Multi-step prediction for nonlinear autoregressive models based on empirical distributions

A multi-step prediction procedure for nonlinear autoregressive (NLAR) model s based on empirical distributions is proposed. Calculations involved in th is prediction scheme are rather simple. It is shown that the proposed predi ctors are asymptotically equivalent to the exact least squares multi-step p redictors, which are computable only when the innovation distribution has a simple known form. Simulation studies are conducted for two- and three-ste p predictors of two NLAR models.