Financial Time Series Forecasting Based On Local Preserving Projection

Financial time series is the external manifestation of financial market's inner fluctuation and change,which are bound to contain the valuable objective information about securities market.Therefore,through the quantitative analysis of time series,we can dig out the important information merged into the background,and then provide theoretical support for risk management and strategy development among the individual or government.At the same time,exploring the relationship between financial regulation and the securities market,the development of the whole market can be promoted healthily.Financial time series has a large amount of historical data,but most of the current forecasting models directly predict the value which ignore the important step of extracting the intrinsic characteristics of the time series.Under these circumstances,except the hidden information,the accuracy of prediction could be reduced.This paper research Shanghai Composite Index and the Nasdaq Index,which represents the index of the Chinese market and the capital market respectively,then aiming at the deficiency of existing feature extraction methods,revolving around locality preserving projection algorithm.Finally constructing hybrid model to study the broader market index forecasting method after the feature extraction.The main research work and innovation of this paper are as follows:(1)In view of the shortcomings of bidirectional two-dimensional principal component analysis algorithm in the financial time series prediction,a bidirectional subspace method of horizontal 2DPCA vertical 2DLPP is proposed.Compared with(2D)~2PCA,this method combines the advantages of 2DPCA and 2DLPP,and extracts both local and global features of financial time series.(2)For the shortcomings of one-dimensional linear feature extraction algorithm and(2D)~2PCA algorithm in financial time series,it is designed to use the bidirectional two-dimensional locality preserving projection algorithm to compare the feature extracting about original data set from the row direction and the column direction.The model gets the bidirectional feature matrix of the time series to achieve dimensionality reduction and feature extraction.The radialbasis function neural networks are used to get the regression prediction.Finally,analyzing the error of actual value and prediction value.(3)Empirical mode decomposition(EEMD)and support vector machines(SVM)are common models of financial time series forecasting.Researching the application of ensemble empirical mode decomposition in removing the noise from the securities exponential time series.Considering that the time series after noise removal are still s high dimensions data set for support vector machine,theoretically speaking,it will increase the complexity of the model,and affect the accuracy of prediction.Therefore,extracting the feature of denoised time series by LPP algorithm in the original EEMD-SVM model,we can reduce the dimension of SVM input data,and finally EEMD-LPP-SVM model is built for the prediction of financial time series.