Encompassing the Forecasts of U.S. Trade Balance Models

Yock Y. Chong and David F. Hendry (1986) propose the concept of forecast encompassing--the lack of additional information in another model's forecasts. The corresponding test statistic is based on the regression of one model's forecast errors on the other model's forecasts. This paper generalizes Chong and Hendry's statistic to include sets of dynamic nonlinear models with uncertain estimated coefficients generating multistep forecasts with possibly systematic biases. Using stochastic simulation, the generalized statistic is applied to forecasts over 1985Q1-1987Q4 from six models of the U.S. merchandise trade balance, revealing misspecification in all models. Copyright 1993 by MIT Press.