Determination of Linear Feedback Between Multiple Time Series

Abstract

Detecting and explaining the temporal co-movements and feedback relationships of economic variables is an important objective of empirical macroeconomics and a prerequisite for successful policy analysis and design. Granger (1963, 1969) introduced a causality concept to economists which is defined in terms of predictability and which led to the development of various univariate approaches to causality testing. Defining more general measures for “linear dependence” and “linear feedback,” Geweke (1982) proposed test procedures for multiple time series. His method is related to an earlier scheme suggested by Caines and Chan (1975, 1976).This paper discusses and compares the two approaches. Using Monte Carlo methods, we investigate their power in statistical decision making. In the context of Geweke's method we also study to what extent the choice of the autoregressive order selection criteria may affect the outcome of the test.