Title: Estimating detection in threshold auto-regressive models via dynamic programming
Authors: Ali Shojaie - University of Washington (United States) [presenting]
Abstract: Threshold autoregressive (TAR) models are used in many scientific applications, from economics and finance to epidemiology, due to their flexibility. However, existing approaches often assume a fixed number of thresholds as well as a pre-specified threshold variable. Both of these assumptions are somewhat arbitrary and often violated in applications. To address these limitations, we develop a dynamic programming approach to estimate the locations of the thresholds and the autoregressive parameters in high-dimensional TAR models where the threshold variable may need to be selected among a (small) set of candidate variables. We establish the consistency of both the estimated thresholds and the autoregressive parameters for high-dimensional TAR models and illustrate the advantages of the method via simulated and real data examples.