This strikes me as odd. The rolling 18 year mean doesn't seem to represent the underlying data terribly well. Is there a lag? Is the window centred on the correct year?If not, that would explain why there appears to be an end-discard in the early years, but the smoothing goes all the way to the end of the data set.And why you don't see the hiatus.
I've done this on yearly (J-D) from that data set, with 18 year Loess fit, which does go through the data points correctly. Any thing I've done has been done better here:https://tamino.wordpress.com/There is no hiatus.
Doug: No, it's just the 18-yr trailing average.
Post a Comment