The "no good entries" in some of your stim_times files is probably the cause of the improper matrix error, since those stimuli won't be represented in the matrix.
"no good entries" arises, in the case of 'global times', when none of the input times lies between 0 and Tmax=(Nt-1)*TR and Nt=number of time points in the (auto-catenated) dataset.
Without further information or samples, it is impossible to tell WHY the program doesn't like your -stim_times files.