Hello Shyam,
I would not suggest using -npregress- in your case. From what you describe you have more than one regressor and very few observations.
Nonparametric methods require more observations than parametric methods and this is compounded as the number of regressors increases.
Also, -npregress- is not using all observations, most likely, because these observations are not identified (the method does one regression per observation) so this is also pointing to a problem with the model/estimator strategy you are using.
P.S: Sorry for the late reply. Note you can always contact technical support and will receive a more prompt answer. I happened to come to the post by chance looking for wording I had used in the past.