i'm currently working on currency misalignment , i have 3 variables the productivity differential and the NFA, as independent variables and REER as dependent variable, but my REER is unfortunately stationary I(0) the other are I(1), how can i run a cointegration or find a long run. By the way i used a DOLS despite having a I(0), and the misalignment results was following the literature for some countries and not following for the others, i'm working on 32 countries
Running cointegration regression for a model with I(0) dependent variable is not recommended. Cointegration regression is suggested when (i) all the variables including the dependent variable are integrated of same order, (ii) the dependent variable is integrated of order 1 and the independent variables contain both I(0) and I(1) variables. In first case, you can use the Johansen cointegration technique and in second case, ARDL cointegration technique may be used.
I guess ARDL may be applied in this situation and a built in feature of this approach is available in Eview 9 or higher
Thanks for your answers, i have another question what is your variable is stationary at none, and non stationary with trend and intercept selection. Which one will you select? Or do you have to select the one who is suitable for your research?
If the dependent variable is I(0) it cannot cointegrate with the independent variables. Cointegration means that a linear combination of variables have a reduced order of integration. Since you cannot have negative orders of integration, you cannot reduce the order of integration of an I(0) variable. If all variables are I(0) there is no reason to apply cointegration methods. Standard regression and notions of correlation with standard critical values for hypothesis testing should be valid. If some of the independent variables are I(1) you might apply a cointegration test to these variables to see if they cointegrate. If they do, a regression of the I(0) dependent variable on the independent variables will be a balanced regression in the sense that both left and right hand sides of the equation are I(0) and standard regression can be applied. The ARDL method strictly assumes that the dependent variable is I(1) and only the regressors can be either I(1) or I(0). The Johansen method would normally only be used with I(1) variables because if you include I(0) variables it will likely find a separate cointegrating equation for each of those variables.
- If you mean that the variable is stationary at 1(0) with trend and intercept, Here you can see the other variables. if its stationary at 1(0) , you can used Johnson test or ARDL
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running