Hi,
Hope anyone with more experience than me are able to help me!
I´m using single equation modelling to find a model for the mortgage rate (endo) and the money market rate (weakly exo). I have found that both of the rates are I(0), and that the long-run relationship also is I(0) by using the Bounds-test in Stata. The bounds-test is done with out the dummies which I used to get a well specified ARDL.
I have to questions:
1. Do I have to use the same number of lags and dummies for both rates when re-parameterizing the ARDL to an ECM?
- When I use the same lags an dummies in both the ARDL and the ECM, the standard specification tests in PcGive does not give the same answer.
2. Do I need to use the same dummies in the ADF-test for the respective variables as I use in the ARDL and ECM?
/Mette Marit :)
In Eviews the number of lags can be selected automatically using AIC, SC or HQ and the ECM is derived from the ARDL equation. You need to look under diagnostic under Eviews10. Try this software. I think in STATA you need to specify but you can log into the STATA forum and ask for help.
https://www.statalist.org/forums/
and they will respond to how to do it in STATA.
In the first case the answer is a resounding no. But your analysis may reflect the fact that x or y use more lags and the converse is tractable if you are convinved that it is not mis specified. Then you can compare both with theoretical models. The one with less lags is probably more tractable because it is structurally intuitive.
In the second case the answer typically is more lags in the ADF test as the ADF tet will recommend the number of lags based on the Augmented Dickey Fullerson results. Your answer should go with what is theoretically more tractable. Also the software implementation of the test in both R and stata is typically fully automated and you need not tweak any settings apart from understanding the output and impacts on your model, if any.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling