Forecasting Exchange Rates in the Short-Term: The Statistical and Machine Learning Approach

We show that regime classification can be a beneficial addition to deep learning models when forecasting exchange rates, especially in volatile environments or highly information efficient currencies. We build on existing applications of Deep Learning technologies in the financial spectrum, and develop a combinational approach to regime classification utilizing know Regime Aware Multiscale Transformer methods and a Mixture of Experts, comparing its performance with econometric and Deep Learning benchmarks. Overall, the findings indicate that the effectiveness of regime-aware Deep Learning is conditional on the underlying market environment and characteristics.