0
$\begingroup$

Question: Let Θ and X be independent random variables each Uniform U(0,1) distributed.

Let Z = X/Θ

What is the marginal density of Z?

I know that: Given Θ = θ, the density of Z is Uniform U(0, 1/θ).

However, I'm not exactly sure how to get the PDF of Z from this conditional PDF of Z given θ.

Would really appreciate any help! Thanks!

  • 1
    Define W=X and find the distribution of (Z, W) using Jacobian theorem (change of variables) – 2017-01-13
  • 1
    Because the denominator of Z can be very near 0, Z must have a very skewed distribution with a very fat tail. – 2017-01-13
  • 0
    @BruceET: the density for $Z \gt 1$ follows a power-law distribution with finite first moment but infinite second moment – 2017-01-13
  • 0
    $\log Z$ follows a Laplace distribution – 2017-01-13

2 Answers 2

0

Start by finding $P(Z \leq z)$, for fixed $z$. You can write this probability as a double integral over the joint density between $\Theta$ and $X$. Then differentiate.

  • 0
    Possibly easier if you deal with the cases $z \le 1$ and $z \gt 1$ seperately – 2017-01-13
  • 0
    Thank you so much! I tried finding the P(Z≤z) = P(X/Θ ≤ z) = P(X ≤ zΘ). I get ½ from taking the derivative of the CDF when z ≤ 1. But I'm not sure what I did wrong for the z > 1 case: I took the double integral from (0 to 1/z) and (0 to zΘ) of 1 (the joint pdf of x and Θ) dx dΘ. I got 1/(2z) and took the derivative but got -1/(2z^2). Since pdf can't be negative, do you mind explaining my mistake? – 2017-01-13
  • 0
    @user406509 You made *no* mistake in the derivation, however a PDF is an *unsigned derivative* of the CDF, so you just needed to use the absolute value of the derivative. – 2017-01-13
  • 0
    I'm getting that $P(X/\Theta \leq z) = 1 - \frac{1}{2z}$ when z > 1. Double-check your limits of integration. The math becomes simpler if you compute $P(Z > z)$ instead, in which case, the integral becomes $$\int_0^1 \int_0^{x/z} d\theta dx$$. – 2017-01-14
0

I know that: Given Θ = θ, the density of Z is Uniform U(0, 1/θ).

So then you know the joint distribution:$$\begin{align}f_{Z\mid \Theta}(z\mid \theta)\cdot f_{\Theta}(\theta) &= (\theta\cdot\mathbf 1_{z\in(0; 1/\theta)})\cdot(\mathbf 1_{\theta\in(0;1)})\\[1ex] &= \theta\cdot\mathbf 1_{z\in(0;\infty)}\cdot\mathbf 1_{\theta\in(0;\min\{1,1/z\})}\end{align}$$

However, I'm not exactly sure how to get the PDF of Z from this conditional PDF of Z given θ.

Just apply the Law of Total Expectation to obtain the marginal for $Z$

$$\begin{align}f_{Z}(z) &= \int_\Bbb R f_{Z\mid \Theta}(z\mid \theta)\cdot f_\Theta(\theta)\operatorname d\theta \\[1ex] &= \int_0^{\min\{1,1/z\}}\theta\cdot\mathbf 1_{z\in(0;\infty)}\operatorname d\theta \\[1ex] & = \end{align}$$

You can take it from there.

PS: Which, by the way, is what you could obtain using vvnitram's suggestion of the Jacobian change of variables theorem.

PPS: Also note Henry's useful tip of considering the cases, $z<1$ and $z\geq 1$.

$$\begin{align}f_{Z}(z) &= \int_0^{\min\{1,1/z\}}\theta\cdot\mathbf 1_{z\in(0;\infty)}\operatorname d\theta \\[1ex] & = \mathbf 1_{z\in(0;1)}\cdot\int_0^1\theta\operatorname d \theta~+~\mathbf 1_{z\in[1;\infty)}\cdot\int_0^{1/z}\theta\operatorname d \theta \\[1ex] &= \end{align}$$

  • 0
    Thanks! Do you mind explaining how you got the integral limit of min{1, 1/z}? – 2017-01-13
  • 0
    @user406509 Sure. You just partition by the cases. $$\min\{1,1/z\}\cdot\mathbf 1_{z>0} ~=~ \begin{cases}1 &:& 0 – 2017-01-13
  • 0
    @user406509 Also the bounds come from support for the joint pdf. $$\begin{align}\bigl\{(z,\theta): 0< z\theta< 1~\wedge~ 0<\theta<1\bigr\} &\equiv \bigl\{(z,\theta): (0 – 2017-01-14