Definition
Expected shortfall at confidence level alpha is the average loss in the worst tail beyond that confidence level, expressed with a loss variable. A definition that integrates loss quantiles handles probability mass at the cutoff more cleanly than an informal average strictly beyond VaR. The horizon, confidence level, loss convention and estimation method must be stated.
How it is calculated
Expected shortfall at alpha = one divided by 1 minus alpha times the integral from alpha to 1 of the loss quantile q_u(L).
| Symbol | Meaning | Unit |
|---|---|---|
| ES | expected shortfall | loss units |
| alpha | confidence level | probability |
| L | loss random variable | currency or return loss |
| q | loss quantile function | loss units |
| u | quantile probability index | probability |
-
Identify the loss quantiles from the confidence level through the upper tail.
-
Integrate those tail quantiles.
-
Divide by the tail probability.
-
Interpret the result as the average tail loss under the stated convention.
Worked example
Synthetic example.
-
inputs:
-
losses: 1, 2, 3, 4
-
confidence level: 0.75
-
-
result:
- expected shortfall: 4
Interpretation
VaR marks a tail boundary; expected shortfall asks how bad the tail is on average. That makes it sensitive to the severity of losses after the boundary, not only to the boundary's location. With discrete data, the quantile integral allocates any atom at the cutoff rather than silently dropping it.
Assumptions and conventions
-
Losses are defined so larger L means worse outcome.
-
The confidence level satisfies 0<alpha<1, and the upper-tail expectation is finite.
-
The quantile convention and tail probability are stated.
-
The estimate has enough tail observations or uses an explicitly modeled tail.
Loss units over the stated horizon; alpha and u are probabilities.
Limitations and common misreadings
-
Expected shortfall is not an average return; it is an average loss in a stated tail.
-
A confidence level without a horizon and loss convention is incomplete.
References
- Carlo Acerbi, Dirk Tasche. (2002). Expected Shortfall: A Natural Coherent Alternative to Value at Risk. Economic Notes. Whole article. doi:10.1111/1468-0300.00091. Source

