So... I'll be honest, I don't know anything about anything Bayesian, this problem being no exception (from the Society of Actuaries' Exam C sample questions):

You are given:

(i) The annual number of claims for a policyholder has a binomial distribution with probability function: $$p\left(x \mid q\right) = {2\choose x}q^{x}(1-q)^{2-x}\text{, }\quad x = 0, 1, 2$$

(ii) The prior distribution is $$\pi(q) = 4q^{3}\text{, }\quad 0 < q < 1$$

The policyholder had one claim in each of Years 1 and 2.

Determine the Bayesian estimate of the number of claims in Year 3.

I have decided not to look at the solution, as I am attempting to study for this exam through reverse engineering of questions (i.e., grab a question and learn about the underlying concepts using a text). Unlike the previous questions for which this method has been successful, I hit a brick wall with this one.

So from what I understand, unlike the frequentist mindset, with Bayesian statistics, you assign a "prior" distribution to some variable. The only other thing I know about Bayesian methods is what a posterior distribution is (by definition) and how it can be motivated through Bayes' theorem.

Guidance with this problem is appreciated.


You have a prior distribution $\pi(q)$, a likelihood function $p(x|q)$ and two observations $x_1=1$ and $x_2=1$. So your posterior distribution for $q$ is $$\pi(q|x_1,x_2)=\frac{\pi(q)p(x_1|q)p(x_2|q)}{\displaystyle \int_0^1 \pi(r)p(x_1|r)p(x_2|r)\; dr}$$ in effect taking the product of the prior and the likelihoods and then turning this into a probability density by dividing by its integral over all possible $q$ (represented here by $r$ as $q$ is also a free variable), though knowing $0 \le q \le 1$.

If you want the expected number of accidents in year $3$, then your next step would be to calculate $$\int_0^1 (2 \times p(2|q) + 1 \times p(1|q) + 0 \times p(0|q)) \pi(q|x_1,x_2)\; dq$$

I will leave the calculus to you.

  • $\begingroup$ Thank you for your response. I have two questions - in your first equation, I noticed that you have $p(x_1 \mid q)p(x_2 \mid q)$ as a product in the numerator; is this because we assume that conditional on $q$, $X_1$ and $X_2$ are independent? Also, is the second equation supposed to have $0 \times p(0\mid q)$ instead of $0 \times p(1\mid q)$? $\endgroup$ – Clarinetist Sep 6 '14 at 19:57
  • $\begingroup$ Yes, and yes ($0$ corrected) $\endgroup$ – Henry Sep 6 '14 at 20:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.