Pitfalls in using Granger causality tests to find an engine of growth

Abstract

This paper discusses the reliability of using a Granger causality test to find an engine of growth. The paper first focuses on growth models' cointegration implications since causality must exist in an error-correction model. As a complementary, Monte Carlo experiments with independently generated I(1) variables also indicate a significant probability for rejecting the Granger non-causality null. Given the persistency and cointegration of variables in growth models, rejecting the non-causality null may reflect a spurious causal relationship, rather than confirm a theoretical causality.