Does machine learning help private sectors to alarm crises? Evidence from China's currency market

With respect to a broad range of well-established early warning systems (EWSs) for financial crises, this study discusses the practical usefulness of currency early warning models in their service to the daily operation of private sectors. Based on 24-year daily and monthly data of China, a complete mixture of classic and emerging crisis early warning paradigms are examined at both the short and long predictive horizons. To answer the question whether and how machine learning algorithms contribute to the institutional assessment of currency vulnerabilities, the models are evaluated in terms of the out-of-sample crisis predictive power as well as the real-time improvements on the performance of currency portfolios. Evidences are found to support the potential of machine learning in outperforming canonical models. Nonetheless, it is stressed that in comparison to the selection of predictive mechanisms, the quantitative definition of crises, that addresses precision and robustness at the same time, is an equally important determinant for the success of an EWS. This study suggests that private sectors are more likely to be benefited from monitoring short-horizon market turmoils at daily frequency.