3
$\begingroup$

I wanted to do a sanity check.

I want to remove the effect of structural breaks on my series.

Therefore, I create a dummy with 1s for the period of structural change and 0s for the rest. (There might be several periods). Let's call it D

I then regress it on yt = b0 + b1D + ytStar, where ytStar is the series with no structural break effect.

Do you see any issues with that?

$\endgroup$
  • $\begingroup$ There might be a typo, your equation states you're regressing yt on yt. Also, you're using one dummy for all potential structural brakes? Why you believe they can all be modeled by the same variable? $\endgroup$ – Lucas Farias Feb 2 '19 at 21:20
  • $\begingroup$ Hi Lucas, thanks for the comment. Somehow the "*" from the equation was not printed, the ytStar is the error term that will contain the filtered series. I would like to model the structural changes in 1 variable to estimate the effect of a change in mean in general instead for a specific period. I have a lot of data (many time series) and I am using the model from Bai & Perron (2003) to detect the structural breaks. Beside the "structural breaks" I am also modelling trend, cycles and holidays. I am using the least square filtering to remove any deterministic effects from my data. $\endgroup$ – Kiril E. Proykov Feb 2 '19 at 22:33
  • $\begingroup$ If you post one of your time series I will try and help you form an ARMAX model that incorporates level/shift structure (one of the i'S) while dealing with pulses , time trends , seasonal pulses, arima structure, changes in model error variance. ![enter image description here](i.stack.imgur.com/iVFoO.png) See stats.stackexchange.com/questions/376222/… $\endgroup$ – IrishStat Feb 3 '19 at 11:56
0
$\begingroup$

I do see possible flaws with your suggested approach as my answer is that you should consider using ARMAX model which can incorporate X's and their lags , ARIMA and pulses, level shifts,seasonal pulses and time trends.

| cite | improve this answer | |
$\endgroup$

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.