Algebra on gamma distributions

Can anyone help me with this Q about algebra on gamma distributions?

If the pdf *f(x)* of *X* is given by a gamma distribution *gamma(shape, scale)*

...and the pdf *g(y)* of *Y* is given by *(A/X^2) * f(A/Y)*

...where Y is X*N

...what I want is a parametric form for *g(y)*. I __think__ it should be an inverse gamma distribution, and I just need to do some algebra to find the parameters.

...now first Q: am I correct in thinking that *f(A/Y)* is an inverse gamma distribution *gammainv(shape, newscale)*

...where *newscale = A/(N*scale)*

...(note this is* f(A/Y), *not *f(A/X) *which would be a gamma distribution *gamma(shape, A/scale)*

...which comes from two properties of gamma distributions, according to wikipedia:

- if X = gamma(scale, shape), 1/X = invgamma(shape, 1/scale)
- if X = invgamma(scale,shape), cX = invgamma(shape, c*scale)

...second Q: how should I now add the pre-multiplication by *(A/X^2)* to this?

Multiply the scale parameter again, directly?

*....(A/X^2) * f(A/Y) = g(y) = **gammainv(shape, newscale2))*

where *newscale2 = (A / (Y/N)^2 ) * newscale*

* = (A / (Y/N)^2) * ( A / (N*scale) ) *

= *A^2 / ( (Y/N) * scale ) *

this could be completely wrong of course...

Ta!

Re: Algebra on gamma distributions

Hey JohnGriffiths.

You have written a lot down and its a bit hard to decipher: Can you just point out in the simplest way what distribution you have (call it X) and what new random variable you want (i.e f(X)) by stating the transformation of the variable itself without adding all the extra stuff.

Re: Algebra on gamma distributions

Check if the following CDF for g(y) is of any use for x>=0:

(Shake)

Re: Algebra on gamma distributions

How about this:

"Of what distribution (gamma, inverse gamma, etc.) is

A/X^2 * f(A/Y)

the pdf, where X is a random variable, A is a constant, Y is a linear function of X and A, and f is a gamma distribution with scale and shape parameters k and theta ? "

*Hey JohnGriffiths.*

You have written a lot down and its a bit hard to decipher: Can you just point out in the simplest way what distribution you have (call it X) and what new random variable you want (i.e f(X)) by stating the transformation of the variable itself without adding all the extra stuff.