Dan: Thanks for the feedback on the AR estimation.

Rich: Our local stats expert here just suggested that the Yule-Walker will do just fine for the more general AR process as Dan Rowe indicated. My limited understanding is that AR(1) is sufficient (at least according to a comment made by Keith Worsley at a talk here last month). However, is there an advantage to estimating the autoregression for a lag of greater than 1?

Keith' Worsley's code for his computation of the autocorrelation can be found in the following file:


[www.math.mcgill.ca]

I hope this helps.

Any further info on the number of degrees of freedom consumed by including the AR(1) or higher correction for a time series?

philippe