5
$\begingroup$

I'm novice in time series analysis and completely lost through reading so.. I have an enormous dataset of discrete time series of aggregated events and I wish to fit each one of them into ARIMA models. Therefore I have to ensure that the time series are stationary, which I do through adf and kpss tests. However I wonder if these tests also ensure that time series do not need to be tested for trend and seasonality. Is there any statistical criterion that reveals that a time series is trending? Visual inspection of time plots is difficult in this case out of the question* since the dataset is way too big.

Thanks in advance, btw I'm an R user ;) -A

$\endgroup$
  • 2
    $\begingroup$ Visual inspection of data is never out of the question: if you can read or analyze them, you can look at them! True, huge datasets can create overwhelming plots, but don't let that stop you: plot small random samples of your data instead. $\endgroup$ – whuber Nov 22 '11 at 20:51
  • $\begingroup$ you're right, I regretted the phrase "out of the question" the minute I pressed the "post" button... $\endgroup$ – Anastasia Nov 22 '11 at 21:42
  • $\begingroup$ You are welcome to edit your own question, Anastasia. If people have already posted replies and the edit is substantial, it's conventional to alert readers that an edit has been made, such as by calling attention to it in the text. $\endgroup$ – whuber Nov 22 '11 at 23:06
1
$\begingroup$

When faced with the daunting task of evaluating a "zillion" number of time series for local time trends one could spend a lot of time reviewing graphs which may or may not have unusual data points potentially obfuscating the trend or an embedded auto-regressive structure which could also defeat your "eyeballing". A more efficient approach might be to perform a computerized search which could be classified as a productivity aid for your bleary eyes. The suggested procedures to detect time trends and other deterministic structure have been all peer-reviewed and have transparency. These are not ad-hoc methods but state-of-the-art procedures to help you in your qwest. Look at my answer and previous references in Fancy detrending of time series

$\endgroup$
  • $\begingroup$ thank you so much for the help, I will check it immediately ;) $\endgroup$ – Anastasia Nov 22 '11 at 21:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.