Estimating the effects of land cover on housing prices with Bayesian model averaging
Investigator: Andrew Bjorn
Policies to protect, maintain and expand green spaces in urban areas can have a number of objectives that are important to improving urban sustainability. These can include such goals as maintaining urban and suburban habitat; supporting recreation opportunities; preserving desirable environmental services such as erosion control, shading, carbon sequestration and acoustic isolation; and providing aesthetic benefits. In regions where these spaces are relatively scarce, unevenly distributed and desirable, renters and home buyers in the housing market may be willing to pay more to live in places where they can enjoy the higher levels of local environmental quality that they provide. Therefore, decisions that influence green space can have important economic and financial impacts, including increases in property values and property tax revenues, changes in development and redevelopment decisions, gentrification or other demographic sorting effects, and so on. Understanding these economic impacts can be an important consideration in developing policies that influence these spaces.
Hedonic price modeling has been established in the literature as one approach to recovering these price effects, by assessing the price effects of individual components of the housing good This approach has been widely used for this application and results from existing studies have suggested that there is a small but significant effect of higher amounts of green space. However, these models are often overspecified, even though there is a high level of uncertainty about which variables should be included in the “correct” model. Certain variables used in the model may also be highly correlated with one another, distorting the estimates of many of the individual effects. In addition, the results may be influenced by the presence of outliers within the data set, which can be a significant problem if the dataset includes a number of properties that do not represent proper “arms’ length” sales in the market.
This project presents robust estimations for the effects of land cover on single-family housing prices in the Seattle, Washington area. Land cover metrics are calculated using satellite raster data for the region, with housing sales data obtained from local tax assessment rolls. The estimation of a hedonic price model is conducted using a reversible jump Markov chain Monte Carlo (MCMC) sampling algorithm. This approach to Bayesian model averaging simulates a search across model space, and represents a numerical approximation of the average of all possible models weighted by the likelihood that they are “true” according to the data. Although this has not been widely used in hedonic price modeling to date, these averaged models better express uncertainty about model selection within the estimation results, provide for more accurate out-of-sample estimations than single hedonic regression models alone, and present results that are more robust for ranges of sample sizes. The results from these price estimates are examined in the context of existing open space policies in the Seattle area to evaluate the price effects associated with efforts to preserve green space. Preliminary results suggest that there is a small but notable marginal effect of green space on property values, which is not constant and tends to differ according to locational and submarket characteristics.