0
$\begingroup$

I have an hourly time series, so when I plot the ACF it shows decimal lags. I can solve this by setting l=frequency of data times the numeber of integer lags I want to study. Ex: acf(data, l=8765*4)

For selecting ARMA order, should i look at integer lags or decimal lags?

Improve this question
$\endgroup$
3
  • $\begingroup$ The ACF should have the same shape regardless of whether the lag is measured in hours or fractions of a year or whatever. However, if you want to explore the possibility of an annual cycle based on hourly data, you'll have a problem! If that's the case, expand the question a little to get more relevant comments and answer. $\endgroup$
    – jbowman
    Commented Feb 27, 2018 at 18:30
  • $\begingroup$ But even if it´s the same shape, the number of lags suggest different models. Decimal ACF suggest data must be differenced, while the ACF with integer lags suggest an MA(0) with AR(p) $\endgroup$
    – Garcher
    Commented Feb 27, 2018 at 18:36
  • $\begingroup$ Not really seeing how you draw that conclusion. The problem you have is that the graphs on the bottom have far more lags, so they aren't immediately comparable. "1.0" on the bottom one corresponds to 1 year, I believe, so it really isn't the same scale. Looking at the PACF I see the first four or five lags look the same, as far as I can tell. If you do the bottom one with say a max of 25 lags that might help. $\endgroup$
    – jbowman
    Commented Feb 27, 2018 at 18:49

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.