A0183
Title: Conditional threshold autoregression (CoTAR)
Authors: Kaiji Motegi - Kobe University (Japan) [presenting]
John Dennis - Institute for Defense Analyses (United States)
Shigeyuki Hamori - Kobe University (Japan)
Abstract: A new time series model is proposed where the threshold is specified as an empirical quantile of recent observations of a threshold variable. The resulting conditional threshold traces the fluctuation of the threshold variable, which can enhance the fit and interpretation of the model. In the proposed conditional threshold autoregressive (CoTAR) model, the existence of threshold effects can be tested by wild-bootstrap tests which incorporate all possible values of nuisance parameters. The estimation and hypothesis testing of the CoTAR model satisfy desired statistical properties in both large and small samples. We fit the CoTAR model to new confirmed COVID-19 cases in the U.S. and Japan. Significant conditional threshold effects are detected for both countries, and the implied persistence structures are consistent with the fact that the number of new confirmed cases in the U.S. is larger than in Japan.
