半参数STAR模型的估计及应用

Estimation of semi-parametric STAR model and its application

文中首次提出了一个新的STAR模型,在保留了转换函数的前提下,让转换变量以非参数的形式进入转换函数,从而有效减少了模型误设的风险,提高了样本内拟合和样本外预测的能力.蒙特卡罗实验的结果显示半参数STAR模型的有限样本拟合结果令人满意.利用1994年1月到2012年7月的人民币实际有效汇率月度数据,将半参数STAR模型和随机游走模型、自回归模型、门限自回归模型、平滑转换自回归模型和人工神经网络模型的样本外预测能力进行比较,结果显示半参数STAR模型在样本外预测能力上具有显著优势.

This paper proposes a new semi-parametric STAR model in which we allow transition vari- ables to enter into the transition function nonparametrically but the transition function itself is a known parametric form. The new model can avoid the risk of misspecification and improve the in-sample good- ness of fit and the out-of-sample forecasts. Monte Carlo simulations show that the estimates have good finite sample performance. Using the monthly real effective exchange rates from January 1994 to July 2012, we find that the new model outperforms other popular models such as the random walk model, the autoregressive model, the threshold autoregressive model, the smooth transition autoregressive model and artificial neural network model in terms of out-of-sample forecasting ability.