Predictive Ability of the Interest Rate Spread Using Neural Networks

Abstract

Interest rates are commonly used as predictors of future economic conditions as measured by industrial production, real gross domestic product and real total business sales (RTBS), as well as through the prediction of recessions in the economy. Recession forecasting is mainly characterized by probit categorical analysis, and there appear to be hardly any neural network research in this area. This paper contributes to the recession forecasting literature using interest rate spreads (the difference between the average yields on 10 year U.S. Treasury bonds and on 3 month U.S. Treasury bills) to forecast the 2007 to 2009 recession with neural network models referenced against regression models. It is shown that neural network models out-performed regression models as evidenced by the R-squared and mean square error performance metrics. Unlike other studies, the change in interest rates is used to compute the interest rate spread. The targeted variable is RTBS. The interest rate spread variable was used to generate three input variables comprising 23, 26, and 29 month leads respectively over RTBS.