半非参数模型选择研究以及在上交所回购利率中的应用

Research on Seminonparametric Model Selection with Application to the Pledged Repo Interest Rate in Shanghai Stock Exchange

首先基于模型动态稳定性的要求,本文改进了半非参数(SNP)模型的选择方法,使所选择的SNP模型既能很好地拟合真实样本又能模拟与真实样本统计特征相近的时间序列。其次,本文得到了二维随机过程情形下赫米特展开项的理论结果。再次,实证结果表明:上交所旧质押式回购利率初始、插值后样本两组数据的最优SNP模型均为Semiparametric AR(1)-GARCH(1,1)(即11118000)模型,但是两者的系数估计值却不相同。最后,本文的实证结果表明了所提出的SNP模型选择改进方法的合理性与稳定性。

Based on model dynamic stability, this paper improves the approach of selecting SNP （Seminonparametric） model, making the SNP model selected better can appropriately fit the true sample and simulate a time series with statistical property similar to the true sample. This paper derive theoretical result for the Hermite expansion terms under 2-dimensional stochastic processes and empirical results show that the optimal SNP model, for both raw data and interpolated data of pledged repo interest rate in the Shanghai Stock Exchange, is coincidently the Semiparametric AR （1） --GARCH （1, 1） model, the 11118000 model, except for different parameter estimates. Finally, the empirical results prove that the way to improve SNP model selection is appropriate and stable.