If you specify a BY statement and you specify the ESTDATA= option in the FIT statement in PROC MODEL, which provides starting values for each BY-group, then incorrect starting values are used in all BY-groups. For the first BY-group, the procedure uses the starting values specified in the last BY-group in the ESTDATA= data set. For all subsequent BY-groups, the procedure uses the final parameter estimates from the previous BY-group as starting values for the current BY-group. This may lead to convergence problems or execution errors.

There is no circumvention to this problem. If you want to specify different starting values for different BY-groups, you will need to run each BY-group separately.