If a Bayesian model is fit in PROC VARMAX and non-integer prior means are specified with the PRIOR=(MEAN=(....)) option in the MODEL statement, then any non-integer priors are truncated to their integer values.

For example, the same prior means will be used in the following two model specifications:

model y1 y2 / p=1 noint prior=(mean=(2.5 0.1 1.5 0.6)); 
model y1 y2 / p=1 noint prior=(mean=(2 0 1 0));

A fix for this issue is planned for the SAS 9.3 release.