0

I am using 2000-2020 quarterly data and want to test if an appreciation of the real exchange rate leads to a decrease in manufacturing output of a country. However, my data seems to be non-stationary using a ADF test.

Since my knowledge in terms of econometrics is fairly limited, my plan was to use an OLS regression while taking first differences. After taking first differences, all variables seem to become stationary. Would this approach lead to valid results or would this lead to a spurious regression? I have also looked into the idea of a cointegration analysis and VECM.

Furthermore, I have several control variables, some of which are stationary before taking first differences. Should I take first differences for those as well or only for the non-stationary variables?

I would appreciate any feedback.

mek1401
  • 3

1 Answers1

0
  • If the series are cointegrated, use VECM; otherwise, difference the integrated series.
  • Do not difference variables that do not have unit roots. (Doing so would result in overdifferencing.)
  • Spurious regression may be a concern when dealing with integrated variables. After differencing, it is no longer a concern.
Richard Hardy
  • 67,272
  • Thank you for the quick reply. It is very helpful. I have another question if i may: Can I use VECM if it turns out that some variables are stationary while some are not? – mek1401 Jun 09 '22 at 18:26
  • @mek1401, you can include stationary variables in a VECM; see https://stats.stackexchange.com/questions/148994/var-or-vecm-for-a-mix-of-stationary-and-nonstationary-variables/149263#149263. I have answered a number of similar questions, you can look them up. – Richard Hardy Jun 09 '22 at 18:53