Why expected shortfall matters
Unlike value at risk, which identifies a loss threshold, expected shortfall incorporates the magnitude of worse outcomes and is useful for nonlinear or fat-tailed portfolios.
How it is applied
Analysts choose a horizon and confidence level, generate a loss distribution using historical, parametric, or simulation methods, and average losses beyond the relevant quantile. Select horizon and confidence level, estimate the loss distribution, identify outcomes beyond the Value at Risk threshold, and average those tail losses. Historical, parametric, and simulation methods should be stress tested. Risk limits often combine expected shortfall with scenarios and liquidity measures.
Portfolio example
If one-day 97.5% value at risk is $5 million and losses in the worst 2.5% of modeled days average $8 million, expected shortfall is $8 million. At 97.5% confidence, suppose the worst 2.5% of modeled one-day outcomes average a 4 million loss. Expected shortfall is 4 million. The corresponding Value at Risk threshold might be only 2.5 million.
How to interpret it
A larger result indicates more severe modeled tail loss, but comparisons require consistent horizon, confidence, data, valuation, and liquidity assumptions. Expected shortfall estimates average severity after entering the chosen tail, whereas VaR provides a threshold. A larger figure indicates more modeled tail exposure. It remains conditional on horizon, confidence, data, and model.
Limitations and common misconceptions
Tail observations are scarce, correlations change in stress, and models can omit liquidity, default, and regime shifts. It is not a maximum possible loss. Tail observations are scarce, correlations change, and historical samples may omit plausible crises. Parametric assumptions can understate jumps and nonlinear losses. The measure does not show the single worst outcome, liquidity need, or path of losses. Backtesting is harder than for a single quantile because the statistic depends on multiple tail observations. Model governance should compare predicted and realized losses, investigate breaches, and use independent stress scenarios that exceed the historical sample. Aggregation benefits can disappear when correlations approach one. Report both currency amount and percentage of capital so exposure is interpretable across portfolio sizes. Results should always identify the loss horizon and confidence level.
Sources and further reading
- Introduction to Risk ManagementCFA Institute