Early warning system for currency crises using long short-term memory and gated recurrent unit neural networks

Abstract

Currency crises, recurrent events in the economic history of developing, emerging, and developed countries, have disastrous economic consequences. This paper proposes an early warning system for currency crises using sophisticated recurrent neural networks, such as long short-term memory (LSTM) and gated recurrent unit (GRU). These models were initially used in language processing, where they performed well. Such models are increasingly being used in forecasting financial asset prices, including exchange rates, but they have not yet been applied to the prediction of currency crises. As for all recurrent neural networks, they allow for taking into account nonlinear interactions between variables and the influence of past data in a dynamic form. For a set of 68 countries including developed, emerging, and developing economies over the period of 1995–2020, LSTM and GRU outperformed our benchmark models. LSTM and GRU correctly sent continuous signals within a 2-year warning window to alert for 91% of the crises. For the LSTM, false signals represent only 14% of the emitted signals compared with 23% for logistic regression, making it an efficient early warning system for policymakers.