Hi, i am working on macro economic data in a bi-variate framework with GDP as dependent variable and inflation as independent variable. I have applied bound test plus ARDL model and it confirms co-integration between the variables. But the problem is coefficient of my only independent variable becomes insignificant in long run. The error correction term is significant. I am using annual data of 28 years and both the variables are integrated of order I(0). Lag length had been selected using AIC criterion. I want to proceed further with ECM model. Moreover i checked for Granger causality but it is not moving in any direction. I am considering my results on the basis of p value.
Both the variables have been used in log form.
Regards.
Monika
Hi, I am also working on a long term analysis with an ARDL-model. I am not sure but ARDL was my only choice while the variables are from different order of integration (dependent variable is I(1)). I think that you can just check a normal regression analysis if there are no structural breakpoints that will influence the stability (view stability diagnostics in eviews). Otherwise add a dummy. I do not know however how to handle multiple breakpoints in analysis, while I am facing this problem myself
Dear Enjo,
First, looking at your sample period, I wish to know the unit root test employed for stationarity of data for the variables. Note that unit root tests are low-power tests. Besides, Ng and Perron unit root test is suitable for small sample period. I recommend you employ it too.
Second, even if the stationarity test result turns out to be same-which I strongly doubt, the evidence of cointegration does not necessarily warrants the significance of the only independent variable.
Third, it seems you have log the Inflation too. You are not clear about that. If so, just transform the dependent variable (GDP) using either ln or log depending on which one that exhibit normal distribution after transformation. This can be determined using Jarque Berra or Normal distribution test. Note, inflation data should remind unlogged.
After all these, there would be improvement in the results. If there is need to expand sample period, kindly do. However, if the sample period is justified on a particular issue, consider the second advice.
Summarily, using large sample periods produce robust results as compared to small sample periods.
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running