The existing vector heterogeneous autoregression (VHAR) does not allow for threshold effects. The threshold autoregressions are well established in the literature, but the presence of an unknown threshold complicates inference. To resolve this dilemma, we propose the vector moving average...