The Prediction Research Of Long-memory Financial Chaotic Sequence Based On Multiple Scales

With the rapid development of modern financial markets,rapid changes in stock price volatility.How to accurately grasp of real-time information on financial fluctuations is one of the core issues in the financial field.For the financial chaotic sequences,this paper introduces ensemble empirical mode decomposition,grasping the law of price changes in financial market accurately,revealing the inherent development of China's financial market changes,and establishes prediction model,having important theoretical and practical significance.The object of study is Shanghai and Shenzhen 300 index day closing price;the main research is as follows:1)Firstly,utilizing the ensemble empirical mode decomposition technology,we can decompose the Shanghai and Shenzhen 300 Index into 10 IMFs and a trend item.Taking usage of the method which proposed by Zhang,we can refactor the IMFs into the short-term fluctuations?medium-long term fluctuations and long-term fluctuations.Then,we make the single fractal analysis on each sequence that we find sequences have own fractal features.2)Multifractal analysis of each sequence performed,we can explore the fractal characteristics of each sequence.The more obvious the fractal feature of the sequence has,the more complex the fluctuation of the sequence.Then,multifractal comparison is performed by randomizing the sequence of each sequence that we can study the causes of multiple fractal features of different sequences.This will help us to further understand the inherent law of sequence fluctuations.3)We study the sequence with chaotic characteristics to determine the chaotic characteristics of each sequence.Firstly,we perform the nonlinear and deterministic test of the sequence with BDS test and recursive graph.If the sequence derives from non-linear and deterministic systems,we can make the phase space reconstruction of the sequence.Here,we determine the embedding delay and embedding dimension utilizing C-C algorithm and compute associative dimension maximum Lyapunov exponent utilizing G-P algorithm and wolf algorithm to identify chaotic characteristics.4)This paper will propose the forecasting model based on the SVR.Firstly,the sequences were normalized separately;secondly,utilizing C-C algorithm to determine the embedding delay and embedding dimension;thirdly,we can make the phase space reconstruction of the sequence,we use SVR to train the sample data based on the characteristics of the reconstructed data.Here,we use the particle swarm optimization to select the parameter of SVR.Finally,we output the predicted value.Utilizing RMSE,MAE,and MAPE as indicators to measure the effectiveness of predictions,we find that the prediction accuracy of the proposed model is superior to other models.