A0202
Title: Matrix autoregressive model with vector time series covariates for spatio-temporal data
Authors: Yang Chen - University of Michigan (United States) [presenting]
Abstract: A new model is presented for forecasting time series data distributed on a spatial grid, using historical spatiotemporal data and auxiliary vector time series data. The historical matrix time series are mapped to the predicted matrix via row and column-specific autoregressive coefficient matrices. The vector predictors are mapped to the predicted matrix by taking a mode-product with a 3D tensor coefficient. Given the high dimensionality and underlying spatial structure, we represent the tensor coefficient by a functional-coefficient from a Reproducing Kernel Hilbert Space. We jointly estimate the autoregressive and functional coefficients under a penalized maximum likelihood estimation coupled with an alternating minimization algorithm. Large sample asymptotics of the estimators are established. The performance of the model is validated with extensive simulation studies. A real data application to forecast the total electron content maps, an important parameter for monitoring space weather impacts, will be presented.
