Home / Risk Metrics / Conditional VaR (Expected Shortfall)
Tail metricAdvanced

Conditional VaR (Expected Shortfall)

Conditional Value at Risk (CVaR), also called Expected Shortfall, is the average loss suffered on the occasions when the loss exceeds the Value at Risk threshold, so it measures the expected severity of the tail that VaR only marks the edge of.

Quick answer: Conditional Value at Risk (CVaR), also called Expected Shortfall, is the average loss suffered on the occasions when the loss exceeds the Value at Risk threshold, so it measures the expected severity of the tail that VaR only marks the edge of.

In simple words

Value at Risk tells you the loss you rarely exceed, but not how bad things get when you do exceed it. Conditional VaR fills that gap: it is the average of all the losses on the bad days beyond the VaR line. If your 95 percent VaR is ₹12,000, the CVaR might be ₹20,000, meaning that on the worst 5 percent of days the average loss is ₹20,000. Because it looks inside the tail rather than at its edge, CVaR is a more honest measure of how much a rare bad outcome can hurt.

Purpose

This page defines Conditional VaR as the average loss beyond the VaR quantile, explains why it captures the tail severity that VaR ignores, and notes both its coherence advantages and its dependence on tail estimation.

Visual explanation

Conditional VaR (Expected Shortfall)

The loss distribution with VaR at the quantile and CVaR as the average of the shaded tail beyond it.

Value at Risk & the Tail (CVaR)VaRCVaRLosses ← 0 → GainsVaR is the threshold; CVaR is the average loss beyond it

Professional explanation

CVaR is the average of the tail, not its edge

Where VaR marks the loss at a quantile, Conditional VaR averages all the losses that lie beyond that quantile. Formally CVaR at a confidence level is the expected loss conditional on the loss exceeding the VaR at that level, E[Loss given Loss is greater than VaR]. Because it integrates over the whole tail rather than reading a single point, CVaR is always at least as large as the corresponding VaR, and usually larger. This single change, from the edge of the tail to the average of the tail, is what makes CVaR sensitive to how catastrophic the rare losses actually are, which is exactly the information VaR discards.

Why VaR's blind spot matters and CVaR fixes it

Two portfolios can share an identical VaR while having wildly different tails: one loses a little more than its VaR on a bad day, the other loses many times its VaR. VaR treats them as equally risky; CVaR does not, because it averages the actual tail losses and so assigns the second portfolio a much higher number. This matters most for payoffs with rare, large losses, such as short options or leveraged positions, where the danger lives entirely in the tail VaR ignores. CVaR was designed precisely to make that hidden tail visible in a single figure.

Coherence and sub-additivity

CVaR is a coherent risk measure in the technical sense: among other properties it is sub-additive, meaning the CVaR of a combined portfolio never exceeds the sum of the parts' CVaRs, so combining positions can only reduce or maintain measured risk, never spuriously increase it. VaR lacks this property in general, which can make it reward concentration or misstate diversification. Sub-additivity matters for aggregating risk across a book and for optimisation, because a coherent measure behaves sensibly when positions are added, and this mathematical soundness is a large part of why regulators and academics favour Expected Shortfall.

Estimation is harder because the tail is data-poor

The strength of CVaR, that it reads deep into the tail, is also its practical weakness: estimating an average over rare events requires data about exactly the outcomes that are, by definition, scarce. A 99 percent CVaR is an average over the worst 1 percent of cases, of which any sample contains few, so the estimate is noisy and sensitive to the handful of extreme observations and to the assumed distribution. Parametric CVaR inherits the normal model's thin tails and understates severity; historical CVaR is hostage to whether the window happened to contain a genuine crash. CVaR is more informative than VaR but not more certain in the deep tail.

Using CVaR alongside VaR and stress tests

In practice CVaR is reported next to VaR rather than instead of it: VaR gives the threshold and its breach frequency, CVaR gives the expected severity beyond it. The regulatory move, embodied in the Basel Fundamental Review of the Trading Book, replaced VaR with Expected Shortfall for market-risk capital for exactly this reason. Even so, CVaR is a model-based expected tail loss, not a guaranteed maximum, so it is still paired with explicit stress tests and scenario analysis that ask what a specific severe event would do. The disciplined reading is that CVaR sharpens the picture of the tail but does not eliminate the uncertainty within it.

