0
$\begingroup$

I have a random variable X (which has some complicated PDF, it could be approximated to exponential), and I'm trying to find the distribution of another random variable Y, which is the sum of j samples of (1/X)

$Y_j=\sum_{i=1}^j(1/Xi)$

is there a way to find the distribution, CDF or CCDF of Y? regards

$\endgroup$
0
$\begingroup$

Following are the steps to be followed,

  1. Find the distribution of $\frac{1}{X}$ by using the transformation $Z=\frac{1}{X} \implies \left|\frac{\partial X}{\partial Z}\right| = \frac{1}{Z^2}$

$$f_{Z}(z) = f_{X}\left(\frac{1}{z}\right)\frac{1}{z^2}$$

  1. Find the moment generating function of $Z$ which is $M_Z(t) = E(e^{tZ})$.

  2. Compute the moment generating function of $Y_j = \sum\limits_{i=1}^{j}Z_i$ where $Z_i$'s are iid random variables following the distribution $f_Z(z)$.

$$M_{Y_j}(t) = E(e^{tY_j}) = E(e^{t\sum\limits_{i=1}^{j}Z_i}) = \prod\limits_{i=1}^{j}E(e^{tZ_i}) = \prod\limits_{i=1}^{j}M_{Z_i}(t) = M_{Z}(t)^j$$

Note that in $3$ I used the independence of $Z_i$'s in third equality and $M_{Z_i}(t) = M_{Z}(t)$ in last equality since both $Z_i$ and $Z$ have same distribution.

Alternative $2$ and $3$ steps if the moment generating function does not correspond to commonly known distributions,

  1. Let $U_1 = Y_j = \sum\limits_{i=1}^{j}Z_i$ and $U_i=Z_i, \forall i \in \left\{2,3,\ldots j \right\}$, then $Z_1 = U_1 - \sum\limits_{i=2}^{j}U_i$ and $Z_i=U_i, \forall i \in \left\{2,3,\ldots j \right\}$. The determinant of Jacobian of $Z$ with respect to $U$ will be equal to $1$. Then,

$$f_{U_1, U_2, \ldots, U_j}(u_1, u_2, \ldots, u_j) = f_{Z_1, Z_2, \ldots, Z_j}(u_1-\sum\limits_{i=2}^{j}u_i, u_2, \ldots, u_j) \cdot |det(J)|$$

$$f_{U_1, U_2, \ldots, U_j}(u_1, u_2, \ldots, u_j) = f_{Z_1}(u_1-\sum\limits_{i=2}^{j}u_i)f_{Z_2}(u_2)\ldots f_{ Z_j}(u_j) \cdot 1$$

The above equation follows from the independence of $Z_i$'s.

  1. Integrate the joint pdf with respect to variables $U_2, U_3, \ldots, U_j$ to get the pdf of $U_1 (=Y_j)$.

$$\int_{u_1, u_2, \ldots u_j}f_{U_1, U_2, \ldots, U_j}(u_1, u_2, \ldots, u_j) \partial u_1 \partial u_2 \ldots \partial u_j = \int_{u_1, u_2, \ldots u_j}f_{Z_1}(u_1-\sum\limits_{i=2}^{j}u_i)f_{Z_2}(u_2)\ldots f_{ Z_j}(u_j)\partial u_1 \partial u_2 \ldots \partial u_j$$

The resulting function in $u_1$ will be the pdf of $U_1 = Y_j$.

$\endgroup$

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.