A STATISTICAL EVALUATION OF THE CAUSE‐EFFECT RELATIONSHIP BETWEEN MONEY AND INCOME*

Abstract

ECONOMIC THEORETICIANS and empirical analysts have long been concerned with the nature of the relationship between the money stock and economic activity or one of its proxies, aggregate income. Four postulates exhaust the possibilities: (1) money and income are not causally related; (2) causality is unidirectional from income to money; (3) causality is unidirectional from money to income; and (4) a two way causal relationship exists between money and income. Few economists assert that money does not matter or that money is purely passive as suggested in the first two postulates. The empirical relevance of the last two postulates constitutes an important issue in monetary economics. Empirical studies of the cause-effect relationship between money and income are based on the assumption that a theoretical connection exists between the two variables. Theoretical models can be made to exhibit any of the connections described by the four postulates. The models which result from alternative simplifying assumptions which produce the various results are often tested empirically. Empirical studies of the money-income nexus have produced conflicting results, however. One or both of two weaknesses are responsible: (1) precise, testable definitions of causality have not been employed; (2) conventional hypothesis-testing procedures have been used to evaluate empirical data which may not meet the assumptions of such tests. The purpose of this study was to test the nature of the causal relationship between money and income by using a procedure designed to deal with both weaknesses. A theoretical causality criterion developed by C. W. J. Granger was adopted and the cause-effect statements were statistically evaluated by ordinary least-squares regression combined with the jackknifing technique. Jackknifing is a general, distribution-free method for reducing the bias in a statistic and for producing approximate confidence intervals. Computer simulation experiments within the study demonstrated the sensitivity of the F and t statistics to nonnormality; the jackknife statistic was found to be an attractive alternative. The insensitive F test allowed the acceptance of the null hypothesis that nominal gross national product (GNP) is purely passive; the test of the jackknife statistic required that this hypothesis be rejected. The alternative hypothesis that causality is bidirectional is consistent with the empirical results. For the period 1947 through 1972, tests on the components of GNP revealed that most of the detectable causal influences of money on nominal GNP worked through the price components of GNP; real GNP was relatively unaffected by money. The feedback effects of income on the money stock were found to operate through both the real and the price components of GNP. The results were generally the same for three alternative measures of money: (1) M1, the narrowly defined money