Eviews Manual
Glendora Starr
This site is designed for readers of the fourth edition of Applied Econometric Time Series. You can download the data sets as *.XLS, EViews or SAS files. You can also download the Supplementary Manual, two variants of the Programming Manual, Powerpoint slides, and some of my working papers.
NOTE: Student manuals for RATS and EViews users can be downloaded here. For obvious reasons I cannot provide the Instructors' Manual to students. Instructors manuals are available from Wiley.
Well we can now have ARDL module in EViews 9 which can replicate same results as compared to what Microfit can do with the advantage that we can have more than two lags and more than 6 variables which currently available demo version of Microfit does not allow. You can download your trial version of EViews 9 at following link
After that we select the variables by pressing control button and selecting the dependent variable first and independent variables after it and right click it and open it as equation. Here in the drop down menu we can see option of ARDL at bottom
Here we can fix some particular lag or use automatic selection within the maximum lags of dependent variable and independent variable. The automatic lag selection criteria can be changed from default in the option window. press ok to see the ARDL model results in the following
these are the basic results see here that there are 4 lags used for the dependent and 2 for the first independent and 3 for the second independent variable using AIC criteria. Now we need the Bounds F test to see if there is cointegration or not, it can be done by pressing view button on the top and going in the coefficient diagnostics
This F test will tell if we can proceed further or not
Here we can see that our F test value of 3.5 is not bigger than any of the I1 bound value hence there is no cointegration among these variables. Since it is a tutorial I will show you further steps. If the F test value is small then we have to change the variables (add or remove) or try adding trend variable. and If we find F test value larger then we can go for the Long run results which can be seen by pressing view button and coefficient diagnostics
Here we can see that there was no cointegration because all the long run coefficients are insignificant and the coefficient of cointEq(-1) is also non negative and insignificant which is with the short run coefficients. These should be significant as they are important. Further diagnostics like hetroskedasticity, Auto-correlation etc can be done by selecting view and residuals diagnostics.
Thank you sir, When we want to test the short run coefficients? What test we should perform for its significance, whether there exists any short run or not? and what coefficients we will take for short run test? Kindly please explain?
My second question is when we are going to explain R square, F statistics, and the other tools, which model we should select to support our model? Because as its important to write with Long run cointegeration form as support for model?
Hello sir, i want to ask question, as when we are going to select ARDL from the drop list menu, we see that it is not taking the first difference of the variable, and the method used in the table is ARDL,as its is must for ARDL model to have the difference of the variables. Can u address this question sir? As in ARDL we are going to take the difference of the dependent variables followed by the difference of the independent variables along with the optimal lags. But here i am seeing that it is not taking the first difference of the dependent variables and the independent variables like. D(dependent Variable) c D(Dependent(-1)) D(independent(-1)) their values in the table should be appear like this? Sir, i need your kind advice and guidance in this regards.
Dear sir, we have more then one short run coefficient, so we use wald test, let suppose they are significant now how we will show all the independent coefficients in an article as we cant show them all ??
This error comes when you mistakenly add variables which are forming identity, like 4 = 2+2 if your dependent is 4 and two independents are 2 each then it will give error. So try to change the variables it will sort it. In Sha Allah
Dear sir!!!How many Lags there should we use for ARDL model? If the results are not good at the lower lags, for better results can we go for higher lags as it gives significant and good results. kindly guide me please
Dear Sir,
Thank you for your reply.I have the following results when I test for long run relation ship in which the F-Statistics is 3.97.Can I use ARDL model.I just use one lag for each variable as lag one have lower SIC and AIC in my case. Pesaran et.al also recommend the maximum lag as 2 and I test 2 and prefer lag 1.
Hi See in this tutorial there is one link provided in which the example is done where the cointegration exists. For the case of second question yes if it was short run equation only then the dependent should have been D(CPI) but if both short run and long run there then mentioning of short run it self means that we can write CPI as dependent on top. For the case of last question I have shared one document here it has shown the method of doing ARDL in eviews 8 in which the results will match Microfit which will also match the results of eviews 9. So you will be able to note why there is minor difference between the Dave Giles manual results and eviews 9 automatic results.
Suppose we have a situation, in which we have a certain ARDL model (lets say (1,1,1,0)) and now we want to see what happens if we include a further variable.
1) If we add this further variable and compute an ARDL, then eviews (9) is going to suggest new lag lengths. Does this distort the effect of the newly introduced variable? (Say now we have something like (2,1,3,1,1).
Should I rather keep the previous lag lengths of the other variables, and if yes how can I do that with eviews since one cannot select manually lag length (only if the variables have all the same length).
3) I know that the error correction term needs to be significant and negative. Is it a problem if the computed value is larger (in absolute terms) than -1? e.g. -1.34
And what is the relation between the error correction term and the long run coefficients?
And a last question concerning the optimal lag length:
4) How can it be, that if I select 3 as maximum lag length the calculated ARDL model shows significant long run coefficients but if I chose 4 as maximum lag length, all long run coefficients become insignificant?
How can I argue for the maximum lag length that I chose? (e.g. perhaps there is some kind of rule of thumb rule that for annual data and a certain number of varibles and observations, the maximum lag lenghts should not exceed a certain number? Or would you say in general 2 are sufficient?).
Following are some insights
1) yes lag order must change when variable changes, and it does not matter what lag length is it as all the results adjust automatically to used lags.
2) the significance changed because of adding a variable can happen as ARDL is a complex model hence likelihood of finding significance result when there are many variables decreases when you add more variables secondly the insignificance might be due to economic theory.
3) there is no consensus among the researchers that ECM term should be between 0 and -1 but if all of your variables are in log form then it is OK to have the coefficient of ECM to be -1.34 it will only show that convergence is more elastic. Long run coefficients tell the long run effect while ECM term just quantifies the strength of equilibrium between the selected variables.
4) there is not hard and fast rule for the lag length the idea of using lag length is to absorb all the past effects hence we should use as much lags which removes all the issues from the model. and I repeat again it does not matter what lag you will use 3 or 4 as the results (t values, f value etc.) will adjust to the senario
Hope it helps
Woa have a look at my earlier post regarding the ARDL model it will guide you and regarding significance of variables it is judged the same way simple OLS is done. If ECM is positive then try to change variables or lag order.
first thing in time series data R square and adjusted R square are not important to report as they are overestimated because of common trend among the variables. That is why they are not explicitly mentioned for both short run and long run. Unfortunately eviews does not have the CUSUM test but it has some other alternatives for the stability testing for that see the eviews help manual to see how to interpret them.
See you can solve this problem by estimating this model in microfit and telling software to not to use the lag of call rate as you already have used it in long run. this way you can get what you want but here you need to provide solid theory that the lag of call rate is a long run determinant. The other way is what you have mentioned that not to use the lag in eviews as usually lags are not long run determinant there are there as there is inertia in the series.
Dear Sir I have one question. When i am performing the Ramsey Reset test for functional stability of the model. As i have two regressors, when i click ramsey test in Eviews then it asks no of fitting variables i put 2. it gives me error the regressors are perfetly corelated. But when i put just one in no of fitting variables it gives me significant results. What should i do sir? As in eviews 9 the cosum test is not available when we are using ARDL option in built eviews 9?
in the case of 1 number of fitting variables it is significant, it means there is missing squared form variable. It might be solved if you use all of your variable in logarithm form. secondly there is not CUSUM test in eviews 9 but there are alternative stability tests available that you can use see the eviews help manual for guidance in interpreting them.
e59dfda104