VAR, test for normality, autocorrelation and heteroskedasticity- should I use stationary first differences for these tests? Please help!

[Anastazja Stachwiuk]

HI, I am checking thhe long-term relationship between unemployment and labor force participation rate. I have a integration order I(1) and I want to run VAR. As far as I understand I need to use first differences in VAR cause they are stationary. Should I perform tests for heteroskedasticity, normality and autocorrelation on them or variables on levels?

[Joro Kolev]

To check whether two variables are cointegrated, you do not necessary need a VAR. You can just regress one on the other, and check whether the residual is stationary.

-var- with a cointegrating relationship is called -vec- (vector error correction model).

When you test for normality, hetero, and autocorrelation, you test whether those are present in the residuals not in the regressors and regressand.

Testing for normality, hetero, and autocorrelation, is not very popular in modern econometrics, because test can be made robust to hetero and auto correlation, and normality is not necessary in large samples.

[Anastazja Stachwiuk]

Thank you very much for your help. As an assignment I have to replicate one paper concerning long-run relationship between unemployment rates and labor force participation rates (unemployment invarience hypothesis). The lecturer asked us to perform these tests (normallity etc.); I have one more question which is probably trivial or does not make sense, should I use residuals of tested variables or residuals of their first differences. Sorry, but it is my first time in doing such an assignment. One more time, I am very grateful for your help!