Comparison of Robust and Varying Parameter Estimates of a Macroeconometric Model

Four estimators of econometric models are compared for predictive accuracy. Two estimators assume that the parameters of the equations are subject to variation over time. The first of these, the adaptive regression technique (ADR), assumes that the intercept varies overtime, while the other, a varying-parameter regression technique (VPR), assumes that all parameters may be subject to variation. The other two estimators are ordinary least squares (OLS) and a robust estimator that gives less weight to large residuals. The vehicle for these experiments is the econometric model developed by Ray Fair. The main conclusion is that varying parameter techniques appear promising for the estimation of econometric models. They are clearly superior in the present context for short term forecasts. Of the two varying parameter techniques considered, ADR is superior over longer prediction intervals.