To create an AR(1) matrix from the estimate, rho, see the first section in "Fast simulation of multivariate normal data with an AR(1) correlation structure."
Let's suppose both matrices are 3x3, although they can be any square dimensions. The following program computes the Kronecker product of an AR(1) correlation matrix with rho=0.8 and a unstructured matrix with estimates UN[1,2]=0.63, UN[1,3]=0.12, and UN[2,3]=0.34:
proc iml;
/* return p x p matrix whose (i,j)th element is rho^|i - j|
See https://blogs.sas.com/content/iml/2018/10/03/ar1-cholesky-root-simulation.html */
start AR1Corr(rho, p);
return rho##distance(T(1:p), T(1:p), "L1");
finish;
/* test on a matrix with rho = 0.8 */
rho = 0.8; p = 3;
AR1 = AR1Corr(rho, p);
UN = {1 0.63 0.12,
0.63 1 0.34,
0.12 0.34 1 };
K = UN@AR1;
print UN, AR1, K;
... View more