基于高斯过程混合模型的国债收益率预测

债券分析的核心问题是发现偿还期限与到期收益率之间的关系,即利率期限结构,而实际上国债利率期限结构是最为重要和基本的模式。目前人们对于利率期限结构的分析主要采用经济理论模型和数量模型进行,但是这两种方法都难于对国债收益率进行有效的预测。基于高斯过程混合模型强大的数据拟合和分析能力,本文将其应用于国债收益率的建模和预测。本文采用国债收益率数据作为输出变量,筛选出对国债收益率影响最强的一组作用因子作为驱动或输入变量,然后利用高斯过程混合模型对数据进行学习和建模,并依此对国债收益率进行建模和分析。实验结果表明高斯过程混合模型能够更好的描述国债利率期限结构。相比于其他机器学习模型和算法,高斯过程混合模型在国债收益率的测试数据上获得了更好准确的预测结果。 

Gaussian Process Mixture Based Prediction of Treasury Bond Yield Rate

The key problem of bond research is how to find the relationship between repayment maturity and yield to maturity, known as the term structure of interest rates. Among all the term structures of interest rates, the most important one is the term structure of the interest rates of treasury bond. Generally, there are two methods using economic theoretical model and quantitative model, respectively, for investigating the curve structure of interest rate, however, they are not so effective to predict the rate of treasure bond yield. With the help of the strong data fitting ability of the mixture of Gaussian processes(MGP) model, we apply it to the forecast of the rate of the treasury bound or national debt yield in this paper. Specifically, we consider the treasury bond yield rate as the output variable, screen out the most important factors as the input or driving variable, and utilize the MGP model to learn and fit the given data for the prediction of the rate of treasury bond yield. It is demonstrated by the experimental results that the MGP model can better describe the term structure of the interest rates of treasury bond. Moreover, in comparison with the other machine learning models, the MGP model can achieve the better prediction result on the test data of bond yield.