Estimating Market Risk Measures: An Introduction and Overview

Welcome to Market Risk Estimation!

Hello! If you’ve made it to FRM Part II, you already know that risk is the "possibility of loss." But in the professional world, "possibility" isn't specific enough. We need to measure it. In this chapter, Estimating Market Risk Measures, we move from theoretical definitions to the actual "how-to" of calculating risk. We’ll explore different ways to estimate Value at Risk (VaR) and Expected Shortfall (ES), and we'll look at the data that fuels these models. Don't worry if this seems heavy on the math—we'll break it down into simple, logical steps!

1. The Foundation: Selecting Your Data

Before we can run a model, we need data. Think of data as the "ingredients" for our risk "recipe." If the ingredients are stale or wrong, the result will be a mess!

Data Windows and Weighting

How much past data should we look at? This is called the window.
- Long Windows: Provide more data points, making the results stable, but they might include "ancient history" that isn't relevant to today's market.
- Short Windows: Are very reactive to recent market shocks but can be "jumpy" or volatile.

The Weighting Game:
Usually, we give every day in our window the same importance (Equal Weighting). However, we can use Exponential Smoothing (EWMA) to give more weight to recent events.
Analogy: If you're trying to guess what the weather will be like tomorrow, today's weather is a better clue than what the weather was like three years ago today!

Quick Review:
- Equal Weighting: Simple, but slow to react to new crises.
- Age Weighting: Newer data counts for more. It helps the model "wake up" faster when volatility spikes.

2. Non-Parametric Approach: Historical Simulation (HS)

The Historical Simulation method is the most popular "non-parametric" approach. "Non-parametric" simply means we don't assume the data follows a specific shape (like a Bell Curve). We just let the past data speak for itself.

How it Works (Step-by-Step):

1. Collect a series of past returns (e.g., the last 500 days).
2. Rank those returns from the worst (biggest loss) to the best (biggest gain).
3. If you want a 95% VaR, you look at the bottom 5% of those ranked returns. The value at that "cutoff" point is your VaR.

Pros and Cons of HS:

- Pro: It’s easy to explain to management. No complex calculus is required.
- Pro: It naturally captures "Fat Tails" (extreme events) because it uses real events that actually happened.
- Con: The Ghosting Effect. If a huge market crash happened 250 days ago and your window is 250 days, tomorrow that crash will "fall off" the window, and your risk measure will suddenly drop, even if the market hasn't changed.
- Con: It assumes the future will be just like the past. It cannot account for events that haven't happened yet.

3. Parametric Approach: The Normal Distribution

In the Parametric Approach, we assume that market returns follow a specific mathematical distribution, usually the Normal Distribution (the Bell Curve).

The Formula

For a portfolio with a mean return of \( \mu \) and a standard deviation of \( \sigma \), the VaR at a confidence level \( \alpha \) is:
\[ VaR = - (\mu + \sigma \times Z_{\alpha}) \]
(Note: We usually express VaR as a positive number representing a loss, which is why there is a negative sign at the front).

Common Z-scores to memorize:
- 95% Confidence: \( Z = 1.645 \)
- 99% Confidence: \( Z = 2.326 \)

The "Fat Tail" Problem

Real-world financial returns almost always have Fat Tails (Leptokurtosis). This means extreme losses happen much more often than the Normal Distribution predicts.
Did you know? If market returns were truly "Normal," the 2008 financial crisis would have been statistically impossible. Because the Normal Distribution underestimates the "extremes," it can give a false sense of security.

Key Takeaway: Parametric models are fast and easy to calculate, but if you choose the wrong distribution (like the Normal Distribution for fat-tailed data), you will dangerously underestimate your risk.

4. Coherent Risk Measures and Expected Shortfall (ES)

Not all risk measures are created equal. To be "good" (or Coherent), a risk measure must follow four mathematical rules:

1. Monotonicity: If Portfolio A always performs worse than Portfolio B, Portfolio A must have a higher risk measure. (Common sense!)
2. Subadditivity: The risk of (A + B) should be less than or equal to the risk of A + the risk of B. This reflects the benefits of diversification.
3. Positive Homogeneity: If you double the size of your position, you double your risk.
4. Translation Invariance: If you add cash (a risk-free asset) to your portfolio, your risk should decrease by that exact amount.

The Big VaR Weakness

VaR is NOT always coherent because it often fails the Subadditivity test. In some weird cases, combining two portfolios can actually make the VaR go up! This is a major reason why regulators have moved toward Expected Shortfall (ES).

Expected Shortfall (ES)

While VaR asks, "What is the minimum I could lose on a really bad day?", ES asks, "If we have a really bad day and exceed the VaR, what is the average loss we should expect?"
- ES is Coherent (it always satisfies subadditivity).
- ES looks at the entire "tail" of the distribution, not just one single point.
- Formulaically, ES is the expected value of the loss, given that the loss is greater than the VaR.

5. Estimating the Shape: Extreme Value Theory (EVT)

If Historical Simulation is too backward-looking and Parametric models are too "perfect," is there a middle ground? Yes: Extreme Value Theory (EVT).

EVT is a "semi-parametric" approach. Instead of trying to model the whole distribution (the gains and the small losses), it only models the "Tail" (the extreme losses).
Analogy: Imagine a doctor who doesn't care about your height or weight, only about your symptoms when you have a 104-degree fever. That's EVT—it focuses solely on the "fever" (the crisis).

Common EVT Approaches:
1. Block Maxima: Taking the largest loss from separate periods (e.g., the worst day of every month). This follows the Generalized Extreme Value (GEV) distribution.
2. Peaks Over Threshold (POT): Looking at every loss that exceeds a certain "scary" level. This follows the Generalized Pareto Distribution (GPD).

6. Summary and Final Tips

Don't get overwhelmed! Here is what you need to remember:
- Historical Simulation is simple and uses real data but suffers from the "ghosting effect."
- Parametric VaR is elegant but often misses "Fat Tails."
- VaR is popular but is not coherent (fails subadditivity).
- Expected Shortfall (ES) is coherent and tells us the average "tail loss."
- EVT is the specialist tool for modeling the "worst-case scenarios."

Common Mistake to Avoid:
Students often think VaR tells you the maximum possible loss. It doesn't! It only tells you the minimum loss you can expect on the worst X% of days. There is always a "tail" beyond the VaR!

Keep practicing these concepts—once you understand the "why" behind the models, the "how" (the math) becomes much easier to manage!