spb/wp4_uqo

1% ==========================================================================2\section{Conclusion}3\label{sec:conclusion}4% ==========================================================================56This paper develops a comprehensive hedonic pricing model for Airbnb7listings across Quebec, Canada, combining OLS estimation, quantile8regression, machine learning, and spatial econometrics. Using 5,0009listings from Inside Airbnb---3,412 with valid prices---we estimate nine10nested specifications, quantile regressions at five percentiles, seven11machine learning models, and three spatial econometric models.1213\textit{Key findings.} The full hedonic model achieves $R^2 = 0.274$14with 50 covariates. Bedrooms command a premium of approximately 213\%15per unit, modulated by a concave log-bedrooms effect. The review score16carries a premium of approximately 11\% per 0.1-point increment. Local17listing density within 500m depresses prices ($-0.8$\% per listing),18while density at 1~km shows a positive agglomeration effect.1920Quantile regression reveals substantial heterogeneity: the entire-home21premium nearly doubles from $\tau = 0.10$ to $\tau = 0.75$, while22review scores matter more at lower quantiles. The pseudo-$R^2$ declines23from 0.100 to 0.020 across quantiles, confirming that luxury pricing24depends heavily on unobserved factors.2526XGBoost achieves the best out-of-sample $R^2$ (0.285, MAPE = 14.3\%),27modestly exceeding the OLS model. The similar performance of parametric28and non-parametric models suggests that the log-linear specification29captures the majority of systematic variation.3031\textit{Practical implications.} The price decomposition framework32allows hosts to benchmark their listing against comparable properties.33The spatial analysis highlights that immediate competitive density34depresses prices, suggesting differentiation on amenities and quality35in dense urban clusters. Quantile-specific estimates are useful for36revenue management: luxury-segment hosts should prioritize entire-home37exclusivity and premium location, while budget-segment hosts gain from38structural capacity and review credibility.3940\textit{Limitations.} The cross-sectional design limits causal41identification. The price variable reflects the listed price, which may42differ from the transaction price. The original dataset contains only4319 variables, requiring extensive feature engineering from listing names44and amenity counts. Certain optional dependencies (CatBoost, pygam,45SHAP, PySAL) were unavailable, necessitating fallback implementations.46The spatial models rely on a fixed $k$-nearest-neighbours weight matrix47($k = 5$); alternative spatial structures may yield different estimates.4849\textit{Future research.} Panel data methods exploiting within-listing50variation would strengthen causal identification. Booking-level51transaction data including cleaning fees and occupancy rates would52permit a more complete analysis. Pre-trained language model embeddings53applied to descriptions and reviews could extract richer text-based54features. Extending the framework to other Canadian provinces would test55external validity.56