VECM standard errors

[ashtongate]

I am trying to write a bootstrap program but I am getting SEs that look too small. I use a model, obtain the fitted values and then resample the sample residuals in the usual way. GIven this is an AR process, am I missing a step? Cheers

Update -- I think I solved it by using addfactor and the resampled residuals, re-solving the model, using the solution variables as the endogenous variables in the VECM and calculating new impulse responses. SEs now look sensible. Is this right?