Quantitative Finance: Expected Shortfall vs VaR

Introduction: Expected Shortfall vs. VaR – Taming the Tail Risk Beast

In the realm of risk management, two titans stand out: Value at Risk (VaR) and Expected Shortfall (ES), also known as Conditional Value at Risk (CVaR). Both are designed to quantify potential losses in a portfolio or investment over a specific time horizon and at a given confidence level. However, their approaches and implications differ significantly, especially when dealing with tail risk – the risk of extreme, rare events. Understanding the nuances between VaR and ES is crucial for making informed decisions, meeting regulatory requirements like Basel III, and ultimately safeguarding against catastrophic losses. This deep-dive will explore the theory, applications, and limitations of both measures, empowering you with the knowledge to choose the right tool for the job.

Theory and Fundamentals: Delving into the Depths

VaR and ES represent different philosophies in risk assessment. VaR attempts to answer the question: "What is the maximum loss I can expect with a certain probability over a given period?" It’s a single quantile of the loss distribution. For instance, a 99% VaR of $1 million means there is a 1% chance of losing more than $1 million over the specified time horizon.

ES, on the other hand, delves deeper into the tail of the distribution. It asks: "Given that I've exceeded the VaR threshold, what is the expected loss?" Therefore, ES considers the average loss beyond the VaR level. It's a conditional expectation, focusing on the severity of losses in the worst-case scenarios.

A crucial distinction lies in the concept of coherent risk measures. A coherent risk measure, according to Artzner et al. (1999), satisfies four properties:

Translation invariance: Adding a guaranteed amount of capital reduces the risk measure by the same amount.
Subadditivity: The risk of a portfolio should be less than or equal to the sum of the risks of the individual assets in the portfolio. This is the key to diversification benefits.
Positive homogeneity: Scaling the portfolio size by a factor scales the risk measure by the same factor.
Monotonicity: If one portfolio always has losses greater than or equal to another portfolio, its risk measure should be greater than or equal to the other portfolio's risk measure.

VaR fails to be subadditive in general, particularly for portfolios with non-elliptical distributions (e.g., portfolios with options or credit risk). This means that VaR can underestimate the risk of a diversified portfolio. ES, however, is a coherent risk measure under most conditions, making it a more robust and reliable tool for risk management.

The lack of subadditivity in VaR stems from its focus on a single quantile. Consider two portfolios, each with a 95% VaR of $1 million, but with different loss distributions beyond that threshold. One portfolio might have a relatively flat tail, while the other has a very heavy tail. Diversifying between them could lead to a worse outcome than either on its own, violating subadditivity. ES, by considering the entire tail, mitigates this issue.

Practical Applications: From Trading Desks to Regulatory Compliance

Both VaR and ES find wide application across various areas of finance:

Risk Management: Financial institutions use VaR and ES to assess the risk of their trading portfolios, investment strategies, and overall balance sheets. They set risk limits based on these measures and monitor them regularly.
- Example: A hedge fund manager uses ES to understand the potential drawdown of a specific trading strategy under adverse market conditions. They then adjust their position size to keep the ES below a pre-defined risk tolerance level.
Regulatory Compliance: Basel III, the international regulatory framework for banks, utilizes ES as a key measure for determining capital adequacy. Banks are required to hold sufficient capital to cover potential losses, and ES is used to quantify these losses.
- Example: A bank calculates its 97.5% ES over a 10-day period. The result informs the amount of regulatory capital the bank must hold to absorb potential losses and maintain solvency.
- Basel III moved to ES as the primary risk measure because of VaR’s inability to capture tail risk and its lack of subadditivity. This ensures that banks are better capitalized and more resilient to financial shocks.
Portfolio Optimization: Investors can use ES to construct portfolios that minimize potential losses in extreme scenarios. This involves incorporating ES as a constraint or objective function in portfolio optimization models.
- Example: An institutional investor uses ES to create a portfolio that limits the expected loss in the worst 5% of cases. This helps protect the portfolio against significant market downturns.
Internal Model Validation: Model validation teams frequently backtest both VaR and ES to assess their accuracy and reliability. Backtesting involves comparing the predicted losses from the model with actual losses experienced over a historical period. Significant deviations indicate model deficiencies that need to be addressed.

