[Note: please see the UPDATE at the end of this post]
In the last post we looked at smoothing data by fitting a polynomial, and by fitting a Fourier series. We found that we could control the degree of “smoothness” in either case (by polynomial degree or by cutoff frequency) and that with the right amount of smoothness both methods did a reasonable (and sometimes outstanding) job over most of the observed time span, but did much worse near the endpoints of observation. Let’s forget about the endpoint mess for the moment, and concentrate on the performance of the smoothing near the middle of the window of observation.
Open Mind 