0
$\begingroup$

I am dealing with time series regression, where I have stationary and nonstationary variables. Can I regress nonstationary I(1) variable on stationary variable when controlling for the lag of the nonstationary I(1) variable?

I know regressing nonstationary variable on stationary variable would lead to bias. However, would it be okay if I control for the lag of the nonstationary variable in my regression.

My regression would be something like this in Stata (where y is I(1), x is I(0) and for illustration I use up to 3 lags of y)

reg  y_t+h    x    l(1/3).y_t
$\endgroup$

1 Answer 1

0
$\begingroup$

No. If you truly have i(1) variables then coefficient at lag will be one. It is a unit root process, so regress the differences on x properly

$\endgroup$
9
  • $\begingroup$ Thanks. I thought about first differencing before. But I will estimate impulse responses afterwards. Will first differencing provide normal impulse responses, or should I estimate cumulative impulse responses? Also, I do not get coefficient as 1 when regressing on lags. $\endgroup$
    – sverdi
    Apr 25, 2021 at 18:17
  • $\begingroup$ If the coeffiicient is not 1 then why do you think it is I(1) process? Run a unit root test on your y, then run with the lag and post what you get. The model you are estimating is called ARMAX(1,0,0). $\endgroup$
    – Aksakal
    Apr 25, 2021 at 20:38
  • $\begingroup$ You can do this with ARDL. See the bound testing approach that allows some variables to be I[1] and others I[0] $\endgroup$
    – user54285
    Apr 26, 2021 at 1:32
  • $\begingroup$ @Aksakal, recall that the distribution of the OLS estimator of AR(1) is peculiar when the process has a unit root. So no wonder the OP is not getting the estimated coefficient at 1 even if the true coefficient is 1. $\endgroup$ Apr 26, 2021 at 5:27
  • $\begingroup$ @Richard Hardy? Can you even run OLS with a unit root (non-stationarity). $\endgroup$
    – user54285
    Apr 26, 2021 at 22:40

Your Answer

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

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