I tried to fit three variables using the non-parametric method(kernel smooth), and those distribution for each of three variables are different, so cholesky descomposition won´t work for replicate the correlation calculated by the historical data when i´m trying to make a simulation. And I need a way to simulate these variables which meets follow requirements:
Each simulated variable should still follow it´s fitted non parametric distribution.
Correlation should be replicated.
Anyone has an idea? I´ll apreciate.
Zhe