Machine learning on Forex Carry Forecasting with Economics Feature

This paper studies the forecasting capability of machine learning models with economic features. The machine learning model constructed is based on Random Forest, Support Vector Machine, Decision Tree with Adaptive Boosting, and Hybrid Model. With the information of past risk metrics, our models signify the predictability of the currency market instability. The predictability comes from the fact that our machine learning model observes the violation of martingale restriction in the currency market portfolio. Furthermore, we apply the resulting outputs from the model to the forex carry trading strategy. The profitability of the corresponding trading strategy is significantly higher than those from a long-term holding strategy and the benchmark strategy constructed by the VIX.