I am not sure where to report this but it seems that when specifying
constantcoef's to a model fit,
numparam in the
bic methods ignore the number of applied restrictions. The dimension of the transformed parameters is used, but this contains the constrained parameters as well.
Perhaps it is better to use the length of