I am trying to understand overdifferencing, if I were to first difference a stationary time series, would that time series remain stationary? Would it become white noise? Does it actually make the time series less stable?
Asked
Active
Viewed 814 times
2
-
A recent thread on overdifferencing appears at http://stats.stackexchange.com/questions/250728. I believe it might answer all your questions. – whuber Dec 15 '16 at 00:27
-
@whuber Saw your answer to the referenced question. Did you get the bounty? The question was about iid variables. It is a special case of an uncorrelated time series that is stationary. My intuitive thinking is that it adds spurious trends to the time series and that could explain the instability . For the OPs benefit I do not think the differenced series would look anything like white noise. – Michael R. Chernick Dec 15 '16 at 00:37
-
@Michael: I think you're right. Look at it in reverse. If a differenced series were white noise, then that original series would be integrated white noise--that is, Brownian Motion--which obviously is not stationary. BTW, that bounty lasts another 5 days. If I were the OP I wouldn't award it until the last moment, because it might continue attracting attention and bring in additional replies. – whuber Dec 15 '16 at 00:44
-
2Does this answer your question? [High autocorrelation when taking the L-th order of difference of a sequence of independent random numbers](https://stats.stackexchange.com/questions/250728/high-autocorrelation-when-taking-the-l-th-order-of-difference-of-a-sequence-of-i) – kjetil b halvorsen Oct 04 '20 at 01:38