利率期限结构预测、国债定价及国债组合管理  

Term Structure Modeling,Prediction and Bond Pricing Based on a Neural Network Approach

作　　者：闫红蕾[1,2] 张自力 Yan Honglei , Zhang Zili

机构地区：[1]北京大学光华管理学院 [2]嘉实基金 [3]嘉实基金人工智能投研中心

出　　处：《统计研究》2018年第3期23-37,共15页Statistical Research

摘　　要：国债利率期限结构是固定收益产品定价和投资组合管理的核心问题。本文利用NARX(Nonlinear AutoRegressive network with e Xogenous inputs)神经网络模型研究利率曲线的运动机制,拟合并预测利率期限结构,在此基础上利用Hermite插值方法构造平滑的利率曲线并计算得到国债理论价格及其预测值。实证分析发现:我国国债定价效率不足,交易价格显著偏离理论价格,但国债理论价格的实际值和预测值均对交易价格具有显著的预测能力。基于上述发现,本文提出了主动国债组合管理策略,通过预测的期限结构得到国债理论价格的预测值构建的多空对冲组合和单边多头组合均能获得显著的收益。Treasury bond term structure is the core of fixed income products pricing and portfolio management. We use NARX （ Nonlinear AutoRegressive network with exogenous inputs） neural network to fit and predict interest rate term structure of treasury bonds, and we use Hermite interpolation to construct smooth yield curve whereby to compute real and predicted theoretical Treasury bond price. Empirical result shows the market prices of the treasury bonds notably deviate from their theoretical prices. However, both the real and predicted theoretical prices can forecast the market price of the treasury bonds. Based on such findings, we propose an active treasury management strategy by computing the predicted theoretical prices through forecasting term structure, and both of the hedging strategy and long only strategy gain remarkable profit while the risk is acceptable. Our research enriches the studies on interest rate term structure modeling, and the active strategy of Treasury bond portfolio management provides practical reference for enhancing pricing efficiency of treasury bonds market through hedging.

关 键 词：利率期限结构 NARX神经网络 国债理论价格 国债投资组合管理 

分 类 号：C812[社会学—统计学]