3
$\begingroup$

In order to draw a sample from an N-dimensional Gaussian copula, we draw N independent standard Gaussian random variables, form a vector, and multiply it by an appropriate matrix (Cholesky and such). Each independent standard Gaussian can be computed by drawing a sample from the uniform distribution on [0, 1] followed by an application of the inverse of the standard Gaussian CDF.

Question: Given an arbitrary N-dimensional copula and N independent samples uniformly distributed on [0, 1], is it possible to demonstrate that there is always a transformation which allows one to obtain a sample from the copula using only the uniform samples?

Update: I’m aware of the Rosenblatt transformation. The inverse of this transformation might be an answer to my question. I don’t have deep knowledge in this area, and I’m looking for a person who could consolidate all ideas and describe the whole process step by step.

Improve this question
$\endgroup$
8
  • $\begingroup$ What kind of "transformation" are you looking for? I am puzzled, because according to the definitions of copula I know (such as in Nelsen's book), it is defined as a distribution $F$ on $[0,1]^N$, so one can "sample" from it with a uniform vector $(x_1,\ldots,x_N)$ merely by computing $F(x_1,\ldots,x_N)$. $\endgroup$
    – whuber ♦
    Commented Dec 10, 2014 at 17:03
  • $\begingroup$ @whuber, I’m looking for a transformation that would turn N independent uniformly distributed random variables into N dependent uniformly distributed random variables according to the copula. $F(x_1, \dots, x_N)$ is just a single value, isn’t it? Or may be I’m missing your point. Can you please elaborate a bit? Thank you. $\endgroup$
    – Ivan
    Commented Dec 10, 2014 at 20:29
  • $\begingroup$ Are you perhaps asking the multidimensional generalization of the question at stats.stackexchange.com/questions/124865/…? $\endgroup$
    – whuber ♦
    Commented Dec 10, 2014 at 20:57
  • $\begingroup$ @whuber, thank you for your time. It’s certainly related, but not quite. I’m trying to find a way to correlate independent variables according to a given copula. Please have a look at the update in the original post. $\endgroup$
    – Ivan
    Commented Dec 11, 2014 at 9:30
  • 1
    $\begingroup$ Yes, I think so. $\endgroup$
    – whuber ♦
    Commented Dec 11, 2014 at 15:15

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.