Time Series Regression Forecasting with Measurement Errors

摘要

Estimation methods can have dramatic impact on the outcome of empirical analysis. In this article, we quantify the effects of estimation on prediction generated from time series regression models with and without measurement errors. The estimators considered are unconditional least squares, maximum likelihood and conditional least squares. The results suggest that although these estimators are asymptotically equivalent, the finite sampling properties of predictors based on those estimators can differ substantially, because of differences in finite-sample estimation efficiencies, and more importantly in residual regeneration methods.