1
$\begingroup$

I am wondering if there is a closed form for finding the expected value or variance for a conditional exponential distribution.

For example: $$ E(X|x > a) $$ where X is exponential with mean $\lambda$.

Same question for variance.

What about for a joint distribution of independent exponentials?

$$ E(X|y > a) $$ where X is exponential with mean $\lambda$, Y is exponential with mean $\theta$ and X & Y are independent.

A sample problem for the actuarial P/1 exam (#124 for those also studying) asks:

The joint probability for $f(x,y) = 2e^{-x-2y}, ~ x > 0, ~ y > 0$. Calculate the variance of Y given $x > 3, ~ y > 3$.

The solution goes like this: (Math on the right, reasoning on the left)

  1. $Var (Y|x>3, y>3) =$
  2. $Var (Y|x>3) = ~~~~~$Independence
  3. $Var (Y + 3) = ~~~~~$Memoryless
  4. $Var (Y) + Var (3) =~~~~~$Independence of Y and 3.
  5. $Var (Y) = ~~~~~ $ Since $Var (3) = 0$.
  6. $0.25 ~~~~~ $Exponential Variance, $\lambda = 2$.

So this says to me that $Var (Y|x>3) = Var (Y)$. Is that true? If so, is it always true? If not, then how does this solution work?

Could one also replace E(Y) for Steps 1 - 4, Use $E(a) = a$ and get $E(Y| y>a) = E(y) + a$?

Shortcuts like this are immensely valuable for a timed test. (Not just faster, but less error prone).

$\endgroup$
1
$\begingroup$

Let $X$ be an exponentially distributed random variable with parameter $\lambda$. We can think of $X$ as the lifetime of a specific Carbon $14$ atom. Given that $X\gt a$, the additional lifetime $Y$ of the atom is exponentially distributed with parameter $\lambda$. This is a key property of the exponential, usually called memorylessness.

Given that $X\gt a$, $X=a+Y$. It follows that $E[X|X\gt a]=a+E[Y]=a+E[X]$.

The second question of the same kind is much simpler. If $X$ and $Y$ are any independent random variables, then $E[X|Y\gt a]=E[X]$.

Back to the exponential. Since the random variable $X|X\gt a$ is just $a+Y$, where $Y$ is exponential, we have $\text{Var}(X|X\gt a)=\text{Var}(a+Y)=\text{Var}(Y)=\text{Var}(X)$.

Since we know that an exponentially distributed random variable $X$ with parameter $\lambda$ has mean $\frac{1}{\lambda}$ and variance $\frac{1}{\lambda^2}$, we can write down with no trouble the mean and variance of the conditional version.

$\endgroup$
1
$\begingroup$

The memorylessness property of the exponential distribution states that for if $X$ is exponentially distributed then for any function $f$ we have $E(f(X)|X>\alpha) = E(f(X+\alpha))$. Essentially this means that the distribution of an exponential conditioned to be greater than $\alpha$ is exactly the same as the distribution of an exponential with $\alpha$ added to it.

So your method works for both the mean and the variance. It's also useful to know what happens to the mean and variance of a random variable when you add a constant. (You add the constant to the mean, but leave the variance alone.)

In a maths exam you would be required to prove that the distribution of $X$ and $Y$ are independent exponentials. I don't know if that would be true in an actuary's exam.

$\endgroup$

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.