Folio Lab/Documentation

Conditional Value at Risk (CVaR)

Conditional Value at Risk, also known as Expected Shortfall (ES), measures the expected loss in the worst-case scenarios beyond the VaR threshold. It is a coherent risk measure that captures tail risk and has been adopted by the Basel III framework as the primary regulatory risk metric.

Overview

CVaR addresses the most significant shortcoming of Value at Risk: its inability to describe the severity of losses beyond the VaR threshold. While VaR answers "What is the minimum loss in the worst of cases?", CVaR answers "What is the average loss in those worst cases?"

The concept was formalized by Artzner et al. (1999) in their foundational paper on coherent risk measures, and its computational properties were developed extensively by Rockafellar and Uryasev (2000, 2002). Their key contribution was showing that CVaR minimization can be reformulated as a linear programming problem, making it tractable for large-scale portfolio optimization.

In 2019, the Basel Committee on Banking Supervision mandated the transition from VaR to Expected Shortfall (at the 97.5% confidence level) for internal models under the Fundamental Review of the Trading Book (FRTB). This regulatory shift reflects the consensus that CVaR provides a more complete picture of tail risk.

Mathematical Formulation

Definition

CVaR at confidence level is the conditional expectation of losses exceeding the VaR threshold:

Equivalently, CVaR can be expressed as the average of all losses beyond the quantile of the return distribution. For a continuous distribution, this is the tail conditional expectation.

Parametric CVaR (Normal Distribution)

When portfolio returns are normally distributed with mean and standard deviation , CVaR has a closed-form solution:

where is the standard normal probability density function (PDF) and is the -quantile of the standard normal distribution. Note that the ratio is always greater than , ensuring that CVaR always exceeds VaR.

Integral Representation

CVaR can also be expressed as the average of all VaR values from the confidence level to 1:

This representation is particularly useful for understanding CVaR as a "weighted average of tail VaRs" and for proving its coherence properties.

Rockafellar-Uryasev Formulation

A key computational result is the optimization-based representation that enables efficient CVaR minimization:

where . The minimizer equals . This formulation converts CVaR minimization into a convex optimization problem that can be solved via linear programming when applied to discrete scenarios.

Coherence Property

A risk measure is called coherent if it satisfies four axioms. CVaR satisfies all four, while VaR fails subadditivity:

1. Subadditivity

Diversification should not increase risk. CVaR satisfies this property, meaning that the CVaR of a combined portfolio is always less than or equal to the sum of the individual CVaRs. VaR famously violates this condition, making it possible for merging two portfolios to appear riskier than holding them separately.

2. Monotonicity

If portfolio X always has worse outcomes than portfolio Y, then X should be measured as riskier.

3. Positive Homogeneity

Doubling the position doubles the risk.

4. Translation Invariance

Adding a certain cash amount reduces risk by exactly that amount.

CVaR vs VaR Comparison

Value at Risk

  • Reports the threshold loss at a given confidence level
  • Ignores the severity of tail losses
  • Not subadditive (not a coherent risk measure)
  • Can be manipulated by restructuring tail exposure
  • Easier to compute and backtest

CVaR / Expected Shortfall

  • Reports the average loss beyond the VaR threshold
  • Fully accounts for tail severity
  • Subadditive (coherent risk measure)
  • Resistant to tail manipulation
  • Can be optimized via linear programming

For a given confidence level , CVaR is always greater than or equal to VaR: . The gap between the two increases with the heaviness of the distribution's tails.

Advantages & Limitations

Advantages

  • Coherent risk measure: Satisfies all four axioms including subadditivity, ensuring consistent risk aggregation across portfolios.
  • Tail sensitivity: Captures the full severity of tail losses, not just the threshold, providing a more complete risk picture.
  • Optimization-friendly: The Rockafellar-Uryasev formulation enables CVaR minimization via linear programming.
  • Regulatory alignment: Adopted by Basel III/FRTB as the primary internal model risk metric.

Limitations

  • Estimation difficulty: Requires accurate estimation of the tail of the distribution, which is inherently data-scarce.
  • Backtesting challenges: Unlike VaR exceedances, CVaR is harder to backtest since it involves averaging over rare tail events.
  • Higher variance: Estimates of CVaR have higher sampling variability than VaR estimates at the same confidence level.
  • Less intuitive:The concept of "average loss in the tail" is harder to communicate than a simple threshold.

References

  1. Artzner, P., Delbaen, F., Eber, J.-M., & Heath, D. (1999). "Coherent Measures of Risk." Mathematical Finance, 9(3), 203-228.
  2. Rockafellar, R. T., & Uryasev, S. (2000). "Optimization of Conditional Value-at-Risk." Journal of Risk, 2(3), 21-42.
  3. Rockafellar, R. T., & Uryasev, S. (2002). "Conditional Value-at-Risk for General Loss Distributions." Journal of Banking & Finance, 26(7), 1443-1471.
  4. Basel Committee on Banking Supervision (2019). "Minimum Capital Requirements for Market Risk." Bank for International Settlements.
  5. Acerbi, C., & Tasche, D. (2002). "On the Coherence of Expected Shortfall." Journal of Banking & Finance, 26(7), 1487-1503.