Financial Time Series Forecasting Based on Sliding Window - Variational Mode Decomposition Deep Learning Model

In view of the time series characteristics of financial data, a data processing method based on sliding window - variational mode decomposition (SW-VMD) is constructed to decompose and reconstruct stock closing prices and return rate time series, transforming nonlinear and non-stationary data sequences into linear and stationary ones. The processed data are then used as input for long short-term memory (LSTM) neural networks to predict future stock closing prices and returns. Empirical analysis adopts trend prediction accuracy as the model evaluation index, thereby reflecting the model’s ability to predict stock price and return fluctuations. The results show that, compared with models without data decomposition, the LSTM model using data decomposition achieves significant optimization in trend prediction accuracy.