A car insurance company has $2,500$ policy holders. The expected claim paid to a policy holder during a year is $1,000$ with a standard deviation of $900$. What premium should the company charge each policy holder to assure that with probability $0.999$ the premium income will cover the cost of the claims?

I'm having trouble with interpreting this problem. I want to correctly denote the random variables to use CLT or Chebyshev's inequality.
So I suppose $X$ to be the cost of the claims, then we have $E[X]=1000$, $Var(X) = 900^2$.
Let $y$ be the cost of the premium, so we need to find $ P(Y-X\geq 0)$ right?


No; you need to look at the aggregate claims random variable. The idea is that the insurer needs to collect sufficient premiums to cover the aggregate claims with at least $0.999$ probability.

Assuming each claim is independent, then the aggregate claims $S = \sum_{i=1}^{2500} X_i$ where $X_i$ is the annual claim size for policyholder $i$, for the book of business is approximately normally distributed with mean $$\operatorname{E}[S] = 2500 \operatorname{E}[X_i] = 2500000,$$ and variance $$\operatorname{Var}[S] = 2500 \operatorname{Var}[X_i] = 2250000.$$ Consequently, $$Z = \frac{S - \operatorname{E}[S]}{\sqrt{\operatorname{Var}[S]}}$$ is approximately standard normal, and we want to find some number $P$ such that $$\Pr[S \le P] = 0.999;$$ equivalently, $$\Pr\left[ Z \le \frac{P - \operatorname{E}[S]}{\sqrt{\operatorname{Var}[S]}}\right] = 0.999.$$ Since the $99.9^{\rm th}$ percentile of the standard normal is approximately $3.09023$, it follows that $$\frac{P - 2500000}{1500} = 3.09023,$$ or $$P = 2504635.35.$$ This represents the total premiums the insurer needs to collect; dividing by $2500$ gives the per-policyholder premium, assuming that all policyholders are rated equally. What is the excess above the expected loss per policyholder?


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.