Convolution and recursive formulas for aggregate claims

Q: What is the basic formula for the aggregate claim amount $S$ in a collective risk model?

$ S = X_1 + X_2 + ... + X_N $, where $N$ is the number of claims and $X_i$ are the individual claim amounts.

Q: What are the typical assumptions regarding independence in the aggregate loss model?

The individual claim severities $X_i$ are assumed to be independent and identically distributed (i.i.d.), and they are also independent of the frequency $N$.

Q: How is the (a, b, 0) class of frequency distributions defined?

A distribution belongs to the (a, b, 0) class if its probability mass function satisfies the recursive relationship: $ \frac{p_n}{p_{n-1}} = a + \frac{b}{n} $ for $n = 1, 2, 3, ...$.

Q: How do you determine the starting value $f_S(0)$ for Panjer's recursion?

$ f_S(0) = P_N(f_X(0)) $, where $P_N(z)$ is the probability generating function (pgf) of the frequency distribution.

Q: How does the Panjer formula change if the individual claim severity cannot be zero ($f_X(0) = 0$)?

The formula simplifies to $ f_S(s) = \sum_{x=1}^{s} (a + \frac{bx}{s}) f_X(x) f_S(s-x) $.

Q: What is the Method of Rounding for discretizing a continuous severity distribution?

Using a step size $h$, the probability mass is assigned to grid points: $ f_0 = F_X(h/2) $ and $ f_j = F_X(jh + h/2) - F_X(jh - h/2) $ for $j=1, 2, ...$.

Q: How do you choose between using the Convolution Method and Panjer's Recursion?

Use Convolution if the number of claims $N$ is fixed; use Panjer's Recursion if $N$ is a random variable from the (a, b, 0) class.

Exam ASTAM – Advanced Short-Term Actuarial Mathematics · SOA (Society of Actuaries) · 5 min read

Welcome to the World of Aggregate Claims!

Hi there! If you've made it to Exam ASTAM, you already know that insurance companies don't just care about one single claim—they care about the total amount they have to pay out over a period. This total amount is called the Aggregate Claim, usually denoted by $S$.

In this chapter, we are going to learn how to calculate the probability distribution of $S$. Why does this matter? Because if an actuary can't predict the likelihood of total claims exceeding a certain amount, the company might not have enough money to pay them! We will focus on two main "toolkits": Convolution (for when we know exactly how many claims occurred) and Recursive Formulas (specifically Panjer’s Recursion, for when the number of claims is a random variable).

Don't worry if this seems heavy on the math at first. We’ll break it down into bite-sized pieces!

1. The Building Blocks: What is $S$?

Before we dive into the formulas, let's remember our core model:

$ S = X_1 + X_2 + ... + X_N $

Where:
• $N$ is the Frequency (the number of claims).
• $X_i$ is the Severity (the cost of the $i$-th claim).
• $S$ is the Aggregate Loss (the total "pile" of money paid out).

In this chapter, we generally assume that all $X_i$ are independent and identically distributed (i.i.d.) and that they are also independent of $N$.

2. The Convolution Method

Imagine you know for a fact that exactly 2 claims occurred ($N=2$). To find the probability that the total cost is \$100, you have to look at all the ways two numbers can add up to 100 (e.g., 20+80, 50+50, 0+100).

Convolution is just the formal mathematical way of "adding up all the combinations."

\n\n

How it works:

If we define $f_X(x)$ as the probability mass function (pmf) of a single claim, then the pmf of the sum of $n$ claims, denoted as $f^{*n}(s)$, is calculated as:

$ f^{*n}(s) = \sum_{x=0}^{s} f_X(x) \cdot f^{*(n-1)}(s-x) $

\n\n

Step-by-Step Breakdown:

1. Start with $n=1$: The distribution of the sum of one claim is just the distribution of the claim itself. $ f^{*1}(s) = f_X(s) $.\n
2. Move to $n=2$: Calculate the "2-fold convolution" by sliding the distribution of $X$ against itself.\n
3. Keep going: For $n=3$, you convolve $X$ with the result of $n=2$.

\n\n

Quick Tip: Convolution is great for small $N$, but it becomes a nightmare to do by hand if $N$ is large. Think of it like manual counting—accurate, but slow!

\n\n

Key Takeaway: Convolution is the method of finding the distribution of a fixed number of independent variables added together.

\n\n

3. Panjer’s Recursive Formula

Now, let's get to the "magic trick" of actuarial science. In the real world, $N$ isn't fixed; it’s a random variable (like Poisson or Negative Binomial). If we used convolution here, we’d have to calculate a sum for every possible value of $N$. That’s too much work!

Harry Panjer developed a recursive shortcut. If the frequency $N$ belongs to the (a, b, 0) class, we can find the probability of the total sum $S$ without needing to calculate every possible $n$ individually.

\n\n

The (a, b, 0) Class Refresh

Recall that for these distributions, the probability $p_n = P(N=n)$ satisfies:\n
$ \frac{p_n}{p_{n-1}} = a + \frac{b}{n} $

Common members include:\n
• Poisson: $a = 0, b = \lambda$\n
• Binomial: $a < 0$\n
• Negative Binomial: $a > 0$

\n\n

The Formula (The Star of the Show)

For a discrete severity $X$ with pmf $f_X(x)$ where $x = 1, 2, 3, ...$, the pmf of aggregate claims $f_S(s)$ is:

$ f_S(s) = \frac{1}{1 - a f_X(0)} \sum_{x=1}^{s} (a + \frac{bx}{s}) f_X(x) f_S(s-x) $

\n\n

Wait! How do I start the recursion?

