Expected shortfall (CVaR)
Expected shortfall, also called conditional value at risk (CVaR), is the average loss on the periods whose loss exceeds the value-at-risk threshold. Where VaR says where the bad tail starts, expected shortfall says how bad the tail is on average.
Senzoukria · Glossary · Updated September 2026
At a glance
- Definition
- Mean loss beyond the VaR at the same confidence level
- Also called
- CVaR, average value at risk, tail loss
- Always
- ≥ VaR at the same level
- Property
- Coherent risk measure (subadditive)
A worked example
Take 100 daily results and look at the five worst: −1,200, −1,500, −1,800, −2,400 and −4,100 dollars. Using the convention that the 95% VaR is the fifth-worst loss, the VaR is 1,200 dollars. The expected shortfall at 95% is the average of the five worst days: (1,200 + 1,500 + 1,800 + 2,400 + 4,100) ÷ 5 = 2,200 dollars.
Now replace the worst day, −4,100, with −8,100. The VaR does not move, because the fifth-worst day is unchanged. The expected shortfall rises to 3,000. That difference is the reason the measure exists: VaR is blind to what happens past its threshold, expected shortfall is not.
Why risk managers prefer it
- It reflects the size of tail losses, which are what break accounts.
- It is subadditive: the expected shortfall of two positions combined never exceeds the sum of their separate values, so diversification is never penalized. VaR does not guarantee this.
- It pairs naturally with scenario thinking: the average of the worst outcomes is a concrete number to size against.
Estimation limits
Expected shortfall averages the rarest observations, so it rests on very few data points. With 100 days and a 95% level, it is the mean of five numbers; one unusual day changes it a lot, and a sample that happens to contain no crisis day understates it. Longer histories, stress scenarios and reshuffling the observed results help, but none of them can produce a loss larger than those in the data unless a model is assumed.
In Senzoukria
The desktop does not compute VaR or expected shortfall; its risk distributions are built on drawdowns. The Gauntlet reports the observed maximum drawdown with the median, the 5th percentile, the 1st percentile and the worst of 10,000 reshuffled orders of the backtest trades, and the prop firm Drawdown zones panel reports how often each level of a drawdown allowance is reached under two reshuffling methods. For a daily expected shortfall, group exported trades by session day and average the worst tail.
Related
In the same section
- Extrinsic value
- Expected move
- Failed auction
- Expectancy
- Fair value
- Exhaustion
- Fair value gap
- Execution report
This page in other languages
Frequently asked questions
- Is expected shortfall the same as the worst loss?
- No. It is the average of the losses beyond the VaR threshold, so it lies between the VaR and the single worst loss. The worst loss is one observation; expected shortfall summarizes the tail.
- Why is my expected shortfall unstable from month to month?
- Because it averages only a few extreme observations. Adding or dropping one bad day changes the average noticeably. Use longer windows and report the sample size alongside the figure.