Comparison Of Different Forecasting Methodologies For Interest Rates

Risk-free interest rate is the key variable of asset pricing in financial market.The reasonable and effective forecast of treasury bond yield can play a good guiding role in fixed income investment and portfolio allocation.Since Chinese bond market is not mature enough,interest rate level is affected by various non-market factors.Besides,the data quality also constraints empirical research.However,with the interest rate marketization and the gradual opening of the bond market,the research on the interest rate curve has become more and more practical.The main content of this paper is divided into three parts: factor decomposition of yield curve,macro-finance model and interest rate derivatives.In the first part,this paper uses two kinds of different ways to decompose the term structure of interest rate: The first type adopts principal component analysis,which shows that three independent factors can represent more than 99% information of curve;the second type adopts Nelson-Siegel model,and obtains parameter variable by fitting method.The factors of these two kinds of models do not have corresponding relationship,but they are similar in the accuracy of the interest rate forecast.The macro-finance model is intended to analyze the relationship between macro factors and the yield curve,and then to incorporate prospective indicators into prediction model.These indicators can be divided into real economy,inflation level,monetary policy and other related financial market,each part has a number of subdivision indicators.Through Granger causality test and Wald statistics of these factors,the linkage between factors is determined,and the first principal component of the real economy and inflation level is finally included in the macro-finance model of interest rate prediction.Interest rate derivatives use data on bond futures and interest rate swaps to compare the price discovery mechanism.Through correlation analysis and Granger causality test,it is found that the bond futures lag behind spot,interest rate swaps and spot synchronization changes.In this paper,the model is simulated by vector autoregressive(VAR)model,and the prediction accuracy significance is determined by Diebold-Mariano test.The results show that the simple yield factor model is more accurate in the short-term forecast,while the macro-finance model is more accurate in the long-term forecast,and the interest rate derivative data cannot improve the accuracy of interest rate prediction.