人民币汇率的半参数预测模型

Semi-parametric forecasting model for USD/CNY exchange rate

利用从2006年1月4日到2008年7月18日人民币对美元汇率中间价的日均数据,同时运用非参数函数系数模型和GARCH模型来分析人民币对美元汇率收益率与波动率的非线性时间序列特征.实证结果表明,半参数组合模型具有较好的拟合以及预测效果,而且汇率管制政策变动的虚拟变量的估计系数显著不为0.跨度为50天的样本外预测显示:96%的收益率真实值都落在2.5%以及97.5%的非参数分位数回归预测线区间之内;参数GARCH(1,1)模型拟合的波动率所显示出的汇率震荡与实际情况一致.

Using the daily USD/CNY exchange rate time series data from January 4th,2006 to July 18th, 2008,this paper proposes a semi-parametric approach to model the conditional mean and conditional volatility simultaneously.A nonparametric functional-coefficient model is employed to estimate the conditional mean,and a GARCH-type model with a policy change dummy is adopted to describe the dynamics of the conditional volatility.Moreover,the corresponding policies play a significant role on both mean and volatility.Finally,a nonparametric quantile regression estimation is applied to compute prediction intervals. The empirical results demonstrate that the proposed semi-parametric model has good performance in terms of in-sample goodness of fit and out-of-sample forecasts.