I am trying to estimate the impact of economic shocks (captured by a dummy variable in a ECB study) on unemployment rate in EU-28. The thing is PMG is so attractive because have the option - short-run coefficients and I can extract the impact of this dummy for each member state. Anyway, I have two variables non-stationary (stationary at I1): unemployment rate and youth unemployment rate. I used youth only to control unemployment and to make feasible the estimation of the economic shocks impact on unemployment. As fixed regressors (I0) I used percentage change of unit labour cost and the dummy variable mentioned above. I am satisfied about my coefficients results. All are significants and the economic reality is justified. My sample size consist in 1680 observation (60*28 cross sections).
Question 1: Can I ignore the non normal distribution of the residuals (JB test rejects the null) - I have a problem with Kurtosis which is high near 7? Even if I add more variables, the JB prob doesn't change.
Question 2: After extracting the residuals I check the cross dependence test and the result is that in my model exist cross section dependence between residuals. Do you think I should try to quit this method and try another one having in regard that in UE exists a significant level of homogeneity between some countries?
Thank you!
Ignoring assumptions means that you can't trust the probability statements from the model. How do you propose to resolve that. ? Can you afford to just ignore it?
I know that ignoring assumptions can lead to biased results but I was impressed by PMG differentiated coefficients for each cross-section and I tried to follow this approach. Any transformations didnt facilitate the normal distribution of the residuals. In time I made two clusters of EU MS (incluzive and extractive) using EGLS panel method and now the results are good. Also the assumptions of the Gauss Markov theorem are confirmed. Thank you a lot for your answer!
Thank you for your answer!
I will try to change lag selection method in my further work.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size