Relationship of Forecast Encompassing to Composite Forecasts with Simulations and an Application

Abstract

This paper examines the role of forecast-encompassing principles in model-specification searches through the use of linear composite forecasts. Based on the results of the pairwise forecast-encompassing test, this paper outlines a conceptual framework to provide some useful insights on cross-model evaluations in econometrics and the selection of predictors in composite forecasts. Second, it offers three different ways of performing the encompassing test and compares their finite sample performance through a Monte Carlo simulation study. Test results guide researchers to choose component forecasts and thus to avoid blind pooling in the combining regression.