Autoregressive conditionally heteroskedastic models

Q: What is the definition of heteroskedasticity in the context of time series?

Heteroskedasticity refers to a situation where the variance (or volatility) of the error terms in a model is not constant over time.

Q: What is volatility clustering?

Volatility clustering is the phenomenon where large price changes tend to be followed by large changes, and small changes tend to be followed by small changes.

Q: Why is the error term squared in the ARCH variance formula?

The error is squared to ensure that the predicted variance is always positive and to treat positive and negative shocks as having the same impact on future volatility.

Q: What are the parameter constraints for \( \alpha_0 \) and \( \alpha_1 \) to ensure a valid ARCH(1) model?

To ensure positive variance and stability, we require \( \alpha_0 > 0 \), \( \alpha_1 \geq 0 \), and for stationarity, \( \alpha_1 < 1 \).

Q: How is the long-run (unconditional) variance of an ARCH(1) process calculated?

\( Var(\epsilon_t) = \frac{\alpha_0}{1 - \alpha_1} \)

Q: What is an ARCH(p) model?

An ARCH(p) model extends the ARCH(1) by allowing the current variance to depend on the squared errors of the previous \( p \) time periods.

Q: Why are ARCH models particularly important for actuaries?

They allow for more accurate risk assessment and capital reserves by recognizing that financial risks often come in clusters of high volatility.

Exam SRM – Statistics for Risk Modeling · SOA (Society of Actuaries) · 5 min read

Welcome to the World of Shaky Data!

Hi there! If you've made it this far in your Exam SRM journey, you’ve already seen how we model the "center" of data (the mean). But in the real world—especially in finance and insurance—the "shakiness" or volatility of data isn't always constant. Sometimes things are calm for a long time, and then suddenly, everything goes wild at once!

In this chapter, we are going to learn about Autoregressive Conditionally Heteroskedastic (ARCH) models. Don't let the long name scare you! By the end of these notes, you'll see that it’s just a fancy way of saying: "If things were crazy yesterday, they are likely to be crazy today."

1. What on Earth is Heteroskedasticity?

Before we look at the math, let's break down that giant word: Heteroskedasticity.

In basic regression, we usually assume "Homoskedasticity," which means the variance (the spread of our errors) stays the same all the time. Heteroskedasticity means the variance changes over time.

Conditional Heteroskedasticity means the variance changes based on past information. If a stock market crash happened yesterday, the "conditional" variance for today will likely be much higher than if yesterday was a boring, quiet day.

The Core Idea: Volatility Clustering
Have you ever noticed that in the stock market, big price swings tend to be followed by more big price swings, and quiet periods are followed by more quiet periods? This is called volatility clustering. ARCH models were built specifically to capture this "moodiness" in data.

Key Takeaway

ARCH models are used when the variance of the error terms is not constant but depends on the size of previous errors.

2. The ARCH(1) Model: A Step-by-Step Breakdown

Let's look at the simplest version: the ARCH(1) model. Here, the "1" means we are only looking back at one time period (yesterday).

In a standard time series, we have an error term, let's call it \( \epsilon_t \). In an ARCH model, we assume this error looks like this:

\( \epsilon_t = \sigma_t Z_t \)

Where:
1. \( Z_t \) is just "white noise" (usually a standard normal distribution with mean 0 and variance 1).
2. \( \sigma_t \) is the conditional standard deviation (the "shakiness" at time \( t \)).

The ARCH(1) equation for the variance is:

\( \sigma^2_t = \alpha_0 + \alpha_1 \epsilon^2_{t-1} \)

Let's translate that into English:
Today's variance (\( \sigma^2_t \)) is equal to a baseline constant (\( \alpha_0 \)) plus a piece of yesterday's squared error (\( \alpha_1 \epsilon^2_{t-1} \)).

Wait, why is the error squared?
Great question! We square the error (\( \epsilon^2_{t-1} \)) because variance must always be positive. Whether the market went up by 10% or down by 10% yesterday, it was a "big move," and it should increase our prediction of volatility for today.

Memory Aid: The "Storm" Analogy

Think of \( \alpha_0 \) as the typical wind speed on a normal day. Think of \( \alpha_1 \) as how much a storm yesterday affects the weather today. If \( \alpha_1 \) is large, and a massive storm (\( \epsilon^2_{t-1} \)) hit yesterday, you can bet today's weather forecast (\( \sigma^2_t \)) will be very volatile!

3. Rules and Constraints

For an ARCH(1) model to make sense mathematically and stay stable, we have a few rules for our parameters \( \alpha_0 \) and \( \alpha_1 \):

1. Positivity: We must have \( \alpha_0 > 0 \) and \( \alpha_1 \geq 0 \). Why? Because variance can't be negative! You can't have "negative shakiness."
2. Stationarity: For the variance to stay stable over the long run, we need \( \alpha_1 < 1 \). If \( \alpha_1 \geq 1 \), the variance would keep growing and growing until it explodes to infinity!

