Co-Integration and Error-Correction Models: The Temporal Causality between Government Taxes and Spending

Abstract

Listen

Translate article

Several interesting papers have recently considered whether government taxes Granger cause government spending or vice versa. The findings, based on standard Granger [6] causality tests, are not consistent. Manage and Marlow [10] conclude that when evidence of one-way causality exists, federal taxes Granger cause federal spending, while Anderson, Wallace, and Warner [1] conclude that the evidence supports one-way causality in the opposite direction. Ram [11], reexamining the issue, concludes that taxes Granger cause spending at the federal level, but that spending Granger causes taxes at the state and local level. Von Furstenberg, Green, and Jeong [13], in another related article, examine the tax and spend, spend and tax causality issue using a more general vector autoregressive (VAR) model. They introduce, in addition to federal taxes and spending, two economic variables-the GNP gap and the inflation rate, decompose federal spending into several categories, and remove the effects of automatic stabilizers when appropriate. They find little evidence that taxes Granger cause spending, and limited evidence that spending causes taxes-a finding which corresponds most closely to that of Anderson, Wallace, and Warner [1]. Our purpose is to reconsider the causality issue with the aid of co-integration and errorcorrection modeling. Our analysis illustrates with a specific example how sources of causality can be neglected in standard Granger causality tests. Moreover, addressing the question of temporal causality between government taxes and spending provides insight as to how different policies might, or might not, help control the growth of government. The results of our estimating various co-integration and error-correction models with quarterly data suggest bi-directional causality between government taxes and spending, both for the federal, and state and local sectors. This finding counters the conclusions of previous studies. When annual data are used, however, our