Docstring says 'params' is optional if the model has not been fit, but the implementation does not allow 'params' to be absent.
Thanks, that statement is outdated by a few versions.
(we dropped attaching results to the model instance because of circular reference)
the model predict always requires params, but x/exog is optional.
Update docstring of RegressionModel.predict()
@chatcannon Do you want to open the pull request for your correction