Research On Futures Price Forecast Based On Rolling Time Window VMD-ISSA-LSTM Model

Predicting the trend of future prices is a.focal point for investors and financial regulators,as it plays a crucial role in national macroeconomic regulation.High-frequency data contains more information compared to low-frequency data,making it easier to identify changes,but it is also accompanied by higher noise levels,making it difficult to extract useful information.Moreover,the increaused complexity of input data makes modeling challenging,leading to poorer model training outcomes.Additionally,difficulties in parameter selection and setting during model construction may result in decreased prediction accuracy and efficiency.To address the aforementioned issues,this paper proposes a hybrid model based on rolling time window variational mode decomposition(VMD),an improved sparrow search algorithm(ISSA),and LSTM neural network for future price time series prediction and analysis.Firstly,a rolling time window VMD algorithm is used to decompose and denoise the original series,reducing the impact of noise on information extraction and preserving the information contained in the original series.This minimizes the impact of highfrequency data.and noise on training and addresses short-term prediction errors and data leakage in futures price decomposition.By continuously updating the data content of the time window,real-time prediction of futures prices can be achieved,preventing the use of future information.Secondly,considering the issues of hyperparameter setting and slow convergence in LSTM models,the LSTM model parameters are proposed as optimization objectives for an improved sparrow search algorithm with fast convergence,high accuracy,and good stability to enhance modeling efficiency and prediction accuracy.The improved sparrow algorithm combines Tent chaotic mappinghybrid sine-cosine algorithm,and Levy flight strategy to improve the population initialization method,discoverer position update formula,and follower position update formula in the sparrow search algorithm.Thirdly,by combining the rolling time window variational mode decomposition algorithm,the improved sparrow search algorithm,and the long short-term memory neural network,we have developed a high-precision VMD-ISSA-LSTM composite prediction model.This model optimizes data preprocessing methods and model parameters,thereby enhancing the predictive performance of the model.To test the feasibility of the model construction and the effectiveness of high-frequency data processing,this paper applies 16 models,including the BP model,RNN model,LSTM model,VMD-LSTM model,SSA-LSTM model.FASSA-LSTM model.ISSA-LSTM model,and VMD-ISSA-LSTM model,to 1-minuteand daily closing prices of No.1 soybeans futures,soybean meal futures:and soybean oil futures from the Dalian Commodity Exchange.The relative mean absolute error(MAPE),mean absolute error(MAE),root mean square error(RMSE),and goodness of fit(R2)are used as evaluation indicators to compare the predictive performance of the 16 models.The empirical results show that the VMD-ISSA-LSTM model has the best predictive performance,and the results obtained from high-frequency data are better than those obtained from low-frequency data.Therefore,the rolling time window-based VMD algorithm and ISSA algorithm simultaneously optimize the LSTM model from two aspects,making it very effective for predicting highfrequency futures prices and greatly improving the prediction quality,providing effective help for controlling financial risks and investment decisions.