The influence mechanisms of news sentiment indicators from AI’s large language models on stock market Index returns: a prospect theory lens

Based on prospect theory, this paper discusses the influence mechanism of news sentiment indicators of AI large language models on stock market index returns. Taking 1 January 2024 to 21 January 2026 as the research window, taking 29,077 Lujiazui financial breakfast news from the Wind database as a sample, the news sentiment index is constructed through the DeepSeek large language model, and the value function, weight function and total subjective value variables are constructed in combination with the prospect theory. The results show that the total subjective value has a positive and significant impact on the return of the Shanghai Composite Index, and the weight function has a negative significant impact, and the model has the problems of heteroscedasticity, multiple collinearity and residual non-normality, and the conclusion is reliable by the robustness test. This study enriches the intersection of behavioural finance and artificial intelligence, and the constructed sentiment indicators can provide reference for investor decision-making and market supervision, and points out the limitations and future expansion directions of research.