green (~2–3 breaches).
- • Slide it to 1.6× — the model underestimates risk, breaches pile up, and the zone turns red.
- • Watch the capital multiplier rise with the zone — bad models cost real money.