Abstract:
IN A RECENT ARTICLE Robert Engle [2] explored the extent of the sin practicing econometricians commit when approximating the true autocorrelation function of the disturbances in a regression by some finite parameter scheme, and finished by giving some rules to attain salvation. Perhaps the major implication of the paper is that ordinary least squares (OLS) is likely to be an efficient estimator in many circumstances; e.g., "Consequently, moving from OLS to ALS, which is itself an inefficient estimator, should not yield much of a gain in efficiency unless serial correlation is substantial"
