經濟學系講座: Time-varying Factor Selection: A Sparse Fused GMM Approach
講者: 崔麗媛教授 / 香港城市大學商學院經濟及金融系副教授
日期：23/10/2024 (三)
時間：14:00 – 15:15
地點: E21B-G002
語言: 英語
內容: Dynamic Mode Decomposition (DMD) is a potential tool for dimension reduction and factorization of complex dynamical systems. However, compared to PCA-based approaches, the theoretical properties of DMD-based methods are less developed. Most DMD-based methods rely on prior knowledge or accurate estimation of the rank of the transition matrix to realize their advantages, which can limit their performance in scenarios with weak signals. To address this limitation, we introduce a truncated-ridge regularized DMD method that provides more accurate and interpretable predictions, along with better estimation of the transition matrix. Additionally, to handle the impact of noise, we propose a bagging truncated ridge DMD method based on block bootstrap, significantly enhancing the robustness of ridge DMD. We establish favorable theoretical guarantees and demonstrate its superior performance through a comprehensive set of simulation studies. In an empirical application, we apply the newly proposed RDMD method to predict inflation rates for 20 countries. By examining the roles of cross-sectional and time-series dependence among different countries, we improve the out-of-sample performance for inflation forecasting.
