Hi there,
I am interested in the optimal lag length for my model. I am taking macroeconomic variables such as GDP and unemployment rates. I know how to implement the test but are not sure what variables I should test. Should I use the original data (downloaded from the database) or take the transformed data (e.g. gdp growth in log) that go into my regression?
Thank you.
Hi Thomas,
What are your goals? Depending on what you are trying to get for your study it is possible to provide you with the answers. There are several lag length tests for several regressions, so this is the first step.
About the second question, it also depends on your model. As you've already mentioned, the log version of a variable changes your understanding about it. The first step here should be figuring out a model, plotting the variables to understand how they perform and accounting for the possibility of outliers.
Your question is not clear, sorry.
but to what I understood you want to determine the optimal lag length of a macroeconomic model (as you are using two macroeconomic variables). Apparently you are performing a VAR model. if it is the case, I suggest you use the Log form to make both variables more uniform and and make data conform more closely to the normal distribution.
Hi Brenno and Helmi,
thank you for your answers. Generally speaking, I am looking for a statistical or theoretical explanation why one should perform certain statistical tests (e.g. testing for the optimal lag length of a macroeconomic model) on the raw or transformed data. The same applies to unit root tests for instance. I have found different views on this so far. My motivation is to test a model with two macroeconomic variables using OLS and accounting for the dynamic nature of it (e.g. including lags). Hope this is clearer now.
Hi Thomas,
From an econometric perspective every test should be performed to the data that will be plugged in to the OLS regression. Nonetheless, it can be easily shown that any monotonic transformation does not affect the outcome of unit root tests and/or lag selection.
Thank you Konstantinos. This helps me. Can you give a reference to your second point (e.g. monotonic transformation does not affect the outcome of unit root tests and lag selection)? I would like to study this more...
Best regards
Thomas
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression