bounds test for cointegration (Pesaran ardl)

Question

I am new to cointegration, particularly in the context of ARDL. Central to that approach is the bounds test where the null is (I believe) no cointegration. What confuses me is to run the test you have to know which of five cases your model consists of. One of the keys to this is if you have a restricted or unrestricted intercept (or no intercept). The second is whether you have a trend (restricted or not restricted). How do know whether you data (or model) has these or not? I have worked with ARDL and not encountered these terms before. Does a trend here mean if the data is stationary or not, these types of models can mix I(1) and I(0).

Nats · Answer

The ARDL R package handles automatically any model (and turns it into an uecm if it is not already) and performs the test on the uecm.
About the selection of cases, I will probably write a blog post about it and I will add a throughout example in the upcoming vignette (I am the developer of the package).
Thank you for pointing into this direction, I will try to create an intuitive way for the selection.

mr.rox · Answer

I do not think there is a theory for including/excluding trends and constants in coitegration analysis. You may want to draw your dependent variable on a graph and see if it trends and if any other variables trend. If they do, include trend, and see if your variables drift, then include a constant so on and so forth. Also, if the trend/constant is not significant, you may not need it. Check if including these improve your model fit. etc. It is data-driven rather than theory.
The video I posted above should be of help.

bounds test for cointegration (Pesaran ardl)

2 Answers

Add your own answers!

Ask a Question