Quick Review: Long-Run Variance
If the model is stationary (\( \alpha_1 < 1 \)), the "unconditional" or long-run average variance is:
\( Var(\epsilon_t) = \frac{\alpha_0}{1 - \alpha_1} \)

Common Mistake Alert!

Students often confuse the mean of the process with the variance. In a basic ARCH model, the mean of the error \( \epsilon_t \) is still 0. It's the variance that changes. The series isn't necessarily trending up or down; it's just getting "noisier."

4. Scaling Up: The ARCH(p) Model

What if the volatility today depends on the last two days? Or the last week? We simply add more terms to the equation! This is called an ARCH(p) model, where \( p \) is the number of past lags we look at.

The formula looks like this:
\( \sigma^2_t = \alpha_0 + \alpha_1 \epsilon^2_{t-1} + \alpha_2 \epsilon^2_{t-2} + ... + \alpha_p \epsilon^2_{t-p} \)

Don't worry if this looks long! It’s the exact same logic as ARCH(1), just looking further back into the past. For this to be stationary, the sum of all the alphas (\( \alpha_1 + \alpha_2 + ... + \alpha_p \)) must be less than 1.

5. Why do we use ARCH models in Actuarial Science?

Did you know?
Standard Linear Regression and even many ARIMA models assume the "noise" is constant. However, if you are an actuary pricing an insurance product that depends on the stock market (like Variable Annuities), assuming constant noise can lead to disaster! If you underestimate the "shakiness" of the market, you might not hold enough capital to pay out claims during a crash.

Real-World Example:
During the 2008 financial crisis, volatility was extremely high for months. An ARCH model would have recognized that "High volatility yesterday = High volatility today" and adjusted risk reserves accordingly. A standard model would have just seen it as a "one-off" outlier.

6. Summary and Final Checklist

Before you move on to the next chapter, make sure you're comfortable with these points:

1. The "Why": ARCH models are used for volatility clustering (periods of high/low variance).
2. The "What": The variance today depends on the squared errors of the past.
3. The Formula: Know \( \sigma^2_t = \alpha_0 + \alpha_1 \epsilon^2_{t-1} \) by heart.
4. The Constraints: Parameters must be positive to keep variance positive; \( \alpha_1 \) must be less than 1 for stability.
5. The Logic: If \( \epsilon_{t-1} \) is large (positive or negative), it will increase \( \sigma^2_t \), making the next error \( \epsilon_t \) likely to be large as well.

Pro-Tip for the Exam:
If an exam question asks which model is best for "changing variance over time" or "volatility clustering," your brain should immediately scream ARCH!

Keep up the great work! You're mastering a concept that earned Robert Engle the Nobel Prize in Economics. You've got this!

Quick check

Can you answer these now?

Open each question to check the key ideas from this chapter.

What is the definition of heteroskedasticity in the context of time series?

Heteroskedasticity refers to a situation where the variance (or volatility) of the error terms in a model is not constant over time.

What is volatility clustering?

Volatility clustering is the phenomenon where large price changes tend to be followed by large changes, and small changes tend to be followed by small changes.

What is the mathematical equation for the conditional variance in an ARCH(1) model?

\( \sigma^2_t = \alpha_0 + \alpha_1 \epsilon^2_{t-1} \)

Why is the error term squared in the ARCH variance formula?

The error is squared to ensure that the predicted variance is always positive and to treat positive and negative shocks as having the same impact on future volatility.

What are the parameter constraints for \( \alpha_0 \) and \( \alpha_1 \) to ensure a valid ARCH(1) model?

To ensure positive variance and stability, we require \( \alpha_0 > 0 \), \( \alpha_1 \geq 0 \), and for stationarity, \( \alpha_1 < 1 \).

How is the long-run (unconditional) variance of an ARCH(1) process calculated?

\( Var(\epsilon_t) = \frac{\alpha_0}{1 - \alpha_1} \)

What is an ARCH(p) model?

An ARCH(p) model extends the ARCH(1) by allowing the current variance to depend on the squared errors of the previous \( p \) time periods.

Why are ARCH models particularly important for actuaries?

They allow for more accurate risk assessment and capital reserves by recognizing that financial risks often come in clusters of high volatility.

Ready to test yourself?

Turn these notes into exam-style practice. Get unlimited AI questions on this topic with instant marking and explanations.

Practice This Topic

More Exam SRM – Statistics for Risk Modeling chapters

* The content provided by thinka is generated by AI and may not always be accurate or up-to-date. Please use it as a supplementary resource and verify with official materials.

Put These Notes into Practice

Reading the notes is a great start. Now practise with unlimited AI-generated questions and get instant feedback. 100,000+ students are already improving their grades.

Start Practising Now View Pricing

Done reading? Test yourself with AI practice questions

Practice This Topic Now