Steven C. Bourassa’s research while affiliated with University of Washington and other places

This paper investigates the usefulness of adding a discrete choice model to the hedonic model via a finite mixture approach. Our approach leads to different hedonic models for different housing market segments based on household information. As such, the proposed method goes beyond measuring the average price of housing attributes. As a case study, we estimate the finite mixture model for the Miami and Louisville metropolitan areas using information on race, ethnicity, and income from the American Housing Survey. We find that the model outperforms the standard hedonic model or a model with linear interaction terms between demographics and housing characteristics. Moreover, market segmentation is based on a complex combination of race, ethnicity, and income. For Louisville, Black households need 2.5 times higher income than White households to advance to a higher market segment and even at high incomes tend to occupy their own segment. For Miami, low-income, non-Hispanic households live in their own segment even if occupying the same dwelling size as households in other segments.

Accurate estimates of land values on a property-by-property basis are an important requirement for the effective implementation of land-based property taxes. We compare hedonic, residual, and matching techniques for mass appraisal of residential land values, using data from Maricopa County, Arizona. The first method involves a hedonic valuation model estimated for transactions of vacant lots. The second approach subtracts the depreciated cost of improvements from the value of improved properties to obtain land value as a residual. The third approach matches the sales of vacant lots with subsequent sales of the same properties once they have been developed. For each pair, we use a land price index to inflate the land price to the time of the improved property transaction and then calculate land leverage (the ratio of land to total property value). A hedonic model is estimated and used to predict land leverage for all improved properties. We conclude that the matching approach is the most promising of the methods considered.

In some applications of supervised machine learning, it is desirable to trade model complexity with greater interpretability for some covariates while letting other covariates remain a “black box”. An important example is hedonic property valuation modeling, where machine learning techniques typically improve predictive accuracy, but are too opaque for some practical applications that require greater interpretability. This problem can be resolved by certain structured additive regression (STAR) models, which are a rich class of regression models that include the generalized linear model (GLM) and the generalized additive model (GAM). Typically, STAR models are fitted by penalized least-squares approaches. We explain how one can benefit from the excellent predictive capabilities of two advanced machine learning techniques: deep learning and gradient boosting. Furthermore, we show how STAR models can be used for supervised dimension reduction and explain under what circumstances their covariate effects can be described in a transparent way. We apply the methodology to residential land and structure valuation, with very encouraging results regarding both interpretability and predictive performance.

This paper examines certain implications from the literature on Tiebout’s model of local government service provision, particularly Hamilton’s extension of the model to include local control of land use and property taxation. Our empirical analysis focused on the use of fiscal zoning to lower property tax rates, a topic that has not been addressed in the extensive literature on Tiebout’s model. Using data for over 100 municipalities in the Miami, Florida, metropolitan area, we specified property tax rates as a function of fiscal zoning measures, other municipal characteristics, and tax mimicking. We conclude that single-family zoning is by far the most important variable explaining municipal property tax rates.

This paper examines certain implications from the literature on Tiebout’s model of local gov-ernment service provision, particularly Hamilton’s extension of the model to include local control of land use and property taxation. Our empirical analysis focuses on the use of fiscal zoning to lower property tax rates, a topic that has not been addressed in the extensive Tiebout literature. Using data for over 100 municipalities in the Miami, Florida, metropolitan area, we specify property tax rates as a function of fiscal zoning measures, other municipal characteristics, and tax mimicking. We conclude that single-family zoning is by far the most important variable ex-plaining municipal property tax rates.

We explore long-term patterns of the house price-income relationship across the 70 largest U.S. metropolitan areas. In line with a standard spatial equilibrium model, our empirical findings indicate that house price-income ratios are typically not stable even over the long run. In contrast, panel regression models that relate house prices to aggregate personal income and allow for regional heterogeneity yield stationary long-term relationships in most areas. The relationship between house prices and income varies significantly across locations, underscoring the importance of using estimation techniques that allow for spatial heterogeneity. The substantial differences across metropolitan areas are closely related to the price elasticity of housing supply.

Big data applications are attracting increasing interest among urban researchers. One unexplored question is whether the inclusion of big data accessibility indices improves the accuracy of hedonic price models used for residential property valuation. This paper compares a big data index with an index derived from a regional travel demand model developed by local transportation planning agencies and traditional measures of accessibility defined as distances to employment centres. Controls for submarkets and a combined spatial autoregressive and spatial error model are also assessed as tools for capturing the value of location. Using single-family residential transactions from the Miami, Florida, metropolitan area, the study’s main conclusion is that the big data accessibility measure does not add meaningful explanatory or predictive power. In contrast, the spatial autoregressive and error model outperforms the other options considered.