This paper develops a cointegrating nonlinear autoregressive distributed lag (NARDL) model in which short- and long-run nonlinearities are introduced via positive and negative partial sum decompositions of the explanatory variables. We demonstrate that the model is estimable by OLS and that...