So I'm checking the relationship between economic growth and environment. The model is stable everything seems ok, cointegrated variables and significant negative error term however all of the long-run coefficients are not significant. similarly, the short-run ones expect one variable. It doesn't make sense. any suggestions?
You need to take a look at your data. You did not tell us your method of estimation and the nature of your data. Tell us more about your data so that we can provide you with adequate advice
The error correction term is significant so your long-run effects should be significant. So you have a contradiction. But tell us about your data and method of estimation
Chuck A Arize firstly thank you for your help. i'm using time series data however, after checking again it seems that i have breaks in the series. thus, i have no idea how to proceed! too many methods to check for breaks and each one give different results.
Are you using annual or quarterly data. How many observations do have and what period does it cover. Simply saying you are using time series data is not enoug information. What method are you using Johansen, ARDL or Engle and Granger etc?
Hi Tara. It is not a problem that you have received insignificant long-term coefficients. That means that in the long run there will be no expected relationships between those variables. You should explain/point that only you have received significance for some coefficients on short-time, i.e, for the coming years, if the other checking tests are OK. So, it happens. It happens to me as well recently. I was using ARDL model and I explained that in a good way. Also have confirmed it with some theories. I hope it helps. Best, Goran
Chuck A Arize I'm using annual data -48 observations-
i tried both Johannes and ARDL since my data is integrated at first difference however I got contradictory results from ardl and johansen!
Tara: Avoid Johansen because it is a system approach and almost everything has to be good before one gets good results. Also, your having 48
observations is not friendly to Johansen. I know of this earlier and that is why I was wanting information from. you. Focus on ARDL it is much better and it fits your 48 data points. Let me know how it goes! Are you using unrestricted or restricted? You may want to try both. E-views has both but MiCRofit uses only unrestricted
Error term and level of significance do not measure thesame thing. Significance level measures how significant the dependent variable is in accounting for any change in the dependent variable. An increase or decrease in the dependent variable by the independent variable might be significant or insignificant. While error term only measures the percentage of errors that have been corrected as we move from the short run to the long run.
Are your data monthly quarterly or annual. The period covered by your data set (number of years ) is more important than the actual number of observations. For example, if a disequilibrium arises in the equilibrium situation your data set must be long enough to allow for returns to equilibrium to be measured. If your data set is quarterly 12 years does not appear to be long enough to model such returns to a long run equilibrium.
My understanding is that the coefficient on your error correction term in the VECM is significant. If you are using EVIEWS it used to produce t-statistics for that coefficient. That t-coefficient has a non-standard distribution. I have seen many people lead astray by assuming that this has a standard t-distribution. You used to be able to test this in eviews using tests of restrictions on \alpha. I would like to know how you decided that the coefficients in your long run error correction term were not significant. Again these all have non-standard distributions and can be tested as restrictions on \beta
If in your VECM more than one of your equations has a significant coefficient then the ARDL process is not valid. You do need a system approach.
In VARs and VECMs one is seldom worried about individual coefficients. You can plot impulse response functions and do Granger Causality Tests 9Be careful as these also may have non-standard distributions.
In a system with just economic growth and some environmental variable(s) I would be concerned that you would have missing variables. You need to do a complete set of specification tests to validate your model.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs