4
$\begingroup$

We have $Z \sim N(0,1)$ and $Y=a+bZ+cZ^2$. What is the variance of $Y$?


This is what I did:

$$\mbox{Var} (Y)=0+b^2 \, \mbox{Var} (Z)+c^2 \, \mbox{Var} (Z^2) = b^2 + c^2 \, \mbox{Var} (Z^2)$$

To get $\mbox{Var} (Z^2)$, I tried to use the definition $\mbox{Var} (Z^2)= \mathbb E(Z^4)-\mathbb E^2(Z^2)$ But I'm having trouble with this part. If this was a odd for example $\mathbb E(Z^3)$ you can say that because of the symmetry of the normal distribution $\mathbb E(Z^3)=0$, but in this is pair.

$\endgroup$
  • 2
    $\begingroup$ If you have a browser and an internet connection, which apparently you do, it's not terribly difficult to find the value of $E(z^4)$. $\endgroup$ – Mark L. Stone Jul 20 '16 at 18:04
  • 1
    $\begingroup$ $Z^2$ will be a chi-squared random variable. $\endgroup$ – Patrick Li Jul 20 '16 at 18:13
  • 5
    $\begingroup$ You overlooked one term in your calculation: $\operatorname{Cov}(bz,cz^2)$. Although it turns out to be zero, it does need to be considered for your derivation to be complete and valid. $\endgroup$ – whuber Jul 20 '16 at 19:24
  • 1
    $\begingroup$ Please add the [self-study] tag & read its wiki. Then tell us what you understand thus far, what you've tried & where you're stuck. We'll provide hints to help you get unstuck. $\endgroup$ – gung - Reinstate Monica Jul 20 '16 at 20:05
  • $\begingroup$ Also asked by the OP on math.SE where it has received three answers including one that the OP has accepted. $\endgroup$ – Dilip Sarwate Jul 21 '16 at 1:32
5
$\begingroup$

Solution 1: look it up.

Write

$$Y = c\left(Z - \frac{b}{2c}\right)^2 + a - \frac{b^2}{4c}.$$

Since $Z-\frac{b}{c}$ has a Normal$(-b/c, 1)$ distribution, its square has a non-central chi-squared distribution with one degree of freedom and noncentrality parameter

$$\lambda = \left(-\frac{b}{2c}\right)^2.$$

Its variance therefore is $2(1 + 2\lambda)$. Consequently the variance of $Y$ is $c^2$ times that, reducing to

$$\operatorname{Var}(Y) = c^2 2(1 + 2\lambda) = b^2 + 2c^2.$$

Solution 2: compute it.

Use the well-known (and easily derived) result that $\mathbb{E}(Z^4) = 3$ as well as the obvious values $\mathbb{E}(Z^3)=\mathbb{E}(Z) = 0, \mathbb{E}(Z^2)=1$. Starting with the definition of variance, expand the expressions into sums of simpler terms and use the linearity of expectation to simplify them:

$$\eqalign{ \operatorname{Var}(Y) &= \operatorname{Var}(cZ^2 + bZ + a) \\ &= \mathbb{E}((cZ^2 + bZ + a)^2) - (\mathbb{E}(cZ^2 + bZ + a))^2 \\ &= \mathbb{E}((c^2Z^4 + 2bcZ^3 + (2ac+b^2)Z^2 + 2abZ + a^2) - (c + 0 + a)^2 \\ &= 3c^2 + 0 + (2ac+b^2) + 0 + a^2 - (a^2 + c^2 + 2ac) \\ &= b^2 + 2c^2. }$$

Solution 3: exploit basic identities.

This answer uses only the basic properties of covariance and independence, the values of the given two moments of $Z$ (its mean and variance), and the fact that uncorrelated jointly Normal distributions are independent.

Let $Z$ and $W$ independently have standard Normal distributions. Then the two variables $Z\pm W$ have jointly Normal distributions with zero mean and variance $2$. This is equivalent to having the same distribution as $Z$, multiplied by $\sqrt{2}$. Thus

$$\operatorname{Var}((Z\pm W)^2) = \operatorname{Var}((\sqrt{2}Z)^2) = \operatorname{Var}(2Z^2) = 4\operatorname{Var}(Z^2).$$

Moreover, since

$$\operatorname{Cov}(Z+W, Z-W) = \operatorname{Var}(Z) - \operatorname{Var}(W) = 1-1=0,$$

$Z\pm W$ are uncorrelated, implying they are independent. Consequently $(Z\pm W)^2$, since they are functions of independent variables, are also independent. Their covariance must be zero. Therefore

$$8\operatorname{Var}(Z^2) = \operatorname{Var}((Z+W)^2 - (Z-W)^2) = \operatorname{Var}(4ZW) = 16 \operatorname{Var}(Z)\operatorname{Var}(W)=16,$$

showing that

$$\operatorname{Var}(Z^2) = \frac{16}{8} = 2.$$

Finally, note that because $(Z, Z^2)$ and $(-Z, Z^2)$ have identical distributions, they have equal covariances; but since their covariances must be negatives of each other, their covariances are both zero:

$$\operatorname{Cov}(Z^2, Z) = -\operatorname{Cov}(Z^2, Z) = 0.$$

Now

$$\operatorname{Var}(cZ^2 + bZ + a) = c^2\operatorname{Var}(Z^2) + b^2 \operatorname{Var}(Z) + 2bc \operatorname{Cov}(Z^2, Z) = 2c^2 + b^2.$$

| cite | improve this answer | |
$\endgroup$

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.