Recognizing the (a,b,0) and (a,b,1) classes and their relationships

Welcome to Frequency Modeling!

In this chapter, we are diving into the "Frequency" side of Actuarial Mathematics. While severity deals with "how much" a claim costs, frequency deals with "how often" claims occur. To make our lives easier, actuaries use two special families of distributions called the (a,b,0) and (a,b,1) classes.

Don't worry if these names sound like secret codes! By the end of these notes, you'll see they are just a clever way to group the most common frequency distributions (Poisson, Binomial, and Negative Binomial) using a simple recursive relationship. Let’s jump in!

The (a,b,0) Class: The Foundation

The (a,b,0) class is a family of discrete probability distributions where the probability of having \(k\) claims, denoted as \(p_k\), can be found by looking at the probability of having \(k-1\) claims.

The "magic formula" that defines this class is:

\[ \frac{p_k}{p_{k-1}} = a + \frac{b}{k}, \text{ for } k = 1, 2, 3, \dots \]

Think of this like a recursive recipe. If you know the starting point (\(p_0\)) and the values for a and b, you can find every other probability in the distribution without needing the full mass function every time!

Meet the Members of the (a,b,0) Family

There are only three primary members of this family that you need to know for Exam FAM. Each has its own specific a and b values:

1. Poisson Distribution (\(\lambda\))
The Poisson is the "gold standard" for random occurrences.
\(a = 0\)
\(b = \lambda\)
Quick Tip: Since \(a=0\), the ratio is simply \( \frac{p_k}{p_{k-1}} = \frac{\lambda}{k} \).

2. Binomial Distribution (\(n, q\))
Used when there is a fixed number of trials (like \(n\) policies) and a probability of claim \(q\).
\(a = \frac{-q}{1-q}\)
\(b = \frac{(n+1)q}{1-q}\)
Notice: For the Binomial distribution, a is always negative!

3. Negative Binomial Distribution (\(r, \beta\))
Used when the frequency is more "spread out" (volatile) than a Poisson.
\(a = \frac{\beta}{1+\beta}\)
\(b = \frac{(r-1)\beta}{1+\beta}\)
Notice: For the Negative Binomial distribution, a is always positive!

4. Geometric Distribution (\(\beta\))
This is just a special case of the Negative Binomial where \(r = 1\).
\(a = \frac{\beta}{1+\beta}\)
\(b = 0\)

Quick Review: How to Identify the Distribution

If you are given \(a\) or the ratio of probabilities, you can immediately tell which distribution you're looking at:
- If \(a < 0\): It's Binomial.
- If \(a = 0\): It's Poisson.
- If \(a > 0\): It's Negative Binomial.

Key Takeaway: The (a,b,0) class covers the "standard" versions of our frequency distributions where the formula starts working right from \(k=1\).

The (a,b,1) Class: The "Modified" Family

Sometimes, real-world claim data doesn't fit the (a,b,0) model perfectly—usually because the number of "zero claims" is much higher or lower than we expect. This is where the (a,b,1) class comes in.

The (a,b,1) class uses the same recursive formula as the (a,b,0) class, but with one big catch: It only starts working for \(k = 2, 3, \dots\).

The probability at zero, \(p_0\), is "arbitrary." It can be whatever we need it to be to fit the data. There are two types of (a,b,1) distributions you need to know:

1. Zero-Truncated Distributions (ZT)

In a Zero-Truncated distribution, we assume that a "zero" is impossible. Perhaps we are only looking at files for people who have already filed at least one claim. In this case, \(p_0^T = 0\).

How to calculate:
1. Take the original (a,b,0) probability, \(p_k\).
2. Divide it by \((1 - p_0)\).
\[ p_k^T = \frac{p_k}{1 - p_0} \]

2. Zero-Modified Distributions (ZM)

In a Zero-Modified distribution, we manually set \(p_0^M\) to a specific value (let's call it \(g_0\)). This is common when there are "too many zeros" in the data (Zero-Inflation).

How to calculate:
1. Find the zero-truncated probability \(p_k^T\).
2. Scale it by the remaining probability: \( (1 - p_0^M) \).
\[ p_k^M = \frac{1 - p_0^M}{1 - p_0} \times p_k \]

Analogy: The Exclusive Nightclub

Imagine a nightclub (the distribution).
- (a,b,0): The rules for entry are the same for everyone, including the first person in line.
- Zero-Truncated (a,b,1): The "zero" spot is blocked off with a velvet rope. No one can be in the "zero claims" category.
- Zero-Modified (a,b,1): There is a special VIP rule for the "zero claims" spot (\(p_0^M\)), but everyone else in line (\(k=1, 2, 3...\)) follows the standard (a,b,0) rules, just scaled down to fit.

Did you know? Every member of the (a,b,0) class is technically a member of the (a,b,1) class, but not every (a,b,1) member is an (a,b,0) member! It's like how all squares are rectangles, but not all rectangles are squares.

Step-by-Step: Solving (a,b,k) Problems

If you encounter a problem asking for a probability in these classes, follow these steps:

Step 1: Identify the class. Is it (a,b,0) or (a,b,1)? Look at whether the recursion holds for \(k=1\) or starts at \(k=2\).
Step 2: Identify the distribution. Use the sign of a (Negative = Binomial, Zero = Poisson, Positive = Negative Binomial).
Step 3: Find the parameters. Use the given \(p_k\) values to solve for \(a\) and \(b\).
Step 4: Adjust for ZT or ZM. If it's an (a,b,1) class, calculate the "standard" \(p_0\) first, then apply the modification formula to find the requested \(p_k\).

Common Pitfalls to Avoid

1. Mixing up a and b: Always remember the formula is \(a + b/k\). If you flip them, your probabilities will be wildly wrong!
2. Forgetting \(p_0\): In the (a,b,1) class, the recursion doesn't tell you what \(p_1\) is relative to \(p_0\). You must use the "Modified" formula to bridge the gap between \(p_0\) and \(p_1\).
3. Negative Binomial 'r' parameter: Remember that in the (a,b,0) class, for the Negative Binomial, \(b = (r-1)a / (a/ (1+\beta))\). If \(r=1\) (Geometric), then \(b=0\).

Key Takeaways for Exam Day

1. (a,b,0) Formula: \( \frac{p_k}{p_{k-1}} = a + \frac{b}{k} \) starting at \(k=1\).
2. Sign of 'a': \(a < 0\) (Binomial), \(a = 0\) (Poisson), \(a > 0\) (Neg. Binomial).
3. (a,b,1) Difference: The recursive relationship only starts at \(k=2\). \(p_0\) is special.
4. ZM Formula: \( p_k^M = \frac{1-p_0^M}{1-p_0} p_k \). This relates the modified probability to the standard one.

Don't worry if this seems tricky at first! The more you practice identifying "a" and "b" from a list of probabilities, the more intuitive it will become. You've got this!