Value at Risk (VaR) tells you the maximum loss you would expect under normal conditions. But what happens when conditions are not normal? That is where CVaR — Conditional Value at Risk, also called Expected Shortfall — becomes essential.
Most retail brokerage platforms do not display Expected Shortfall. This is a significant gap, because VaR systematically underestimates the risk that matters most: tail risk.
VaR vs CVaR: A Critical Distinction
VaR (95%) answers: "What is the worst loss I would expect 95% of the time?"
CVaR (95%) answers: "When losses exceed that VaR threshold, how severe do they actually become on average?"
VaR tells you where the danger zone starts. CVaR tells you how deep into the danger zone you are likely to go.
Why Normal VaR Underestimates Tail Risk
Standard VaR calculations assume returns follow a normal (Gaussian) distribution. Market data consistently demonstrates they do not:
- Fat tails: Extreme events occur 3–10x more frequently than normal distributions predict. The 2020 COVID crash, 2008 financial crisis, and 2022 rate shock all exceeded 4-sigma events under Gaussian assumptions — a one-day 4-sigma loss that a normal distribution expects roughly once every 31,500 trading days, or about 125 years.
- Skewness: Returns are not symmetric. Downside moves tend to be larger and faster than upside moves.
- Volatility clustering: Large moves cluster together — a bad day is more likely to be followed by another bad day than a calm one.
A portfolio with 95% VaR of −5% could have tail losses of −8% or −30% — standard VaR cannot distinguish between these scenarios.
A Concrete Example
Consider a portfolio with a 95% VaR of −5%:
- VaR says: "95% of the time, your daily loss will not exceed 5%"
- CVaR says: "In the worst 5% of days, your average loss is 8.5%"
The CVaR reveals that when losses breach the VaR threshold, they are significantly worse — information that VaR alone conceals.
When KlirInvest Uses Student-t Distributions
The standing VaR/CVaR view uses the disclosed Gaussian baseline. When Tail Risk Mode is enabled for Monte Carlo, KlirInvest switches to a Student-t distribution to model fatter tails than the normal baseline.
The practical difference: Gaussian CVaR at 95% might estimate a −7% tail loss. Student-t CVaR for the same portfolio might estimate −11%. The Student-t estimate more closely matches what portfolios actually experience during stress events.
KlirInvest displays your portfolio's Expected Shortfall at 95% confidence next to your Risk Score — providing a clear view of both typical and tail risk in a single dashboard. Expected Shortfall is shown as a separate risk metric; it is not an input to the V3 Risk Score itself, which is a volatility-percentile score.
How to Use CVaR in Practice
- Compare portfolios: A portfolio with lower CVaR has less tail risk, even if VaR is similar
- Set risk budgets: Use CVaR to define the maximum tail risk you are willing to accept
- Detect fragility: If your CVaR is 3x your VaR, your portfolio has fat tails — it is more fragile than standard metrics suggest
- Evaluate changes: Before adding or removing positions, compare the CVaR impact to understand how the change affects tail risk
CVaR Is Mathematically Superior to VaR
CVaR satisfies a property called subadditivity: the risk of a combined portfolio is never greater than the sum of individual risks. VaR does not satisfy this property, which means VaR can produce the paradoxical result that combining two portfolios appears to increase risk — a mathematical inconsistency that makes VaR unreliable for portfolio construction.
This is why institutional risk managers, Basel III banking regulations, and Solvency II insurance frameworks have moved toward Expected Shortfall as the primary risk measure.
See how your portfolio scores — run your free analysis at [KlirInvest](/free-tools?tab=analysis).