1. How to run panel var model?
Professor Rapelanoro Nady commented as such > You should go to professor inessa love
home page, you will find 2 package for stata dedicated to the panel Var model. The first one
is a code and the new one is a complete program alloWing you to do the panel Var procédure
( including diagnostic tests and more option like using gmm or ols). You will need a good
computer since the panel var procedure needs à lot of memory to do the subroutine (i used it
and it took me one day to obtain results, using the built in program, so it is better to use the
code). Also please check at one of my previous post here if you want an estimation strategy
for the Pvar model i posted à link to my working paper using à Pvar model.
2. EVIEWS 9.0, EVIEWS 9.5 and panel data.
Professor Burcu Özcan commented about first generation panel data and also cross section
independence test as such >> Eviews-9 or Eviews 9.5 run first generation panel tests. Also
you can easily do cross section independece tests by these Eviews version
3. Mundlak test helps to decide between fixed effect and random effect model
Professor Abu Subhi commented>If I'm not mistaken, there is no built-in function in EViews to
do the test. However, If you want to use the Mundlak test, follow these steps:
1. Create time averages variables of all your time-varying independent variables.
2. Estimate an auxiliary regression by regressing dependent variable on independent
variables plus the variables you created in (1).
3. Test the joint significance of all additional variables using Wald test.
4. If the variables are jointly signiﬁcant, then use FE estimates. If the variables are jointly
insigniﬁcant, use RE estimates. The Mundlak test can help you to decide between FE and RE
4.In case of I(1) and I(0) mixture of series (Panel)
Professor Afolabi Luqman commented > When you have mixed result result you can go for
panel ARDL ( pool mean group, mean group or dynamic fixed effect ) . The most interesting
thing about the technique is that you will have both the short and long run result together
including panel ECM. Also if go for full option it will show you the regression result per
country. xtpmg d.dv d.ivs lr(l. dv ivs) ec(ec) pmg or mg or dfe......findit xtpmg to install.......
5.Minimum time period (T) in panel cointegration test
Professor Abu Subhi commented as such > The answer to this question is very subjective.
Most papers that applied panel cointegration analysis in decent journals have at least T = 20
6. I wanna ask about panel regression. When the best model is Random Effect Model
(by using hausman test and chow test) that estimated by GLS, can we just ignore the
small durbin watson value (it's near 0 and probably have positive autocorrelation)?
Muhammad Rizky Septian posted the following figure.
Abdullah Sonnet commented> You can add lagged dependent variable to address the
Muhammad Rizky Septian commented > thank you sir for the answer any other suggestion?
because i just want to use static panel model.
Abdullah Sonnet commented> You can also use some standard errors which will be robust
against autocorrelation......but making the model dynamic would be the best one....
7. Steps for Panel Data modelling.
Professor Ijaz Khan commented as such > The routine panel data modeling and treatment
1. Check for Normality
2. Chck for Multicollinearity
3. Check for Heteroscedasticity
4. Select the best model from Pooled OLS, Fixed effect and Random effect models
5. Normality can be checked through Jarqur Bara test,
6. Multicollinearity through Variance Inflation factor (VIF)
7. Heteroscedasticity through Breush Pagan test if ur data is normal and White test if not as
white test do not assume normality of data
8. To select the best model u have to run two comparisons (A) and (B) below:
A. Pooled VS Fixed effect (Select through Wald Joint Significance test) if Pooled is best then
stop here if not go for another comparison
B. Random VS Fixed (Selection through Hausman test).
In case u have the problem of Multicolinearity, You can take log of variables or u can run
ridge regression. If u have Hetero problem u can simply run Robust regression to control for
the problem during regression this is what usually panel data is treated.
But it depends on what u want to find out in your thesis. If u want to check for panel
cointegration or run ARDL model on panel then that is a part of dynamic panel data modeling
and in that case treatment will be different.
Meo School of Research
Hossain Academy Note