Skip to content
On this page
  1. Key Takeaways
  2. What It Is
  3. The Intuition
  4. How It Works
  5. Worked Example
  6. Frequently Asked Questions
  7. Common Mistakes
  8. Sources
  9. Disclaimer
← All concepts
RiskAdvanced5 min read

Conditional Value at Risk: What Happens Beyond VaR

Conditional Value at Risk is the average loss you expect on the bad days that breach your Value at Risk threshold. It answers the question VaR refuses to answer: not just "how often" you lose more than X, but "how much" on average when you do.

Key Takeaways

  • Conditional value at risk (CVaR) is the average loss in the worst 5% (or 1%) of days, not just the threshold loss that VaR identifies.
  • Two portfolios with identical VaR can have very different CVaR, one might average $1.1M in tail losses, the other $4M, making them materially different risk profiles.
  • CVaR is a coherent risk measure (sub-additive), meaning diversifying two books can never make total CVaR larger than the sum of the parts, a property VaR lacks.
  • Estimating CVaR reliably at 99% on a 500-day window means averaging only 5 observations, so sample size is a real practical limitation.

Key Takeaways

  • Conditional value at risk (CVaR) is the average loss in the worst 5% (or 1%) of days, not just the threshold loss that VaR identifies.
  • Two portfolios with identical VaR can have very different CVaR, one might average $1.1M in tail losses, the other $4M, making them materially different risk profiles.
  • CVaR is a coherent risk measure (sub-additive), meaning diversifying two books can never make total CVaR larger than the sum of the parts, a property VaR lacks.
  • Estimating CVaR reliably at 99% on a 500-day window means averaging only 5 observations, so sample size is a real practical limitation.

What It Is

Conditional Value at Risk (CVaR), Expected Shortfall (ES), and sometimes Conditional Tail Expectation (CTE) or average value at risk are used interchangeably in most sources. They all refer to the same idea: the mean loss conditional on the loss exceeding the VaR threshold.

For a 95 percent VaR, CVaR at the same 95 percent level is the average of all losses that fall in the worst 5 percent of the distribution. It is a number about the inside of the tail, not the edge of it. Since the global financial crisis, regulators and many practitioners have pushed expected shortfall as the preferred tail-risk measure because it captures severity, not just frequency.

The Intuition

VaR tells you the edge of the cliff. CVaR tells you how far you tend to fall once you go over.

Imagine two portfolios, both with a 95 percent one-day VaR of 1,000,000 USD. Portfolio A, when it breaches that threshold, loses on average 1,100,000 USD. Portfolio B, when it breaches, loses on average 4,000,000 USD. VaR says they are identical. CVaR says one of them is roughly four times more dangerous in a genuine stress event. That distinction is exactly why Basel III's Fundamental Review of the Trading Book replaced VaR with expected shortfall at a 97.5 percent confidence level for market-risk capital.

CVaR also has a property VaR lacks: it is a coherent risk measure, meaning it behaves sensibly when you combine portfolios. Diversifying two books can never make the total CVaR larger than the sum of the individual CVaRs. VaR, which can violate this property, can actually penalise diversification in odd cases.

How It Works

Start with the quantile. Let L denote portfolio loss and alpha the confidence level (0.95, 0.99, and so on).

VaR_alpha = smallest value x such that P(L <= x) >= alpha

Expected shortfall is the conditional mean of losses above that quantile:

ES_alpha = E[ L | L >= VaR_alpha ]

For a continuous loss distribution this can be written as an average of the VaRs across the tail:

ES_alpha = (1 / (1 - alpha)) * integral from alpha to 1 of VaR_u du

For a normal distribution with mean mu and volatility sigma, there is a closed form. Let phi be the standard-normal density and Phi^-1 the inverse CDF:

ES_alpha = -mu + sigma * phi(Phi^-1(alpha)) / (1 - alpha)

The three estimation methods used for VaR carry over directly:

  1. Historical. Sort the losses, identify the worst (1 - alpha) share, and average them.
  2. Parametric. Plug the estimated mean and volatility into the closed-form expression above, assuming a distribution such as normal or Student-t.
  3. Monte Carlo. Simulate paths, compute losses, and average those that exceed the simulated VaR.

Rockafellar and Uryasev showed in 2000 that CVaR can also be written as the solution to a convex optimisation problem, which makes it tractable for portfolio construction. That is a practical advantage VaR does not share.

Worked Example

You have 1,000 simulated one-day losses for a portfolio, sorted from worst to best. You want 95 percent VaR and 95 percent CVaR.

