On 28 May 2015, at 10:17 , Martin Maechler <maech...@stat.math.ethz.ch> wrote:
> > On the other hand, to me, it did look a bit strange or "ugly" to always > perform to different lm()s and the coefficients of one and the > residuals of the other. Yes, but the whole point was that if you do OLS on an integrated series, the only thing you get right is the number of observations (which you, slightly oddly, obtain as the length of the residuals) - the coefficients of that analysis will be inconsistent or at least severely inefficient. If you do OLS of the differenced series, you get coefficients that are good enough for starting values, but might not use all observations. A WLS would get both right, but you need to fit the model to find the weight matrix (which in general is N x N, non-sparse)... -- Peter Dalgaard, Professor, Center for Statistics, Copenhagen Business School Solbjerg Plads 3, 2000 Frederiksberg, Denmark Phone: (+45)38153501 Office: A 4.23 Email: pd....@cbs.dk Priv: pda...@gmail.com ______________________________________________ R-devel@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-devel
