# Sum of t distributed variables

• Jan 9th 2014, 08:42 AM
Naena
Sum of t distributed variables
Hi, I am looking for a class of fat-tailed probability distributions $\displaystyle$\mathcal{D} which would satisfy the following:
Let $\displaystyle$X,Y\in\mathcal{D}$.$ be two uncorrelated random variables (not necessarily independent). Then it is true that also $\displaystyle$Z\in{\mathcal{D}}$$where \displaystyle Z=\sqrt{\beta}X + \sqrt{1-\beta}Y \ \ \ \forall\beta\in[0,1] In other words, I am trying to find a fat-tailed distribution that X,Y would be from for which no matter the value of beta, Z would always follow the same distribution. For example normal distribution satisfies this property. I tried t-distribution. However, for t-distributed X and Y, distribution of Z depends on \displaystyle \beta$$.

I was thinking about multivariate t-distribution such that X,Y are not independent but only uncorrelated but I cannot seem to get it right. Any ideas? Links? Papers?

Thanks guys :)
• Jan 9th 2014, 06:08 PM
chiro
Re: Sum of t distributed variables
Hey Naena.

You may have to introduce an extra transformation to get what you want. There is an area that is right up your alley called ancillary statistics:

Ancillary statistic - Wikipedia, the free encyclopedia

Take a look at this and see if you can adapt the knowledge and theory of this area to your needs.