You can generate correlated multivariate normal with arbitrary correlations,
then transform each dimension from normal to uniform using pnorm.
Note that the correlation of the uniforms is somewhat smaller than
the correlation of the corresponding normal variates. You can adjust the
correlation of the normals upward to get the output correlation you want.
More generally, look at the literature on copulas. Alexander
McNeil has created a library with a variety of copulas
http://www.math.ethz.ch/~mcneil/book/QRMlib.html
to accompany his book "Quantitative Risk Management: Concepts,
Techniques and Tools".
Tim Hesterberg
>Is their such a thing like multivariate UNIFORM distribution?
> I mean the multivariate distrubution which is marginally uniform in [0,1]
> for all the components and having a prescribed correlation matrix
> (non-diagonal)?
> How to generate such a thing? For a time being I would probably be satisfied
> if all the cross correlations are identical.
>
> Thank you
>
> Jewel Bright
========================================================
| Tim Hesterberg Research Scientist |
| timh@insightful.com Insightful Corp. |
| (206)802-2319 1700 Westlake Ave. N, Suite 500 |
| (206)283-8691 (fax) Seattle, WA 98109-3044, U.S.A. |
| www.insightful.com/Hesterberg |
========================================================
Download the S+Resample library from www.insightful.com/downloads/libraries
|