In modelling smooth transition regression, the transition variable is supposed to be deterministic trend, lagged variable, or sometimes exogenous variable.
My question is what is the theory behind using such exogenous? what is the criterion behind using it?