- When there is an issue of autocorrelation in an ARDL or ECM model, are we allowed to use AR(1) to correct for the problem?
- What is the difference between ARDL model and ECM model?
Asked
Active
Viewed 1,514 times
1

Richard Hardy
- 54,375
- 10
- 95
- 219

Okoye Arinze Francis
- 11
- 2
-
[Here](http://stats.stackexchange.com/questions/229250/difference-between-distributed-lags-and-var-models/229754#229754) is an answer for what is the difference between ARDL and VAR. – Richard Hardy Aug 30 '16 at 05:29
-
These are two pretty separate questions which should be posted separately, IMHO. – Richard Hardy Aug 30 '16 at 10:04
-
I was going through my old answers and noticed this one was not accepted. Do you perhaps need further clarification? – Richard Hardy Feb 20 '17 at 19:31
1 Answers
1
If the residuals are autocorrelated in an ARDL or an ECM model, you may choose a different lag order to remove the autocorrelation, so there is no need to model the residuals as an AR(1) process. However, I do not see a technical reason why allowing for autocorrelated errors and modelling the autocorrelation explicitly would fail. It would just be somewhat unorthodox.
ARDL can handle both I(0) (nonintegrated) and I(1) (integrated) variables at the same time, as per Dave Giles blog post "ARDL Models - Part II - Bounds Tests". ECM is normally used for I(0) variables (which could also be first differences of I(1) or cointegrating combination of I(1)).

Richard Hardy
- 54,375
- 10
- 95
- 219