Research On Prediction Method Of Time Series Based On Nonlinear Paradigm

In the context of big data,the complexity of data structure and its internal pattern and feature is increasing.How to scientifically and effectively extract,filter,and screen out valuable information from mass financial data to better provide support for socioeconomic decision-making activities has become one of the important trends in the era.As the core of the modern market economy,the financial industry is undergoing profound changes under the background of big data.How to effectively manage financial risks plays an important role in the healthy development of the financial industry.To improve the risk management capabilities of the financial industry,this paper takes three representative complex financial time series as research objects,and uses computational intelligence methods to carry out their predictive modeling research,which is conducive to reducing the risk of financial decision-making,improving the information efficiency of the financial market,and promoting the healthy and stable development of the financial market,so the research work has an important practical significance.However,due to the complex characteristics of financial time series,such as randomness,non-linearity,and long memory,making its accurate and effective predictions has always been a very challenging research work.For a long time,scholars at home and abroad have carried out a lot of research on financial time series prediction from the perspective of linear and nonlinear methods.The research results obtained have improved the prediction accuracy and stability of financial time series to a certain extent.However,most of research results ignore the importance of data preprocessing and feature selection,model parameter optimization,and deep learning techniques,which have largely restricted the further improvement of the prediction accuracy of financial time series.To improve the current research status and further improve the accuracy of financial time series prediction,this paper has carried out a series of research work combining data feature testing,data preprocessing,feature selection,neural network and computational intelligence theory and algorithms,and proposed three novel non-linear financial time series prediction models,including point prediction models and interval prediction models.Specifically,the point prediction model is used to infer the future value of financial time series,aiming to provide valuable decision-making information for macro financial market regulation and micro investment activities.The role of interval prediction is to characterize the uncertainty of financial time series,which plays an important role in risk management and control of financial markets.To verify the validity of the proposed prediction models,this paper applies the proposed models to the prediction problems of financial volatility,stock&fund price indexes,and crude oil price index,respectively.Therein,financial volatility is an effective representation of the price volatility of financial assets,which can measure and reflect the uncertainty of asset returns,precise prediction of which can help guide asset pricing and allocation and improve the scientific nature of the risk management of financial markets;the stock&fund price index is a systematic expression of the macro and micro financial markets,which can effectively reflect the changing trend of the stock market and the fund market.Effective prediction research on the stock&fund price index can provide valuable guidance information for financial investment decisions;the crude oil price index is a key indicator reflecting the supply relationship of the international crude oil market and energy-finance market,which plays an important role in guiding the oil trade of various countries.Accurate prediction of crude oil prices helps to scientifically regulate the international crude oil market and reduce the risk of crude oil trade.To sum up,the research work in this paper has both strong theoretical research value and high practical significance.In detail,the research ideas for the prediction of financial volatility,stock&fund price index,and crude oil price index are as follows.This paper first uses a series of testing methods to study the system characteristics of the above three types of financial time series and utilizes the BDS test and improved surrogate data method to validate the non-linear and non-linear characteristics of the three types of studied financial time series;the long-memory characteristics of the three types of financial time series are tested using the Hurst index;the recurrence plot and recurrence quantification analysis are utilized to analyze the recursiveness of the three types of financial time series.Then,according to the specific characteristics of the research data,this paper proposes the predictive solutions for the above three types of financial time series,as follows:for univariate financial volatility,this paper constructs a novel volatility prediction model based on data decomposition algorithms,clockwork recurrent neural networks,and the improved multi-objective gray wolf optimization algorithm;for the multivariable stock&fund price indexes,this paper proposes a two-stage feature selection model and a deep learning model;for the crude oil price index,this paper combines the linear prediction model,the nonlinear prediction model,and an improved gray wolf optimization algorithm to establish an effective ensemble interval prediction model.Finally,the performance of the three models mentioned above is evaluated and verified based on a series of statistical indicators.The research content of this paper is composed of eight parts:Chapter one introduces the research background and practical significance of this paper,explains the research problems and their solutions in this paper,builds the structural framework of the paper,and emphasizes the research innovation and defects of the paper;Chapter two sorts out the current research results on financial time series prediction,summarizes the advantages and disadvantages of the existing research results,and points out the corresponding improvement directions,verifying the basis and rationality of the topic selection of this paper;Chapter three systematically introduces the relevant theoretical basis of this paper in detail,mainly including the calculation logic of neural network models,the basic knowledge of recurrent neural networks,and the theory of computational intelligent