You need a starting value, $f_S(0)$. This is the probability that the total aggregate claim is zero. This happens if $N=0$, or if $N>0$ but all claims happen to be \$0.
For the (a, b, 0) class, the formula is:
$ f_S(0) = P_N(f_X(0)) $

Where $P_N(z)$ is the probability generating function (pgf) of the frequency $N$.

Did you know? If the severity $X$ cannot be zero (i.e., $f_X(0) = 0$), the formula becomes much simpler because the leading fraction becomes 1!

Step-by-Step Panjer Process:

1. Identify $a$ and $b$: Based on the given frequency distribution ($N$).
2. Calculate $f_S(0)$: Use the pgf of $N$.
3. Solve for $f_S(1)$: Use the formula with $s=1$.
4. Solve for $f_S(2)$: Use the formula with $s=2$ (which will use your result from $f_S(1)$).
5. Repeat until you reach the desired value of $s$.

Common Mistake: Forgetting that the sum in the Panjer formula starts at $x=1$, not $x=0$. If you include $x=0$ in the sum, you’ll end up with $f_S(s)$ on both sides and get very confused!

Key Takeaway: Panjer's recursion is a powerful way to find the distribution of $S$ by using previously calculated values of $f_S$.

4. Dealing with Continuous Severities: Discretization

Panjer’s formula only works if the severity $X$ is discrete (meaning claims are \$1, \$2, etc., not \$1.2345...). But in the real world, claims are often continuous. What do we do?

We discretize the distribution! We "chop" the continuous curve into buckets of size $h$.

Method of Rounding (The Simplest Way)

In this method, we assign the probability of an interval to the nearest "grid point." If our step size is $h$:

• $ f_0 = P(0 \le X < h/2) = F_X(h/2) $
• $ f_j = P(jh - h/2 \le X < jh + h/2) = F_X(jh + h/2) - F_X(jh - h/2) $

Analogy: Imagine you have a bag of flour (continuous). Discretization is like putting that flour into 1lb bags (discrete). It's not exactly the same, but it's much easier to count!

5. Quick Review & Exam Tips

Memory Aid: The "A-B-S" of Panjer
Remember the term in the sum: (a + bx/s).
• a and b come from the frequency.
• x is the value of the claim you are looking at right now.
• s is the total sum you are trying to calculate.

Quick Summary Table:
• Frequency $N$ is fixed: Use Convolution.
• Frequency $N$ is (a, b, 0): Use Panjer’s Recursion.
• Severity $X$ is continuous: Discretize it first, then use Panjer.

Final Encouragement

The first time you see a Panjer table, it looks like a mountain of numbers. But notice the pattern: each row uses the answers from the rows above it. Practice calculating $f_S(1)$ and $f_S(2)$ a few times, and you'll see the rhythm. 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 basic formula for the aggregate claim amount $S$ in a collective risk model?

$ S = X_1 + X_2 + ... + X_N $, where $N$ is the number of claims and $X_i$ are the individual claim amounts.

What are the typical assumptions regarding independence in the aggregate loss model?

The individual claim severities $X_i$ are assumed to be independent and identically distributed (i.i.d.), and they are also independent of the frequency $N$.

What is the formula for the $n$-fold convolution of a discrete claim distribution?

$ f^{*n}(s) = \sum_{x=0}^{s} f_X(x) \cdot f^{*(n-1)}(s-x) $

How is the (a, b, 0) class of frequency distributions defined?

A distribution belongs to the (a, b, 0) class if its probability mass function satisfies the recursive relationship: $ \frac{p_n}{p_{n-1}} = a + \frac{b}{n} $ for $n = 1, 2, 3, ...$.

What is the Panjer recursive formula for the probability mass function of aggregate claims $f_S(s)$?

$ f_S(s) = \frac{1}{1 - a f_X(0)} \sum_{x=1}^{s} (a + \frac{bx}{s}) f_X(x) f_S(s-x) $

How do you determine the starting value $f_S(0)$ for Panjer's recursion?

$ f_S(0) = P_N(f_X(0)) $, where $P_N(z)$ is the probability generating function (pgf) of the frequency distribution.

How does the Panjer formula change if the individual claim severity cannot be zero ($f_X(0) = 0$)?

The formula simplifies to $ f_S(s) = \sum_{x=1}^{s} (a + \frac{bx}{s}) f_X(x) f_S(s-x) $.

What is the Method of Rounding for discretizing a continuous severity distribution?

Using a step size $h$, the probability mass is assigned to grid points: $ f_0 = F_X(h/2) $ and $ f_j = F_X(jh + h/2) - F_X(jh - h/2) $ for $j=1, 2, ...$.

How do you choose between using the Convolution Method and Panjer's Recursion?

Use Convolution if the number of claims $N$ is fixed; use Panjer's Recursion if $N$ is a random variable from the (a, b, 0) class.

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 ASTAM – Advanced Short-Term Actuarial Mathematics 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

Welcome to the World of Aggregate Claims!

1. The Building Blocks: What is \(S\)?

2. The Convolution Method

How it works:

Step-by-Step Breakdown:

3. Panjer’s Recursive Formula

The (a, b, 0) Class Refresh

The Formula (The Star of the Show)

Wait! How do I start the recursion?

Step-by-Step Panjer Process:

4. Dealing with Continuous Severities: Discretization

Method of Rounding (The Simplest Way)

5. Quick Review & Exam Tips

Final Encouragement

Can you answer these now?

Ready to test yourself?

More Exam ASTAM – Advanced Short-Term Actuarial Mathematics chapters

Put These Notes into Practice