I am testing a cointegration between two variables using Johansen test and VECM framework (2 variables, optimal lag length selected by AIC/BIC/HQC on VAR). Johansen test indicates existence of 1 cointegration vector, but when I test for model mis-specification non-normality and presence of ARCH in residuals is detected. How should I proceed with the analysis/what method to use to get a correct estimates of cointegration and VECM?

  • $\begingroup$ Engle-Granger two-stage procedure could perhaps be more robust to such problems than Johansen's analysis; the former relies on OLS which is robust to conditional heteroskedasticity and non-normality. I cannot back this up by a reference, though. Engle-Granger two-stage procedure is fine when you have just two variables. $\endgroup$ – Richard Hardy Mar 6 '15 at 14:05
  • $\begingroup$ I was trying to find references regarding Johansen test's sensitivity w.r.t. non-normality but with little luck. However, Silvapulle & Podivinsky "The effect of non-normal disturbances and conditional heteroskedasticity on multiple cointegration tests" (2000) should be relevant for you. $\endgroup$ – Richard Hardy Apr 9 '15 at 14:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.