Conditional Probability: FRM Part I
Conditional probability updates the probability of an event after observing information.
- Bayes’ theorem combines the prior, likelihood, and total probability of the evidence.
- P(A given B) is generally different from P(B given A).
- Name the conditioning group before choosing the denominator in a Bayes calculation.
On this page8 sections
Definition and multiplication rule
Conditional probability P(A|B) is the probability of event A given that event B has occurred, provided P(B) is positive. It is calculated as P(A and B) divided by P(B). The denominator is the probability of the condition.
The multiplication rule follows: P(A and B) = P(A|B)P(B). It also equals P(B|A)P(A). These forms are useful when the joint event is easier to compute through one sequence.
Independent events satisfy P(A|B) = P(A), so learning B does not change the probability of A. Independence is different from mutual exclusivity. Mutually exclusive events cannot occur together, while independent events generally can.
Check the event definitions. The phrase “given a positive signal” changes the denominator to all cases with a positive signal, not all cases that are distressed.
Bayes’ theorem
Bayes’ theorem reverses conditioning: P(A|B) = P(B|A)P(A) / P(B). The denominator can be expanded using the law of total probability, such as P(B|A)P(A) + P(B|not A)P(not A).
The prior P(A) represents the base rate before observing B. The likelihood P(B|A) describes how often the signal appears when A is true. The posterior P(A|B) combines both with the false-positive rate.
A common error is treating P(B|A) as P(A|B). Even a highly sensitive warning signal can have a modest posterior probability when the underlying event is rare and false positives are common.
Frequency tables often make the denominator clear. Translate probabilities into counts for a hypothetical population, separate true positives from false positives, then calculate the proportion of positive signals that correspond to A.
Worked base-rate example
Imagine 1,000 exposures: 100 are distressed and 900 are healthy. A signal appears for 80 distressed exposures and 90 healthy exposures. There are 170 signals total, of which 80 are associated with distress.
P(distressed | signal) = 80/170 ≈ 47.06%. P(signal | distressed) = 80/100 = 80%. These are different conditional probabilities with different denominators.
Bayes’ theorem gives the same posterior: sensitivity 0.80 times prior 0.10, divided by total signal probability (0.80×0.10 + 0.10×0.90) = 0.08/0.17 ≈ 47.06%.
If distress prevalence were only 1% while sensitivity and false-positive rate stayed the same, a hypothetical group of 10,000 would have 80 true positive signals and 990 false positives. The posterior would be 80/1,070, or about 7.48%. Lower prevalence reduces the positive predictive value.
Common distinctions and pitfalls
P(A|B) is not generally equal to P(B|A). Equality can occur in special cases but should not be assumed. Verify the denominator every time.
Independent events are not the same as mutually exclusive events. If A and B are independent and both have positive probability, they can occur together. Mutually exclusive events with nonzero probabilities cannot be independent.
For mutually exclusive alternatives A1 through An that cover all cases, total probability is P(B) = sum of P(B|Ai)P(Ai). If the alternatives omit a possible case, the denominator is incomplete.
Use exact event definitions. “A signal” could mean one threshold, any of several flags, or a specific test outcome. A different definition changes sensitivity, false-positive rate, and posterior.
In a portfolio, daily losses may be correlated because they share market factors. Assuming independence can understate the chance of a joint stress event. Use independence only when the problem states it or the model explicitly assumes it.
Exam workflow
Define A and B in words. Identify which event is the condition and which probability the question asks for. Then write the prior, likelihood, and evidence denominator.
If counts are given, create a 2-by-2 table. Fill true positives, false positives, false negatives, and true negatives. The desired conditional probability is a row or column fraction, depending on the condition.
Check that the denominator contains all cases satisfying the “given” event. Ensure probabilities remain between zero and one and sum appropriately across exhaustive alternatives.
When explaining a result, identify the base rate and the direction of conditioning. This is often the conceptual point behind a Bayes problem.
Use a probability tree or table when there are groups
A tree starts with mutually exclusive groups, such as default and no default, then branches into alert and no alert. Multiply along each path to get joint probabilities, then add paths that lead to the observed event. A two-way table performs the same accounting with counts. Both methods make clear that the denominator for a conditional probability is the set compatible with the known observation.
Consider a portfolio divided into corporate and sovereign exposures. If the task asks for the probability an exposure is corporate given that it has breached a limit, count all breaches first. Add corporate breaches and sovereign breaches to form the denominator, then use corporate breaches as the numerator. Using all corporate exposures as the denominator would answer the reverse question: the probability of a breach given corporate exposure.
If data are expressed as percentages, keep them on a common scale. A 4% event rate combined with a 75% conditional signal rate gives 0.04 × 0.75, or 0.03 joint probability. Do not mix 4 with 0.75. If using 100,000 hypothetical observations instead, scale every branch consistently and confirm the counts sum back to the total population.
Check conditional probability in risk applications
Credit-risk questions may distinguish probability of default given a rating, probability of a rating given default, and probability of default after a signal. Market-risk questions can condition a return distribution on a volatility regime. In each case, conditioning changes the population or information set being considered. Write the conditioning event in words to keep the interpretation connected to the application.
A model can also report a conditional probability based on estimated parameters. A mathematically correct calculation may still be unreliable if the base rate, sample selection, or signal quality is poor. For instance, a default model calibrated to a stable period may not represent a recessionary portfolio. Probability arithmetic does not validate the data or the assumptions that produced it.
When reviewing a result, check whether the observed event has been selected after looking at the data. Repeatedly testing many conditions can produce patterns that appear predictive by chance. The exam may focus on calculation, but the interpretation should preserve the distinction between a conditional estimate and a causal claim.
Check a joint-probability calculation with a two-way table
| Expression | Meaning | Denominator |
|---|---|---|
| P(A|B) | A given B occurred | All outcomes in B |
| P(B|A) | B given A occurred | All outcomes in A |
| P(A and B) | Both occur | Entire sample or probability space |
| P(B) | Total probability of B | All mutually exclusive routes to B |
The same table can answer other questions. The signal rate among distressed exposures is 80/100; the overall signal rate is 170/1,000; and the probability of no signal among healthy exposures is 810/900. Each quantity has a different denominator. Write the verbal question beside the fraction before calculating.
If counts are weighted or sampled unevenly, raw frequencies may not represent the population probabilities. In that case, use the stated priors and conditional rates, and apply total probability. A table remains useful, but its counts should reflect the weights or the problem’s probability assumptions.
Common questions
How do I read P(A|B)?
It is the probability of A among outcomes where B has occurred. The conditioning event B defines the denominator group.
Are independent events mutually exclusive?
Usually not. Independent events do not change each other’s probabilities; mutually exclusive events cannot occur together. These are different relationships.
Why does the base rate matter in Bayes theorem?
The prior probability and the event rate among alternative groups both contribute to the denominator. A strong signal can still have modest predictive value when the event is rare.
How can I avoid reversing a conditional probability?
Write the target in words, identify the conditioning group, and use its probability as the denominator before substituting counts.