• <Source Information> Jin Seo Cho, Matthew Greenwood-Nimmo and Yongcheol Shin (2024): Working Paper.

  • <Abstract> The nonlinear autoregressive distributed lag (NARDL) model is a single-equation error correction model that has been widely applied to accommodate asymmetry in the long-run equilibrium relationship and the short-run dynamic coefficients with respect to positive and negative changes in the explanatory variable(s). The NARDL model exhibits an asymptotic singularity issue that frustrates efforts to derive the asymptotic properties of the single-step estimator. We propose a two-step estimation in which the parameters of the long-run relationship are estimated by the fully-modified least squares estimator before the short-run dynamic parameters are estimated by OLS. We establish that the two-step NARDL estimators follow a limiting normal distribution, the validity of which is confirmed by Monte Carlo simulations. We demonstrate the utility of our approach with an application to the asymmetric relationship between R&D intensity and investment in the U.S.