Forecasting Central Bank Policy Rates Using Machine Learning and Deep Learning Approaches

Accurate forecasting of central bank policy rates is critical for guiding monetary policy, shaping market expectations, and maintaining macroeconomic stability. In emerging economies such as Mongolia, conventional econometric approaches, including the Taylor Rule, ARIMA, and SVAR, often struggle to capture nonlinear dynamics, temporal dependencies, and structural breaks. This study addresses these limitations by developing and evaluating modern forecasting methods that combine machine learning and deep learning models within hybrid frameworks. The analysis employs a comprehensive monthly dataset of 26 macroeconomic indicators spanning January 2008 to December 2024. Seven models are constructed and assessed using RMSE, MAE, and R² metrics. The empirical results show that hybrid approaches, particularly XGBoost combined with Gradient Boosting and LSTM integrated with XGBoost, deliver the highest predictive accuracy, with the leading model reaching an R² of 0.9355. These hybrid methods consistently outperform both traditional econometric and standalone ML or DL models in capturing complex macroeconomic patterns and structural changes. The findings provide a robust data-driven framework to support evidence-based monetary policy in Mongolia and offer a transferable methodology for other emerging markets facing similar economic challenges.