spb/wp17_uqo

1% ============================================================================2\section{Introduction}3\label{sec:introduction}4% ============================================================================56The arrival of generative artificial intelligence has been the defining corporate-narrative shift7of the 2020s. In the span of barely three years following the public launch of ChatGPT in late82022, the market capitalization of AI-linked firms swelled by trillions of dollars, ``AI'' vaulted9past every other theme in frequency on earnings calls, and the language of corporate disclosure10filled with references to large language models, generative tools, intelligent automation, and the11compute infrastructure that powers them. A technology that had been the preserve of research labs12and back-office analytics suddenly became the organizing story of public equity markets. This paper13asks a question that is simple to state but surprisingly demanding to answer well: did the14\emph{language} of AI in firms' mandatory disclosures carry information about their subsequent stock15returns---and if so, when did that information begin to be priced, and through what economic channel?1617Answering the question convincingly requires three ingredients that have only recently become jointly18feasible. The first is a large, machine-readable corpus of corporate disclosure. The second is a19measurement technology that captures the \emph{meaning} of text rather than the mere frequency of20keywords---one that recognizes that a firm describing how it ``deploys machine-learning models to21optimize its supply chain'' is talking about AI even when the three-letter acronym never appears. The22third is a clean, look-ahead-free mapping from disclosure to tradeable returns. We bring all three to23bear. Our laboratory is the universe of S\&P~500 constituents over fiscal years 2010--2026. Our raw24material is the full text of their annual reports (Form~10-K), parsed section by section. Our25measurement engine is the EDGAR Research Database, which embeds every text chunk with the26\texttt{all-MiniLM-L6-v2} sentence transformer and scores it by cosine similarity to a set of27carefully written ``anchor'' passages that define eight economic concepts. Three of these anchors28isolate distinct facets of AI engagement, each corresponding to a different way the technology bears29on firm value: \emph{internal AI adoption} (using AI to run the business more productively),30\emph{AI infrastructure exposure} (capital and revenue tied to the compute build-out), and \emph{AI31product-disruption risk} (the danger that AI commoditizes or displaces the firm's own offerings). The32result is a continuous, comparable, firm-year measure of how intensively---and in what sense---each33firm engages with AI in its own words, available for the entire index and spanning a decade and a34half.3536We pair these semantic scores with monthly adjusted prices for all 500 tickers from Yahoo Finance37and with the Fama--French five factors plus momentum. To avoid look-ahead bias we adopt the standard38portfolio-formation convention. Because a 10-K covering fiscal year $Y$ is filed and public by early39in year $Y{+}1$, we form signals at the end of June of $Y{+}1$ and measure the forward twelve-month40return from July of $Y{+}1$ to June of $Y{+}2$. The roughly three-month cushion between the typical41filing date and portfolio formation guarantees that the signal is in the public information set42before the holding period begins, even for late filers. This yields 6{,}852 firm-years with a43strictly tradeable forward return, spanning formation years 2010 through 2024.4445Our headline finding is that the pricing of AI language is \emph{not} a stable factor premium but a46regime-dependent re-rating that switched on with the generative-AI boom. For most of the 2010s,47sorting S\&P~500 firms on the intensity of their AI disclosure produced only a faint return pattern:48the equal-weighted spread between the highest and lowest quintiles averaged about $+2.4$ percentage49points per year---economically modest and statistically fragile---and a risk-adjusted long--short50portfolio earned an alpha that was both small and statistically insignificant. The information was51there in the filings, but the market was not paying for it. After the launch of ChatGPT in November522022, the very same sort produced a spread of $+14.6$ percentage points per year, a more than sixfold53jump. This is not a subtle tilt buried in a long-run average; it is a sharp, datable change in how54the cross-section prices a fixed characteristic. The re-rating is visible from three independent55angles: it appears in every post-2022 formation cohort taken on its own, in a cumulative long--short56wealth curve that drifts sideways for a decade and then inflects sharply upward at the end of 2022,57and in a year-by-year Fama--MacBeth slope that hovers near zero through 2021 before turning large and58positive.5960A natural worry is that this pattern simply reflects the well-known rally of a handful of technology61mega-caps---that we have rediscovered the Magnificent~7 dressed in textual clothing. We confront this62concern head-on and reject it. The post-2022 spread is $+14.1$ percentage points \tstat{1.9} when