We apply machine learning methods to granular environmental variables and test if there is a strong positive relation between environmental sustainability and future stock returns. A long-short portfolio that longs stocks with high forecasted returns and sells stocks with low forecasted returns earns large abnormal returns, and it performs better when climate concerns in the media are more intense. Further diagnosis shows that various dimensions of environmental sustainability help return predictions. High forecasted returns are associated primarily with strong environmental operational performance. The return prediction based on a customized transformer model is similar.

We find that expected return is related to trading volume positively among underpriced stocks but negatively among overpriced stocks. As such, trading volume amplifies mispricing. Our results are robust to alternative mispricing and trading volume measures, alternative portfolio formation methods, and controlling for variables that are known to have amplification effects on mispricing. By attributing trading volume to investor disagreement, we show that our results are consistent with the recent theoretical model of Atmaz and Basak (2018) in that investor disagreement predicts stock returns conditional on expectation bias.