Expected Shortfall

Expected shortfall (ES) — synonymous with conditional-value-at-risk — is the average loss a portfolio experiences in its worst-case scenarios, specifically the mean of losses when they exceed the value-at-risk threshold. It directly addresses the key weakness of value-at-risk by measuring the magnitude of tail losses, not just the probability.

This entry is a synonym and detailed treatment of expected shortfall. For the VaR threshold it builds on, see value-at-risk; for broader exposure to tail losses, see tail-risk.

Expected Shortfall — key facts

Expected shortfall captures the average loss in tail scenarios.


What it is	Average loss when losses exceed the VaR threshold
Synonyms	CVaR, conditional value-at-risk, expected tail loss
Calculation	Average of worst X% of outcomes (e.g., worst 1% for 99% confidence)
Regulatory standard	Basel III and modern frameworks; replacing VaR
Key insight	VaR tells you the threshold; ES tells you how bad the tail actually is
Advantage	Direct measurement of tail severity; not just a percentile
Mathematical property	ES ≥ VaR; equality only if distribution is thin-tailed

The simple idea behind expected shortfall

Value-at-risk is a threshold: the 99% VaR of $1M means there is a 1% chance of losing more than $1M. But it does not tell you what happens in that 1% tail. The loss could be $1.5M or $10M.

Expected shortfall answers: “If the worst happens (you are in the 1% tail), what is the average loss?” For example, a 99% ES of $1.5M means that when losses exceed $1M, they average $1.5M.

This directly measures tail risk severity, which value-at-risk does not.

How expected shortfall is calculated

Method 1: Empirical (historical)

Sort all past daily returns from worst to best. For 99% expected shortfall on 1,000 days of data:

Identify the worst 1% of days (roughly 10 days).
Calculate the average loss over those 10 days.
That average is the 99% expected shortfall.

Method 2: Parametric

If returns follow a normal distribution with mean μ and standard deviation σ:

The 99% VaR is at x = μ - 2.33σ.
The 99% ES is μ - σ × φ(z) / (1 - Φ(z)), where φ and Φ are the PDF and CDF of the standard normal, and z = 2.33.

This formula is more complex than VaR, but it is standard.

Method 3: Monte Carlo

Simulate thousands of scenarios, calculate portfolio loss in each, sort by loss magnitude, take the average of the worst 1%.

Expected shortfall is always at least as bad as VaR

By definition, ES is the average of tail losses, and VaR is the threshold. The average of tail losses cannot be better (less negative) than the threshold itself.

Mathematically: ES(α) ≥ VaR(α) for any confidence level α.

If the distribution is normal (thin tails), ES ≈ VaR, so the difference is small.
If the distribution has fat tails (which real markets do), ES » VaR, and the tail risk is much worse than VaR implies.

This is why the shift from VaR to ES exposed many institutions to much higher capital requirements in post-2008 regulations. Suddenly, tail risk was being measured directly rather than ignored.

Example: ES reveals hidden tail risk

A portfolio’s returns over 100 days:

Scenario A: Normal-ish returns

Days 1-95: Returns between -0.5% and +0.5%
Days 96-99: Returns of -1.5%, -2.0%, -2.5%, -3.0%
Day 100: Return of -3.5%

95% VaR: The worst 5% is days 96-100, with returns ≥ -3.5%. The fifth-worst day is -1.5%, so VaR is -1.5%. 95% ES: Average of worst 5% = (-1.5 - 2.0 - 2.5 - 3.0 - 3.5) / 5 = -2.5%.

ES is worse, but not dramatically so.

Scenario B: Fat-tail returns

Days 1-95: Returns between -0.5% and +0.5%
Days 96-99: Returns of -1%, -1%, -1%, -1%
Day 100: Return of -30%

95% VaR: The worst 5% includes day 100 (-30%), so VaR is the fifth-worst, which is -1%. 95% ES: Average of worst 5% = (-1 - 1 - 1 - 1 - 30) / 5 = -6.8%.

Here, VaR says -1%, but ES reveals the true tail severity: -6.8%. This is the power of ES: it reveals what VaR hides.

Regulatory adoption of expected shortfall

After the 2008 financial crisis, regulators realized that value-at-risk was systematically too low. Banks held capital for 2-3% daily VaR, but experienced losses of 10%+ in bad months.

Basel III, adopted globally in 2013+, shifted to expected shortfall as the primary risk metric. Banks now calculate:

10-day 97.5% ES (instead of 1-day 99% VaR).
And back-tested against actual daily losses.

This increased capital requirements for banks with fat-tailed portfolios (like trading desks and hedge funds) while leaving those with thin-tailed portfolios largely unchanged.

Practical use of expected shortfall

For a portfolio manager:

A 1-day 99% ES of $2M means: on the worst 1% of days (about 2-3 per year), the portfolio loses an average of $2M.
This directly tells the manager: “I need at least $2M buffer, and probably more, to be safe.”
They can stress-test further to understand the absolute worst case.

For a risk manager:

ES forces explicit modeling of tail risk.
It cannot be dismissed as “unlikely” because it is anchored to the probability (worst 1%) and the magnitude (average of those scenarios).

Limitations of expected shortfall

Despite its superiority to VaR, ES has limitations:

Tail data sparsity. For 99% ES on 1,000 days, only 10 tail observations inform the estimate. Those 10 days might not represent future tails.
Parameter uncertainty. If volatility or correlation estimates are wrong, ES will be wrong.
Does not capture absolute worst case. ES is the average; the absolute worst could be worse.
Black swans still missed. ES reflects historical data or distributional assumptions. A new type of tail event is not captured.
Computational complexity. ES is harder to calculate than VaR, especially for portfolios with complex instruments.

Despite these, ES is the modern standard and is far superior to VaR alone.