A1567
Title: Robust change point detection in VARX models via recursive CUSUM tests
Authors: Sangjo Lee - Inha University (Korea, South) [presenting]
Abstract: A robust change point detection procedure for Vector Autoregression with Exogenous variables (VARX) models is developed in the presence of bounded non-stationarity in Exogenous policy variables, such as level shifts or regime changes. The proposed test is based on Recursive residuals from the Recursive Least Squares (RLS) algorithm. Since these residuals are constructed as one-step-ahead prediction errors, they preserve the Martingale Difference Sequence (MDS) property with respect to the natural filtration, regardless of the persistence properties of the Exogenous variables. Under the null hypothesis, the Recursive LSCUSUM statistic converges to the supremum of a Brownian bridge. Simulation results demonstrate that the method performs reliably under policy shifts and retains strong power for detecting genuine structural changes. Its practical value is illustrated through an application to U.S. financial market data.
