avas vs. lowess

[Vadim Ogranovich <vograno@arbitrade.com>]

Dear All, I modeled y = f(x) + noice using two estimators:
a)lowess(x, y, f=1/3)
b)avas(x, y, linear=0)

To my surprise the prediction from avas was uniformly (and non-trivially)
bigger than that from lowess, which suggest that one of the procedures is
biased for the dataset at hand.

Could anyone give a hint why this may be so?

Thanks,
Vadim