Glossary/Risk

Conditional Value at Risk

Also known as CVaR, Expected shortfall, Average Value at Risk

Conditional Value at Risk estimates the average loss in outcomes that are worse than the Value at Risk threshold at a chosen confidence level. It focuses on the tail of the loss distribution and is commonly treated as equivalent to expected shortfall under standard conditions.

Editorially reviewed 2026-07-29

Why conditional value at risk matters

VaR identifies a threshold but ignores the magnitude of losses beyond it. Conditional Value at Risk addresses that gap by averaging tail outcomes, making portfolios with rare catastrophic losses appear riskier than portfolios whose breaches are modest. It is used in portfolio optimization, limits, capital assessment, and comparison of strategies with asymmetric payoffs. The measure encourages attention to the severity, not just frequency, of extreme loss.

How it is applied

An analyst estimates or simulates the portfolio loss distribution, finds the VaR percentile, and averages losses at or beyond that boundary. Historical calculation uses the worst observations, while Monte Carlo and parametric methods rely on modeled scenarios or distributions. The horizon and confidence must match the decision. For optimization, expected returns can be balanced against conditional tail loss, with constraints for liquidity, leverage, and concentration.

Formula

CVaRα = E[L | L ≥ VaRα]
CVaRα
Average tail loss at confidence level α
L
Portfolio loss over the selected horizon
VaRα
Value at Risk threshold at confidence level α

Portfolio example

A portfolio has one-day 95% VaR of $1 million. Across simulated outcomes worse than that threshold, the average loss is $1.8 million, so conditional VaR is $1.8 million. A second portfolio has the same $1 million VaR but $4 million conditional VaR. The second has much more severe modeled tail exposure even though an ordinary VaR report makes them look identical.

How to interpret it

Conditional VaR should normally be at least as large as VaR because it averages losses beyond the cutoff. A rising gap between the two can indicate a heavier modeled tail. Comparisons require the same confidence, horizon, valuation, and scenario methodology. The result remains an estimate, not the maximum possible loss. Users should examine which scenarios populate the tail and whether they represent economically coherent stresses rather than accepting the summary alone.

Limitations and common misconceptions

Tail estimates are data-hungry and highly uncertain because extreme events are scarce. Historical samples may omit relevant crises, while simulations depend on distribution and correlation assumptions. Illiquid prices and nonlinear instruments complicate valuation. Different conventions at a discontinuous percentile can yield slightly different figures. Optimization against conditional VaR may still produce fragile positions if scenarios are incomplete. Reverse stress tests and judgment remain essential.

Sources and further reading

← All glossary terms