My good friend: Sorry, I have had computer problems today, nevertheless, Johansen estimates are robust to departures from normality (Cheung & Lai, 1993, Johansen, 1995) and to heteroskedasticity (MacDonald & Tayor, 1991; Johansen, 1995). I will use standardized residuals, to identify the ith data point and then I will add a dummy ( 1 and 0 otherwise) or (dummy variables) variable to remove the normality problem. Then, I will see if ARCH residuals were solved by this mediating approach. If it isn't, I will try alternative coding of the dummy, such as 1 and -1 or 1 and 1. Johansen'(1995: 29) "the normality assumption is not so serious for the conclusion, but the ARCH effect may be). Here , you may consider introducing as exogenous variables such as oil prices and the delta of it OR even interest rate while keeping your eyes on the ARCH test.
If it isn't, I will try alternative coding of the dummy, such as 1 and -1 or 1 and 1. Johansen'(1995: 29) "the normality assumption is not so serious for the conclusion, but the ARCH effect may be). Here , you may consider introducing as exogenous variables, such as, oil prices and the delta of it OR even interest rate while keeping your eyes on the ARCH test.
Thanks a lot my friend! I use the Rahbek and Mosconi (1999) approach to some other exogenous variables, that is adding the accumulated exogenous variables into the cointegration space. How do I treat the (exogenous) dummy variables you propose in relation to this? I can't imagine they too should be entering the cointegration space in accumulated form, right?
My recommendation is to obtain the normalized VECM and save the residual, get the SEE of that equation , then create standard_ res= residuals/SEE. Then, list the Standard _res to locate the ith data (pick based on absolute values). Then create a dummy and place it OUTSIDE the .COINT. space. Could you tell me how you accumulated exogenous variables
I do not believe, it is fine to add INSIDE the cointegration space.Am assuming you have CENTERED seasonal dummies, and they do not enter the coint. space. The same treatment for my dummies. If they enter the coint. space the critical values are no longer appropriate. You talked about ARCH and normality-- are they solved?
I do not believe, it is fine to add INSIDE the cointegration space.Am assuming you have CENTERED seasonal dummies, and they do not enter the coint. space. The same treatment for my dummies. If they enter the coint. space the critical values are no longer appropriate. You talked about ARCH and normality-- are they solved?
(Sorry about double entry, it was computer problem error). In your response, please tell me about the sign and statistical significance of ECM(-1) in your normalized vector.
I'm experimenting with different models. I know the ECM(-1) should be negative and significant. I send you the paper by Rahbek and Mosconi (1999). It would be very interesting to hear you opinion on it.
They cannot do what they intend to do without creating new critical values since they have violated the coint. space. I simply need you to tell me what you are referring to as accumulated I(0) variables. Also, comment on the normality & ARCH. Finally, Yes, the coefficient on the ECM(-1) must be negative and significant.
They referr to critical values in Harbo et al. (1999). My two endogenous variables are the profit share of functional incomes and the bank debt to-GDP-ratio. I was using real interest rate and real GDP grwoth as exogenous variables. However, I'm considering skipping these control variables, since ot seems they create spurrious cointegration. Hm. I identified three outliers in the residuals, and imposed dummies as "1" for them as you proposed. I still get ARCH at 5% however, when 1 lag is chosen for ARCH. How many lags should I have in for my ARCH test?
Could you enter them as individual dummies? Experiment with other types of dummies I mentioned earlier. If it passed at the !% that will be okay and normality passed then you accept the model since you can support your work with the literature I noted earlier. Even Johansen had to ignore autocorrelation in his own paper.
Btw, can you please point me to a reference on the dummy method of obtaining normality & ARCH? It would be nice to referr to in my paper. Also, is it totally necessary to pass the ARCH test for all VAR models? I read that it is mostly used for financial data, such as stock prices and such.
It is much better to reference a whole volume "Special Section on Exogeneity, Cointegration, and Economic Policy" Edited by Ericsson, Neil R Journal of Business and Economic Statistics, Vol. 16 (4) 369-459. October 1998.
You can reference "Special issue : Practical Issues in Cointegration Analysis," Edited by Les Oxley and Michael McAleer in Journal of Economic Surveys, Vol 12(5) December 1998.
You may want to see Terrence C. Mills and Geoffrey E. Woods, Applied Economics 2002, 34 2143-2149 or Even Juselius K. Journal of Policy Modelling (1992). The above will suffice.
I have been on vacation. When we resume inJune, I will be glad to be more specific. Lars,
CAN YOU SEND ME OECD INDUSTRIAL PRODUCTION INDEX quarterly or annual data, 1975 - 2015? Looking forward to it. It will be of help to one of my doctoral students. Thanks in advance.
Thank you! I will loook for the OECD index later today. When you are back form your vacation, may I trouble you with another question? I got an idea: Since it is really not my target equation in the VAR/VEC system that suffers most from non-normality (and ARCH), may I still be able to use it for valid inference?
Thank you very much. Please include OECD real GDP and OECD industrial production for 1975 through 2015. Lars, I am here until Friday so you can ask me the questions. Please, bear in mind that my response rate will not be automatic because I I will be making travel arrangements. Nevertheless, I will respond the same day.
Since it is not your normalized vector, you can please, see page 405 four lines from the top right corner. (see page 415 -422 for Juselius's,1992 approach) Special Section: Exogeneity, Cointegration and Economic Policy Edited by Ericsson Neil, Journal of Business & Economic statistics. I gave this citation earlier to you
I can't seem to find the right paper by Ericsson Neil. I found one similar, but I'm not sure it's the same. But I take it from your answer that it is sufficient if normality is only ashieved only the VAR/VEC equation associated with my normalized vector?
I think that all you need is Journal of Business & Economic Statistics Vol. 16 (4) October 1998. and then you go to the pages I have noted. It among the A+ journal in the field. The reason you are not seeing may be that you are looking for Ericsson. IHe was the special edition editor.