Lag lengths in Johansen, VECM and granger causality
Posted: Fri Mar 25, 2011 1:29 pm
Hi,
I am having trouble figuring out what lag length to use when using Johansen to test for cointegration.
I have found using the VAR lag length criteria selection that the optimal lag length is 1. For Johansen I then minus 1 (as VAR estimation is in levels, whilst as johansen is calculated in differences), is this correct? If so I type 0 0 into the lag length when carrying out Johansen?
If I find no cointegration, I calculate granger causality using a differenced VAR framework, however if I find evidence of cointegration then I use a VECM framework, is this correct?. When I then estimate a VECM, are the lag lengths then also 0 0?. This is where I run into difficulties as the existance of no lags removes all coefficients, and also means that I cannot test for granger causality. Is this correct?, if so is there a way to test for granger causality (or short run causality).
Any help is greatly appreciated,
Many Thanks
fb
I am having trouble figuring out what lag length to use when using Johansen to test for cointegration.
I have found using the VAR lag length criteria selection that the optimal lag length is 1. For Johansen I then minus 1 (as VAR estimation is in levels, whilst as johansen is calculated in differences), is this correct? If so I type 0 0 into the lag length when carrying out Johansen?
If I find no cointegration, I calculate granger causality using a differenced VAR framework, however if I find evidence of cointegration then I use a VECM framework, is this correct?. When I then estimate a VECM, are the lag lengths then also 0 0?. This is where I run into difficulties as the existance of no lags removes all coefficients, and also means that I cannot test for granger causality. Is this correct?, if so is there a way to test for granger causality (or short run causality).
Any help is greatly appreciated,
Many Thanks
fb