Exogeneity.

Working paper

In spite of the importance of exogeneity in econometric modeling, an unambiguous definition does not seem to have been proposed to date. This lack has not only hindered systematic discussion, it has served to confuse the connections between "casuality" and "exogeneity". Moreover, many existing definitions have been formulated in terms of disturbances from relationships which contain unknown parameters, yet whether or not such disturbances satisfy certain orthogonality conditions with other observables may be a matter of construction or may be a testable hypothesis : a clear distinction between these situations is essential. To achieve such an objective, we formulate definitions in terms of the distributions of the observable variables, distinguishing between exogeneity assumptions and causality assumptions, where causality is used in the sense of Granger (1969). Following in particular Koopman's pioneering article (1950), exogeneity will be related to the statistical completeness of a model. In short, a variable will be considered exogenous for a given purpose if a statistical analysis can be conducted conditionally on that variable without loss or relevant sample information

Authors: Engle, R; Hendry, D; Richard, J

• Publisher: Department of Economics (University of Warwick)
• Series: The Warwick Economics Research Paper Series (TWERPS)
• Publication date: 1979-01-01
• Language: English