Formula

CVaR = E[Loss | Loss > VaR] = the average of all losses that exceed the VaR at the chosen confidence level

E[Loss | Loss > VaR] = the conditional expectation, the average loss given that the loss is greater than the VaR threshold. VaR = the Value at Risk at the chosen confidence level (e.g. 95 or 99 percent) and horizon. In a historical estimate, CVaR is the mean of the losses worse than the VaR quantile (e.g. the average of the worst 5 percent of days for a 95 percent CVaR). CVaR is always at least as large as the VaR at the same confidence, because it averages losses that are all beyond VaR.

VaR vs CVaR on the same portfolio

AspectVaRCVaR (Expected Shortfall)
What it reportsThe loss at the quantileThe average loss beyond the quantile
Tail severityIgnoredCaptured directly
Distinguishes two tails with equal VaRNoYes
Coherent / sub-additiveNo, in generalYes
Estimation in deep tailReads one pointAverages scarce extreme data; noisier

Practical example

Illustrative example (Indian market)

Take the earlier ₹5,00,000 portfolio with a one-day 95 percent VaR of about ₹12,340. Suppose the historical worst 5 percent of days, the days beyond that VaR, average a loss of ₹21,000; that average is the 95 percent CVaR, ₹21,000, roughly 1.7 times the VaR. So while VaR says most bad days stay within about ₹12,300, CVaR says that when a bad day does breach it, the typical loss is around ₹21,000. If the portfolio were instead a short-option book with the same VaR but a fatter tail, its CVaR might be ₹45,000, exposing a danger the identical VaR completely masked. This is why comparing two strategies on VaR alone can be misleading and why the CVaR reveals which one truly carries the heavier tail.

A retail trader selling out-of-the-money Bank Nifty options can show a modest VaR because losses are small on most days, but the CVaR is far higher because a sharp expiry or event move produces losses many times the premium collected. The gap between a comfortable VaR and an alarming CVaR is the quantitative signature of a strategy that is short the tail.

Advantages

  • Measures the average severity of losses beyond VaR, not just the threshold
  • Distinguishes portfolios that share a VaR but differ in tail depth
  • Coherent and sub-additive, so it behaves sensibly when risks are combined
  • Exposes short-tail payoffs such as naked options that VaR flatters
  • Adopted by the Basel framework as Expected Shortfall for market-risk capital

Limitations

  • Blind spot: it is an estimated expected tail loss, not a guaranteed maximum, and a true outlier can still exceed it
  • Estimating an average over rare events is data-poor and noisy in the deep tail
  • Parametric CVaR inherits the normal model's thin tails and understates severity
  • Historical CVaR depends on whether the sample window contained a real crash
  • More complex to compute and communicate than a single VaR number

Why it matters in practice

  • Reveals tail-heavy strategies that a comfortable VaR would hide
  • Its coherence makes it the sounder basis for aggregating portfolio risk

Common mistakes

  • Treating CVaR as a guaranteed maximum loss rather than an expected tail average
  • Using parametric CVaR and assuming it captures fat tails it actually understates
  • Trusting a historical CVaR whose window never contained a genuine crash
  • Ignoring CVaR and judging a short-option strategy on its flattering VaR
  • Reporting CVaR without its confidence level and horizon
  • Assuming a low VaR implies a low CVaR when the tail can be much fatter

Professional usage

Modern risk practice favours Expected Shortfall, and the Basel Fundamental Review of the Trading Book replaced VaR with it for market-risk capital because it captures tail severity and is coherent. Desks report CVaR beside VaR, use it to flag strategies that are short the tail, and prefer it for risk-based optimisation because sub-additivity makes it well behaved. They still treat CVaR as a model-based estimate of the tail rather than a certainty, pairing it with stress tests that ask what specific severe scenarios would cost.

Key takeaways

  • CVaR (Expected Shortfall) is the average loss when the loss exceeds VaR
  • It captures the tail severity that VaR, a single quantile, ignores
  • CVaR is always at least as large as the VaR at the same confidence
  • It is coherent and sub-additive, but still an estimate, noisy in the deep tail

