Nonlinear autoregressive models with optimality properties
提出一类新的非线性自回归模型，基于预测似然函数的得分更新参数，具有信息论最优性质，并在美国宏观经济时间序列中验证了预测性能。
We introduce a new class of nonlinear autoregressive models from their representation as linear autoregressive models with time-varying coefficients. The parameter updating scheme is subsequently based on the score of the predictive likelihood function at each point in time. We study in detail the information theoretic optimality properties of this updating scheme and establish the asymptotic theory for the maximum likelihood estimator of the static parameters of the model. We compare the dynamic properties of the new model with those of well-known nonlinear dynamic models such as the threshold and smooth transition autoregressive models. Finally, we study the model’s performance in a Monte Carlo study and in an empirical out-of-sample forecasting analysis for U.S. macroeconomic time series.