The optimum lag length selected using the command "varsoc" in STATA and the maximum lag length to be included in underlying VAR model while estimating the cointegrating rank of VECM, are they both same?
I have the set of four annual time series all are I(1) (time period 1983-2002). While I try to estimate the cointegrating rank using the STATA command "vecrank" then STATA produces weird results (having all eigenvalues equal to 1.00000 and no value for trace statistics). But when I only have 3 variables in the model, then there is no such problem. Why can't I use more than three variables while estimating such model?
You can do that by depending on the smallest value for some criteria such AIC, SBC,..., as we use in the following example by choosing the lag that its row has more stars that is mean more small criteritions.
There are AIC, SBC and Hanna criterion and SIM's test. They are all aimed at non-residual correlation with just enough lags. Too lags creates another disease.
You can do that by depending on the smallest value for some criteria such AIC, SBC,..., as we use in the following example by choosing the lag that its row has more stars that is mean more small criteritions.