Study Of Nonlinear Characteristics Of China’s Short-term Interest Rate

Dynamic behaviors in short-term interest rates have been the hot issue in the field of financial economics. As one of the basic variables in the financial markets, short-term interest rate plays a pivotal role in the pricing of financial derivatives. The accurate estimation of dynamic behaviors not only can provide useful information to investors for asset allocation to reduce investment risk, but also provide reference for policy formulation and implementation of the Central Bank. Given the importance of short-term interest rate, the research of its dynamic behaviors has important practical significance.With the advancement of financial econometrics, the research of dynamic model of interest rates has made great progress, the traditional dynamic models continue to be expanded. The existing studies have shown that dynamic behavior of short-term interest rates has obvious nonlinear characteristics. Domestic studies on short-term interest rates in China have also made considerable progress. Many studies have adopted different nonlinear time series model to capture the nonlinear dynamic characteristics of the interest rate. For example, Liu Jinquan and Zheng Tingguo (2006) introduced Markov regime switching in the CKLS model; Wu Jilin and Tao Wangsheng (2009) developed Smith (2002) regime switching stochastic volatility model, introducing the nonlinear drift; Pan Wanbin, Tao Libin and Liao Boqi(2008) considered threshold effect in the drift of CKLS model when modeling the nonlinear drift in interest rates dynamic model.But there is no consistent conclusion about which nonlinear model could portray the dynamic behavior of short-term interest rates better. Therefore, using the Markov regime switching model, the smooth transition autoregressive model (STAR) and contemporaneous smooth transition threshold autoregressive model (CSTAR) three main non-linear models, we conduct empirical analysis of SHIBOR overnight lending rate from January2007to February2012, and then compare the pros and cons of each model from sample goodness of fit and out-sample predictive ability.Our findings show that considering the regime switching model can portray the dynamic characteristics of short-term interest rates better than the linear model. Additionally, the regime switching models can capture the structural changes and non-linear characteristics of the dynamic changes, and have a more accurate out-sample prediction. On the sample goodness of fit, the Markov regime switching model is slightly better than the CSTAR model and CSTAR model is superior to LSTAR model. Looking at out-sample forecast, when the forecast number of days is1-2days, the predictive ability of Markov regime switching model is higher than the CSTAR model and the LSTAR model;When the number of forecast days is3-6days, CSTAR model and LSTAR model have higher prediction accuracy than Markov regime switching model; the predictive ability of CSTAR model is always superior to LSTAR model in1-6days.