Multivariate time series forecasting based on BP-SVR

Author

Fan, Xinwei ; Li, Fengyuan

Author_Institution

Coll. of Quality & Safety Eng., China Jiliang Univ., Hangzhou, China

fYear

2011

fDate

27-29 June 2011

Firstpage

3340

Lastpage

3343

Abstract

The characteristics of financial time series : (1)the selection process is random, complex; (2) contain most of the noise; (3) between the data with strong non-linear. The traditional prediction technologies cannot disclose the inherent rule of stock market. In this paper briefly introduces the basic theory of Support Vector Regress (SVR), and applies SVR combined with neural network (BP-SVR) to create a model, which also can be used for forecasting the multivariate time series. The result of simulation shows that the new model is the least in the mean squared error, which demonstrates that the BP-SVR model has a good ability to generalize.

Keywords

backpropagation; economic forecasting; mean square error methods; neural nets; regression analysis; support vector machines; time series; BP-SVR model; financial time series; mean squared error; multivariate time series forecast; neural network; prediction technology; stock market; support vector regression; Artificial neural networks; Computer languages; Forecasting; Statistical learning; Stock markets; Support vector machines; Time series analysis; Data mining; Support Vector Regress (SVR); neural network; time Series;

fLanguage

English

Publisher

ieee

Conference_Titel

Computer Science and Service System (CSSS), 2011 International Conference on

Conference_Location

Nanjing

Print_ISBN

978-1-4244-9762-1

Type

conf

DOI

10.1109/CSSS.2011.5974111

Filename

5974111