The impossibility of causality testing

• Published in: Agricultural Economics Research Vol. 36; no. 3; p. 1
• Main Authors: Conway, R.K, Swamy, P.A.V.B, Yanagida, J.F, von zur Muehlen, P
• Format: Journal Article
• Language: English
• Published: Washington, D.C U.S. Dept. of Agriculture, Economic Service, etc 01.01.1984
• Subjects: ANALISIS ESTADISTICO; ANALYSE STATISTIQUE; ENSAYO; MODELE; MODELOS; TESTAGE
• ISSN: 0002-1423; 1043-3309

Extract: Causality tests developed by Sims and Granger are fatally flawed for several reasons. First, when two variables, X and Y, are uncorrelated, X has no linear predictive value for Y; but X and Y may be nonlinearly related unless they are statistically independent, in which case X and Y are not related at all. The right-hand side variables in a regression equation are exogenous if they are mean independent of the disturbance term. Mean independence is stronger than uncorrelatedness. The proofs for deriving causality-exogenity tests imply weaker results than statistical or mean independence. Second, transformations such as the Box-Cox transformation and Box-Jenkins stationarity-inducing transformations are not causality preserving. Third, counterexamples constructed by Price have invalidated the Pierce-Haugh theorem on instantaneous causality. Fourth, omission of other variables influencing those tested renders any test results spurious. Finally, causality tests are inconsistent because they are based on underidentified models. We provide a logically valid method of building models which does not use causality tests