Welcome to Severity Models: Comparing Distribution Tails!

In the world of insurance, we aren't just worried about the average fender-bender; we are worried about the massive, "once-in-a-career" catastrophes. Comparing distribution tails is how actuaries determine which mathematical models predict more frequent or more severe "big losses."

In this chapter, we will learn how to look at two different loss distributions and decide which one is "heavier" (more likely to have huge claims). Don't worry if this seems a bit abstract at first—we'll use plenty of analogies to keep things grounded!

1. What Exactly is a "Tail"?

Imagine a graph of a loss distribution. The "tail" is the far right side of the curve—the part that represents very high dollar amounts. When we say a distribution has a heavy tail, we mean it stays "higher" for longer as you move to the right. This suggests that large claims are more probable than in a light-tailed distribution.

Think of it like this: A "light-tailed" distribution is like a sprinter who runs out of breath quickly—the probability drops to zero very fast. A "heavy-tailed" distribution is like a marathon runner—the probability stays active even at very high loss amounts.

Quick Review: Why do we care?
If you use a light-tailed model for a heavy-tailed risk (like hurricane damage), you will significantly underestimate the cost of big claims. That’s how insurance companies go broke!

2. Method 1: Comparing Survival Functions

The most direct way to compare two distributions, let's call them \(X\) and \(Y\), is to look at their Survival Functions, \(S_X(x)\) and \(S_Y(x)\), as \(x\) approaches infinity.

We look at the limit of their ratio:
\[ \lim_{x \to \infty} \frac{S_X(x)}{S_Y(x)} \]

  • If the limit is \(\infty\): Distribution \(X\) has a heavier tail than \(Y\).
  • If the limit is \(0\): Distribution \(X\) has a lighter tail than \(Y\).
  • If the limit is a positive constant: The tails are considered to have similar weight.

Pro-Tip: If the Survival Functions are hard to work with, you can use the Density Functions (\(f(x)\)) instead! The result of the limit will be the same.

Did you know? This is often where L'Hôpital's Rule becomes your best friend. If you get an indeterminate form like \(\frac{\infty}{\infty}\), just take the derivative of the top and bottom!

3. Method 2: The Hazard Rate Function

The Hazard Rate, \(h(x) = \frac{f(x)}{S(x)}\), tells us the "instantaneous" rate of failure (or claim occurrence) given that the claim has already reached size \(x\).

Here is the rule of thumb for tail weight using Hazard Rates:
Lower Hazard Rate = Heavier Tail

Why? If the hazard rate is decreasing, it means as the loss gets bigger, the "pressure" for the claim to stop growing actually decreases.

Common Classifications:

  1. Decreasing Hazard Rate (DHR): The tail is heavy (e.g., Pareto).
  2. Constant Hazard Rate: This is the Exponential distribution. It is our "benchmark."
  3. Increasing Hazard Rate (IHR): The tail is light (e.g., Normal distribution).

Key Takeaway:

If \(\lim_{x \to \infty} h_X(x) = 0\), the distribution \(X\) is generally very heavy-tailed. If the limit is \(\infty\), it is very light-tailed.

4. Method 3: Mean Residual Life (MRL)

The Mean Residual Life, \(e(x)\), is the expected additional loss amount, given that the loss has already exceeded \(x\).

\[ e(x) = E[X - x | X > x] = \frac{\int_{x}^{\infty} S(t) dt}{S(x)} \]

How this relates to tails:

  • If \(e(x)\) increases with \(x\): The tail is heavy. (The more you've lost, the more you're expected to continue losing!)
  • If \(e(x)\) is constant: This is the Exponential distribution again.
  • If \(e(x)\) decreases with \(x\): The tail is light.

Analogy: Imagine a "never-ending" meeting. If the longer the meeting goes, the longer you expect it to keep going, that meeting has an "increasing mean residual life"—and a very heavy tail!

5. Comparing Common Distributions (The "Pecking Order")

For Exam ASTAM, you should recognize the relative "heaviness" of the standard distributions. From Heaviest to Lightest:

1. Pareto: The king of heavy tails. Its hazard rate decreases to zero. It has "power-law" tails.
2. Lognormal: Also heavy-tailed, but generally considered slightly "less" heavy than a Pareto with a small \(\alpha\). Its hazard rate also decreases to zero.
3. Weibull (with \(\tau < 1\)): Heavy-tailed.
4. Exponential: The neutral boundary. Hazard rate is constant (\(1/\theta\)).
5. Weibull (with \(\tau > 1\)): Light-tailed. Hazard rate increases.
6. Gamma: Generally light-tailed (hazard rate increases to a constant \(1/\theta\)).

Memory Trick: "Please Let Everyone Go" (Pareto > Lognormal > Exponential > Gamma). This covers the most common comparison questions!

6. Common Mistakes to Avoid

  • Mixing up \(h(x)\) and \(S(x)\): Remember, a smaller hazard rate means a heavier tail. Don't flip them!
  • Ignoring Parameters: Some distributions, like the Weibull, can be heavy OR light depending on their parameters. Always check the \(\tau\) (shape) parameter.
  • Calculus Errors: When taking limits of ratios, students often forget to simplify the expression before applying L'Hôpital's Rule. Simplify first to save time!

7. Final Summary Table

Heavier Tail Characteristics:
- Survival Function Ratio \(\to \infty\)
- Hazard Rate \(h(x) \to 0\) or is Decreasing
- Mean Residual Life \(e(x)\) is Increasing
- High probability of extreme values

Lighter Tail Characteristics:
- Survival Function Ratio \(\to 0\)
- Hazard Rate \(h(x) \to \infty\) or is Increasing
- Mean Residual Life \(e(x)\) is Decreasing
- Low probability of extreme values

Keep practicing these limits! The more you compare these functions, the more intuitive the "heaviness" of the tails will become. You've got this!