Companies that emphasize quantitative AI work in regulatory filings earn higher long-term market valuations, reflected in Tobin’s Q and enterprise-value multiples; by contrast, AI rhetoric and sentiment do not generate short-term stock gains, suggesting investors treat AI as a durable intangible rather than a speculative signal.

Main Finding

Higher quantitative AI focus in firms’ 10‑K disclosures (frequency/proportion of AI-related terms) is associated with significantly higher long-run firm valuations—measured by Tobin’s Q and enterprise-value multiples—while equity-based ratios (e.g., P/E, P/B) show no consistent relation. By contrast, the qualitative tone (management sentiment) of AI language in MD&A does not produce meaningful short-term abnormal returns around 10‑K filings. Overall, capital markets appear to price AI as a long-term intangible strategic asset rather than as a driver of short-run returns from narrative sentiment.

Key Points

• AI focus (how much firms talk about AI in 10‑Ks) predicts higher long-run market valuation metrics (Tobin’s Q, EV multiples).
• AI sentiment (how positively or negatively management frames AI) does not produce statistically significant cumulative abnormal returns (CARs) in the short window around 10‑K release dates.
• The findings imply that markets reward durable, disclosure‑measured AI commitment rather than short-term PR/tone around AI.
• Robustness checks include lagged AI measures, exclusion of influential observations (e.g., Nvidia), and multiple model specifications; main result on valuation holds across these checks.
• Tests for sentiment effects include cross‑sentiment pairwise tests (Mann–Whitney, Welch’s t) and industry analyses; sentiment impacts are broadly null or economically small.

Data & Methods

• Sample: S&P 500 firms; annual 10‑K filings (documented coverage 2017–2024 in the project).
• Textual measures:
  • AI anchors / keyword list: curated AI-related keywords used to identify AI mentions (Keywords‑in‑Context, KWIC).
  • AI focus: normalized count/frequency of AI keywords in firm-year 10‑Ks (used as proxy for strategic AI commitment).
  • Management stance / sentiment: KWIC windows around AI anchors classified with a dictionary‑based approach (finance‑specific lexicon such as Loughran & McDonald and bespoke classification) into positive/neutral/negative tones.
• Valuation analysis:
  • Outcome variables: Tobin’s Q, enterprise-value multiples, and equity-based ratios (P/E, P/B).
  • Econometrics: OLS regressions (with standard controls—firm size, leverage, profitability, industry and year effects—and diagnostic testing for heteroskedasticity, multicollinearity, influential points).
  • Robustness: lagged AI measures, excluding outliers/influential firms, alternative specifications.
• Event study (short-term returns):
  • Event: 10‑K filing/release date.
  • Method: calculation of Abnormal Returns (ARs) and Cumulative Abnormal Returns (CARs) using a market model/CAPM baseline; statistical testing of CARs by sentiment category and across industries.
  • Validation: nonparametric tests (Mann–Whitney), pairwise comparisons, normality and heteroskedasticity checks.
• Key methodological note: both analyses use the same base textual extraction but operationalize different dimensions—intensity (how much) vs. tone/context (how).

Implications for AI Economics

• Valuation of AI as an intangible: Markets appear to incorporate disclosed AI emphasis into firm valuations through long-term value channels (Tobin’s Q, EV multiples). This supports the view that AI investments are perceived as durable, firm-specific intangible capital rather than ephemeral hype.
• Disclosure and signaling: Firms’ substantive disclosure of AI activity (frequency/extent) matters more to investors than upbeat managerial language; credible, documentable commitment may be more valuable than positive spin.
• Corporate strategy and reporting: Managers aiming to capture valuation benefits from AI should prioritize demonstrable commitment (investments, integration, scaled applications) and transparent reporting in regulated filings rather than relying solely on rhetorical framing.
• Implications for asset pricing and modeling: Intangible capital models and valuation frameworks should account for AI exposure as a persistent productivity/option-like driver of enterprise value; short-window event pricing models may understate the value of narrative content when it signals long-term investments.
• Policy and research directions: Results encourage further work to (i) causally identify AI investment → value channels (e.g., instrumenting for AI adoption); (ii) link disclosure metrics to realized productivity and cash‑flow outcomes; and (iii) extend analysis beyond large U.S. firms (cross‑country, SME samples) and to alternative AI proxies (patents, hiring, CapEx).

Limitations to keep in mind: textual proxies may misclassify mentions; endogeneity between valuation and disclosure (firms with higher valuations may disclose more); sample limited to large U.S. listed firms; and short-run null sentiment results do not rule out longer-horizon effects of tone on investor beliefs.