High Frequency Stock Price Forecasting Research Based On CEEMDAN-SSA-LSTM

Forecasting the trend of stock price is not only closely watched by investors,but also provides an important basis for financial control and has a profound impact on national macro-control.Compared with low frequency data,high frequency data contains more effective information and can identify more patterns of change,but similarly,high frequency data can be noisier,and it is more difficult to extract and identify effective information.On the other hand,due to the increased complexity of the input data,it is more difficult to model,the training effect of the model becomes worse and less efficient,and it is difficult to extract the effective information from the data.Moreover,there are still problems such as difficulties in selecting parameters and adjusting parameters for the model parameters,which often result in poorer forecasting results and lower forecasting efficiency.To address the above issues,this thesis adopted a combination of theoretical analysis and empirical testing.Firstly,the Complete EEMD with Adaptive Noise(CEEMDAN)algorithm was used to decompose the original sequence,reduce the noise,and reorganize according to the sample entropy value of each component after decomposition,so as to reduce the impact of noise on the extracted information while retaining the information contained in the original sequence to the maximum extent.Thus,the impact of high frequency stock data with noise on training is reduced.Secondly,to address the problems of difficult hyperparameter adjustment and slow convergence of the LSTM model,the Sparrow Search Algorithm(SSA),which had fast convergence,high accuracy and good stability in the process of modeling and prediction was used to optimize the parameters of the LSTM model to improve the modeling and analysis efficiency and prediction accuracy of the model.Thirdly,CEEMDAN,SSA and LSTM were combined to construct a CEEMDAN-SSA-LSTM high-precision hybrid prediction model,while data pre-processing methods and model parameters were optimized,so as to improve the prediction capability and prediction efficiency of the model.In order to test the feasibility of the models constructed in this thesis and the effectiveness of processing high-frequency data,the LSTM model,CEEMDAN-LSTM model,SSA-LSTM model and the constructed CEEMDAN-SSA-LSTM model were applied to the 1-minute closing price and daily closing price of the CSI 300 index,respectively.The prediction quality of these eight models was compared usingR~2,MAPE,MAE,and RMSE as evaluation metrics.The final empirical results found that the CEEMDAN-SSA-LSTM model had the best prediction results,and the final results using high-frequency data were better than those using low-frequency data.Therefore,the model constructed by using CEEMDAN algorithm and SSA algorithm to optimize LSTM from two aspects simultaneously was very effective in forecasting stock prices based on high frequency,and it can significantly improve the prediction quality and fitting effect,which can provide effective help for financial risk management and investment decision.