A Shock-Aware Machine Learning Framework for Adaptive Cryptocurrency Hedging in Exchange Rate Risk Management: Evidence from Post-Subsidy Nigeria

In the unstable foreign exchange environment that emerged in Nigeria following subsidy removal, exchange rate risk management and financial decision-making have become increasingly difficult. Under such conditions, traditional forecasting methods and static hedging strategies may not provide sufficient protection against sharp exchange rate movements. This paper examines whether machine learning predictions can be transformed into practical decision rules for exchange rate risk management through dynamic cryptocurrency hedging. Daily data on the USD/NGN exchange rate, Bitcoin, and Ethereum from 1 January 2018 to 1 January 2025 are employed, with particular emphasis on the post-subsidy period beginning on 1 June 2023. Three forecasting models—linear regression, random forest, and extreme gradient boosting—are evaluated, and their outputs are integrated into four decision strategies: FX-only exposure, static hedge, adaptive hedge, and shock-only hedge. The results show that conventional forecasting accuracy remains weak, as reflected in low directional accuracy and strongly negative R-squared values. However, once the forecasts are embedded into hedging decisions, their economic value becomes much clearer. Among the strategies considered, the adaptive hedge is found to be the most effective, while the linear forecasting model delivers the strongest return–risk trade-off. By contrast, static hedge and shock-only hedge strategies are substantially less effective. The findings suggest that cryptocurrencies can support exchange rate risk management in post-subsidy Nigeria when they are embedded in a dynamic and shock-aware hedging framework.