Exploring the nonlinear impact of air pollution on housing prices: A machine learning approach

Air pollution has profoundly impacted residents’ lifestyles as well as their willingness to pay for real estate. Exploring the relationship between air pollution and housing prices has become increasingly prominent. Current research on housing prices mainly uses the hedonic pricing model and the spatial econometric model, which are both linear methods. However, it is difficult to use these methods to model the nonlinear relationship between housing price and its determinants. In addition, most of the existing studies neglect the effects of multiple pollutants on housing prices. To fill these gaps, this study uses a machine learning approach, the gradient boosting decision tree (GBDT) model to analyze the nonlinear impacts of air pollution and the built environment on housing prices in Shanghai. The experimental results show that the GBDT can better fit the nonlinear relationship between housing prices and various explanatory variables compared with traditional linear models. Furthermore, the relative importance rankings of the built environment and air pollution variables are analyzed based on the GBDT model. It indicates that built environment variables contribute 97.21% of the influences on housing prices, whereas the contribution of air pollution variables is 2.79%. Although the impact of air pollution is relatively small, the marginal willingness of residents to pay for clean air is significant. With an improvement of 1 $\mu$g/m${}^3$ in the average concentrations of PM_2.5 and NO${}_2$, the average housing price increases by 155.93 Yuan/m${}^2$ and 278.03 Yuan/m${}^2$, respectively. Therefore, this study can improve our understanding of the nonlinear impact of air pollution on housing prices and provide a basis for formulating and revising policies related to housing prices.