##  [Conditional Value at Risk](/conditional-value-risk-0) 

 Definition

A tail-risk statistic equal to the expected loss conditional on losses exceeding the Value at Risk (VaR) threshold at a specified confidence level for a given loss distribution and loss definition; it quantifies the average severity of losses in the tail beyond the VaR cutoff.

 

 

 

 

 

 





## Principle

Principle

By conditioning on exceedance of the VaR threshold, CVaR captures average tail severity and thus provides a coherent, convex measure of downside risk that is sensitive to the shape of the loss tail rather than only to a quantile.

 

 

 

 

 





## Demonstration

Demonstration

Illustrative computation: for a portfolio loss distribution and confidence level 97.5%, identify the VaR such that 2.5% of losses exceed it, then compute the mean of losses conditional on being greater than that VaR; this conditional mean is the CVaR (expected shortfall) at 97.5%.

 

 

 

 

## Misapplication

Misapplication

Confusing VaR with CVaR or treating CVaR as a complete description of tail risk without specifying the distributional model, aggregation method, or whether losses are liquidatable; the semantic error is assuming CVaR alone characterizes all relevant tail dependence or capital needs.

 

 

 

 

 





## Consequence

Consequence

CVaR is used for regulatory and internal capital allocation, optimization and stress assessments because it penalizes extreme losses more consistently than VaR; however, its numeric value depends on modeling choices, loss aggregation and the confidence level, affecting capital and portfolio decisions.

 

 

 

 

## Reversal

Reversal

For discrete, bounded or poorly estimated tails, or when loss distributions are subject to model uncertainty, the conditional expectation beyond a quantile may be unstable or undefined under some definitions; alternative tail metrics or robustification may be required.

 

 

 

 

 





## Boundary

Boundary

Applies to loss distributions and requires specification of confidence level, loss definition (e.g., mark-to-market, liquidation), and aggregation method; it does not by itself prescribe allocation rules, regulatory treatment or recovery assumptions.

 

 

 

 

 





## Semantic Tension

Semantic Tension

Conservatism versus model sensitivity: CVaR is more conservative and informative about tail severity than VaR but is more sensitive to tail-model assumptions and estimation error, creating a trade-off between prudence and numeric stability.

 

 

 

 

 





## Synthesis

Synthesis

CVaR provides a principled, expectation-based summary of tail losses beyond a chosen quantile, making it preferable to VaR for optimization and coherent-risk frameworks, but its practical use requires careful specification of model, aggregation and confidence choices and attention to estimation risk.