In this paper, we analyze the stock market effects of text-based ESG proxies. Based
on fine-tuned versions of the FinBERT language model for ESG and sentiment clas-
sification, we derive text-implied ESG metrics for firms based on information from
their annual reports. We show that the text-based ESG proxies are strongly related to
their Ratings-based counterparts. We also demonstrate that the composite textual ESG
metric as well as the environmental (E) subcomponent are priced in the cross-section
of US stock returns over the full sample. More precisely, stocks with higher textual E
and ESG scores tend to earn higher one-month ahead realized returns. However, this
pricing effect concentrates in a subsample which starts in November 2012. This aligns
with the results of the literature, showing that more recently greener stocks started to
outperform brown stocks.