Abstract: In the application of large scale econometric models for forecasting, the single-equation ordinary least squares (OLS) method is often used to estimate parameters in each model equation. This paper investigates the properties of the parameter estimates under single-equation estimation methods. Since the disturbance of a time series regression model is seldom a white noise process, it is found that bias is almost inevitable as long as contemporaneous endogenous variables are present in a model equation. This paper proposes a model identification method based on reduced form linear transfer function (LTF) models that can avoid or reduce bias of transfer function weight estimates under rather practical assumptions. It is found that forecasts can be greatly improved if appropriate models are identified and employed.
Key words and phrases: Econometric models, reduced form models, structural form models, simultaneous transfer function models, linear transfer function models, serial correlations, consistent estimates, LTF analysis.