spb/wp14_uqo

1% ============================================================================2\section{Conclusion}3\label{sec:conclusion}4% ============================================================================56We have estimated a hedonic price model for Quebec houses that augments the standard7set of structural, locational, and amenity attributes with two machine-generated8feature layers: structured specifications extracted from free-text descriptions by a9large language model, and objective visual attributes extracted from façade10photographs by a vision--language model. Schema-constrained decoding renders these11outputs as ordinary, reproducible econometric variables, obtained at a marginal cost12of a fraction of a cent per listing. The central message is methodological:13general-purpose AI models, disciplined by strict output schemas, function as a14low-cost feature-engineering technology that measurably expands the characteristics15set available to hedonic analysis.1617The gains are economically meaningful. The two AI layers add nearly four points of18in-sample $R^2$ to an already rich log--linear specification (0.717 to 0.755 with19living area observed), and they raise the out-of-sample accuracy of every estimator20in our ladder, from OLS to gradient boosting and an embedding-based residual neural21network; the best blend attains an out-of-sample $R^2$ of 0.779 with a median22absolute prediction error of 14.0\%. The façade-vision layer alone contributes the23largest increment of any feature block (2.7 points for gradient boosting, 5.8 for24OLS), confirming that image-derived quality is genuinely orthogonal to text and25structured fields. Crucially for hedonic practice, the extracted attributes carry26interpretable implicit prices---curb appeal, renovation level, luxury cues, and27exterior materials all price sensibly and survive municipality-clustered28inference---so the technology augments the classical regression rather than29replacing it with a black box.3031Three extensions are natural. First, the same schema-constrained approach scales32from a single façade image to a small panel of interior photographs (kitchen,33bathroom, living space), which our design already accommodates. Second, the34DuProprio coordinates permit continuous spatial features---distance to the urban35core, to water, or to amenities---that would refine the geography beyond discrete36city effects. Third, and most important for external validity, linking listings to37registry transaction prices would establish how much of the AI-measured quality38premium survives bargaining. For appraisal, automated valuation, and property-tax39assessment---applications that consume hedonic models at scale---the practical40implication is immediate: the unstructured majority of a listing's information41content is no longer out of reach.42