spb/wp14_uqo

1% ============================================================================2\section{Introduction}3\label{sec:introduction}4% ============================================================================56The hedonic price framework decomposes the market value of a heterogeneous good7into the implicit prices of its characteristics \citep{lancaster1966new,8rosen1974hedonic}. In housing markets it remains the workhorse tool for appraisal,9index construction, property-tax assessment, and policy analysis. Yet the empirical10performance of hedonic models is bounded by the richness of the characteristics an11analyst can measure. A large share of what determines a home's value---its state of12repair, the quality of its finishes, its curb appeal, the character of its13street---is recorded not in structured fields but in the free-text description and14the photographs that accompany each listing. These modalities have historically been15discarded because they were costly to encode at scale, leaving hedonic estimates16exposed to omitted-quality bias and hedonic predictions short of their potential17accuracy.1819This paper treats general-purpose artificial-intelligence models as a20\emph{feature-engineering technology} for hedonic analysis, and quantifies the21explanatory power that technology adds. We assemble a cross-section of 19{,}12322single-family houses and townhouses listed for sale in Quebec, pooling the23brokerage-listed MLS inventory from Realtor.ca with the for-sale-by-owner inventory24from DuProprio. On top of standard structural, locational, and amenity attributes,25we build two machine-generated feature layers. First, a large language model (LLM)26reads each free-text description and returns a fixed set of property27specifications---state of repair, renovation of the kitchen and bathroom, flooring28and countertop materials, heating system, view, and income potential---under a29strict JSON schema that guarantees valid, reproducible output. Second, a30vision--language model inspects the façade photograph of each property and returns31objective visual attributes: overall condition, curb appeal, renovation level,32exterior materials, landscaping, setting, and a luxury indicator. Because both33extractions are schema-constrained, deterministic, and blind to the listing price,34their outputs behave as ordinary tabulated regressors rather than as noisy text.3536Our contribution is threefold. First, we show that AI-based feature engineering37delivers economically meaningful gains in explanatory power over an already rich38hedonic specification: the two machine-generated layers raise in-sample $R^2$ from390.717 to 0.755 (nearly four points) in our preferred specification, and raise the40out-of-sample $R^2$ of every estimator we consider, with the façade-vision layer41alone contributing the largest single increment of any feature block (2.7 points for42gradient boosting, 5.8 points for OLS). Second, we show that the extracted43attributes carry interpretable implicit prices---a one-point increase in visual curb44appeal is associated with a 2.3\% higher asking price, a visible luxury cue with455.2\%, brick or stone façades with premia of 3--4\% relative to vinyl siding, and a46renovated kitchen (read from the text) with 2.6\%---so the approach augments rather47than replaces the classical hedonic regression. Third, we document a complete,48reproducible pipeline from raw scraped listings to an analysis-ready dataset,49including rules-based outlier cleaning, unit harmonization that reconstructs living50area from room-level dimensions, and the schema-constrained extraction protocol,51with a total AI cost of about \$60 for the full sample---roughly a third of a cent52per listing.5354Our main findings can be summarized as follows. A log--linear specification with 19555municipality fixed effects, property-type fixed effects, and56municipality-clustered standard errors explains 75.5\% of the variation in log57asking prices when living area is observed, and 72.5\% on the full sample when it is58not. The estimated implicit prices are economically sensible: the living-area59elasticity is 0.36, each additional bathroom is associated with a 7.6\% premium, and60the age profile traces the familiar depreciation-then-heritage pattern. The61municipality effects span a wide range---from roughly $+231$\% in Westmount to about62$-30$\% in remote municipalities---and order the province's markets as expected.63Benchmarking a ladder of estimators on the identical feature set, an ensemble blend64of a robust embedding-based neural network and gradient boosting attains an65out-of-sample $R^2$ of 0.779 with a median absolute prediction error of 14.0\%. We66interpret all estimates as \emph{predictive} associations rather than causal67effects, since listing content is chosen by sellers and correlates with unobserved68determinants of value, and we discuss the asking-price nature of the dependent69variable at length in Section~\ref{sec:robustness}.7071Beyond the specific Quebec application, the exercise speaks to a general measurement72problem in housing economics. Text- and image-derived quality measures have until73now required bespoke supervised models trained on labeled data74\citep{nowak2017textual,shen2021information,johnson2020curb}. General-purpose75foundation models with schema-constrained decoding lower the fixed cost of that76measurement step to nearly zero, produce attributes that are human-readable rather77than latent, and can be audited field by field. Our results suggest that this78technology moves the effective characteristics frontier of hedonic analysis by79enough to matter for appraisal, automated valuation, and assessment applications.8081The remainder of the paper is organized as follows. Section~\ref{sec:literature}82reviews the related literature. Section~\ref{sec:data} describes the data sources83and cleaning. Section~\ref{sec:methodology} presents the feature-engineering84pipeline and the econometric specification. Section~\ref{sec:results} reports the85main results. Section~\ref{sec:benchmark} maps the predictive frontier across86estimators. Section~\ref{sec:robustness} discusses robustness and limitations.87Section~\ref{sec:conclusion} concludes.88