Transformations with Inverse Gamma RVs

96 Views Asked by At

I have two independent random variables:

  1. $X_{1}$ is governed by an Inverse Gamma(a, b).
  2. $X_{2}$ is governed by an Inverse Gamma(a, b).

Given the transformation $Y = X_{1}/X_{2}$, I want to find the marginal distribution for $Y$, $f_{Y}(y)$; however, I am a little confused how to proceed with the transformation. Here is how I tried thinking of it:

To make things easier, I tried assigning $W = X_{2}.$

$$Y = X_{1}/X_{2} \rightarrow Y = X_{1}/W$$

$$W = X_{2} \rightarrow X_{2} = W$$

I am unsure if this is correct and how to proceed.

1

There are 1 best solutions below

0
On BEST ANSWER

To simplify the notation let's set $Z=\frac{X}{Y}$


Let's set

$$\begin{cases} z=\frac{x}{y}\\ u=x, \end{cases}\rightarrow\begin{cases} x=u\\ y=\frac{u}{z}, \end{cases}$$

The jacobian is $|J|=\frac{u}{z^2}$ and thus the joint density $(U,Z)$ is

$$f_{UZ}(u,z)=\frac{z^{a-1}}{\Gamma(a)\Gamma(a)}\cdot \frac{b^{2a}e^{-\frac{b}{u}(z+1)}}{u^{2a+1}}$$

This can be rewritten as

$$\frac{\Gamma(a+a)}{\Gamma(a)\Gamma(a)}z^{a-1}(1+z)^{-a-a}\cdot \frac{b^{2a}e^{-\frac{b}{u}(z+1)}}{\Gamma(2a)}\Bigg(\frac{1+z}{u}\Bigg)^{2a+1}\cdot\frac{1}{1+z}$$

To get your $f_Z(z)$ you must integrate

$$\frac{\Gamma(a+a)}{\Gamma(a)\Gamma(a)}z^{a-1}(1+z)^{-a-a}\cdot\underbrace{\int_0^{\infty} \frac{b^{2a}e^{-\frac{b}{u}(z+1)}}{\Gamma(2a)}\Bigg(\frac{1+z}{u}\Bigg)^{2a+1}d\Bigg(\frac{u}{1+z}\Bigg)}_{=1}$$

Thus your density is a Beta prime as the integral is an integral of an Inverse gamma

$$ \bbox[5px,border:2px solid red] { f_Z(z)=\frac{\Gamma(a+a)}{\Gamma(a)\Gamma(a)}z^{a-1}(1+z)^{-a-a}\cdot\mathbb{1}_{[0;+\infty)}(z) \ } $$


If you do not like to deal with Inverse Gamma rv's you can rewrite you $Z$ in the following way

$$Z=\frac{X}{Y}=\frac{\frac{1}{Y}}{\frac{1}{X}}$$

and thus you can use Gamma distributions if you are familiar with them.

In fact, if $X\sim Gamma$ then $\frac{1}{X}\sim InverseGamma$