I am performing a model and the variables are I(1), then I am thinking in a VECM model. I read that the methodology (I read the Chapter 17 of Asteriou and Hall applied econometrics a modern approach using Eviews and Microfit) indicates the estimation of a VAR en levels as a first step in order to find the optimal lag. I have read that I should check for residual correlation in that VAR. However, I have a doubt, because the selection of the proper test in this specific VAR is not clear for me. As far I read and I understand from a web page (because it is not an especialized site in econometrics or university, I feel not sure if this is correct or not), it seems that Portmanteau test is not proper because that VAR is specify in levels (the variables are not stationary) and the Portmanteau test tends to reject the null hypothesis with a very high frecuency. Is this corrrect? If yes, then I should follow the LM test, right?.
With regard this, I want to say that this seems to be happenning to me in this research because I have tried the VAR including and removing some variables (and changing also the lags) and the Portmanteau test yields strong rejections of the null hypothesis and with the LM test there are some rejections, but mostly clear acceptances.
In addition, I am working with yearly series (and I have just 55 observations) and because of that I think it is not proper to test this VARs with many lags (lets say 9 lags). Consecuently, this seems to be another element indicating that the best option is LM test because Portmanteau works better with large lags and LM is recommended when the amount of lags is small. Is this also correct?
Last question, I am performing the model in Eviews and the output of the LM autocorrelation test shows the p-values for each lag. Should be all p-values higher than 0,05 to say that there is not autocorrelation? If it is possible to consider that there is not autocorrelation if out of (lets say) 8 lags just one or 2 have autocorrelation, then appear another question. If I include 8 lags and I have autocorrelation in the 1rst and the 3rd, but not in the other 6 lags, Can I say that there is not autocorrelation? What about if the output shows autocorrelation in the 7th and 8th lags?
Very much thank you in advance.
Ah, sorry the series are I(1) what I checked was the unit root test. The problem is that, as far I read,, I need to stimate this VAR in order to obtain the optimal lag that I will use in the cointegration test. Now I don´t know if the series are cointegrated
- Badges
- Science topic