Detection of Common Long-Range Dependence Component for Multivariate Time Series

• Main Authors: Ying-Shen Chen, 陳英森
• Other Authors: Nan-Jung Hsu
• Format: Others
• Language: en_US
• Published: 1999
• Online Access: http://ndltd.ncl.edu.tw/handle/43264190534525540518

碩士 === 東海大學 === 統計學系 === 87 === In this paper, we proposed a new method to identifying common long-range dependent component in a multivariate time series. A common long-range dependent component exists if individual series are all long-range dependent but there exists a particular linear combination of the process which does not have the long-memory property. We first find the linear combination by the two-stage least squares procedure and then test the long-memory property for the transformed data using the method proposed by Geweke and Porter-Hudak (1983). The performance of the proposed test is investigated via Monte Carlo simulation and compared with the previous method based on the canonical correlation analysis. The new approach has better identifying ability than the previous method in detecting common long-range dependence component for multivariate time series. Moreover, it also produces the accurate estimation for the factor loading matrix. For illustration, the squared returns of the daily weighted stock indices are used for three electronic companies in Taiwan--the Acer Incorporated Company, the Compal Electronic Incorporated Company and the Taiwan Semiconductor Manufacturing Company.