Theory and applications of TAR model with two threshold variables

A growing body of threshold models has been developed over the past two decades to capture the nonlinear movement of financial time series. Most of these models, however, contain a single threshold variable only. In many empirical applications, models with two or more threshold variables are needed. This article develops a new threshold autoregressive model which contains two threshold variables. A likelihood ratio test is proposed to determine the number of regimes in the model. The finite-sample performance of the estimators is evaluated and an empirical application is provided.