Expected Shortfall: Average Loss Beyond VaR

Key Takeaways

• Expected shortfall is the average of all losses worse than the VaR cutoff, not just the threshold.
• It is a coherent risk measure, so it rewards diversification, unlike VaR in general.
• The common error is reading VaR as a worst case, when expected shortfall reveals the deeper tail.
• Basel adopted expected shortfall over VaR for trading-book capital because it captures tail depth.

What It Is

Value at risk answers, "How bad is the loss at the edge of my tail?" Expected shortfall answers the next question, "If I land in that tail, how bad is it on average?" It is the mean of all losses that exceed the VaR level at a given confidence.

Rockafellar and Uryasev formalized CVaR and showed it can be computed and optimized with linear programming, which made it practical for portfolio construction. The term expected shortfall is now standard in regulation and risk management, and CVaR, tail VaR, and expected shortfall refer to the same idea for continuous loss distributions.

The Intuition

VaR has a famous blind spot. Two portfolios can share the same 99 percent VaR while one loses a little more beyond it and the other loses catastrophically more. VaR cannot tell them apart because it only reports the threshold, not the depth.

Expected shortfall fixes this by averaging everything in the tail. It also has a property VaR lacks in general: it is subadditive, meaning the risk of a combined portfolio is never greater than the sum of its parts. That makes expected shortfall a coherent risk measure and means it always rewards diversification, which is why regulators and quantitative managers prefer it.

How It Works

For a confidence level alpha (say 95 percent), expected shortfall is the average loss over the worst (1 minus alpha) fraction of outcomes:

ES_alpha = average of all losses that exceed VaR_alpha

In a historical or simulated sample, the steps are direct.

1. Compute VaR at the chosen confidence level.
2. Collect every outcome with a loss worse than VaR.
3. Take the average of those tail losses.

For 95 percent expected shortfall on 1,000 sorted outcomes, you average the worst 50. Under a normal distribution there is also a closed form, where expected shortfall equals volatility times the height of the normal density at the z-score, divided by (1 minus alpha). Because expected shortfall always sits below VaR in the loss distribution, it is always a larger loss number than the matching VaR.

Worked Example

You hold a 1 million dollar portfolio and run 1,000 simulated 1-day outcomes. Your 95 percent VaR is the 50th worst result, a loss of 24,000 dollars.

To get expected shortfall, you do not stop at the 50th worst. You average all 50 of the worst outcomes, from the 1st worst through the 50th worst.

95% VaR            = 24,000 dollars (the threshold)
95% Expected Shortfall = average of worst 50 losses = 38,500 dollars

The expected shortfall of 38,500 dollars is far larger than the 24,000 dollar VaR. That gap is the information VaR hides. It tells you that when a bad day arrives, the typical loss is closer to 38,500 dollars, which is the number a risk manager should size capital against.

Common Mistakes

1. Treating VaR and expected shortfall as interchangeable. VaR is a single threshold. Expected shortfall is the average beyond it and is always the larger, more conservative figure.
2. Estimating the tail with too few observations. Expected shortfall averages the extreme tail, so it needs more data than VaR to be stable. A short window gives a jumpy estimate.
3. Assuming normality. A normal-curve expected shortfall understates real tail depth because returns are fat tailed. Pairing it with extreme value theory gives a sturdier number.
4. Ignoring that it is harder to backtest. VaR is easy to backtest by counting breaches. Expected shortfall is a conditional average, which complicates statistical validation.
5. Confusing it with maximum loss. Expected shortfall is the average of the tail, not the worst possible outcome. Losses can still exceed it.

Frequently Asked Questions

What is expected shortfall CVaR in simple terms? Expected shortfall, or CVaR, is the average loss you would expect on the worst days beyond the VaR cutoff. Where VaR marks the edge of the bad tail, expected shortfall tells you how deep that tail typically runs.

How does expected shortfall affect investment decisions? It gives a fuller picture of downside than VaR, so managers size capital and limits against the average tail loss rather than a single threshold. Because it is coherent, it can be minimized directly in portfolio optimization.

What is a real-world example of expected shortfall? A 1 million dollar portfolio has a 95 percent VaR of 24,000 dollars, but averaging the worst 50 of 1,000 outcomes gives a 95 percent expected shortfall of 38,500 dollars. The larger figure reflects the depth of the tail.

How can investors use expected shortfall effectively? Report it alongside VaR at the same confidence level, use a long data window, and combine it with fat-tailed models so the tail average is not understated.

How is expected shortfall different from conditional VaR? They are the same measure for continuous loss distributions. Conditional VaR, CVaR, tail VaR, and expected shortfall are different names for the average loss beyond the VaR threshold.

Sources

1. Rockafellar, R.T. & Uryasev, S. Conditional Value-at-Risk for General Loss Distributions. SSRN. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=267256
2. Basel Committee on Banking Supervision. Minimum Capital Requirements for Market Risk. https://www.bis.org/bcbs/publ/d457.htm
3. CFA Institute. Measuring and Managing Market Risk. https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/measuring-managing-market-risk
4. Investopedia. Value at Risk (VaR). https://www.investopedia.com/terms/v/var.asp

Disclaimer

This article is educational content only and is not financial advice. Nothing here is a recommendation to buy, sell, or hold any security. Consult a licensed advisor before making investment decisions.