I want to estimate two models, Vector Autoregression (VAR) and Vector Error Correction Model (VECM).
What is the correct way to choose lag length / lag order?
For the VAR model, I have selected the lag ($p=4$) using:
- variables at first difference
- statistical criterion AIC
- no evidence of autocorrelation or heteroscedasticity in the residuals.
For the VECM model, I have selected the lag ($p=1$) using:
- VAR with variables at level,
- statistical criterion of AIC,
- no evidence of autocorrelation or heteroscedasticity in the residuals of VAR,
- lag of VECM equal lag of VAR in level minus 1, VECM(0).
Is this methodology correct or in error?