Unanticipated Money Growth and Unemployment: Post-Sample Forecasts

Abstract

In the late 1970s, Robert J. Barro published what has since become a popular model of business fluctuations. Barro's [3] findings provide evidence advocating the short run neutrality of money under rational expectations, forging an important empirical link to the theoretical studies of Lucas [19; 20] and Sargent and Wallace [33]. The original Barro unemployment model has received additional support from further statistical testing [5; 6; 15]. Despite criticisms and modifications, some of which are considered in Barro's [6] updated model, these equations have inspired numerous empirical studies analyzing unanticipated effects of macroeconomic variables. This paper examines post-sample forecasts of Barro's original [3] and revised [6] unemployment equations for the 1974-84 period. Mayer [23; 24], Harvey [13, 179-80], and Lovell [18], among many, contend that the ability to successfully forecast outside of the sample period is an important criterion for evaluating model robustness. This view is also reflected in a substantial body of studies, including [10; 25; 26; 29; 39], which investigate the post-sample predictive performance of macroeconomic models. One year ahead forecasts indicate that the Barro [3; 6] equations accurately predict unemployment for the 1974-79 period. During 1980-84, however, the predictive performance of both models sharply deteriorates. Augmenting the Barro models with interest rate volatility, in accordance with the findings of Evans [11] and Tatom [38], considerably improves the overall forecasting ability. This modification also rectifies the 1980s structural instability of the Barro unemployment equations. Furthermore, the basic conclusions of the seminal Barro work are retained with this augmented model. The results reveal that increases in interest rate volatility exert significantly adverse effects upon unemployment, corroborating with the findings of Evans [11] and Tatom [38]. Still, interest rate volatility could be overlooked in previous estimations of the Barro, or other, models of business fluctuations due to the variable's relative constancy prior to 1979.