spb/wp14_uqo

1% ============================================================================2\section{Related Literature}3\label{sec:literature}4% ============================================================================56This paper connects four strands of research.78The first is the classical \textbf{hedonic pricing} literature. Building on9\citet{lancaster1966new}, \citet{rosen1974hedonic} formalized the market for10differentiated goods as a market for their underlying characteristics, providing the11theoretical foundation for interpreting regression coefficients as implicit prices.12The empirical housing literature that followed is vast; \citet{sirmans2005composition}13survey the characteristics most commonly included in hedonic house-price models and14document substantial heterogeneity in specification, \citet{malpezzi2003hedonic}15provides a critical review of functional-form and identification choices, and16\citet{hill2013hedonic} surveys the use of hedonic methods for house-price17measurement. A recurring theme in this literature is omitted-variable bias from18unmeasured quality: the state of repair, the standard of finishes, and the visual19quality of a dwelling are priced by the market but rarely recorded in the structured20fields available to the econometrician. Our contribution speaks directly to this21measurement problem. Our log--linear specification, with logged continuous attributes22and a dependent variable in logs, follows the mainstream of this literature.2324The second strand concerns \textbf{spatial structure, functional form, and25prediction}. \citet{can1992spatial} and \citet{anselin1988spatial} emphasize that26unobserved locational amenities induce spatial dependence that, if ignored, biases27coefficient estimates; we absorb the first-order component of this problem through a28rich set of municipality fixed effects and cluster inference at the same level.29On the prediction side, \citet{bin2004prediction} compares parametric and30semi-parametric hedonic predictors, \citet{mullainathan2017machine} frame31machine-learning methods as estimators of the conditional expectation32$\mathbb{E}[P\mid\mathbf{z}]$ well suited to prediction tasks, and33\citet{kok2017bigdata} document the rise of automated valuation models in industry.34This literature generally finds that flexible learners beat linear hedonics on35accuracy at some cost in interpretability; our estimator ladder36(Section~\ref{sec:benchmark}) reproduces that pattern on a common feature set, and37our focus is on how much of the gap is closed not by a better \emph{estimator} but by38better \emph{measurement}.3940The third strand is the use of \textbf{unstructured listing content}---text and41images---in real-estate valuation. \citet{gentzkow2019text} survey text-as-data42methods in economics. In real estate specifically, \citet{nowak2017textual} show that43words in MLS remarks predict transaction prices and reduce hedonic pricing errors,44and \citet{shen2021information} extract quality signals from property descriptions45with supervised machine learning, finding that the text carries information about46unobserved quality that is capitalized into prices. On the image side,47\citet{ahmed2016house} combine visual and textual features for price estimation;48\citet{poursaeed2018vision} show that learned measures of interior and exterior49quality carry independent predictive power; \citet{glaeser2018vision} use street-level50imagery of Boston houses to measure exterior quality and study its interaction with51incentives; \citet{johnson2020curb} construct a curb-appeal index from Google Street52View images and estimate its capitalization into sale prices; and53\citet{lindenthal2021architecture} classify architectural styles from photographs and54document style premia. Relative to this work, our approach differs in the55\emph{technology of measurement}: rather than training bespoke supervised models on56labeled data---which requires labels, task-specific engineering, and yields opaque57scores---we use general-purpose large language and vision--language models with58schema-constrained decoding to produce a fixed set of interpretable, human-readable59attributes (e.g., ``renovated kitchen,'' ``brick façade,'' a 1--10 curb-appeal score)60at a marginal cost of a fraction of a cent per listing, and we embed those attributes61in a transparent hedonic regression whose coefficients can be read as implicit62prices. The models never observe the price, so the extracted attributes cannot63mechanically encode the outcome.6465The fourth strand is the \textbf{microstructure of listing markets}, which frames two66institutional features of our data. Our sample pools a brokered MLS platform67(Realtor.ca) with a for-sale-by-owner platform (DuProprio); \citet{hendel2009mls}68compare exactly these two market forms and document differences in seller69composition, marketing, and outcomes, motivating our platform indicator and70platform-specific robustness checks. And because we observe asking rather than71transaction prices, the interpretation of our dependent variable rests on the72list-price literature: the asking price is a strategic choice that anchors search and73bargaining \citep{horowitz1992list,knight2002listing,han2015microstructure}. We74return to this limitation in Section~\ref{sec:robustness}.75