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

Abstract: 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.

Key words: Interest Rate Term Structure, NARX, Theoretical Price of Treasury Bond, Treasury Bonds Portfolio Management