A1058
Title: High-dimensional regularized additive matrix autoregressive model
Authors: Nilanjana Chakraborty - Indian Institute Of Management Udaipur (India) [presenting]
Debika Ghosh - Indian Institute of Management Udaipur (India)
Samrat Roy - Indian Institute of Management Ahmedabad (India)
Abstract: High-dimensional time series have diverse applications in econometrics and finance. Recent models for capturing temporal dependence have employed a bilinear representation for matrix time series, or the Tucker-decomposition-based representation in the case of tensor time series. A bilinear or Tucker-decomposition-based temporal effect is difficult to interpret on many occasions, along with its computational complexity due to the non-convex nature of the underlying optimization problem. Moreover, the existing matrix case models have not sufficiently explored the possibilities of imposing any lower-dimensional pattern on the transition matrices. A regularized additive matrix autoregressive model is proposed with additive interaction of row-wise and column-wise temporal dependence, which offers more interpretability, less computational burden due to its convex nature, and estimation of the underlying low rank plus sparse pattern of its transition matrices. The issue of identifiability of the various components in the model is addressed, and subsequently, a scalable alternating block minimization algorithm is developed for estimating the parameters. A finite sample error bound is provided under high-dimensional scaling for the model parameters. Finally, the efficacy of the proposed model is demonstrated on synthetic and real data.
