Empirical Study Of Chinese Interest Rate Dynamic Process Based On Volatility Models

Interest rate is one of the most important and fundamental parameters in financial market. It plays a fundamental role in asset pricing, risk management and macro-economy analysis. As for the importance of interest rate, a lot of literature has been devoted to the studies of interest rate and its dynamic process. There are few researches on Chinese interest rate earlier, but as the interest rate gradually liberalizes in China, the research about Chinese interest rate will become more and more.Usually the volatility of financial time series is time-varying, so volatility models are introduced to one-factor CKLS model in this paper. These volatility models include stochastic volatility model (SV model) and GARCH model. Level effect, jump factor and non-normal distribution are also introduced to the traditional volatility models to capture the level effect, jump process, fat tail and mean-reverting effects in Chinese short rates.This paper employs Markov Chain Monte Carlo (MCMC) method to estimate SV model, GARCH model and CKLS model and compares their performance in modeling Chinese repo rates using deviance information creteria. The comparison method between SV model and GARCH model is not consolidated and this paper applies a brand-new way. The way the paper estimates SV model and compares these models is not involved in domestic research, which makes an improvement.This paper applies WinBUGS to complete parameter estimating and model comparison. The empirical results show that SV model is better at capturing Chinese repo rates than GARCH model and CKLS model. This result is quite robust when level effect, jump and non-normal errors are also introduced into the model. This has strong implications for asset pricing and risk management. This paper also suggest that mean-reverting, level effect and volatility clustering are significant for Chinese short rates. Finally, this paper simulates bond pricing to test whether different types of models would show significantly different results in asset-pricing.