Degree
Doctor of Philosophy (PhD)
Department
Department of Economics
Document Type
Dissertation
Abstract
This dissertation develops novel econometric approaches to improve macroeconomic and financial forecasting in the presence of structural instability. Motivated by the observation that traditional factor-augmented regression models often assume time-invariant relationships, which are inadequate during periods of gradual economic change, this research advances dynamic forecasting methods that incorporate smooth structural shifts in high-dimensional environments. The first essay introduces a time-varying extension of the three-pass regression filter (3PRF) by incorporating locally adaptive rolling window estimation. This approach addresses the instability of forecast coefficients by allowing them to evolve smoothly over time. A data-driven method for selecting the optimal window size is developed, balancing forecast bias and variance. The resulting model demonstrates superior out-of-sample forecasting performance compared to static and full-sample alternatives in both simulated and empirical settings. The second essay proposes a time-varying factor-augmented regression model that allows for smoothly changing factor loadings and forecast coefficients. Using boundary-corrected kernel principal component analysis, latent factors are estimated in a localized manner, enabling the model to adapt to evolving economic regimes. Theoretical results establish the consistency and asymptotic normality of the proposed estimators. Empirical applications to U.S. macroeconomic data highlight the advantages of accommodating smooth parameter changes in both the factor structure and the predictive relationship. Together, these essays contribute to the growing literature on forecasting with high-dimensional data under structural change. By integrating time-varying estimation techniques into dynamic factor models, this dissertation provides robust tools for real-time forecasting in nonstationary environments and offers practical guidance for empirical economists and policymakers.
Date
7-10-2025
Recommended Citation
Liu, Wei, "Essays in Dynamic Factor Models with Smooth Structural Changes" (2025). LSU Doctoral Dissertations. 6837.
https://repository.lsu.edu/gradschool_dissertations/6837
Committee Chair
Zhou, Qiankun
DOI
10.31390/gradschool_dissertations.6837