FMOLS Model
FMOLS Model
For panel cointegrated regression models, the asymptotic properties of the estimators of the
regression coefficients and the associated statistical tests are different from those of the time
series cointegration regression models. Some of these differences have become apparent in
recent works by Kao and Chiang (2000), Phillips and Moon (1999) and Pedroni (2000, 2004).
The panel cointegration models are directed at studying questions that surround long-run
economic relationships typically encountered in macroeconomic and financial data. Such a long-
run relationship is often predicted by economic theory and it is then of central interest to estimate
the regression coefficients and test whether they satisfy theoretical restrictions. Chen, McCoskey
and Kao (1999) investigated the finite sample proprieties of the OLS estimator the t-statistic, the
bias-corrected OLS estimator, and the bias-corrected t-statistic. They found that the bias-
corrected OLS estimator does not improve over the OLS estimator in general. The
Results of Chen et al. suggested that alternatives, such as the fully modified (FM) estimator or
Dynamic OLS (DOLS) estimator, may be more promising in cointegreted panel regressions
(BALTAGI)
Phillips and Moon (1999) and Pedroni (2000) proposed an FM estimator, which can be seen as
a generalization of Phillips and Hansen (1990). Recently, Kao and Chiang (2000) proposed an
Alternative approach based on a panel dynamic least squares (DOLS) estimator, which builds
upon the work of Saikkonen (1991) and Stock and Watson (1993).
Kao and Chiang also investigated the finite sample properties of the OLS, FM and DOLS
estimators. They found that (i) the OLS estimator has a non-negligible bias in finite samples,
(ii) the FM estimator does not improve over the OLS estimator in general, and (iii) the DOLS
Estimator may be more promising than OLS or FM estimators in estimating the cointegrated
panel regressions((BALTAGI ECONOMATRIC ANALYSIS OF PANAL DATA.).
PRECONIDTIONS
1. Variables must be stationary on same order, like X is stationary at order 1st then Y also
should be sationarity at order 1st
2. If variable are cointegreted then we apply FMOLS model otherwise not
3. We apply FMOLS for panel data
4. For long run relationship
So here is data file let suppose my al variables are stationary at first difference, so next step is to
check cointegration in variables,
1
Here i have my variables, there are
few steps for panel cointegration,
the first step is, to check the
sationarity of data, if the entire
variables are cointegreted in same
order then we can employ
cointegration.
Remember I have check the sationarity of my data and now I want to know about the cointegration in
variables
Steps === go to quick===group statistic===cointegration test
Steps === go to
quick===group
statistic===cointegr
ation test
When you will enter cointegression following window will open, and write the name of all your variables in
the blow box and ok
2
Here I wrote my dependent
variable first and then all
independent variables. press
ok
1: pedoroni
2: kao
3: fisher
3
Here in the above pic, we have three test types for checking cointegration: find from drop down button
1: pedoroni
2: kao
3: fisher
You can use all the test type all are beneficial
Let suppose I select pedroni, and pedroni have three deterministic trend specifications, you can use any
specification if you select pedroni test type and other things remain same (advance users can chose
optimal lags, informational criteria etc) I select pedroni and individual intercept and ok
And these are the results of cointegration using pedroni and individual intercept
4
Here we have 11 probability values, and with the help of these values we can
make decision about the cointegration, null hypothesis for cointegration is that
no cointegration, so what is decisional criteria for accepting or rejecting null
hypothesis?
If the majority of these values we find significant then we reject null hypothesis
and accept alternative hypothesis which is there is cointegration among variables.
Or simple if majority of these values we find less than five percent then we
conclude that there is cointegration like in the given results I have total 11 values
and I have 7 values significant means majority of values is significant so we can
say there is cointegration ,
NOTE:I have confirmed that there is cointegration but for double check ill run
again cointegration test but this time ill include individual intercept and individual
trend and for further conformation using pedroni ill run again cointegration but
involve no intercept no trend and compare results of these there shapes ,,
remember using individual intercept I have confirmed that there is cointegration
but for mind satisfaction we use other these two options,
In above results we have 11 probability values, and with the help of these values we can make
decision about the cointegration, null hypothesis for cointegration is that no cointegration, so
what is decisional criteria for accepting or rejecting null hypothesis?
If the majority of these values we find significant then we reject null hypothesis and accept
alternative hypothesis which is there is cointegration among variables. Or simple if majority of
these values we find less than five percent then we conclude that there is cointegration like in
the give results I have total 11 values and I have 7 values significant means majority of values is
significant so we can say there is cointegration ,
NOTE:I have confirmed that there is cointegration but for double check ill run again
cointegration test but this time ill include individual intercept and individual trend and for
further conformation using pedroni ill run again cointegration but involve no intercept no
trend and compare results of these there shapes ,, remember using individual intercept I have
confirmed that there is cointegration but for mind satisfaction we use other these two options,
you also can use Kao etc.
If you are using pedroni test and in pedroni test if individual intercept and no intercept and
trend are telling that there is cointegration but one shape is tell that is individual intercept and
individual intercept that there is no cointegration ,, here we can conclude that there is
cointegration because majority of shapes telling cointegration.
5
OTHER METHOD OF KNOWING ABOUT
COINTEGRATION (KAO)
Now I select kao rather than pedroni and ok results are
blow
Quick, group
statistics,
Johnson
cointegration
and then I
select Kao ok
6
Now I have found that there is a cointegration and now next step is to know about long run
relationship for this I ll run panel FMOLS MODEL
Steps == quick = estimation equation,
Go to quick, estimate
equation, write
dependent variable
first and then all
independent
variables, and form
method select
cointeg- Cointegrating
regression
7
From here I selected
Fmols
So when you will press ok a resulted window will be open which is for long run relationship and this will
be window of FMOLS model, make decision on the base of probability value.