Conditional Probability and Bayes' Theorem for Actuaries

Conditional Probability Basics

Conditional probability measures the likelihood of event A given that event B has occurred: P(A|B) = P(A and B) / P(B), provided P(B) > 0. In insurance, this arises constantly. For example, the probability that a claim exceeds $10,000 given that it has already exceeded $5,000 is a conditional probability.

The multiplication rule follows directly: P(A and B) = P(A|B) * P(B). Two events are independent when P(A|B) = P(A), meaning knowledge of B does not change the probability of A. Independence is a key assumption in many actuarial models.

You can extend the multiplication rule to multiple events: P(A and B and C) = P(A) * P(B|A) * P(C|A and B). This sequential conditioning is useful when drawing items without replacement. For instance, if 3 defective policies exist among 20, the probability that the first two drawn are both defective is (3/20) * (2/19) = 6/380 = 3/190.

The Law of Total Probability

If B1, B2, ..., Bn form a partition of the sample space (mutually exclusive and exhaustive), then P(A) = sum of P(A|Bi) * P(Bi) for i = 1 to n. This is heavily tested on Exam P. A typical problem might partition policyholders into risk classes (low, medium, high) and ask for the overall probability of a claim.

For example, suppose 50% of drivers are low risk with P(claim) = 0.02, 30% are medium risk with P(claim) = 0.05, and 20% are high risk with P(claim) = 0.10. The overall claim probability is 0.50(0.02) + 0.30(0.05) + 0.20(0.10) = 0.045.

The law of total probability also applies to expected values: E[X] = sum of E[X|Bi] * P(Bi). This is sometimes called the law of iterated expectations or the tower property. For variance, a related formula holds: Var(X) = E[Var(X|B)] + Var(E[X|B]), known as the law of total variance or the Eve\'s law. Both formulas are tested frequently on Exam P.

Bayes\' Theorem

Bayes\' Theorem reverses the conditioning: P(Bi|A) = P(A|Bi) * P(Bi) / P(A). Using the example above, given that a claim occurred, the probability the driver was high risk is P(claim|high) * P(high) / P(claim) = 0.10 * 0.20 / 0.045 = 0.444. This "posterior" probability is much higher than the prior 0.20, reflecting the information gained from observing a claim.

On Exam P, tree diagrams are an efficient way to organize Bayes\' calculations. Draw branches for each class, then sub-branches for the event of interest, and compute posterior probabilities from the resulting products.

The denominator P(A) in Bayes\' Theorem is computed using the law of total probability: P(A) = sum of P(A|Bi) * P(Bi). This means you always need the full partition to apply Bayes\' Theorem. A systematic approach is to compute all the joint probabilities P(A and Bi) = P(A|Bi) * P(Bi) first, then sum them for P(A), and finally divide each by P(A) to get the posteriors.

Worked Example with a Table

An insurance company has two underwriting offices. Office 1 processes 60% of applications and has a 5% error rate. Office 2 processes 40% of applications and has an 8% error rate. An application is found to contain an error. What is the probability it came from Office 2?

Step 1: Define events. Let E = error found, O1 = Office 1, O2 = Office 2. We know P(O1) = 0.60, P(O2) = 0.40, P(E|O1) = 0.05, P(E|O2) = 0.08.

Step 2: Compute the total probability of error. P(E) = P(E|O1) * P(O1) + P(E|O2) * P(O2) = 0.05 * 0.60 + 0.08 * 0.40 = 0.030 + 0.032 = 0.062.

Step 3: Apply Bayes\' Theorem. P(O2|E) = P(E|O2) * P(O2) / P(E) = 0.032 / 0.062 = 0.516. So there is about a 51.6% chance the error came from Office 2, even though Office 2 only processes 40% of applications. The higher error rate shifted the posterior upward.

Multiple Updates and Sequential Bayes

Bayes\' Theorem can be applied sequentially. After observing one piece of evidence, the posterior becomes the new prior for the next observation. Suppose in the risk class example, a driver has a claim in year 1. The posterior probabilities become the new priors: P(high|claim) = 0.444, P(medium|claim) = 0.333, P(low|claim) = 0.222. If a second claim occurs in year 2, you apply Bayes\' Theorem again using these updated priors.

This sequential updating is the foundation of credibility theory, which you will encounter in later actuarial exams. For Exam P, focus on single-step Bayes\' problems, but understanding the sequential logic helps build intuition.

Common Pitfalls

The most common mistake with conditional probability is confusing P(A|B) with P(B|A). These are generally not equal. For example, P(has disease | tests positive) is very different from P(tests positive | has disease). The second is the test\'s sensitivity; the first is the predictive value, which depends on the disease prevalence via Bayes\' Theorem.

Another frequent error is assuming independence without justification. Two events can appear unrelated but still be dependent when conditioned on a third event. Always verify independence using the definition P(A and B) = P(A) * P(B) rather than relying on intuition.

Students also sometimes forget that P(A|B) is only defined when P(B) > 0. On Exam P, this is usually satisfied, but be careful with continuous distributions where conditioning on a single point requires the use of conditional densities rather than conditional probabilities.

Study Tips

For Exam P, practice organizing Bayes\' problems using a table with columns for the partition, prior probabilities, likelihoods, joint probabilities, and posterior probabilities. This systematic approach reduces errors and speeds up your work. Aim to solve Bayes\' problems in under three minutes on exam day. Work through at least 15 practice problems covering different setups: risk classification, diagnostic testing, and quality control scenarios. Pay special attention to problems where the partition has three or more categories, as these require more careful arithmetic.