Time-dependence or instability in the threshold so far in the literature may be implemented through either a change-point mechanism or a function of candidate variables. Bessec (2003) proposes a Self-Exciting Threshold Autoregressive (SETAR) model with time-varying thresholds using a one-time break where the change point is not estimated but corresponds to the time of the change in the official margins, and finds that the data do not reject the time-varying threshold specification. Yu and Fang (2019) proposes a threshold regression with a threshold boundary, in which the threshold is modeled as a function of a dummy variable. Dueker et al. (2013) present a model in which the threshold varies as a function of other observable variables, and demonstrate that models with constant thresholds have been outperformed by models with a time-varying threshold both in terms of in-sample fit and out-sample forecast accuracy