Dear All,
I am confronting a problem regarding cointegration / vector error correction modelling and would appreciate your thoughts and help very much.
Estimating a VECM (vector error correction model) I take the following steps:
-lag length test in levels
-test for cointegration (Johansen´s ML approach)
-estimation of VECM
So, if I find that for example 3 lags are appropriate as lag length, I test for cointegration following Johanson´s ML approach. If I find a cointegration relationship I can go on estimating the VECM. Doing so, I include the number of lags I found most appropriate and include deterministic elements in the cointegration vector and / or VAR (EVIEWS cases 1 to 5) as in the test for cointegration. NOW my problem starts:
I know that certain assumptions are met regarding the cointegration test, especially white noise residuals.
So I need to test for these assumptions and do diagnostic tests, e.g. AR LM Test, test for normalty, ARCH-LM Test, stability.
Two problems arise:
First: When to test? Initially after the VECM is estimate – so for example before I test for weak exogenity and include exogenous variables – or do I do these test only for the final model (with additional exongenous etc.)
Second: What if the testsfail – e.g. I detect autorcorrelated residuals? What is the consequence especially of autocorrelated resides on the Johansen procedure – are the results of the cointegration test still valid – despite ar in resdis (so - in contrary the ADF test, where ar resides are a killer for the test results?
What do you do when you face ar resides in a VECM?
I appreciate your help very much and look forward to all of your answers.
hum
VECM / Cointegration diagnostics
Moderators: EViews Gareth, EViews Moderator
Return to “Econometric Discussions”
Who is online
Users browsing this forum: No registered users and 26 guests