Comparing the effectiveness of traditional vs. mechanized identification methods in post-sample forecasting for a macroeconomic Granger causality analysis

Abstract

We identify forecasting models using both a traditional, partially judgmental method and the mechanized Autometrics method. We then compare the effectiveness of these two different identification methods for post-sample forecasting, in the context of a relatively large-scale exemplar of macroeconomic post-sample Granger causality testing. This example examines the Granger causal relationships among four macroeconomically important endogenous variables–monthly measures of aggregate income, consumption, consumer prices, and the unemployment rate–embedded in a six-dimensional information set which also includes two interest rates, both of which are taken to be weakly exogenous in this context. We find that models indentified by the traditional method tend to have better post-sample forecasting abilities than analogous models identified using the mechanized method, and that the analysis done using the traditional identification method generates stronger evidence for post-sample Granger causality among the four endogenous variables.