stationarity

[Viktoria]

Hi, I badly need your help, plss))
I am trying to run a regression, but got stuck with the problem of stationarity. I have gdp growth (%) and inflation rates (annual % change of cpi index) in my equation which are not stationary. Is it right to take the first difference of this variables (d(gdp_growth), d(inflation))? My worry is that taking the 1st difference of a variable results in its growth, right? than wouldn't it be wrong taking taking the first difference of a variable showing growth? And finally do the variables need to be of the same order integrated to run a regression?

[trubador]

As long as you are guided by the economic theory, you really do not have to worry about the stationarity of the variables. Given the endogeneity of your dependent variables, you can go with a VAR or an ARDL model.

[startz]

As long as you are guided by the economic theory, you really do not have to worry about the stationarity of the variables. Given the endogeneity of your dependent variables, you can go with a VAR or an ARDL model.

Make sure you carefully read Trubador's second sentence. If you are simply trying to describe the joint dynamics of your variables, do as he says. But if you were planning on running a least squares regression for causal inference, you would need to worry about stationarity.

[Viktoria]

Thank you very much trubador and startz for your reply! Actually I am going to run a least squares regression. What do I need to do in this case?

[trubador]

Least squares regressions in the form of y = a +bx usually lead to identification of "spurious" relationships among the variables, if they are nonstationary. But dynamic models (like VAR, ARDL, etc.) can handle this problem elegantly. Also, if you have a sound theory supporting that y = a +bx should be the true causal relationship, then you can estimate that either. Again, it depends on the context of your study and the method you use:

1- What is your research question?
2- What does the theory say?
3- What is the form of your equation?
4- What are your dependent and independent variables?
5- Are you going to use lags of variables?
6- How do you decide the stationarity of a variable?
7- Did you use unit root tests?
8- Did you take into account the outliers and structural breaks when testing for stationarity?

[Viktoria]

Thanks Trubador for further clarifications and for devoting your time))
I am trying to evaluate a simple monetry rule according to which the central bank should change the nominal interest rate in response to changes in inflation and output. I'm using OLS regression. The equation is i=c+aπ+by , where "i" is the central bank's interest rate, "π" is the rate of inflation (annual CPI change in %) and "y" is gdp growth (%).
I decide the stationarity of the variables by Augmented Dickey-Fuller test. I'm taking maximum lags of 12 as the data is quarterly. Actually I calculated the gdp growth as gdp change over the same quarter of previous year, and the data series is not stationary. The same problem is with inflation rate. As I understand I have 3 options: use ARDL model, take the 1st differences of the series(if it's not wrong in this case, which was my initial question) or take the raw data as they are.