Forecasting Location of New Housing in Integrated Models of Land Use Perspective of Developers in the Portland, Oregon, Region

Most housing market submodels in extant integrated models of land use have focused on the demand for dwelling units by household, with scant attention paid to the supply side of the market represented by individual home developers. This study tried to fill this gap by developing and comparing three models that forecast the development of single-family homes. The three models were characterized separately on the basis of the specifications for location of the new housing choice: (a) assumed market homogeneity and atomization of development projects (Model 1), (b) deterministic market segmentation and synthesis of projects (Model 2), and (c) probabilistic market segmentation and synthesis of projects by use of a latent class approach (Model 3). Examination of the forecast results showed that all three models could successfully capture the basic spatial pattern of single-family-home developments in the region. Although Models 2 and 3 were more sophisticated and more theoretically appealing, they did not produce better forecast results than Model 1 because of some practical issues, including the lack of developer information for forecast years, the small sample size of large projects, the physics of forecasting a small number of large projects across a large number of location alternatives, the need to sample large numbers of alternatives when nonmultinomial logit models were estimated, and the difficulty of using dummy variables in latent class models. In this particular context, the simpler model specification proved to be both easier to implement and more accurate. Models 2 and 3, however, were expected to perform better when those practical issues are solved, at least partially, in further research.