The worst 5 percent is the 50 worst observations. Read off the 50th worst loss: suppose it is 180,000 USD. That is your VaR.

Now average the 50 worst losses. Suppose they range from 180,000 up to 620,000 USD, with a mean of 265,000 USD. That mean is your CVaR.

Two headline takeaways:

  • On 95 percent of days the loss stays under 180,000 USD.
  • On the worst 5 percent of days the average loss is 265,000 USD, and a few of those days hit 600,000 USD or worse.

A risk committee looking only at the first number would underweight how bad the tail actually is. The second number closes that gap.

Frequently Asked Questions

Q: What is conditional value at risk in simple terms? CVaR is the average loss on your worst days, not the threshold loss that VaR marks, but the mean of everything beyond it. If your 95% VaR is $180,000, your CVaR is the average loss across all the days that exceeded that figure.

Q: How does conditional value at risk affect investment decisions? CVaR-based portfolio optimisation leads to smaller positions in fat-tailed or skewed strategies compared to mean-variance optimisation. It also drives Basel III's capital rules for trading books, where expected shortfall at 97.5% has replaced VaR.

Q: What is a real-world example of conditional value at risk? From 1,000 simulated portfolio losses, the worst 50 (the 5% tail at 95% confidence) average $265,000. The VaR threshold is the 50th worst loss at $180,000. CVaR is $265,000, nearly 50% higher than VaR alone suggests.

Q: How can investors use CVaR instead of or alongside VaR? Report both numbers together. VaR tells you the threshold; CVaR tells you the average severity once that threshold is breached. For option-heavy, credit, or alternative-strategy portfolios, the gap between the two is the most informative number in the risk report.

Q: How is conditional value at risk different from VaR? VaR tells you the edge of the cliff. CVaR tells you the average fall once you go over. VaR answers "how often will I lose more than X?" CVaR answers "how much will I lose on average when that happens?" For fat-tailed assets, the difference between the two can be several multiples of the VaR figure.

Common Mistakes

  1. Treating CVaR as a guarantee. It is still an estimate from a finite sample or a model. If the model's tail is wrong, CVaR is wrong in the same direction. Fat-tailed distributions will give very different answers from normal ones on the same data.
  2. Confusing VaR and CVaR because the numbers look similar. For thin-tailed distributions the two can be close, which tempts practitioners to treat them as interchangeable. For fat-tailed distributions they diverge sharply, and that divergence is exactly where CVaR adds value.
  3. Estimating the tail on too little data. CVaR uses only the observations beyond VaR. At 99 percent CVaR on a 500-day window, you are averaging five observations. That is a noisy estimator. Longer windows, bootstrapping, or extreme-value-theory fits help, but the sample problem never goes away.
  4. Ignoring horizon scaling issues. The square-root-of-time rule that people apply to VaR also applies poorly to CVaR when returns are autocorrelated or volatility clusters. For multi-day CVaR, simulate multi-day paths rather than rescaling a one-day number.
  5. Optimising a portfolio on CVaR without out-of-sample tests. CVaR-minimising portfolios can overfit the historical tail. Walk-forward testing and stress scenarios are essential sanity checks before trusting an optimised weight vector.

Sources

  1. CFA Institute Research Foundation. "Risk Management: A Review." https://www.cfainstitute.org/sites/default/files/-/media/documents/book/rf-lit-review/2009/rflr-v4-n1-1-pdf.pdf
  2. Corporate Finance Institute. "Value at Risk." https://corporatefinanceinstitute.com/resources/career-map/sell-side/risk-management/value-at-risk-var/
  3. Basel Committee on Banking Supervision. "Explanatory Note on the Minimum Capital Requirements for Market Risk." BCBS d457. https://www.bis.org/bcbs/publ/d457_note.pdf
  4. Rockafellar, R.T. and Uryasev, S. "Conditional Value-at-Risk for General Loss Distributions." Journal of Banking and Finance, 2002. https://sites.math.washington.edu/~rtr/papers/rtr187-CVaR2.pdf
  5. Bank Policy Institute. "Why is the FRTB Expected Shortfall Calculation Designed as It Is?" https://bpi.com/why-is-the-frtb-expected-shortfall-calculation-designed-as-it-is/

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.

Back to your knowledge path

The IWP Substack

You understand the concept. Now see it applied.

The Investing With Purpose Substack turns ideas like this into research and risk-managed trade plans on real stocks, updated every week.

Read on Substack (opens in a new tab)

Related concepts