spb/wp17_uqo

1% ============================================================================2\section{Related Literature and Hypotheses}3\label{sec:literature}4% ============================================================================56Our study sits at the intersection of five strands of research: textual analysis in finance, the7pricing of corporate disclosure, technology and innovation in asset pricing, investor attention and8narrative economics, and the nascent literature on the economics of artificial intelligence. Each9contributes a building block---a measurement tool, an object of study, a candidate risk factor, a10behavioral mechanism, and a comparison case---and our contribution is to combine them: we measure a11technology theme from firms' own words, price it in the cross-section, and show that its pricing turns12on precisely when a behavioral mechanism predicts it should. We take the strands in turn and close13with the hypotheses they jointly motivate.1415\subsection{Textual analysis and natural language processing in finance}16A now-large literature extracts economic signals from financial text. The foundational work of17\citet{tetlock2007} showed that the tone of a \textit{Wall Street Journal} column forecasts18aggregate returns, and \citet{tetlock2008} extended this to firm-level news. \citet{lm2011}19demonstrated that finance-specific sentiment dictionaries materially outperform general-purpose word20lists in 10-K analysis, launching a wave of dictionary-based studies and the widely used21Loughran--McDonald word lists; \citet{lm2016} survey this field. A second generation moved from22counting words to modeling them: \citet{gentzkow2019} review text-as-data methods in economics,23while \citet{kelly2019textregression} and \citet{bybee2024narrative} use machine-learning24representations of news to construct interpretable factors and a ``state of the economy'' from text.25Our paper belongs to the most recent, embedding-based generation of this work. The limitations of26keyword counting are acute for a theme like AI: a naive count cannot tell ``we do not currently use27artificial intelligence'' apart from ``we are an artificial-intelligence leader,'' it misses the28paraphrase of a firm that describes machine learning without ever writing ``AI,'' and it treats29boilerplate and substance identically. We instead measure the \emph{semantic similarity} of30disclosure to anchor concepts using a sentence transformer \citep{reimers2019sbert}, which maps31passages into a vector space where meaning, not surface form, governs proximity. This follows a32growing practice of using contextual embeddings to quantify disclosure content33\citep{cong2022textfactor}, and is closest in spirit to similarity-to-concept designs that score34documents against curated reference passages.3536\subsection{The pricing of corporate disclosure}37A parallel literature asks whether the \emph{content} and \emph{changes} of mandatory disclosure38predict returns. \citet{cohen2020lazy} show that firms that change the language of their 10-K39relative to the prior year (``lazy prices'' when they do not) experience predictable future returns40and operating outcomes. \citet{hoberg2010product} and \citet{hobergphillips2016} use 10-K product41descriptions to build text-based industry classifications and measures of product-market fluidity42that carry asset-pricing content. \citet{lopezlira2023risk} extracts risk exposures from 10-K43risk-factor sections. The general lesson is that the qualitative portions of filings contain44value-relevant information that markets process incompletely and with delay. We contribute a45focused, concept-level measure---AI engagement---and show that its pricing is strongly46state-dependent, a dimension the disclosure-pricing literature has largely left implicit.4748\subsection{Technology, innovation, and the cross-section of returns}49Asset pricing has long sought to connect technological change to risk and return.50\citet{hirshleifer2013innovative} show that innovative efficiency predicts returns;51\citet{kogan2017technological} build market-based measures of innovation from patents and link them52to growth and reallocation; \citet{koganpapanikolaou2014} show that investment-specific technology53shocks shape the cross-section. A more recent literature studies AI exposure directly:54\citet{babina2024aiinvestment} measure firm-level AI investment from job-posting and resume data and55document faster growth among AI-adopting firms; \citet{eisfeldt2023generative} construct a56firm-level measure of \emph{generative}-AI exposure from the task content of occupations and show57that a generative-AI factor was priced precisely in the weeks around ChatGPT's release.58\citet{acemoglu2020wrong} and \citet{webb2020} study the labor-market exposure of firms to AI and59automation. Our contribution relative to this work is to measure AI exposure from the firm's60\emph{own disclosure}---a self-reported, narrative signal available for the entire index at annual61frequency---and to characterize the full time profile of its pricing across a decade and a half, not62only the event window. Where \citet{eisfeldt2023generative} use occupation-task exposure, we use the63language firms choose, providing a complementary and highly granular measure.6465\subsection{Investor attention, salience, and narrative economics}66Why might value-relevant disclosure be priced only after a catalyst? The behavioral literature67offers a clear mechanism: limited investor attention. \citet{barberodean2008} show that retail68investors are drawn to attention-grabbing stocks; \citet{da2011} construct a direct attention proxy69from Google search and link it to prices; \citet{dellavigna2009} document under-reaction to70information released when attention is low; and \citet{hirshleifer2009driven} show that earnings news71is underweighted when investors are distracted. \citet{shiller2017narrative} argues that contagious72narratives drive major asset-price movements, and \citet{bakerwurgler2006} show that sentiment73differentially affects hard-to-value, speculative stocks. Our results provide a sharp illustration:74the information about firms' AI engagement existed in filings for years but was repriced abruptly75when a salient consumer event---ChatGPT---focused investor attention on AI. The asymmetry we document76(AI rewarded, ESG penalized after 2022) is itself consistent with a shift in the prevailing market77narrative.7879\subsection{ESG and climate disclosure as a comparison case}80Finally, our placebo design connects to the literature on the pricing of ESG and climate disclosure.81\citet{pastor2021sustainable} model how sustainability preferences depress the expected returns of82green assets in equilibrium, and \citet{pastor2022dissecting} show that green stocks' strong83\emph{realized} returns through 2021 reflected unexpected increases in climate concern rather than84high expected returns. \citet{bolton2021risk} document a carbon premium. Consistent with the85post-2021 reversal in this literature, we find that high-ESG and high-climate disclosure earned86\emph{negative} spreads after 2022, exactly when AI disclosure began to earn large positive87ones---underscoring that the post-2022 re-rating reflects a thematic rotation toward AI rather than a88generic disclosure effect.8990\subsection{Hypotheses}91\label{subsec:hypotheses}92Our analysis is organized around four hypotheses.93\begin{description}[leftmargin=2.4em,style=nextline]94\item[H1 (Unconditional pricing).] If AI exposure carries a stable risk premium or persistent95mispricing, then sorting on AI-disclosure intensity yields a positive and significant risk-adjusted96return spread over the full 2010--2026 sample.97\item[H2 (Regime dependence).] If instead AI exposure is repriced only once AI becomes salient, the98return spread is small before and large after the late-2022 generative-AI catalyst, with a99detectable structural break.100\item[H3 (AI-specificity).] If the post-2022 spread reflects a genuine AI repricing rather than a101mechanical or market-wide artifact, it appears for AI-related semantic factors but not for unrelated102factors (e.g.\ ESG, climate, macro risk).103\item[H4 (Cross-sectional, not within-firm).] If the market repriced \emph{which} firms are104AI-exposed rather than rewarding incremental wording, the effect is concentrated in the105cross-section and attenuates under firm and sector fixed effects.106\end{description}107\noindent The data, as we show, decisively favor H2--H4 over H1.108