An Application of Spatial Econometrics in Relation to Hedonic House Price Modeling

Article excerpt


This paper applies spatial econometrics in relation to hedonic house price modeling. Some basic spatial model alternatives are used for a battery of relevant tests. Geographically-weighted regression, semiparametric analysis, and the mixed spatial Durbin model are also applied. The purpose is to detect missing spatial variables, misspecified functional form, and spatial heterogeneity in estimated parameters. Such misspecifications have been shown to give spurious results in relation to some frequently used directional based tests. Significant model improvement is achieved, so the paper should be of general interest as an example of practical econometric modeling within the field.

Location characteristics are important determinants of housing prices. Even so spatial modeling is not prevalent in mainstream empirical research on housing markets (Kim, Phipps, and Anselin, 2003). Over the past thirty years, spatial econometrics has advanced from the fringe to a fledgling discipline (Goodchild, 2004). Being a fledgling discipline, and also given the technical and interpretational difficulties of alternative spatial models and test results, the field is not easily available to applied econometricians. The purpose of this paper is to expose researchers in the field to important tools that are available when testing for and incorporating spatial effects into hedonic house price models. Such an approach is timely, due to the widespread use of hedonic modeling. Building on previous research, this paper analyzes some acknowledged and relative basic spatial econometric model alternatives and estimators. It also gives an overview of a battery of tests that may be used in the hedonic modeling process, and the order in which the tests could be conducted. The relevant tests are frequently performed to get consistent estimates on implicit prices and to increase the reliability of ordinary significance hypothesis testing on individual parameters. Some commonly encountered problems will be discussed. The overall problem is one of diagnostics and significance testing, model specification, and selection. This is an important issue when predicting the partial effects of individual attributes on housing prices, since the choice of spatial model will have an effect on the economic interpretation of the estimated coefficients.

In addition to Dubin (1998), other papers with similar purposes as this are Pace, Barry, and Sirmans (1998), Dubin, Pace, and Thibodeau (1999), Beron, Hanson, Murdoch, and Thayer (2004), and Bourassa, Cantoni, and Hoesli (2010). A short review of the preceding relevant literature is found in Besner (2002). This paper is, however, more comprehensive, naturally at the expense of the depth of each topic. The model discussed here is a significant improvement. The analysis is hence of general interest as an example of practical spatial econometric modeling in relation to hedonic house price studies. Finally, the paper may help implement the replication methodology developed in Lai, Vandell, Wang, and Welke (2008) for estimating property values.

In the next section an overview of important ways of modeling spatial effects in econometrics in relation to hedonic house price models will be given. Subsequently, there is a description of the study area, the data, and the preliminary base model. This part will not be extensive, since the main focus in the empirical analysis is on the relevant tests and the spatial modeling process itself. On the basis of these tests, geographically-weighted regression (GWR) and a semiparametric estimator are used to explore the potential for missing spatial variables, spatial trends, or spatial heterogeneity in our initial model.1

Spatial Effects and Important Spatial Econometric Models

Spatial data is characterized by two key features: spatial autocorrelation and spatial heterogeneity (Anselin, 1988). Together these features are labeled spatial effects. …