The real question behind every VaR problem
FRM candidates rarely struggle because they forgot that VaR is a quantile. They struggle because they apply the wrong measurement framework to the wrong portfolio.
A risk manager might ask:
- What is the likely one-day loss under ordinary market variation?
- What happens if the portfolio is nonlinear and must be fully repriced?
- How bad are losses once the VaR threshold has already been breached?
- What if markets move in a way the recent sample barely contains?
- What if management wants a severe but plausible scenario rather than a probability-based tail metric?
Those are different questions. One metric will not answer all of them well.
flowchart TD
A["Start with the portfolio and the decision"] --> B{"Mostly linear exposures and quick daily estimate?"}
B -->|Yes| C["Delta-normal VaR"]
B -->|No| D{"Can you reprice the current portfolio on historical factor moves?"}
D -->|Yes| E["Historical simulation VaR"]
D -->|No| F{"Need flexible nonlinear revaluation under modeled distributions?"}
F -->|Yes| G["Monte Carlo VaR / ES"]
F -->|No| H["Revisit model design"]
C --> I{"Need more information beyond the cutoff loss?"}
E --> I
G --> I
I -->|Yes| J["Add Expected Shortfall"]
J --> K{"Tail data sparse or unusually heavy?"}
K -->|Yes| L["Consider EVT overlay"]
K -->|No| M["Standard tail estimation may be enough"]
A --> N{"Need severe scenario narrative rather than quantile?"}
N -->|Yes| O["Stress testing"]