Identifying Us Business Cycle Regimes Using Factor Augmented Neural Network Models

We propose a factor augmented neural network model to identify the current state (instead of future) of the US business cycle. Dynamic factors are extracted from a large-scale data set consisted of 122 variables. Then, these dynamic factors are fed into neural network models for predicting the current business cycle regime. First, we show that our proposed method determines US business cycle regimes quite accurately in sample and out of sample without taking account of the historical data availability. Then in a pseudo real time exercise, we also show that our neural network models identify business cycle regimes in a timely and accurate manner. Finally, our results indicate that neural network models outperform probit models.