Comments on: Times Series Methods versus Recurrence Relations

By: Manos Parzakonis

Manos Parzakonis — Fri, 11 Dec 2009 18:20:16 +0000

Now, what are the odds that this post was a project for one of my tutorees last year?

The intercept is a bit tricky in R. Other issues to pay attention when fitting time series model can be found here,
> http://www.stat.pitt.edu/stoffer/tsa2/Rissues.htm

By: John Myles White

John Myles White — Fri, 11 Dec 2009 14:03:20 +0000

Thanks for such a complete explanation, Rob. I had foolishly assumed that the sequence would be demeaned and then manipulated to induce stationarity as preprocessing steps.

By: Rob J Hyndman

Rob J Hyndman — Fri, 11 Dec 2009 06:35:19 +0000

The problem is that the recurrence relation is non-stationary, but the yw method assumes stationarity. So the coefficients can’t take the correct values. But using lm, you have no stationarity condition imposed and the coefficients are to take the correct values.

The intercept issue arises because ar() fits the model to (y-m) where m is the mean of the data. As the mean of fibs is 106.4, that is added back in as the intercept when method ols is used. See the help file on ar().

By: John Cook

John Cook — Fri, 11 Dec 2009 04:34:07 +0000

I forgive you. :-)

You’re putting the Fibonacci sequence to good use.