Frequently asked questions

What is Conditional VaR?
Conditional VaR, or Expected Shortfall, is the average loss on the occasions when the loss exceeds the Value at Risk threshold. Where VaR marks the edge of the tail, CVaR averages the losses inside it, so it measures how bad the bad days actually are on average.
How is CVaR different from VaR?
VaR reports the loss at a quantile, a single threshold rarely breached. CVaR reports the average of all losses beyond that threshold. So VaR tells you where the tail begins and CVaR tells you how deep it goes, which is why CVaR is always at least as large as the VaR.
What is the formula for CVaR?
CVaR is the expected loss conditional on the loss exceeding VaR, written E[Loss given Loss greater than VaR]. In a historical estimate it is simply the average of the losses worse than the VaR quantile, for example the mean of the worst 5 percent of days for a 95 percent CVaR.
Why is CVaR always larger than VaR?
Because CVaR averages losses that are all, by definition, beyond the VaR threshold. Since every value in that average exceeds VaR, their mean must be at least as large as VaR, and it is strictly larger whenever any tail loss exceeds the VaR level, which is normally the case.
Why do regulators prefer Expected Shortfall?
Because VaR ignores the severity of losses beyond the threshold and is not sub-additive, while Expected Shortfall captures tail severity and is coherent. The Basel Fundamental Review of the Trading Book replaced VaR with Expected Shortfall for market-risk capital for these reasons.
What does it mean that CVaR is coherent?
A coherent risk measure satisfies properties including sub-additivity, meaning the risk of a combined portfolio never exceeds the sum of the parts. CVaR is coherent; VaR is not in general. Coherence makes CVaR behave sensibly when positions are added and in risk-based optimisation.
Is CVaR the worst-case loss?
No. CVaR is the average of the losses beyond VaR, not the maximum. An individual outlier can be worse than the CVaR, because CVaR is an expected tail loss, not a cap. It is more honest than VaR about the tail but still an estimate, not a guarantee.
Why is CVaR harder to estimate than VaR?
Because it averages over rare tail events, and there are few observations that far into the tail. The estimate is sensitive to the handful of extreme data points and to the assumed distribution, so CVaR is more informative than VaR but not more precise in the deep tail.
How does CVaR expose risky option-selling strategies?
A short-option book can show a modest VaR because most days lose little, but its CVaR is high because a rare adverse move produces losses many times the premium. The large gap between a comfortable VaR and an alarming CVaR is the signature of a strategy that is short the tail.
Does a low VaR mean a low CVaR?
Not necessarily. Two portfolios can share a VaR while one has a much fatter tail and therefore a much higher CVaR. Judging risk on VaR alone can hide a dangerous tail, which is exactly why CVaR is reported alongside it.
What confidence level is used for CVaR?
Commonly 97.5 or 99 percent; the Basel Expected Shortfall standard uses 97.5 percent. As with VaR, the confidence level and horizon must be stated for a CVaR figure to be meaningful, and higher confidence pushes the average further into the sparse tail.
Is CVaR the same as Expected Shortfall?
Yes, for practical purposes. Conditional VaR, Expected Shortfall and expected tail loss all refer to the average loss beyond the VaR quantile. Minor technical differences arise for distributions with jumps, but in common usage the terms are interchangeable.
Should I use CVaR instead of VaR?
Use both. VaR gives the threshold and its breach frequency in a simple form; CVaR gives the expected severity beyond it. Reported together they give a fuller picture than either alone, and CVaR should never be the only measure, since stress tests remain essential.
Can CVaR be computed from historical data?
Yes. Historical CVaR is the average of the actual losses worse than the historical VaR quantile, requiring no distributional assumption. Its reliability depends on whether the sample window contained genuine stress, since it cannot average tail events that never occurred in the data.

Voice search & related questions

Natural-language questions people ask about Conditional VaR (Expected Shortfall).

What is Conditional VaR in simple terms?
It is the average loss on your worst days, the ones beyond your Value at Risk line. It tells you how bad the bad days really get, not just where they start.
How is CVaR different from VaR?
VaR tells you a loss you rarely cross. CVaR tells you the average loss when you do cross it. CVaR looks inside the tail; VaR only points at its edge.
Is CVaR always bigger than VaR?
Yes. Since it averages losses that are all worse than the VaR, its number is always at least as big, and usually bigger.
Why do regulators like Expected Shortfall?
Because it captures how deep the tail goes, which VaR ignores, and it behaves sensibly when you combine positions. Basel moved to it for those reasons.
Does CVaR show the danger in selling options?
Yes, very well. Option selling can look calm on VaR but shows a much higher CVaR, because the rare loss is many times the premium you collected.
Is CVaR my worst possible loss?
No. It is the average of your bad-day losses, not the maximum. A single extreme event can still be worse than your CVaR figure.

Sources & references

    Last reviewed 12 July 2026. Educational content only — not investment advice. Markets and rules change; verify current conventions with SEBI, NSE/BSE and your broker.

    ← Previous
    Value at Risk (VaR)
    Educational content only — not investment advice. Examples use illustrative numbers and simplified models. Risk-management techniques reduce but never remove risk, and trading derivatives involves substantial risk of loss. See our Risk Disclosure and SEBI Disclaimer.