Quantifying the Energy Transition: Exposure, Tone, and Equity Returns

Abstract

This thesis studies whether text extracted from firms' annual 10-K filings can measure energy-transition exposure and transition tone in a way that helps explain future stock returns in the broad U.S. equity market. Two firm-level signals are constructed from filing text: a Transition Language Score (TLS), which captures substantive transition-related disclosure net of generic ESG boilerplate, and a FinBERT-based sentiment measure, which captures how transition-relevant sentences are framed. Using the Russell 1000 over 2015–2024, the results show that TLS is a positive medium-horizon characteristic, with a pooled coefficient of 0.0260 at one year and 0.1392 at two years (both p<0.0001), while sentiment loads with the opposite sign, with a pooled coefficient of -0.0286 at one year (p=0.0001) and -0.0444 at two years (p=0.0243). When both signals enter the same pooled one-year regression, they remain jointly informative with opposite signs, with TLS at 0.0318 (p<0.0001) and sentiment at -0.0220 (p=0.0037), indicating that exposure and tone are distinct rather than redundant disclosure objects. Translating the characteristics into long-short factor portfolios produces a narrower result: the sentiment factor survives standard benchmarks most cleanly, with monthly alpha of 0.502% under FF3 and 0.493% under FF5 (both p<0.001), whereas the TLS factor remains positive but only borderline significant under FF5 at 0.206% (p=0.091). Overall, the evidence shows that legally binding corporate disclosure text contains economically meaningful transition information, and that the quantity and tone of that disclosure carry different asset-pricing implications in the Russell 1000.