A1450
Title: Dimension reduction in VAR models via informative lag selection
Authors: Wiranthe Herath - Drake University (United States) [presenting]
S Yaser Samadi - Southern Illinois University Carbondale (United States)
Abstract: The increasing dimensionality of multivariate time series data creates significant challenges for traditional vector autoregressive (VAR) models, frequently resulting in overfitting, inefficient estimation, and poor forecasting ability. To address these issues, two novel VAR models are developed that allow for targeted dimension reduction by focusing on the most informative lagged predictors. These models introduce a simple structure that not only improves estimation efficiency but also forecasting accuracy compared to traditional VAR approaches. Extensive simulation results and empirical analyses using finance and macroeconomic data demonstrate consistent performance gains in both parameter estimation and predictive outcomes, emphasizing the practical value of our techniques in multivariate time series contexts.
