上交所国债收益率的聚类结构分析

对上海证券交易所2004年7月1日到11月12日的17种记账式国债的日收益率时间序列进行预分析,在此基础上对日收益率时间序列做了分离趋势修正。提出了一种基于各个国债收益率之间相关性的测度距离的方法,并根据这种新方法对修正后的国债日收益率时间序列进行了聚类分析。通过对采用联接算法计算出的阈值向量和层次树进行权衡比较,将17种国债分成了5类。分类结果表明,我国国债的日收益率之间的相关性程度并不是很明显地依赖于它们的到期时间,因此存在可能进行套利的机会。

The Clustering Analysis of Chinese Treasury Bonds Returns

The paper first analyzes daily returntime series of Chinese treasury bonds market from July 1 2004 to November 12 2004.And on the basis of that,the daily return time series are detrended.Then,a new metrical approach of distance by using the correlation of bond daily returns is proposed and the clustering analysis for the bond daily returns is conducted with the modified series of daily returns according to the new method.At last,the 17 bonds are classified into five categories by trading off between threshold value and hierarchical tree.The classification result indicates that the correlation of Chinese treasury bonds daily return does not evidently depend on their maturities,thus there may be an opportunity to arbitrage.