Housing rental submarkets in hedonic regression: econometric arguments and practical application

In the Walloon region, in Belgium, according to the recent Housing tenancy decree (2018), “the official rent calculator” should be developed to estimate “the reference rent” for any residential dwelling rented on the market. The “calculator” should be market-oriented and be based on individual property attributes and location. The rent estimation mechanism should geographically cover the region. Homogenous zones should be delimited according to rent levels observed on the market. For this purpose, the regional Rent Survey 2018 had collected data on rents and dwelling attributes for 4.112 observations. This geographically representative sample of the regional rental market provides data for the delineation of rental submarkets. The paper deals with the geographical aspect of the regional hedonic regression model of rents. The rental submarkets are delineated with a combination of the “location value response surface”, clustering and expert approach. The purpose of the study is, similarly to Leishman et al. (2013), to find the best way to model the identified submarkets. Three approaches are applied, namely (1) the overall model with dummy variables for submarkets, (2) the set of submodels for the submarkets and, finally, (3) the “location value” in the submarkets, i.e. the ratio of the observed rent to the rent predicted with a hedonic model without location attributes. Actually, the classical ordinary least squares (OLS) methodology is applied. The study should be developed using also the geographically weighted regression (GWR). On the one hand, the three approaches are compared according to standard econometric indicators, i.e. the explanatory power of models, the spatial autocorrelation of their residuals and the accuracy of ex-sample predictions. On the other hand, the practical motivation of the study implies the acceptance of the principles and the results of the methods by stakeholders, i.e. landlords, tenants and the regional government. While the set of submodels for the submarkets provides the best econometric performance according to the majority of indicators, this approach has an important drawback unacceptable for stakeholders: in submodels, some crucial structural variables, such as those for building type or building age, are often either insignificant or have an unexpected sign. This problem caused, at least partly, by limited size of sub-samples, manifest itself much less in the overall models with location dummies or with “location value”. The practical and econometric advantages and disadvantages of the two latter models are discussed.