Multi-Frequency Combination Model Of Financial Time Series And Its Application

Financial market is a huge dynamic system with complex movement patterns in the era of global economic integration.Financial time series(FTS)is a very important data expression form of financial market and the external expression of the market internal operation rule.The prediction and analysis of FTS can explore the potential signal characteristics and operation rules of the market,provide valuable basis for the financial decisions,and facilitate practitioners to supervise and manage the market and prevent financial risks.Therefore,the analysis and prediction of FTS have important theoretical value and practical significance.FTS has the characteristics of non-linear and non-stationary,which lead to the complexity of the FTS prediction and the undesirable effects of traditional prediction methods.In order to improve the prediction performance,a Multi-Frequency Combination Model(MFCM)is proposed,which simplifies the complex sequence to facilitate analysis and modeling.In view of the existing MFCM,an improvement and application of MFCM is proposed,the main work is as follows:1.The existing decomposition method is mainly the Empirical Mode Decomposition(EMD)series method,which relies on integral transformation.The number of screenings for such methods is difficult to determine,and the integrated model introduces noise causing data contamination.Extreme-point Symmetric Mode Decomposition(ESMD)abandoned the traditional idea of integral transformation and creatively put forward the "direct interpolation method",which improved the defects of a series of EMD methods.However,research in the field of finance is still in its infancy.In this paper,the ESMD method is applied to FTS,and compared with the existing MFCM decomposition method,the decomposition effect is significantly improved.2.Existing MFCM reconstruction methods only focus on specific data characteristics,and ignore the other data features.To solve this problem,a reconstruction method based on the comprehensive data analysis of the mode components is proposed,the mode components are reconstructed into the reconstruction components which are easy to model and have certain economic significance under the premise of ensuring the component information,so as to optimize performance of models.3.In the part of components prediction,to solve the problem of insufficient information mining in Support Vector Regression(SVR),a multi-task prediction model based on constructing data sets of adjacent points is proposed.The multi-task training set is obtained by constructing sub-data sets with different input and output intervals,and the related information of adjacent time points is introduced technically.Based on the multi-task training set,multiple tasks are trained at the same time,and the information sharing between tasks can play a correction role,which fully mining the relationship between adjacent points.Then the prediction accuracy of reconstructed components is obviously improved.4.The improved MFCM model is applied to the empirical study of gold price series.The experimental results show that compared with the existing MFCM,the improved MFCM has significantly improved the decomposition,prediction and overall model effect.The improved MFCM method proposed in this paper can provide a new idea and reference for the prediction and analysis of FTS.