• MENTOR introduces a multi-agent framework integrating teacher–student iterative reasoning for event and narrative trend prediction.
• It outperforms existing baselines (StkFEP and SEP) in event prediction and industry ranking tasks on Chinese KOL and English financial news datasets.
• Portfolio-level backtests demonstrate that MENTOR's improved forecasts translate into practical gains in investment performance.
• The framework establishes a principled link between narrative dynamics and financial market outcomes through structured reasoning and multi-agent feedback.