A1959
Title: Parameter estimation under high-dimensional dynamic effects with an application to multilayer network connectedness
Authors: Shiyun Cao - Jinan University (China)
Wei Shi - Jinan University (China) [presenting]
Abstract: The estimation of regression coefficients in high-dimensional time series with complex dependencies from lagged outcome variables is examined. The outcome variable is structured as a multidimensional array and forms a tensor-valued time series. The interaction matrix is inferred from the data without using a predetermined one. Complex dynamic effects are modeled by assuming that the autoregression coefficients exhibit a low-rank structure. The low-dimensional parameters are estimated using orthogonal moments and sample splitting, with estimates shown to be $\sqrt{NT}^{-1}$ consistent and asymptotically normal. Simulations indicate that the estimators perform well in finite samples. As a practical application, the proposed method quantifies connectedness among interconnected multilayer networks by application to the study of the impact of economic development on biodiversity in China, accounting for the spatio-temporal dependence among species.
