What information shapes investors’ beliefs about the higher moments of stock returns? We construct subjective expectations for variance and skewness from analysts’ scenario-based price targets across over 180,000 reports from ten major brokerages (2007–2024). These expectations predict future realized moments incrementally over option-implied and past realized moments and generate profitable option strategies (annualized Sharpe ratios of 1.3 and 0.7). Hedge funds increase their option positions in stocks for which analyst expectations signal higher variance, consistent with sophisticated investors acting on this information. Because each report simultaneously provides beliefs about the three return moments, we can test classical risk-return trade-offs free from the mismatch between ex-ante expected values and ex-post outcomes: we find a positive subjective variance-return and a negative subjective skewness-return relationship. Standard market signals and firm characteristics explain only 34% and 7% of the variation in variance and skewness expectations. Using large language models to extract topics from scenario narratives and machine learning to identify their importance, we show that incorporating these narrative topics raises out-of-sample R² to 54% and 29%, with the information sets driving variance versus skewness being distinct.