We study whether causal narratives in retail investor discourse affect beliefs and asset prices. Using 2.35 million posts from Guba, China’s largest stock forum, we apply a large language model to extract narrative content and classify its implied economic channel. Narrative composition predicts future returns, volatility, and trading volume after controlling for sentiment, attention, and their interactions, while narrative dispersion forecasts higher volatility and turnover. To identify mechanism, we run an experiment with matched stock-forum stimuli that hold sentiment and keywords approximately constant. Causal narratives generate larger shifts in expected returns, confidence, and portfolio allocations, especially when firm-specific narratives are less dispersed.