The properties of some two step estimators of ARMA Models

This paper analyzes the large sample properties of several two step estimators that have recently been suggested for estimating autoregressive moving-average models. As these estimators typically involve generated regressors, the generated regressor literature suggests that, in general, they will be inefficient and their estimated formula standard errors will be inconsistent estimates of the true standard errors. Deriving the covariance matrix of the true disturbances in these models enables consistent estimates of the true standard errors and efficient generalized least squares estimators to be computed.