The Data-Driven Research And Empirical Analysis On Hybrid Models Of Financial Time Series Forecasting

Finance is the core of the modern economy,which relates the overall situation of the national economic and social development.In today's era of the globalization of economic development and the integration of international financial markets,the fluctuations of local regional financial markets are likely to trigger chain reactions affecting the world.At the same time,with the rapid development of China's economy,the continuous improvement of the market economic system,the improvement of people's living level,and the increasing people's investment consciousness,more and more funds have poured into the financial markets.Therefore,it is very crucial to deeply understand and predict the operation rules of the financial markets for the countries,financial institutions and investors.Influenced by many factors,a financial market is a huge,highly complex and dynamic system.So it is very difficult to predict the financial markets.A financial time series is the comprehensive external manifestation of the complex financial market.Through the analysis and prediction of financial time series,we can explore the potential internal relationship and rules of financial markets,and provide important reference for investors and decision makers in financial markets.With the development of information science and computer technology,the acquisition and storage of data are more convenient.A large number of financial data are gathered,which provides the basis for the research of financial time series forecasting.In recent years,the development of big data,artificial intelligence and modern mathematics provides a new research direction for time series forecasting.The research on financial time series forecasting has become a hot spot in academic and practical circles.A large number of literatures on financial time series forecasting show that a single forecasting model,whether a linear traditional statistical model or nonlinear computational intelligence method,can not describe all the characteristics of financial time series.By combining different types of methods or models in a certain way to establish a hybrid model,the accuracy and stability of financial time series forecasting can be effectively improved.Based on these,this paper studies the financial time series hybrid forecasting model,and mainly research on the latest research results in the fields of computational intelligence methods,control theory and data processing methods.Then we construct a series of financial time series hybrid forecasting models to predict the daily trend of financial products prices in the financial market.The main research contents and results of this paper can be divided into the following several sections:Firstly,this paper uses a hybrid model combining the autoregressive integrated moving average(ARIMA)model with the Taylor expansion forecasting(TEF)based on a tracking differentiator(TD).The ARIMA-TEF model is compared with the single ARIMA and TEF models.Adopting empirical study to show the ARIMA-TEF model owning higher prediction accuracy than the single model,which shows the superiority of hybrid model once again.Secondly,the TEF model is improved and a more reasonable ARIMA-TEF hybrid prediction model is proposed.Specifically,the initial state will appear the "peak" problem when the constant gain tracking differentiator estimates derivatives.While the variable gain tracking differentiator will weaken the "peak" problem.Therefore,this paper introduces a variable gain tracking differentiator to replace the constant gain tracking differentiator.The different orders of Taylor expansion model are analyzed in theory and simulation experiment,and a Taylor expansion prediction model with higher prediction accuracy is constructed.Finally,in order to verify the effectiveness of the proposed hybrid forecasting model,the neural network model and the hybrid models combined with neural network are taken as the benchmark models.The real time series data with gold and silver futures prices are selected for empirical analysis.The empirical results show that the improved ARIMA-TEF hybrid model has better prediction performance,and the short-term prediction effect is obviously better.Thirdly,in order to improve the prediction performance of the model,the ensemble empirical mode decomposition(EEMD)technology is applied to financial time series forecasting.A hybrid forecasting model combining EEMD,ARIMA and TEF is established.It is found that the prediction results will be significantly improved by properly processing the data before establishing a model and then establishing the model.The EEMD decomposition technology is very suitable for dealing with non-linear and non-stationary data.Therefore,this paper applies this technology to the hybrid prediction model and establishes the EEMD-ARIMA-TEF hybrid model.In order to illustrate the effectiveness of the hybrid model,the ARIMA-TEF hybrid model and the hybrid model combining empirical mode decomposition(EMD),phase space reconstruction(PSR)and extreme learning machine(ELM)(i.e.EMD-PSR-ELM)with excellent prediction ability are used as the benchmark models.The empirical analysis on several stock index data shows that the EEMD-ARIMA-TEF hybrid model is more close to the real value than other models in different prediction periods,which shows that the hybrid model has better performance.Finally,the latest variational mode decomposition(VMD),ARIMA model and TEF model is combined,and establishes a VMD-ARIMA-TEF hybrid prediction model.To verify the effectiveness of the proposed model,two hybrid models based on empirical mode decomposition(EMD),ARIMA model and support vector machine(SVM)are used as benchmark models.The real daily closing price data of the stock index are used for empirical analysis.The empirical results show that the proposed hybrid model has better prediction performance than other benchmark prediction models in terms of prediction error and direction accuracy.