Hello everyone.
I have got a little question on selecting a proper
bandwidth for kernel regression.
As you all know, for bandwidth selection in a
regression case you can use the averaged squared error
as a criterion for goodness of fit, but for some
problems (e.g. the bandwidth h approaches to zero), it
is better to use the cross-validation criterion in
combination with some penalty function (Generalized
Cross-Validation, Shibata's Model Selector, Akaike's
Information Criterion, Rice's T,...).
So my problem:
Has got anyone any idea or a tip, where I can find
some existing R-functions, that satisfy my needs?
Or has anyone any idea how such a function would look
like, for example to compute a bandwidth selection
with the (Generalized) Cross-Validation or Rice's T,
as I mentioned above?
My aim is to fit in a kernel regression to a time
series with the function ksmooth(), but there is still
the bandwidth to select.
Have a nice weekend and thanks a lot.
Sincerely
Andreas Klein.
