ECONOMETRIC METHODS FOR LAND USE MICROSIMULATION

Typically, urban development models have been based on aggregate principles. UrbanSim (Waddell et al. 2003) is among a new breed of models that use microsimulation in an effort to overcome the limitations of earlier models and provide a more dynamic and detailed paradigm. The advantages and disadvantages of using microsimulation are not within the scope of this chapter, but the main implication is that more data, as well as more detailed ones are required for microsimulation than for aggregate models. In the context of the SustainCity project (http://www.sustaincity.org), three European cities (Brussels, Paris and Zürich, described in other chapters of this handbook) have been modelled using the land use microsimulation platform UrbanSim. This platform relies on various models interacting with each other, to predict long–term urban development. The aim of this chapter is to provide some econometric insight into this process. A common set of notations and assumptions are first defined, and the more common model structures (linear regression, multinomial logit, nested logit, mixed MNL and latent variable models) are described in a consistent way. Special treatments and approaches that are required due to the specific nature of the data in this type of applications (i.e. involving very large number of alternatives, and often exhibiting endogeneity, correlation, and (pseudo–)panel data properties) are discussed. For example, importance sampling, spatial econometrics, Geographically Weighted Regression (GWR) and endogeneity issues are covered. Specific examples of the following models: (i) household location choice model, (ii) jobs location/firmography, (iii) real estate price model, and (iv) land developmentmodel, are demonstrated in the context of the case studies in Brussels, Paris and Zürich. Finally, lessons learned in relation to the econometric models from these case studies are summarized