Maximum of two correlated random variables

AI Thread Summary
The discussion focuses on finding the maximum of two correlated non-Gaussian random variables, A and B, and seeks an analytical or approximate solution for their bivariate distribution. The user mentions existing work that assumes Gaussian variables but requires a method applicable to their scenario. They propose using a quadratic Taylor polynomial to approximate the maximum operation, specifically questioning how to analytically express the absolute difference, abs(A-B). The user provides the forms of A and B in terms of correlated parameters x and y, and requests guidance on deriving their joint probability density function (PDF). The conversation emphasizes the challenge of working with non-Gaussian distributions and the need for a suitable approximation method.
touqeerazam
Messages
6
Reaction score
0
Hi all,

I want to find maximum of two random variables which are correlated and are non gaussian too. Baiscally I need an analytical orr approximate solution to their bivaraite distribution with means and varaince of resulting distribution. There is some work by Clark 'The greatest of finite set of random variables' but that assumes gaussian correlated variables.

so if A & B are two correlated random varaibles. I need C=Max(A,B)?

one other method is to use quadratic taylor polynomial for A & B. and use Max (A,B)=(A+B+abs(A-B))/2. But I don't know can i approximate abs(A-B) by quadratic polynomial (without regression). In other words, if I can get any method to approximate abs(A-B) by analytical expression. This will also give me Max operation (what I really need).

Sorry for long question

I will be very grateful to you if anyone could figure out solution or any directions

cheers
Touqeer
 
Physics news on Phys.org
Are you given the joint distribution f of Aand B? then you could try and compute the distribution of C=max(A,B) as

<br /> \mathbb{P}(C\leq c)=\int_{-\infty}^c{da\,\int_{-\infty}^a{db\,f(a,b)}}+\int_{-\infty}^c{db\,\int_{-\infty}^b{da\,f(a,b)}}.<br />
 
Hi,

Thanks Pere. I don't have their joint dis 'f'. Its difficult to get JPDF of correlated non gaussian variables (not sure how to get). What all I have is varaible A & B given as,

A=a0+a1x+a2x^2+a3y+a4y^2
B=b0+b1x+b2x^2+b3y+b4y^2

where x & y are two parameters which are correlated non gaussian, and I have their PDFs. ai's and bi's are just coefficients.

Can you please point out any solution?

Thanks
Touqeer
 

Similar threads

Back
Top