I had some trouble with a similar model, but did eventually get it to sample, and that was with missing data thrown in as well.
Looking over your model, you might want to non-center intercept[l]
and beta_t[l]
? For instance,
intercept[l] = intercept_state[ss[l]] + intercept_sigma_state * intercept_tilde[l];
with intercept_tilde[l]
being standard normal.