That gap is the reason expected shortfall exists. Value at risk (VaR) is a smoke alarm threshold: it tells you where the alarm trips. Expected shortfall (ES), also called conditional value at risk (CVaR), is the fire damage report: it tells you how much burns once the alarm is already ringing. In this guide you will learn the definition, the math, a worked example, why Basel regulators switched to it, and the mistakes that trip up even experienced analysts.
The Basel Committee's market risk standard, captured 15 October 2026 from bis.org. It builds the internal models approach on expected shortfall.
What Is Expected Shortfall?
Expected shortfall is the average loss you suffer in the worst slice of outcomes. If you pick a 97.5% confidence level, ES looks at the worst 2.5% of days (or weeks, or months) and tells you the average loss across all of them. Wikipedia's definition puts it plainly: the expected return on the portfolio in the worst q% of cases. You can read the full formal treatment on Wikipedia's expected shortfall page.
Compare that with VaR. VaR at 97.5% gives you one point: the loss you expect to be exceeded on only 2.5% of days. It is a cliff edge. ES is the average height of everything past the cliff. Two portfolios can share an identical VaR and have wildly different ES, because VaR ignores what happens beyond the threshold.
Several names, one idea. Expected shortfall is also called CVaR, average value at risk, tail value at risk, and expected tail loss. These labels are used interchangeably in most practical work, though definitions can differ for odd loss distributions.
Why Expected Shortfall Matters
There are two reasons professionals care. The first is mathematical. In 1999, Artzner, Delbaen, Eber and Heath defined a set of properties a sensible risk measure should have and called measures that satisfy them coherent. VaR can fail the subadditivity test, meaning the VaR of a combined portfolio can be larger than the sum of its parts, which punishes diversification instead of rewarding it. Expected shortfall satisfies all the coherence properties.
The second reason is regulatory. The Basel Committee's Minimum capital requirements for market risk, first published in January 2016 and revised on 14 January 2019, built its internal models approach on expected shortfall instead of VaR. Banks now measure trading-book risk at a 97.5% ES confidence level rather than the old 99% VaR. The standard also scales losses by liquidity horizons of 10, 20, 40, 60 and 120 days, so illiquid positions carry more capital.
2.34 standard deviations is the loss multiplier for 97.5% ES under a normal distribution, almost identical to the 2.33 multiplier for 99% VaR. Regulators changed the measure without changing the calibration for calm markets. The difference only shows up when tails get fat.
Retail investors benefit too. If you size positions or set stop levels with VaR alone, you inherit its blind spot. A portfolio of short-volatility trades can show a tiny VaR for years and then lose everything in a week. ES forces you to ask the second question: when it goes wrong, how wrong.
How to Calculate Expected Shortfall
There are three standard ways to estimate ES, and they mirror the three standard ways to estimate VaR. Pick the one that matches your data and your patience. If you have not read it yet, start with our value at risk guide, because ES builds directly on it.
Step 1: Choose confidence level and horizon
Decide how deep into the tail you want to look. Common choices are 95%, 97.5% and 99%, over one day for trading or ten days for regulatory work. A higher confidence level looks at fewer, more extreme observations, so it is noisier. For a personal portfolio, 95% over one week is a workable starting point.
Step 2: Estimate the tail with one of three methods
Historical simulation sorts your past returns from worst to best, cuts off the worst 2.5% (or 5%), and averages them. With 1,000 daily returns at 97.5%, you average the worst 25. It needs no distribution assumption but is only as good as the history you feed it.
Parametric (variance-covariance) assumes returns follow a distribution, usually normal, and applies a closed-form multiplier. For a normal distribution, ES equals volatility times the normal density at the VaR point divided by the tail probability. The multipliers are 2.06 at 95%, 2.34 at 97.5% and 2.67 at 99%.
Monte Carlo simulates thousands of possible paths from a model, then averages the worst slice. It handles options, correlations and fat tails best, at the cost of more setup. Our guide to Monte Carlo simulation for investors walks through the mechanics.
Step 3: Apply the formula
In loss terms, ES at level alpha is the expected loss given that the loss exceeds VaR at that level. The screenshot below shows the closed-form versions for continuous distributions.
Closed-form ES formulas for continuous loss distributions, captured 15 October 2026 from Wikipedia.
Step 4: Compare ES with VaR at the same confidence
Always report both. The ratio of ES to VaR is a quick fat-tail detector. Under a normal distribution at 99% the ratio is about 1.15. When it climbs toward 1.4 or higher, your tail is heavier than a bell curve assumes.
Worked example. Take a $100,000 portfolio with 1.2% daily volatility and a zero average return. This is an illustration, not a forecast.
Normal vs fat-tailed daily risk on $100,000 (1.2% daily volatility)
Notice the pattern. At 97.5%, the normal and fat-tailed VaR numbers are almost the same ($2,352 versus $2,356), so VaR cannot tell the two worlds apart. ES can: $2,805 versus $3,389, a gap of 21%. By 99% the fat-tailed ES is $4,430, nearly 40% above its own VaR.
