ARDL dignostic checks_got confused in real world application

[ethan1989]

Dear, all
I recently utilised ARDL specification to estimate a model with 31 obs. Inspired by the work of Bloch et al. (2015), cited as below: Bloch, H., Rafiq, S., & Salim, R. (2015). Economic growth with coal, oil and renewable energy consumption in China: Prospects for fuel substitution. Economic Modelling, 44, 104-115. doi:http://dx.doi.org/10.1016/j.econmod.2014.09.017, I planned to apply their approach to my dataset.

However, I have several problems with the diagnostic checks [ I employed EVIEWS 9]
1. Having got the regression results of ARDL estimation, we can use the drop-down menu to run some diagnostic tests such as the JB normality test, BP autocorrealtion test, ARCH effect test, Ramsey misspecification test.....and associated CUSUM and CUSUM test. Please have a quick look of table 4 of Bloch et al (2015) (I attached the table4, see as below), they produced various diagnostic statistics. One simple question, are these statistics are just from those drop-down menu of ARDL estimation >>view>>residual diagnostics?

2. In table 5, Bloch et al. (2015) reported results of their long-run and short-run regression results. To the best of my knowledge, this table could be done in Eviews 9 by clicking coeficient diagnostic test>> then cointegration and long-run form (right?). However, there are no diagnostic checks provided for the long-run and short-run estimated results. How can I get them ? If I want to produce some diagnostic checks such as R-sq and adj R-sq, DW statistics etc. How can I do it manually?
Any answers would be highly appreciated.
Regards
Ethan

[EViews Gareth]

1. That's a question that is probably better asked to the authors - we have no way of knowing what they did.
2. EViews doesn't produce those summary statistics automatically. But the longrun /cointegrating form are just auxiliary regressions that you can run manually.

[ecofin]

1)i have download the article, probably the authors have used microfit 4 or 5 and some others software.
2) for the result the microfit 4 and 5 estimate ARDL with 3 cases and with some output statistics (no constant, constant, constant and trend), you can see the user manual "Time series Econometrics using micrifit 5.0 pp 564-565, for the 3 cases", and chapter 6 pp 104-114.
and you can see Time Series and Panel Data Econometrics by M. Hashem Pesaran, OUP.
3) but EViews 9 estimate ARDL with 4 cases (non constant, rest constant, unrest constant, rest linear) the 5th case probably will be added (no idea).
what I mean by all this: that EViews 9 estimate ARDL will all cases you can see the article: BOUNDS TESTING APPROACHES TO THE ANALYSIS
OF LEVEL RELATIONSHIPS, H. PESARAN et al, JOURNAL OF APPLIED ECONOMETRICS, but the microfit can not! So eviews wins
you will understand when you reed the user manual and the article, and see the link below
I thank Mr Igor That he Explained me the difference http://forums.eviews.com/viewtopic.php?f=8&t=12239.

Best Regards. Econometrics and Finance. :)

[ethan1989]

Thanks Gareth and econfin
One additional question, how can I do the point 2 suggested by Mr Gareth ? Those summarised statistics are the last step of my research (I am new to time series analysis, my Ph.D project rests on panel data modelling, mainly depending on STATAl; )
Thanks for any help guys.
Regards
Ethan

[ecofin]

1-EViews 9 can estimate PMG-ARDL (PMG Models in EViews/Pooled Mean Goup-AR Distributed Models) http://www.eviews.com/help/helpintro.ht ... 056.2.html.
2-and the article of M. Hashem Pesaran "Pooled Mean Group Estimation of Dynamic Heterogeneous Panels" http://www.jstor.org/stable/2670182?seq ... b_contents
3- as he said Mr Gareth for the statistics produced by Harry Bloch et al, you can use EViews 9 to do this manually or write a program to do all this, EViews is the powerful software it is the easiest software to manipulate data and make statistic. Hurray for EViews! :)

[ethan1989]

Thanks ecofin very much.
Forgive my poor English. Let me state my question in other way. I can perform ARDL test and then Bounds test, and then get the co-integration and long-run form (with 31 yearly observation). Similar to Bloch et al., I want to report statistics such as R-sq adj R-sq, F-statistics etc for the long-run model and short-run ECM model. Mr Gareth suggested do it manually, but how?
Can I extract estimated residuals for both the shor-run model and the long-run model? And then using them to construct summarised statistics?
For example, we use the estimated coefficients to get the predicted values for the dependent variable (say y for the long run and D(y) for the short-run).
Thanks for your help ecofin :) :) :) :) :) :) :) :)
Regards
Ethan

[ecofin]

1- Dave Giles blogspot:
http://davegiles.blogspot.ca/2013/03/ar ... art-i.html
http://davegiles.blogspot.ca/2013/06/ar ... tests.html
http://davegiles.blogspot.com/2015/01/a ... ews-9.html

2- some video from youtube:
ARDL Approach To Cointegration.avi
https://www.youtube.com/watch?v=d9E8BKsocis
ARDL approach to co-integration using General-to-Specific procedure
https://www.youtube.com/watch?v=NYZOXr-zWMA
Forecasting using ARDL approach to Cointegration
https://www.youtube.com/watch?v=gVd21H9UWTM

3- Sayed Hossain:
ARDL Model. Model Two. EVIEWS
https://www.youtube.com/watch?v=mTa7n1KY9iQ

[ethan1989]

Thank you so much for all what you have done for me. ecofin.
Regards
Ethan