Formulas and Calculations: Getting Down to the Numbers

Calculating VaR and ES involves different approaches, depending on the underlying data and assumptions. Here are some common methods:

1. Historical Simulation:

This non-parametric method uses historical data to simulate future losses.

VaR: Sort the historical losses from worst to best. The VaR at a confidence level c is the loss at the (1-c) percentile.
- Example: With 1000 days of historical data and a 99% confidence level, the VaR is the 10th worst loss.
ES: Calculate the average of all losses that exceed the VaR.
- Example: Using the previous example, the ES is the average of the 10 worst losses.

2. Variance-Covariance (Parametric) Method:

This method assumes that portfolio returns follow a normal distribution.

VaR:

Where:
- is the portfolio's expected return
- is the portfolio's standard deviation
- is the z-score corresponding to the desired confidence level (e.g., for 99% confidence, )
- is the portfolio's initial value
ES:

Where:
- is the probability density function of the standard normal distribution evaluated at

3. Monte Carlo Simulation:

This method uses random sampling to generate a large number of possible future scenarios.

VaR: Sort the simulated losses and find the loss at the desired percentile.
ES: Average the losses exceeding the VaR threshold.

Numerical Example:

Consider a portfolio with an initial value of $1,000,000. Assume that the portfolio return follows a normal distribution with a mean of 10% and a standard deviation of 20%. Calculate the 95% VaR and ES using the variance-covariance method.

(from standard normal table)

VaR:

ES: (from standard normal table or calculator)

This means there is a 5% chance of losing more than $229,000, and if the loss exceeds $229,000, the expected loss is $312,400.

Risks and Limitations: Navigating the Pitfalls

While both VaR and ES are valuable tools, they have limitations:

Model Risk: Both measures rely on models and assumptions, which may not accurately reflect reality. Incorrect assumptions about the distribution of returns or correlations between assets can lead to inaccurate risk estimates.
Data Dependency: Historical simulation depends on the availability and quality of historical data. If the historical period is not representative of future market conditions, the results may be misleading.
Non-Stationarity: Financial markets are constantly evolving. The statistical properties of assets and portfolios can change over time, making it difficult to rely on past data to predict future risks.
Liquidity Risk: Neither VaR nor ES explicitly accounts for liquidity risk – the risk of being unable to sell an asset quickly at a fair price. In stressed market conditions, liquidity can dry up, leading to much larger losses than predicted by these measures.
Tail Dependence: Both VaR and ES may underestimate the risk of extreme events if they do not adequately account for tail dependence – the tendency for assets to become more correlated during market crashes.
ES Still Requires Model: Although ES is a more coherent measure than VaR, its calculation still relies on a model of the loss distribution. Thus, while the ES measure itself is better behaved, its accuracy depends on the accuracy of the underlying model.

Conclusion and Further Reading: Charting Your Course

VaR and ES are essential tools for risk management, each with its own strengths and weaknesses. VaR provides a simple and intuitive measure of potential losses, while ES offers a more comprehensive and coherent assessment of tail risk. The choice between the two depends on the specific application, the availability of data, and the desired level of rigor. As evidenced by Basel III, the movement towards ES is a recognition of its superiority in capturing tail risk and ensuring the stability of the financial system.

To further deepen your understanding, consider exploring the following resources:

Artzner, P., Delbaen, F., Eber, J. M., & Heath, D. (1999). Coherent measures of risk. Mathematical Finance, 9(3), 203-228.
Acerbi, C., & Tasche, D. (2002). On the coherence of expected shortfall. Journal of Banking & Finance, 26(7), 1487-1503.
Dowd, K. (2005). Measuring Market Risk. John Wiley & Sons.
Hull, J. C. (2018). Risk Management and Financial Institutions. John Wiley & Sons.
The Basel Committee on Banking Supervision (BCBS) publications on regulatory capital and risk management.

By mastering the concepts of VaR and ES, you will be well-equipped to navigate the complexities of risk management and make informed decisions in an increasingly volatile and uncertain world. Remember, a robust understanding of tail risk is paramount to protecting against potentially devastating financial losses.