1
$\begingroup$

I have 120 IDs. For each ID I have hourly data (90 hours per ID). My data is perfectly balanced. I run xtunitroot tests and all the different versions (llc, ips, demean, trend, lags etc.) and I strongly reject the null hypothesis of a unit root. However, when I run 200 different dfuller tests per ID, I can only strongly reject the null hypothesis in about half of them.

my questions to you are:

  1. Which test approach should I trust more? xtunitroot for the whole panel or dfuller for each different ID?
  2. Does it make sense to check unit root for each ID? And if yes, how should I now proceed when around half of the data/IDs is non-stationary?
$\endgroup$

1 Answer 1

0
$\begingroup$

The two approaches test different null hypotheses. Hence, we cannot say which we should trust "more", and the results are by no means contradictory.

The panel tests have the null hypothesis that all series in the panel have a unit root. The alternative, i.e. the "negation" of the null is that at least one series in the panel is stationary. A panel test unfortunately does not tell you which (akin to an F-test in a regression).

Individual tests tell you for which series you face a rejection (but are affected by multiple hypothesis testing and may lack power as combining series into a panel leads to larger sample sizes).

$\endgroup$
2
  • $\begingroup$ Many thanks! So given that I can reject the nul hypothesis in the individual tests in just 87 out of 200 IDs, my overall conclusion,should be that I have stationarity, right? This means, I should then use first differences ? $\endgroup$
    – Mike S.
    Mar 7 at 11:04
  • $\begingroup$ Unfortunately, the individual tests suggest that your panel is mixed, i.e. some units are stationary and others are not. If you are willing (see above, multiple testing and low power imply that such a conclusion would be optimistic) to trust the individual results, you would only first-difference for those units for which you did not reject (those units for which the test gives no indication for stationarity). $\endgroup$ Mar 7 at 13:59

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.