Econometric Rent Modeling in a Highly Regulated Market

Although there exists some scientific literature concerned with pricing on the Austrian real estate market (Helbich, W. Brunauer, et al., 2014; Kuntz and Helbich, 2014; W. Brunauer, Lang, and Umlauf, 2013), there is surprisingly little quantitative research on housing rents, their spatial structure and drivers. Expanding the existing literature beyond house price prediction is especially important in the Viennese case, where according to the Austrian Mikrozensus, 77.5% of the population live in a rented flat (Statistik Austria, 2020). Thus, quantitative research on the Austrian housing market cannot primarily focus on price formation regarding private real estate property but needs to also consider rental markets. W. A. Brunauer et al. (2010) provide a notable exception from the lack of quantitative research on the Viennese rental case as well as an interesting approach using a Generalized Additive Model with spatial scaling. Fortunately, there is also a growing body of econometric literature on housing rents and their drivers onthe international stage where one could draw ideas from. Recent examples can be found in Tomal (2020); McCord et al. (2014) or Efthymiou and Antoniou (2013.) Usually in the spirit of hedonic house price models, housing rents are regressed onto certain characteristics of the respective flat as well as some indicators measuring the quality of location. However, the Viennese housing market has several special features which have not been properly addressed in housing rent modeling up till now. An important feature of the Viennese accommodation market is the fact that roughly 43% of households live in a flat provided by the social sector, which consists of municipal as well as cooperative (non-profit) housing. The private rental sector on the other hand accommodates about a third of the households (Tockner, 2017). Thus, dynamics in the comparatively small free market segment cannot be adequately understood if considered independent from the larger social sector (Kemeny,Kersloot, and Thalmann, 2005). A further important aspect is the strong regulation of rent prices through the Mietrechtsgesetz (MRG) that basically constitutes two regulatory regimes within the private market segment. On the one hand flats located in buildings erected before 1945 or built with state subsidies experience strong price controls. On the other hand, flats that do not fulfill the previously mentioned criteria as well as single family houses do not experience any such price controls. However, the introduction of location bonuses to the price-controlled segment led to spatially very uneven price increases over the last years (Kadi, 2015). We propose a hierarchical generalized additive model (HGAM) to model squaremeter prices by smooth functions of flat characteristics such as size, age, and time within the sample. Additionally, various dummy variables enter the model as linear predictors and random effects are used to model subdistrict specific location bonuses in the baseline model. Going beyond the existing hedonic housing rent literature this study also proposes several extensions to model, addressing the aforementioned special features of the Viennese rental market. Thus, the baseline HGAM is modified to incorporate regulatory-regime heterogeneity in its parameters as well as spatial heterogeneity with respect to time trends in order to properly address the differences in the development of location bonuses. Varying degree of competition from the social sector in each subdistrict is also tested as a potential impacting factor onto rents. As spatial autocorrelation might be an issue given the spatial nature of the data, we do not only use random effects for the location bonuses but also add a spatially structured predictor using markov-random-fields. The Data available for this study was kindly provided by the DataScience Service GmbH and consists of over 84,000 observations of flats offered on the Viennese rental market between 2012 and 2020 with a very high coverage rate during the more recent years. Asking prices, GIS data, very detailed real estate characteristics including size, age, furnishing and many more, as well as a multitude of socio-economic variables on the respective area such as share of academics, proximity to medical infrastructure or accessibility of public transport are available for the given dataset. Due to the constant excess demand on the accommodation market in Vienna for the given period, asking prices are assumed to hardly deviate from market prices and can be seen, as a legitimate approximation. First results suggest that all proposed extensions to the baseline model add significant predictive power.