I have Panel Data (T=41, N=16, T>N)- 16 banks with 41 quarters, Obs=656
Out of five explanatory variables, 3 are stationary at the level I(0) and 2 are stationary at the 1st Diff I(1) (NPL-R, and ROA-R)
Data have Heteroscadascacity and Serial correlation
I want to try both "dynamic" and "static" models.1- How could I use fixed effects in term of above-stated info i.e. (Mixture of I(0) and I(1))2- What kind of dynamic model would be appropriate?Thank you
You should first state your theoretical model independent of (the quality) of your data. If you formulate this model in transformed variables, e.g. differences or percentage changes, you mey find that these data are better suited for an econometric estimation. If the statistical quality of an estimate is quite good, you need not much care about stationarity etc.
For mixed staionarity i advice you make use of ARDL model. For the non-stationary variables; at times some data from some countries suffer from this challenge but you can adopt more than one unit root testing approach, by so doing you can include the breaking point unit root testing to take care of the problem, if it is from structural break problems.