Open AccessTime series forecasting of stock market index based on ENWMPCA model
Author(s) -
Zhiqiang Guo,
Xiao Wang,
Yali Zeng,
Jie Yang
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1237/5/052029
Subject(s) - eigenvalues and eigenvectors , stock market , principal component analysis , normalization (sociology) , time series , artificial intelligence , computer science , series (stratigraphy) , stock market index , support vector machine , dimensionality reduction , algorithm , mathematics , econometrics , data mining , machine learning , paleontology , physics , horse , quantum mechanics , sociology , anthropology , biology
Aiming at the high dimension of factors affecting the stock market index, and the problem that MPCA algorithm cannot distinguish the contribution of different eigenvectors to financial time series forecasting, eigenvalue normalization weighted multilinear principal component analysis (ENW-MPCA) is put forward. Due to the existence of correlation between stock markets, 34 technical indexes of 7 stock markets are selected to construct a three-dimensional tensor model. Eigenvectors of different eigenvalues have different contributions to the prediction accuracy of MPCA algorithm, thus adopt ENW-MPCA algorithm for the feature extraction, which can distinguish the different contribution of eigenvectors corresponding to different eigenvalues to the prediction while reducing dimension. Then using support vector machine for regression prediction, the prediction value of financial time series is obtained. Experimental results on the Hang Seng Index shows that, compared with the MPCA, the forecasting error of ENW-MPCA is smaller and the prediction accuracy is improved to a certain extent. This indicates that the proposed algorithm can fully retain the internal structure of stock time series, proving its validity and practicability.