A Generalized Nearly Isotonic Regression and its Applications

• Main Authors: FANG, I-TING, 方奕婷
• Other Authors: HUANG, SHIH-FENG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/s9f2p2

碩士 === 國立高雄大學 === 統計學研究所 === 107 === This study proposes a non-parametric approach, called Generalized Nearly Isotonic Regression (GNIR), to describe the dynamics of data. The GNIR is capable of depicting the up/down fluctuation of data automatically. An algorithm for the GNIR is proposed and its convergence property is derived. The daily Value-at-Risk (VaR) of 13 global financial markets during 2006-2017 are employed for our empirical study. The numerical results reveal that the GNIR is capable of depicting the dynamics of VaR sequences well. In addition, the sequence of risk status for each market is defined by the fitted GNIR and predict the risk status for the 13 markets. Furthermore, we use the association rules to establish the associations among the risk status of the 13 financial markets. The empirical results show that applying the associations among financial markets is capable of improving the accuracy in predicting market risk status.