January 21st, 2010, 10:00 pm
Or, use large sample theory to obtain the standard errors. There exists large sample theory of maximum likelihood estimators, OLS, the method of moments, GMM, etc. Bootstrap is also good, but as soon as you try to analyze time series data, you have to be careful about using bootstrap. You can't just resample non-parametrically because this will destroy the time dependencies in your data, so you have to use some kind of a reasonable parametric bootstrap scheme or use block bootstrap.