Hedonic Models of Real Estate Prices: GAM Models; Environmental and Sex-Offender-Proximity Factors

Abstract

We investigate the use of a P-spline generalized additive hedonic model (GAM) for real estate prices in large U.S. cities, contrasting their predictive efficiency against commonly used linear and polynomial-based generalized linear models (GLM). Using intrinsic and extrinsic factors available from Redfin, we show that the GAM model is capable of describing 84% to 92% of the variance in the expected ln(sales price), based upon 2021 data. In contrast, a strictly linear GLM accounted for 65% to 78% of the variance, while polynomial-based GLMs accounted for 82% to 88%. As climate change is becoming increasingly important, we utilized the GAM model to examine the significance of environmental factors in two urban centers on the northwest coast. While the results indicate city-dependent differences in the significance of environmental factors, we find that inclusion of the environmental factors increases the adjusted R2 of the GAM model by less than 1%. Thirdly, our results indicate that the importance of sex offender residence proximity as a pricing factor is strongly influenced by state sex offender residence regulations.