Your statement "I convolved my regressor files (without any timing info) using wav –GAM" is somewhat confusing. What are the regressor files but the encoding of the stimulus timing?
In any case, you should do exactly 1 of the following:
(a) use "waver -tstim `cat fred`" to generate the response model from a list of stimulus start times in file fred, which you then input to 3dDeconvolve with "-stim_file 1" and "-stim_minlag 1 0" and "-stim_maxlag 1 0"
OR
(b) use 3dDeconvolve with "-stim_times 1 fred GAM" to generate the response model directly
Here, file 'fred' is assumed to be a list of the stimulus start times, relative to time=0 at the first input point of the input dataset. 'fred' is NOT a 0-1 marker file indicating the absence-presence of stimulus.