I think a lot of "smoothing" algorithms don't work at the ends of the data. Smoothing usually takes in past and future data. At the end of the graph there is no future data. (Maybe that's not the case for this type of smoothing?)
I've always thought the smoothed line should be truncated (so that it doesn't show present-day estimates where data is lacking.)
It depends on the type of signal. If it's non-linear and chaotic, sure. If its periodic, or has sustained trends, this approach is the basis for most engineering and science.
I've always thought the smoothed line should be truncated (so that it doesn't show present-day estimates where data is lacking.)