All approaches discussed above and further presume
the imposition of the target correlation structure
only by matrix manipulations. The task can be
understood as the simulation from the multivariate
distribution model consistent with the prescribed
marginals and covariances. Most existing models for
random vectors, however, are restricted to the
bivariate case and/or can only describe the small
correlation between variables. Two models based on
the earlier works of Nataf and Morgenstern are recommended
by Liu and Kiureghian (1986).