Value at Risk (VaR) answers one question: How much can I lose, with a given confidence level, over a given time horizon?
It is compact, standardized, and embedded in regulatory frameworks from Basel III to internal risk policies worldwide. But its simplicity is also its biggest risk.
The Three Main Methods
Historical Simulation takes the actual distribution of past returns and asks: what loss would the portfolio have suffered at the 95th or 99th percentile? No distributional assumptions — but it is entirely backward-looking.
Parametric (Variance-Covariance) assumes returns are normally distributed. It is fast and analytically clean, but fat tails in real markets make it systematically underestimate tail risk.
Monte Carlo simulates thousands of future scenarios drawn from assumed distributions (or estimated factor models). The most flexible — and the most computationally expensive.
Why VaR Alone Is Not Enough
VaR says nothing about what happens beyond the confidence threshold. A 99% VaR of €10M means: on 1% of days you lose at least €10M. It could be €11M or €110M — VaR won't tell you.
Expected Shortfall (CVaR) addresses this by averaging losses in the tail beyond the VaR threshold. It is now required under Basel's Fundamental Review of the Trading Book (FRTB).
Practical Takeaways
- Always pair VaR with stress testing and scenario analysis.
- Backtest your model: compare predicted VaR breaches against actual losses.
- Communicate limitations clearly to stakeholders — VaR is a measure of normal market conditions.