Abstract

The role of money in guiding or predicting the business cycle is a recurring, if not a cyclical, theme in economics. Different econometric approaches, ranging from narrative to vector autoregressions, have been proposed to assess the validity of the money matters proposition in the economy. It is argued here that most empirical work on the effectiveness of monetary policy in explaining the business cycle is based on an incorrect premise that the business cycle can be adequately represented by a linear time-series model. This thesis demonstrates that the business cycle is inherently nonlinear and that this nonlinearity can be captured by regime-switching models and, in particular, by STAR models. STAR models applied to the logarithm of the U.S. real GNP have, however, not produced very satisfactory results. One reason advanced for this lack of success is that the growth rate of GNP does not possess enough variability. The position taken here is that the disappointing results generated by STAR models of real GNP are not due to insufficient variability but simply to the choice of the switching variable that defines the states of nature. My contention is that interest rate instruments and, more specifically, the federal funds rate describe the environment in which the economy functions. This thesis demonstrates that a STAR model of real GNP, with the regimes being determined by the quarterly difference in the funds rate, outperforms linear representations of real GNP by substantially reducing the standard error. Furthermore, this thesis examines whether the growth rate of money explains the U.S. business cycle when the latter is modeled as a regime-switching process. It is shown that the growth rate of M1 and M2, which in the usual bivariate linear setting does not Granger-cause output for the 1960-1993 sample period, would 'cause ' output when the business cycle is modeled as a STAR process. This 'causality ' result holds for the difference stationary, trend stationary and Hodrick-Prescott representations of the business cycle. In particular, the growth rate of M1 has greater effects on the real economy in the low growth regime, while the alternate regime features money neutrality. Finally, this thesis examines whether the Markov regime-switching model of Hamilton, based on an unobservable state of nature, outperforms STAR models based on an observable state of nature when applied to the growth rate of U.S. real GNP. It is shown that a STAR model of real GNP outperforms Hamilton's changing intercept model for the 1960-1993 sample period and compares favorably with further refinements of the Hamilton model