Predicting systemic financial crises with recurrent neural networks

We consider predicting systemic financial crises one to five years ahead using recurrent neural networks. We evaluate the prediction performance with the Jorda-Schularick-Taylor dataset, which includes the crisis dates and annual macroeconomic series of 17 countries over the period 1870-2016. Previous literature has found that simple neural net architectures are useful and outperform the traditional logistic regression model in predicting systemic financial crises. We show that such predictions can be significantly improved by making use of the Long-Short Term Memory (RNN-LSTM) and the Gated Recurrent Unit (RNN-GRU) neural nets. Behind the success is the recurrent networks' ability to make more robust predictions from the time series data. The results remain robust after extensive sensitivity analysis. (C) 2020 Published by Elsevier B.V.