A1237
Title: Dynamic shrinkage and selection for exchange rate forecasting
Authors: Zheng Fan - University of Melbourne (Australia) [presenting]
Abstract: I propose a novel methodology that integrates dynamic sparsity and the dynamic shrinkage process into a coherent framework. This approach introduces an adaptive dynamic variable selection mechanism to achieve time-varying sparsity, ensuring that variables contributing no meaningful variation are deactivated during specific periods. This feature enhances interpretability by highlighting key variables, removing noise, and maintaining model parsimony. The mechanism is seamlessly incorporated into the dynamic shrinkage process, which balances global and local shrinkage on coefficient drift to better capture evolving coefficient patterns. By dynamically identifying relevant predictors based on their changing importance over time, this methodology adapts to shifting economic conditions and mitigates the potential risk of overfitting.
