Welcome to the World of Moments!

Welcome, future actuary! Today, we are diving into a crucial part of the Severity, Frequency, and Aggregate Models section: Characterizing distributions by the existence of moments.

If you have ever wondered, "Can a loss be so unpredictable that we can't even calculate an average?" then you are in the right place. In this chapter, we learn how to tell if a distribution is "well-behaved" or if it has a "heavy tail" that makes it risky for insurance companies. Don't worry if the math seems daunting; we will break it down step-by-step!

1. What is a "Moment" and why does it "Exist"?

Before we jump into the deep end, let's look at the basics. In actuarial math, the \(k\)-th raw moment is defined as the expected value of the random variable \(X\) raised to the power of \(k\). We write this as \(E[X^k]\).

  • The 1st moment (\(k=1\)) is the Mean (Average loss).
  • The 2nd moment (\(k=2\)) is used to find the Variance (Volatility).

The "Existence" Concept: When we say a moment "exists," we mean that when we do the math (usually an integral), the result is a finite number. If the math leads to "infinity," we say the moment does not exist.

An Everyday Analogy

Think of an insurance distribution like a giant mountain. Most mountains have a peak and then slope down to the ground. If the slope is very steep (light tail), we can easily measure the mountain's volume. But if the slope stretches out forever and never quite touches the ground (heavy tail), the "volume" might be so large that we can't give it a single number. That is a distribution where moments don't exist!

2. The Pareto Distribution: The "Star" of this Chapter

In Exam FAM, the Pareto distribution is the most common example used to test your knowledge of moment existence. The Pareto is famous for having a "heavy tail," meaning it predicts a higher chance of very large, catastrophic losses.

For a Pareto distribution with parameters \(\theta\) (scale) and \(\alpha\) (shape):

The Rule: The \(k\)-th moment \(E[X^k]\) exists if and only if \(k < \alpha\).

Why is this important?
If a Pareto distribution has \(\alpha = 1.5\):
- The 1st moment (Mean) exists because \(1 < 1.5\).
- The 2nd moment (Variance calculation) does NOT exist because \(2\) is not less than \(1.5\).
- This means the losses are so volatile that the variance is technically infinite!

Memory Aid: The "Alpha is the Boss" Rule

Think of \(\alpha\) as the Amount of "room" the distribution has for moments. If you want the \(k\)-th moment, it has to fit under the \(\alpha\) ceiling. If \(k\) tries to go above \(\alpha\), the moment "explodes" to infinity.

3. Comparing Tail Heaviness

One of the main reasons we check if moments exist is to determine how "heavy" the tail of a distribution is. This tells an actuary how risky a policy is.

  • Light-Tailed Distributions: Distributions like the Exponential or Normal have all moments. You can calculate the 100th moment, the 1,000th moment, etc. They are generally less risky for extreme losses.
  • Medium-Tailed Distributions: The Log-normal distribution is a special case. All its moments exist (like light tails), BUT it does not have a Moment Generating Function (MGF). This makes it heavier than an Exponential but lighter than a Pareto.
  • Heavy-Tailed Distributions: The Pareto or Burr distributions only have a finite number of moments. These are the "riskiest" because large losses are more frequent.

Quick Review Box:
- All moments exist: Gamma, Exponential, Normal, Log-normal.
- Only some moments exist: Pareto, Burr, Log-logistic.

4. The Step-by-Step Test for Existence

When you are sitting in the exam and you need to determine if \(E[X^k]\) exists for a survival function \(S(x)\), use this logic:

Step 1: Look at the behavior of the distribution as \(x\) goes to infinity (\(x \to \infty\)).

Step 2: Use the formula for the raw moment: \(E[X^k] = \int_0^\infty k \cdot x^{k-1} S(x) dx\).

Step 3: Check if the integral converges. For the Pareto, \(S(x)\) looks like \(x^{-\alpha}\). When you multiply by \(x^{k-1}\), you are looking at \(x^{k - \alpha - 1}\). The integral only converges if the exponent is less than \(-1\), which leads us back to our rule: \(k < \alpha\).

Did you know?

The Log-normal distribution is often a "trick" question on Exam FAM. Students often think that because its tail is "heavy," some moments must not exist. But actually, every power of \(X\) has a finite expected value for the Log-normal! The "heaviness" is shown by the fact that its MGF is infinite, not its moments.

5. Common Mistakes to Avoid

1. Mixing up \(\alpha\) and \(k\): Always remember that the distribution parameter (\(\alpha\)) is the limit, and the moment you want (\(k\)) is the challenger. The challenger must be smaller than the limit.

2. Forgetting that moments must be positive: We are usually dealing with loss distributions where \(X > 0\). We only care about \(k > 0\).

3. Ignoring the Scale Parameter: While \(\theta\) (the scale) is important for calculating the value of the moment, it has no effect on whether the moment exists. Only the shape parameters (like \(\alpha\)) decide that.

6. Summary and Key Takeaways

Key Takeaway 1: Moments tell us about the characteristics of a distribution (Mean, Variance, etc.).

Key Takeaway 2: If a moment is infinite, it "does not exist." This happens in heavy-tailed distributions like the Pareto.

Key Takeaway 3: For the Pareto distribution, \(E[X^k]\) exists if \(k < \alpha\). This is the most likely concept to be tested.

Key Takeaway 4: Tail heaviness is a spectrum. Existence of moments helps us place distributions on that spectrum to manage insurance risk.

Keep practicing! Determining the existence of moments is like being a detective—you are looking for the clues in the parameters to see how the distribution behaves at its extremes. You've got this!