Currency crisis early warning signal mechanisms based on dynamic machine learning

The primary aim of this study is to investigate whether credit default swaps (CDS) serve as an early warning indicator for currency crises. This is done by examining both stock and flow variables, including the external debt stock and reserves (comprising foreign currency and gold), within the context of free exchange rate regimes. An original aspect of the study, which differs from other studies, is the machine learning methods used and the inclusion into the model of both one lag and lag values of the CDs variable, which is an inclusive crisis indicator. The CDS variable was not detected as a strong signal by the logistic regression model. However, the best-performing XGBoost and GB algorithms show the differenced, and one-lagged values of the CDS variable produce significant signals in forecasting currency crises. Consistent with theoretical underpinnings of study on currency crises, this implies that central banks proactively reacted by increasing monetary policy interest rates and the non-current value CDS but its lagged value performed strong early warning signal that is a follower or supplementary indicator of the credibility of monetary authorities and policies. These results demonstrate that the high and rising interest rate signifies that domestic currencies are being supported against speculative attacks. Copyright © 2024 Inderscience Enterprises Ltd.