基于跳辨识-MCMC组合算法的人民币汇率跳扩散模型参数估计问题

Parameter estimation of CNY exchange rates of jump-diffusion model using mixed algorithm of jump identification and MCMC

针对人民币汇率收益率时间序列数据存在的跳变特征,采用跳扩散模型对其时间序列数据进行描述.为识别跳变规律并解决模型的参数估计问题,提出了基于跳辨识-MCMC的组合算法:即结合Lee-Mykland的跳辨识方法与MCMC(蒙特卡罗马尔可夫链)方法形成组合算法,利用仿真实验,通过误差分析得出组合算法在跳扩散模型参数估计方面效果明显优于单一MCMC方法.以人民币/美元日汇率数据为样本进行实证分析,结果表明组合算法不但能较为准确地识别出汇率收益率的跳变时刻及规律,而且其模型参数估计的有效性大大提高.

Since the returns of the CNY exchange rates present some jump features, this paper uses the jump-diffusion model to describe their sequential data. For identifying the law of jumps and solving the parameter estimation of the model, the paper proposes the mixed algorithm based on the jump identification and MCMC methods, i.e., combination of the Lee-Mykland jump identifying method and MCMC. By the simulation experiment, the error analysis shows that the mixed algorithm is superior to the simple MCMC method in estimating the parameters of the jump diffusion model. Also, empirical results sampling by the CNY/USD exchange rates of return show that the mixed algorithm can not only identify the accurate dates and laws of jumps of exchange rate returns, and the estimation validity of the model parameters improves considerably.