Predicting US recessions with leading indicators via neural network models

This paper examines the relevance of variance financial and economic indicators in predicting US recessions via neural network models. We share the view that business cycles are asymmetric and cannot be adequately accommodated by linear constant-parameter single-index models. We employ a novel neural network (NN) to recursively model the relationship between the leading indicators and the probability of a future recession. The out-of-sample results show that via the NN model indicators such as interest rate spread, Department of Commerce leading index, Stock and Watson index, and S&P500 index are useful in predicting US recessions, including the most recent one in the early 1990s. Furthermore, when the out-of-sample forecasting period is divided into three subperiods, we find that the relevance of various leading indicators may change from time to time.