Value at risk
VaR answers: what is the minimum loss we expect with probability p over horizon T? Example: “1-day 5% VaR of $185,000” means on 5% of days the loss should be at least $185,000.
Three methods
- Parametric (variance-covariance): assume normal returns. VaR = (z × σ − μ) × value. Simple; needs means, volatilities and correlations; poor for options and fat tails.
- Historical simulation: apply past daily returns to today’s portfolio and read off the loss at the chosen percentile. No distribution assumed; depends on the window chosen.
- Monte Carlo: simulate thousands of outcomes from assumed distributions; flexible, handles options, computationally heavy and model-dependent.
Extensions
- Conditional VaR (CVaR, expected shortfall): average loss beyond VaR.
- Incremental VaR: change in VaR from adding or removing a position.
- Marginal VaR: change in VaR from a small change in a position; used for risk budgeting.
Limitations
- Sensitive to method and parameters.
- Underestimates tail risk if returns are fat-tailed.
- Calm periods produce low VaR just before crises.
- Ignores liquidity: positions may not be closable at quoted prices.
Sensitivity measures
- Equities: beta.
- Bonds: duration and convexity (and credit spread duration).
- Options: delta, gamma, vega.
Scenario analysis and stress tests
- Historical scenarios: replay events such as 2008 or 2020.
- Hypothetical scenarios: designed shocks, such as a 200bp rate rise with a 20% equity fall.
- Reverse stress tests: find scenarios that would break the firm.
Constraints and limits
- Risk budgeting: allocate total risk across desks or managers.
- Position limits, liquidity limits, VaR limits, scenario limits, stop-loss limits.
- Capital allocation based on risk-adjusted returns.
Different users emphasize different measures: banks (liquidity, leverage, VaR), asset managers (tracking error, drawdown), pension funds (surplus at risk), insurers (solvency).
Exam focus
Compute parametric VaR, compare the three methods, explain CVaR, incremental and marginal VaR, and choose stress tests and limits for a situation.