基于MCMC算法AR-GARCH模型中国出口集装箱运价指数波动性研究

摘要

我国是集装箱货物进出口大国，集装箱班轮运价的剧烈波动使货主和班轮公司面临巨大的风险。为研究我国出口集装箱运价波动风险，对中国出口集装箱运价指数（China Containerized Freight Index, CCFI）建立基于Griddy-Gibbs抽样MCMC算法的贝叶斯AR-GARCH模型。针对1998年4月至2013年12月的CCFI总指数的去均值周收益率数据，建立残差基于正态分布和T分布的AR-GARCH模型，运用WinBugs软件和MH算法进行贝叶斯参数估计，发现AR(3)-GARCH(1,1)模型拟合效果最好；参数估计结果表明，波动具有较强的持续性，不存在“风险溢价”和“杠杆效应”。经对比，发现AR-GARCH-T模型拟合效果更好；对比ML方法，发现MCMC算法估计结果的样本内拟合优度较差，而样本外预测能力较强。%China has a large number of importing and exporting containerized cargo. Shippers and shipping companies face enormous risks from liner freight rates volatility. An AR-GARCH model is proposed to capture dynamic volatility of CCFI with Griddy-Gibbs sampling applied to simulate in WinBUGS. CCFI weekly is from April 1998 to December 2013. The empirical results of MCMC algorithm to a Bayesian inference show that the AR(3)-GARCH(1,1) model well fit the data. The strong persistence of volatility is reflected by the estimations, but no risk-premium or leverage effects. Results show that AR-GARCH-T model has better fitting effect. The AR-GARCH-T model estimated by ML within the sample is more fitting, while the counter party inferred by Bayesian beyond the sample is more predictable.