Quantifying the Housing Price Premium of Walkable Urban Nature in Shanghai: Evidence from Explainable Boosting Models

Abstract How much do walkable distances to urban nature capitalize into housing prices once core built-environment factors are controlled? We study Shanghai listings from 2023--2025 and fit transparent hedonic models with Explainable Boosting Machines (EBMs)—generalized additive learners that recover smooth univariate effects and a small number of theory-guided pairwise interactions. The target is log price per m\((^2)\); controls include floor-area ratio, greening ratio, property management fee, surface parking fee, and completion year. Distance-based regressors capture straight-line proximity to the nearest park, green open space, and blue amenity, plus count measures. Model selection uses 3-fold cross-validation and final fits bag six bootstraps for stability; all metrics and effects are computed on held-out test sets. Across years the EBM attains R\((^2\!≈\!0.67)\)--\((0.70)\) with RMSE \((≈)\) 19--21k RMB/m\((^2)\). Average treatment effect (ATE) calculations indicate that being 1 km closer to a park is associated with a +2.5%--+3.4% price premium; proximity to blue amenities yields +1.2%--+1.6%; while green space shows small negative semi-elasticities (about \((-)\)1.5%--\((-)\)1.7%). Permutation importance and partial-\((R^2)\) ablations confirm that park proximity carries the largest incremental explanatory power. The resulting effect shapes and interactions are directly interpretable and reproducible, supporting planning decisions around amenity provision and walkable access.