Determine the characteristics of jointly distributed random variables

Welcome to Joint Distributions!

In our previous studies, we looked at random variables (RVs) in isolation—like the number of claims a single policyholder makes. But the real world is rarely that simple! Actuaries often need to look at two or more variables at the same time. For example, we might want to know the relationship between the age of a driver and the cost of their insurance claim. This is where jointly distributed random variables come in.

Don't worry if this seems a bit abstract at first. We’re essentially just moving from looking at a single number to looking at pairs (or groups) of numbers. Think of it as moving from a 2D line graph to a 3D map.

1. Joint Probability Distributions

When we have two random variables, \(X\) and \(Y\), we want to know the probability that they take on specific values simultaneously.

Discrete Case: The Joint Probability Mass Function (PMF)

If \(X\) and \(Y\) are discrete, we use a joint PMF, denoted as \( p(x, y) = P(X = x, Y = y) \).
The Golden Rule: The sum of all probabilities in the joint distribution must equal 1.
\( \sum_{all x} \sum_{all y} p(x, y) = 1 \)

Continuous Case: The Joint Probability Density Function (PDF)

If \(X\) and \(Y\) are continuous, we use a joint PDF, \( f(x, y) \). Instead of a table, imagine a "surface" or a "hill" over a flat plane. The volume under this surface represents the probability.
The Golden Rule: The total volume under the surface must equal 1.
\( \int_{-\infty}^{\infty} \int_{-\infty}^{\infty} f(x, y) \,dx \,dy = 1 \)

Quick Review: The Joint CDF

The Joint Cumulative Distribution Function (CDF), \( F(x, y) \), is the probability that \(X\) is less than or equal to \(x\) AND \(Y\) is less than or equal to \(y\):
\( F(x, y) = P(X \le x, Y \le y) \)

Key Takeaway: Joint distributions describe how two variables behave together. For discrete variables, think of a table; for continuous, think of a 3D surface.

2. Marginal Distributions

Sometimes, we have the joint distribution (the "big picture"), but we want to zoom back into just one of the variables. This is called finding the Marginal Distribution.

The "Summing Out" Trick:
To find the marginal distribution of \(X\), you effectively "ignore" or "sum out" \(Y\).
- Discrete: \( p_X(x) = \sum_{all y} p(x, y) \)
- Continuous: \( f_X(x) = \int_{-\infty}^{\infty} f(x, y) \,dy \)

Memory Aid: Why is it called "Marginal"?
Think of a probability table. If you sum up the rows and write the totals in the margins (the edges) of the paper, you’ve found the marginal distribution!

Key Takeaway: To get the marginal distribution of one variable, integrate or sum over the other variable's entire range.

3. Conditional Distributions

In actuarial work, we often ask "What is the probability of \(X\) given that we already know the value of \(Y\)?". This is the Conditional Distribution.

The formula is based on the basic probability rule \( P(A|B) = \frac{P(A \cap B)}{P(B)} \):
\( f(x|y) = \frac{f(x, y)}{f_Y(y)} \), provided \( f_Y(y) > 0 \)

Analogy: Imagine you are looking at the heights (\(X\)) and weights (\(Y\)) of a population. The marginal distribution of height is the height of everyone. But the conditional distribution of height given that a person weighs 100kg is the height distribution of only that specific heavy-weight group.

Common Mistake: Forgetting which variable goes in the denominator! Always divide by the marginal distribution of the variable you are given (the one after the vertical bar |).

4. Independence of Random Variables

Two random variables are independent if knowing the value of one tells you absolutely nothing about the other.

Mathematically, \(X\) and \(Y\) are independent if and only if their joint distribution is simply the product of their marginals:
Discrete: \( p(x, y) = p_X(x) \cdot p_Y(y) \)
Continuous: \( f(x, y) = f_X(x) \cdot f_Y(y) \)

Helpful Tip: If the range (limits) of \(X\) depends on the value of \(Y\) (e.g., \( 0 < x < y < 1 \)), the variables are not independent!

5. Expectations and Covariance

How do we describe the "average" behavior and the "connection" between two variables?

Expectation of a Function \( g(X, Y) \)

To find the expected value of a function of two variables, we multiply the function by the joint probability and sum/integrate:
\( E[g(X, Y)] = \int \int g(x, y) f(x, y) \,dx \,dy \)

Covariance

Covariance measures the linear relationship between \(X\) and \(Y\).
\( Cov(X, Y) = E[(X - E[X])(Y - E[Y])] \)
A more practical formula for calculations is:
\( Cov(X, Y) = E[XY] - E[X]E[Y] \)

• If \( Cov(X, Y) > 0 \): They move in the same direction (both up or both down).
• If \( Cov(X, Y) < 0 \): They move in opposite directions.
• If \( X \) and \( Y \) are independent, then \( Cov(X, Y) = 0 \). (Note: The reverse is not always true!)

Correlation Coefficient (\( \rho \))

Covariance can be any number, which makes it hard to compare. Correlation scales the covariance to a range between -1 and +1:
\( \rho(X, Y) = \frac{Cov(X, Y)}{\sigma_X \sigma_Y} \)

Where \( \sigma_X \) and \( \sigma_Y \) are the standard deviations.

Key Takeaway: Covariance shows the direction of a relationship; Correlation shows both the direction and the strength on a standard scale.

6. Conditional Expectation and Variance

These are "big hitters" in the CS1 exam. They involve the Law of Total Expectation and the Law of Total Variance.

Law of Total Expectation (Adam's Law)

\( E[X] = E[E[X|Y]] \)
This means the overall average of \(X\) is the average of all the conditional averages.

Law of Total Variance (Eve's Law)

\( Var(X) = E[Var(X|Y)] + Var(E[X|Y]) \)
Memory Aid: "Expectation of Variance plus Variance of Expectation." (E-Var + Var-E).

Why is this useful?
Imagine \(X\) is the total claim amount for a year, and \(Y\) is the number of accidents. It might be hard to calculate \(Var(X)\) directly, but easy to calculate the variance given a specific number of accidents. These laws allow you to "build up" to the total variance using those simpler conditional pieces.

Key Takeaway: These laws are essential for breaking down complex problems into smaller, manageable parts based on "given" information.

Summary Checklist

• Can you identify if a joint distribution is discrete or continuous?
• Do you remember to "sum out" the other variable to find a marginal?
• Is your correlation coefficient between -1 and 1? If not, check your math!
• Have you practiced using \( E[X] = E[E[X|Y]] \) for compound distributions?

Don't worry if these double integrals or conditional laws feel heavy right now. With practice, you'll start to see them as simple "recipes" for combining data!