Dear R-users,
I am fitting a kernel regression model of the form y ~ x1 + factor(x2)
+ factor(x3) and am using the function npregbw in the np-package to
find the optimal bandwidths.
My dataset is relatively large and the optimization takes quite long.
When testing different specifications I have noticed that the optimal
bw for x3 is always very close to zero (around 10^-12 or so). I am
wondering whether it is possible to hard code the bandwidth related to
x3 to 0 and limit npregbw's choice of bw's those related to x1 and x2?
My intuition suggests that this would reduce the number of parameters
to be optimized from 3 to 2 and thus make the computations quicker.
Furthermore the theoretical literature (e.g. [1]) seems to suggest
that this might be a good idea with categorical variables and big
datasets.
Any comments?
[1] Racine, J.S. and Q. Li (2004), "Nonparametric estimation of
regression functions with both categorical and continuous Data,"
Journal of Econometrics, 119, 99-130.
Best regards,
----
Otto Kassi
University of Helsinki
Dept. of Economics
