spb/wp17_uqo

1% ============================================================================2\section{Conclusion}3\label{sec:conclusion}4% ============================================================================56Using a semantic, embedding-based measure of artificial-intelligence language in the SEC current7and quarterly reports of the 500 current S\&P~500 constituents, linked to look-ahead-free forward8returns and standard factor data over fifteen formation years, we have examined whether and when9AI language was priced in the cross-section of returns. For the formation years 2010--2021 the10answer is essentially no: the top-minus-bottom quintile spread averaged $+2.5$ percentage points11per year, the six-factor alpha of the long--short portfolio was 1\% per year and insignificant, and12the annual cross-sectional slopes were of mixed sign. For the formation years 2022--2024 the13average spread was $+14.6$ points per year, the monthly long--short earned a Sharpe ratio of 1.32,14firms with high pre-launch AI language earned $+10$ points of additional abnormal return in the15year after ChatGPT, and sustainability language run through the same pipeline was penalized.1617The contribution we would emphasize, however, is the anatomy of that post-2022 evidence. The18average spread rests on three cohorts and on one of them in particular: the 2024 cohort earned19$+38$ points, the 2022 cohort $+9$, and the 2023 cohort $-4$, and two pre-launch cohorts20exceeded the 2022 value. Half of the pooled post-period spread comes from a few memory and storage21manufacturers whose prices multiplied in 2025--2026. The long--short return loads heavily on a22technology-sector factor that absorbs most of its post-period alpha, sector controls halve the23cross-sectional slope and leave it insignificant, firm fixed effects absorb the effect entirely,24and a pre-determined exposure measure earned significant premia in 2015 and 2016 as well as in252022 and 2024. The event-study difference is significant at one year under tests that allow for26cross-sectional dependence, but not at sixty days. The results are therefore consistent with a27sector-level re-rating of persistent AI exposure during the AI investment cycle, not with a stable28premium on AI language nor with a sharp attention shock to firm-specific disclosure.2930Two messages emerge. The economic message is that the pricing of a disclosure characteristic can31lie dormant for a decade and then appear, and that when it appears in a short window it is easy to32mistake a sector cycle, or a single cohort, for a new factor; the discipline of cohort-by-cohort,33leave-one-out, winsorized, sector-adjusted, and dependence-robust inference is what separates the34two. The methodological message is that embedding-based measurement of disclosure content is a35productive tool, but one whose scores inherit the composition of the corpus on which they are36computed; documenting that composition, and re-estimating on homogeneous subsets of it, is part37of the method. Several next steps follow. The most important is to score the Form~10-K annual38reports, which the database has downloaded but not parsed, and to re-run the analysis on the39annual narrative that the original question concerned. The second is to reconstruct a40point-in-time index membership. The third is to add post-2024 cohorts as they become available,41which is the only way to learn whether the 2024 result was the beginning of a regime or an42episode. Beyond these, extending the panel to mid- and small-cap firms, where attention frictions43are larger, and incorporating earnings-call transcripts and patent text as complementary measures44of AI exposure, would sharpen the distinction between firm-level and sector-level repricing that45the present data cannot draw.46