Research on SSE 50 Index Prediction Based on Wavelet Denoising and BP Neural Network with Textual Information Integration

Stock market forecasting, as an important component of the economic system, has long been a significant research hotspot in financial studies. Traditional stock price prediction models rely on econometric methods, assuming linear markets, stable data and returns of normal distribution. However, the real market exhibits nonlinearity, non-steadiness, noise, thick-tail return distribution and investor irrationality, making the traditional model insufficient. This has prompted researchers to adopt machine learning and deep learning methods to simulate complex market behaviors and integrate insights from behavioral finance. In this study, wavelet denoising technology is adopted to preprocess the original indicators, effectively eliminating the interference of high-frequency noise and improving the data quality. Through the text analysis of the relevant financial news of the SSE 50 and CSI 300, the investor sentiment index is constructed. Subsequently, two BP neural network models are developed: an emotion enhancement model containing emotion indices and a benchmark model excluding emotion factors. Both of these models are designed to predict closing prices, logarithmic returns, and index differences. Empirical results show that the sentiment enhancement model exhibits statistically superior performance in all three prediction objectives, confirming the incremental explanatory power of investor sentiment in understanding stock price fluctuations.