algorithms;Chapter four analyzes the internal complex characteristics of the research data based on a series of test methods(including BDS test,improved surrogate data method,Hurst index,recurrence plot,and recurrence quantitation analysis);Chapter five builds a univariate financial volatility prediction model and studies its convergence and sensitivity;Chapter six constructs a novel stock&fund price index prediction model.In addition,Chapter six also studies the convergence and sensitivity of the model;Chapter seven establishes an interval prediction model of crude oil price index.The research focus of this chapter is to effectively quantify the uncertainty in crude oil prices and improve the scientificity of risk management in the energy-finance market;Chapter eight summarizes the research work and important research conclusions of this paper,and points out the future development direction of the research work.Summarizing the full paper,the important research conclusions and main points are listed as follows:(1)Based on a series of data feature testing methods,the paper found that the financial time series data involved in this paper has obvious non-linear characteristics,long memory and recurrence.Based on the conclusions above,this paper adopts a non-linear paradigm model(i.e.,recurrent neural network)to predict the future trend of financial time series.The experimental results show that compared with the linear model,the models based on the non-linear paradigm show obvious advantages for the prediction of complex financial time series.In addition,compared with single prediction models,hybrid prediction models have higher prediction accuracy.(2)This paper conducts the research of deterministic prediction modeling on univariate financial volatility and multivariable stock&fund price indexes.Specifically,for the univariate financial volatility,this paper constructs a volatility prediction model based on the data decomposition algorithm,the improved multi-objective gray wolf optimization algorithm,and the clockwork recurrent neural network.This study shows that the data decomposition algorithm and the improved grey wolf optimization algorithm can significantly improve the prediction accuracy of the clockwork recurrent neural network.For the multivariable stock&fund price index,this paper first proposes a two-stage feature selection model,which effectively combines the advantages of filtered feature selection and wrapper-based feature selection.This model was used to perform feature selection for the studied multivariable dataset.Then,based on the results of feature selection,this paper combines three recurrent neural networks to establish a deep learning model.The experimental results show that the proposed two-stage feature selection model can effectively identify the key features in the multivariate dataset,and can further improve the generalization and prediction ability of the proposed deep learning model.In addition,this paper establishes an error correction model for the proposed deep learning model.The experimental results show that the model can further improve the accuracy of the deep learning prediction model and has high feasibility.(3)Compared with the deterministic forecasting model,the uncertainty prediction model can effectively quantify the uncertainty in financial time series using the prediction interval,which helps to improve the ability and efficiency of financial risk management.Therefore,this paper proposes an uncertainty prediction model of the crude oil price index based on a linear forecasting model,a nonlinear forecasting model,and an improved grey wolf optimization algorithm.Theoretically,the uncertainty prediction model proposed in this paper can capture both the linear component and the non-linear component of the crude oil price index.The experimental evaluation results show that compared with the benchmark models,the uncertainty prediction model proposed in this paper can construct the prediction interval with higher comprehensive quality,which greatly helps to improve the information efficiency of the crude oil market and enhance the ability of risk management in crude oil market and energy-finance market.The main innovations of this paper include:(1)Due to the problems of slow convergence and "population stagnation" of the original grey wolf algorithm,this paper proposes an improved grey wolf optimization algorithm and improved multi-objective grey wolf optimization algorithm in combination with the adaptive Cuckoo search algorithm.and applies the above two algorithms to the financial volatility prediction,stock&fund price index,and the uncertainty modeling of crude oil price index,respectively.The experimental results verify the effectiveness of the algorithm.(2)For the multivariate stock&fund price index,this paper proposes a novel two-stage feature selection model.The empirical results show that the model can effectively identify key features in the multivariate stock&fund price index,which is largely conducive to improving generalization and accuracy of the prediction model.In addition,for the multivariate stock&fund price index,this paper builds a deep learning model that integrates three recurrent neural units.The study found that the model can more accurately predict the future development trend of stock&fund price index,as compared to the benchmark prediction models.(3)Combining the improved grey wolf optimization algorithm,this paper proposes an ensemble interval prediction model based on the linear prediction model and the non-linear prediction model.The study finds that compared with the single interval prediction model considered,the proposed ensemble interval prediction model can produce the prediction interval with higher quality,which can effectively quantify uncertainty in the trend of the crude oil price.The shortcomings of this paper include:(1)This paper proposes three nonlinear financial time series prediction models based on recurrent neural networks and grey wolf optimization algorithms.However,in practice,in addition to recurrent neural networks and grey wolf optimization algorithms,there are many other forms of non-linear prediction models and computational intelligent algorithms.Due to the limited space,these models and algorithms have not been taken into account.(2)This paper only conducts research on financial volatility,stock&fund price index,and crude oil price index.The applicability of the proposed model to other financial data needs further verification in future research work.