topic Simulation from a multivariate normal distribution in SAS/IML Software and Matrix Computations

Simulation from a multivariate normal distribution

JennyQLi — Thu, 15 Nov 2012 01:53:27 GMT

I'm trying to simulate data from a multivariate normal distribution with mean=MU and variance=MVNCOV:

For example, if we want to run 1000 simulations from a multivariate normal with MU={6.3 1.8 3.7} and

MVNCOV={2, 3, 1,

3, 8, 5,

1, 5, 4}

Are the following codes correct? Thanks!

proc iml;

nobs=1000;

MU={6.3 1.8 3.7);

MVNCOV={2 3 1,

3 8 5,

1 5 4};

MVS=ROOT(MVNCOV);

RandSeed=floor(RANUNI(round(time(),16.0))*10000000); /* simulate random seed from system time */

x=J(nobs,1,1)*MU + NORMAL(J(nobs,3,RandSeed))*MVS;

create _simparm0_ from x(| COLNAME={P1 P2 P3} |); append from x;

quit;

run;

PGStats — Thu, 15 Nov 2012 02:27:52 GMT

I can't help you with proc IML. But you might want to look at proc SIMNORMAL. Its sole purpose is the generation of correlated normal random numbers.

Rick_SAS — Thu, 15 Nov 2012 11:54:08 GMT

Use the built-in RandNormal function: