A Short Note on a Multi-Gaussian Model for Multivariate Estimation and Simulation

Abstract

We show that multivariate non-linear features can be reconstructed in a simple and straightforward way by mapping the original p-variate cumulative distribution function with a p-variate Gaussian distribution, equipped with a proper prior covariance matrix. We present theoretical development that allows a proper fitting of such covariance matrix, allowing to continue with the decorrelation of variables by the use principal components analysis.