Intepret CUSUM-squared test in Stata (and compare with Eviews)

[Mi Nguyen]

Hi everyone,
After building an ARDL model, I used cusum and cusum-sq test to check the model stability. I would like to ask how to intepret the cusum-squared chart in Stata. What confused me is that the upper and lower bound of the cusum-sq plot in Stata seemed to be wider (thus more lenient) than the bound in Eviews.

In eviews, cusum-sq indicates model instability (picture 1). But in stata, the result is better because somehow the bound of the significance is wider (picture 2)

I used cusum9 for getting cusum-sq plot in Stata. cusum6 will also give the same result. Pls help!! Thank you so much.

[George Ford]

I think the confidence intervals are the same (95%), so standard errors are computed differently. hard to say without a detailed description of each method, though you can scan the ado file in stata to see what's what.

[Sebastian Kripfganz]

cusum6 and cusum9 are community-contributed commands. They have been superseded since Stata 15 by the official post-estimation command "estat sbcusum". Have you compared your results with that?

[Mi Nguyen]

Hi Sebastian Kripfganz thank you for your reply. I did try estat sbcusum, but estat sbcusum is can only conduct cusum test but not the cusum-sq test as far as I know, while I think if either cusum or cusum-sq does not pass then model is instable. Please correct me if I am wrong.