A0898
Title: Two sample and ANOVA test for high-dimensional non-stationary time series
Authors: Yunyi Zhang - The Chinese University of Hong Kong, Shenzhen (China) [presenting]
Abstract: Testing for differences in population means is a fundamental topic in statistics literature and has been extensively studied in the context of independent samples. However, when the data exhibit high dimensionality and dependence simultaneously, the bias introduced by dependence may invalidate classical test procedures. A new test statistic for two-sample and ANOVA test of high-dimensional time series is introduced that mitigates bias by excluding main diagonals. In addition, a dependent wild bootstrap algorithm is proposed to facilitate statistical inference. Theoretical results, including a Gaussian approximation theorem and the validity of the bootstrap procedure, are established. Numerical experiments demonstrate that the proposed test maintains accurate size and achieves good power performance compared to the existing methods. Given the prevalence of high dimensionality and non-stationarity in modern-era time series data, the proposed test offers a reliable approach for comparing sample means in such complex settings.
