1. Definition
- Value-at-Risk (VaR) is a quantile-based risk measure that estimates the worst expected loss over a given time horizon at a specified confidence level.
Formally:
For confidence level $\alpha$ (e.g., 95% or 99%), $VaR_\alpha = – \inf \{x : P(L \leq x) \geq \alpha \}$
where $L$ = loss distribution.
Intuition:
- “At most, we expect to lose VaR amount with probability $1-\alpha$.”
2. Example
- A bank holds a portfolio.
- 1-day 95% VaR = \$10M.
- Interpretation:
- With 95% probability, daily losses will not exceed \$10M.
- With 5% probability, losses may exceed \$10M.
3. Key Components
- Confidence level: usually 95% or 99%.
- Time horizon: e.g., 1 day, 10 days.
- Loss distribution: estimated using historical data, variance–covariance, or simulation.
4. Methods to Calculate VaR
Parametric (Variance–Covariance) Method
- Assume returns are normally distributed.
- If portfolio return $N(\mu, \sigma^2)$, $VaR_\alpha = \mu – z_\alpha \sigma$ where $z_\alpha$ is the standard normal quantile (e.g., 1.645 for 95%).
Historical Simulation
- Use actual historical returns.
- VaR = empirical quantile of losses at level $1-\alpha$.
Monte Carlo Simulation
- Simulate many possible portfolio return paths from a model.
- Compute loss distribution and take quantile.
5. Limitations
- VaR gives only a threshold, not the size of losses beyond it.
- Example: if VaR = \$10M, it does not tell you if the actual loss could be \$11M or \$100M.
- Not subadditive → risk of portfolio may appear larger than sum of components.
- Led to development of Expected Shortfall (CVaR), which looks at the average loss beyond VaR.
6. Applications
- Banking regulation: Basel Accords use VaR for capital requirements.
- Risk management: portfolio risk, trading desks.
- Stress testing: understanding downside risk exposure.
Summary:
- VaR (Value-at-Risk) = quantile of loss distribution at a chosen confidence level.
- Answers: “How much could I lose, with X% confidence, over Y time horizon?”
- Calculated via parametric, historical, or simulation methods.
- Limitation: ignores losses beyond threshold → often complemented by Expected Shortfall (CVaR).