$1,074 is the extra loss ES reveals at 99% when you swap the normal assumption for a fat-tailed one ($4,430 minus $3,198 on a $100,000 book).
Real Examples: When the Tail Showed Up
History gives you the cleanest test of any risk measure. Here are four real S&P 500 stress points and what a bell-curve model with 1% daily volatility would have said about each.
Real S&P 500 tail events versus a normal-model verdict
56.8% was the S&P 500's peak-to-trough fall in the global financial crisis, measured on closing prices, and the index needed until March 2013 to recover.
A parametric normal VaR is blind to every row. A historical ES that includes these days would have flagged the danger, which is why risk teams stress-test with ES on windows that contain crises, not on calm stretches. For the slower, cumulative side of the same risk, see our guide to maximum drawdown.
Edge cases and limits of ES
Common Mistakes With Expected Shortfall
Mistake 1: Assuming a normal distribution
The easiest way to calculate ES is also the easiest way to fool yourself. Parametric ES under normality produces a modest number, and markets repeatedly produce days that number says are impossible. If you use the closed-form shortcut, treat it as a floor, not an estimate.
Mistake 2: Treating ES as a worst case
ES is an average of bad outcomes, not the worst one. If your 97.5% ES is $2,805 on a $100,000 book, some days inside that tail will lose far more. Pair it with a stress scenario so you see the single ugliest outcome as well as the average.
Mistake 3: Using too little data at extreme levels
At 99% with 250 daily observations you are averaging two or three days. One new data point can swing the estimate by double digits. Either extend the history or step down to 97.5%, the level Basel chose partly because it is statistically sturdier.
Mistake 4: Ignoring liquidity and horizon
A one-day ES is meaningless for a position you cannot sell in a day. Regulators use horizons from 10 to 120 days for a reason. Match the horizon to how long it would actually take you to exit.
Mistake 5: Reading ES in isolation
ES measures size of tail loss, not reward. Combine it with return-based metrics. The Sortino ratio focuses on downside deviation, and the Calmar ratio relates return to drawdown. Together they give you a fuller risk picture.
MoneyFlock's value at risk guide, captured 15 October 2026. Read it first if VaR is new to you.
Frequently Asked Questions
What is the difference between expected shortfall and VaR?
VaR gives the loss threshold you expect to exceed only a small percentage of the time. Expected shortfall gives the average loss on those exceedance days. VaR is a doorway, ES is the room behind it.
Is CVaR the same as expected shortfall?
In most practical settings, yes. CVaR, expected shortfall, tail value at risk and average value at risk are used interchangeably for continuous loss distributions. Small technical differences appear only for distributions with jumps.
Why did Basel replace VaR with expected shortfall?
The Basel Committee wanted a measure that captures tail risk and rewards diversification. ES does both, which is why the market risk standard published in 2016 and revised in 2019 builds the internal models approach on it, calibrated at 97.5%.
What confidence level should I use for expected shortfall?
Use 97.5% as a sensible default because it is the regulatory standard and gives enough tail observations. Use 95% for small data sets and 99% only with long histories or Monte Carlo.
Can I calculate expected shortfall in a spreadsheet?
Yes. Sort your historical returns, find the worst 2.5% of them, and take the average. With 1,000 returns that means averaging the 25 lowest values. No special software is required.
Key Takeaways
- Expected shortfall is the average loss in the worst slice of outcomes, so it measures the fire, not just the smoke alarm.
- ES is also called CVaR, tail value at risk and expected tail loss.
- It is coherent, rewards diversification and sees beyond the VaR threshold, which VaR cannot do.
- Basel's 2016 market risk standard, revised in 2019, moved bank models to ES at a 97.5% confidence level.
- Under normality, 97.5% ES (2.34 sigma) is nearly the same as 99% VaR (2.33 sigma). With fat tails, ES pulls ahead fast.
- In the worked example, 99% ES on a $100,000 book rises from $3,198 to $4,430 when tails get fatter.
- Always report ES next to VaR, and treat the ES to VaR ratio as a fat-tail alarm.
- Match confidence level, data length and horizon to what you can actually trade.
What to Watch Next
- > Does the ES to VaR ratio on your own portfolio stay near 1.15, or drift toward 1.4 or higher?
- > Do your tail estimates change when your data window includes 2008 and March 2020?
- > Does your 97.5% ES stay inside the loss you can tolerate in a single week?
- > Do correlations between your holdings jump in stress, shrinking diversification when you need it most?
- > Do regulators and index providers keep adopting ES in risk reporting?
Remember the smoke alarm and the fire report. VaR tells you when the alarm trips. Expected shortfall tells you how much is burning. This article is educational and is not financial advice.