Can Monetary Policy Uncertainty Predict Exchange Rate Volatility? New Evidence from Hybrid Neural Network−GARCH Model

This paper aims to examine the capacity of the US Monetary Policy Uncertainty (MPU) Index in accurately forecasting the volatility of foreign exchange rates for the Dollar Index, Euro/US$, and Yen/US$. To achieve this, we introduce several hybrid Artificial Neural Networks (ANN)−GARCH models, namely GARCH, ANN−EGARCH, and ANN−GJR−GARCH, which incorporate MPU as the exogenous variable (X). In addition to that, a significant challenge in ANN modeling is choosing the appropriate activation function. Therefore, we consider and compare various forms of activation functions, including logistic, Gompertz, Tanh, ReLU, and leakyReLU. Our results demonstrate that incorporating MPU improves the forecasting performance of the benchmark ANN−GARCH−type models both in- and out-of-sample. In particular, we find that incorporating MPU into the ANN−EGARCH model yields the largest forecasting gains compared to all other variants of the ANN−GARCH−type models. Additionally, our findings reveal that ReLU is the best activation function for predicting Dollar and Yen volatility, while Gompertz performs the best for predicting Euro volatility.