Minor Oral: "Inference of breaks for high-dimensional time series"

Abstract: We consider a new inference method for high dimensional time series. We target at dense or clustered cross-sectional signals, where we adopt an L2 aggregated statistics in the cross-sectional dimension. In the temporal dimension, we aggregate via the maximum norm. Gaussian Approximation results are provided under weak cross-sectional dependence assumptions which facilitates the inference of the breaks. Simulations show the power enhancement in presence of with clustered signals relative to the maximum statistics over the temporal dimension.

Hosts: Soumendra Lahiri and Likai Chen