the63``Magnificent~7'' are excluded and $+14.5$ points when the ten largest firms by market capitalization64are dropped: removing the very names that dominate the headline rally barely moves the estimate, so65the effect is broad-based across the index rather than an artifact of Nvidia, Microsoft, or their66peers. It is equally robust to how the signal is built and measured---to using the simple mean, the67section-weighted mean, or the top-5\% tail of similarity scores; to six-, twelve-, and twenty-four-month68holding horizons; and to replacing the three correlated AI signals with their first principal69component, a single index that alone explains 87\% of their common variation.7071Establishing that the spread is large and well-timed is only half the task; we then deploy three72complementary tests to pin down what the result is---and, equally important, what it is not. The73first is a \emph{placebo} exercise that re-runs the entire research design on the five non-AI semantic74factors in the database (ESG, climate, cybersecurity, automation/labor, and macro risk). The logic is75simple: if the post-2022 jump merely reflected a generic ``more disclosure earns more'' phenomenon, or76a mechanical artifact of our embedding-and-sorting pipeline, every factor should display the same break.77They do not. The post-2022 spread is large and positive for the AI factors, essentially zero for78macro-risk language, and---most tellingly---\emph{negative} for ESG and climate disclosure, meaning79high-ESG firms \emph{under}performed after 2022. The post-ChatGPT re-rating is specific to AI-related80content, not a property of disclosure intensity in general. The second test is a market-model81\emph{event study} around the 30~November 2022 ChatGPT launch. Firms in the top tercile of pre-event82AI disclosure---measured from their FY2021 10-K, the last report filed before anyone could have83anticipated the event---earned cumulative abnormal returns $+10.1$ percentage points higher than the84bottom tercile over the following year \tstat{1.97} and $+14.6$ points higher over eighteen months.85Because abnormal returns are purged of each firm's market-beta exposure, this is a clean86cross-sectional response to an unanticipated narrative shock rather than a reflection of high-AI firms87simply being higher-beta in a recovering market. The third test uses two-way fixed effects and88difference-in-differences to localize the mechanism within the cross-section. The premium is fully89absorbed by firm fixed effects and roughly halved by sector fixed effects, which tells us that the90durable, priced signal is the persistent \emph{level} of a firm's (and its sector's) AI exposure91relative to peers, not the year-to-year wiggle in its wording.9293Taken together, the four pieces of evidence---the datable break, the AI-specific placebo, the94event-study timing, and the fixed-effects anatomy---converge on a single interpretation: an95attention- and narrative-driven repricing. The textual evidence of firms' AI engagement accumulated96quietly in their filings for a decade and was barely priced. A salient catalyst---the consumer97breakthrough of generative AI---then drew investor attention to that pre-existing information, and98capital flowed toward the firms and sectors whose disclosures had already revealed the greatest AI99exposure. What looks like new information was in large part old information newly attended to. This is100the asset-pricing signature of limited attention and narrative economics, observed here not through an101indirect sentiment proxy but directly in the semantic content of mandatory disclosure.102103\paragraph{Contribution.} We make four contributions. \emph{First}, methodologically, we104demonstrate the value of embedding-based semantic measurement of disclosure for asset pricing: a105cosine-similarity-to-anchor design captures economically meaningful, multi-faceted exposure106(adoption vs.\ infrastructure vs.\ risk) that keyword counts cannot, and it does so at scale across107the entire S\&P~500 and a decade and a half of filings. \emph{Second}, substantively, we provide108text-based, look-ahead-free evidence that the cross-sectional pricing of AI exposure is a post-2022109regime phenomenon rather than a long-standing premium. \emph{Third}, we discipline the finding with110a placebo across non-AI factors and an event study, distinguishing a genuine AI-specific repricing111from a mechanical or market-wide artifact. \emph{Fourth}, we document the \emph{anatomy} of the112effect---its concentration in smaller firms, its operation substantially through a sector channel,113and its disappearance under firm fixed effects---which sharpens the economic interpretation and114warns against naive trading strategies.115116The remainder of the paper is organized as follows. Section~\ref{sec:literature} situates the study117in five literatures and states our hypotheses. Section~\ref{sec:data} describes the data and the118construction of the semantic measures. Section~\ref{sec:methodology} lays out the empirical119methodology. Section~\ref{sec:results} presents the main results, and Section~\ref{sec:robustness}120the robustness and identification tests. Section~\ref{sec:discussion} discusses mechanisms and121limitations, and Section~\ref{sec:conclusion} concludes.122