CMStatistics 2018: Start Registration
View Submission - CFE
Title: Leverage subsampling for vector autoregression Authors:  Shuyang Bai - University of Georgia (United States) [presenting]
Ping Ma - University of Georgia (United States)
WenXuan Zhong - University of Georgia (United States)
Rui Xie - University of Georgia (United States)
Zengyan Wang - University of Georgia (United States)
Abstract: When performing statistical analysis of streaming multivariate time series data, the computational capacity often cannot afford to allow all the available data to estimate the model parameters. The sample size needs to be reduced and a subsample must be selected. However, a naive subsample selection approach may lose statistical efficiency. In the context of vector autoregression, we propose a subsample selection method based on leverage score, which is shown to achieve a high efficiency compared with some naive approaches. Furthermore, the method can be adapted to an online decentralized computing setup.