Concept

Bayes Theorem

Bayes theorem updates a prior probability after a signal arrives. In actuarial work, it explains why base rates matter in underwriting flags, claim triage, credibility, and classification models.

What Bayes theorem answers

Bayes theorem answers a reversal question. You may know how often a signal appears among high-risk policies, but the business question is often the reverse: among flagged policies, how many are actually high risk?

The reversal depends on two ingredients: the signal accuracy and the base rate of the event before the signal. Ignoring the base rate is the fastest way to overstate a flag, test, or model score.

Core formulas

The numerator combines the signal rate inside the event group with the event base rate. The denominator is the total probability of seeing the signal across all groups.

Bayes theorem
P(AB)=P(BA)P(A)P(B)P(A\mid B)=\frac{P(B\mid A)P(A)}{P(B)}
Law of total probability denominator
P(B)=iP(BAi)P(Ai)P(B)=\sum_i P(B\mid A_i)P(A_i)

Flag worked example

Suppose 2% of policies are high risk. A screening flag catches 80% of high-risk policies and falsely flags 10% of the other policies. Among flagged policies, the high-risk share is about 14%.

The 80% catch rate is not the answer. The flag group also contains false positives from the much larger non-high-risk population.

Flag posterior
0.80(0.02)0.80(0.02)+0.10(0.98)0.140\frac{0.80(0.02)}{0.80(0.02)+0.10(0.98)}\approx 0.140
Runnable Python

Run a Bayes flag calculator

Change the base rate, catch rate, or false-flag rate and watch the posterior move. This is useful for seeing why rare-event signals can still have modest posterior probabilities.

Expected output
Flag rate: 11.400%
High-risk share among flagged policies: 14.035%

Actuarial interpretation

Bayes theorem is a small formula with a large modeling lesson. A risk signal can rank policies well and still have a modest posterior probability when the target event is rare.

That is why underwriting and claims teams need calibrated probabilities, not just labels. A flag that means 14% high risk may still be useful, but it should not be communicated as if it means 80% high risk.

Exam traps

The most common Bayes error is treating the signal accuracy as the posterior probability. The second is leaving false positives out of the denominator.

  • Write down the base rate before touching the signal rates.
  • Put every way to get the observed signal in the denominator.
  • Check whether the question asks for event given signal or signal given event.
  • If the target event is rare, expect the posterior to be much lower than the catch rate.

Where Bayes connects next

Empirical Bayes and credibility theory extend the same idea: combine prior information with observed experience. Classification metrics and predictive-model calibration use the same base-rate logic in a modern analytics setting.

Practice

Original exam practice

3 questions built from syllabus outcomes and released-exam patterns. The prompts and answers are original, so they train the skill without copying official exam text.

Bayes Theorem Base-Rate Drill

Original Bayes checks for posterior probabilities, base-rate interpretation, and false-positive denominators.

Exam P - 15 min
Source pattern: SOA Exam P Bayes theorem outcomes; original screening and classification prompts.
  1. Question 1/Calculation

    Posterior from a flag

    Two percent of policies are high risk. A flag catches 80 percent of high-risk policies and falsely flags 10 percent of other policies. Among flagged policies, what percent are high risk?

    Corebayes-theorembase-ratesscreening-flagsexam-pBayes TheoremLaw of Total Probability
    Solution and grading points

    High-risk flagged policies contribute 0.80 times 0.02, or 0.016. All flagged policies contribute 0.016 plus 0.10 times 0.98, or 0.114. The high-risk share among flagged policies is about 14.0 percent.

    1. Compute the true-positive route by multiplying the catch rate by the high-risk base rate.
    2. Compute the false-positive route by multiplying the false-flag rate by the non-high-risk share.
    3. Divide the true-positive route by the total flagged group.
    • Uses the high-risk base rate.
    • Includes false positives from the other policies.
    • Reports the posterior probability, not the 80 percent catch rate.
  2. Question 2/Written Answer

    Catch rate is not posterior

    A model catches 90 percent of policies that later have a large claim. Explain why a flagged policy does not automatically have a 90 percent chance of a large claim.

    Solution and grading points

    The 90 percent figure describes how often the model flags large-claim policies. To find the chance that a flagged policy has a large claim, you also need the base rate and the false-positive rate.

    • Identifies the direction of the 90 percent statement.
    • Mentions the base rate.
    • Mentions false positives or the total flagged group.
  3. Question 3/Written Answer

    Denominator routes

    In a two-class Bayes problem with a screening flag, what belongs in the denominator?

    Solution and grading points

    The denominator should include every route into the flagged group: flagged high-risk policies plus flagged non-high-risk policies.

    • Includes true positives.
    • Includes false positives.
    • Frames the denominator as the total flagged group.

References and official sources