Expected shortfall
Also called ES, conditional value at risk, CVaR
The average loss in the worst cases beyond the value-at-risk threshold. Basel rules for bank trading books use it at a 97.5% confidence level because it captures tail losses that VaR misses.

On this page· 5 sections
Value at risk draws a line and says losses should stay on the safe side of it on most days. It is silent about the days they do not. Expected shortfall answers that question: when the line is crossed, how bad is the loss on average?
The Basel Committee defines it as the average of all potential losses exceeding the VaR at a given confidence level[1]. Acerbi and Tasche put the same idea as the “average of the 100p% worst losses” in a sample of returns, where p is the confidence level[3].
How it works
With a history of results, the steps are:
- Sort the results from worst loss to best.
- Take the worst slice. At 99% confidence, that is the worst 1% of observations.
- Average the losses in that slice.
The VaR figure marks the edge of the slice. Expected shortfall is the mean inside it.
Basel’s trading-book rules fix the settings. Banks using internal models must compute expected shortfall daily, bank-wide and for each trading desk, at a 97.5th percentile, one-tailed confidence level[1]. The calculation starts from a base horizon of 10 days and scales it up for risk factors that take longer to exit, with liquidity horizons running from 10 to 120 days[1]. It is calibrated to the most severe 12-month period of stress available in the data[1].
The reason for the switch is stated in the Committee’s own summary: expected shortfall makes up for VaR’s shortcomings in capturing the risk of extreme losses[2]. Acerbi and Tasche add a mathematical point. VaR fails sub-additivity, the property that a combined portfolio risks at most the sum of its parts, so the VaR of a combined portfolio can exceed the sum of the VaRs of its parts. They present expected shortfall as a coherent alternative[3].
Example
Say two portfolios each have 1,000 daily results, and the 99% VaR of both is $1,000. That is the tenth-worst loss, one matched or exceeded on 10 days.
- Portfolio A: the ten worst losses are $1,350, $1,300, $1,300, $1,300, $1,250, $1,200, $1,150, $1,100, $1,050 and $1,000. They add up to $12,000, so the expected shortfall is $1,200.
- Portfolio B: the ten worst are $7,600, $5,000, $4,500, $3,300, $2,600, $2,000, $1,600, $1,300, $1,100 and $1,000. They add up to $30,000, so the expected shortfall is $3,000.
VaR cannot tell the two apart. Expected shortfall shows that B’s bad days cost two and a half times as much.
Watch out
It is an average, not a ceiling. In portfolio B the worst day was $7,600 against an expected shortfall of $3,000.
The number depends on the confidence level. A 97.5% figure averages a wider slice than a 99% figure, so the two are not interchangeable[1].
The tail holds few observations. In the example, a 1,000-day history leaves only ten days to average, so one new extreme day moves the result noticeably.
Sources
- International bodyBank for International Settlements (Basel Committee on Banking Supervision)Minimum capital requirements for market risk (BCBS d457, January 2019)Checked November 9, 2026
- International bodyBank for International Settlements (Basel Committee on Banking Supervision)The market risk framework: In brief (revised market risk framework, January 2019)Checked November 9, 2026
- ResearcharXiv (Cornell University)Expected Shortfall: a natural coherent alternative to Value at Risk (Carlo Acerbi, Dirk Tasche; Economic Notes 31(2), 2002)Checked November 9, 2026
