Martin Faulques (p), Jean Bonnet, Sébastien Bourdin

Paloma Taltavull De La Paz

The impact of real estate prices is one of the crucial points on the social acceptability issues faced by biogas facilities in France. To date, we have only identified three studies on the price of real estate. The one by MODICA (2017) analyzes the impact of biogas plants on the average property value per municipality based on revealed preference data in Piedmont (Italy), and finds no evidence of impact. However, PECHROVA and LOHR (2016) find a negative impact of biogas plants on house prices based on eight biogas plants in the Jihomoravsky region of the Czech Republic using the hedonic pricing method. As for the study conducted by ZEMO et AL. (2019), they highlight that large-scale farms and biogas facilities have an adverse effect on property values while small-scale facilities have a positive effect. More importantly, the acceptability of biogas facilities can vary from country to country (SCHULACHER AND SCHULTMANN,2017) and even from region to region within a country (KORTSCH et AL., 2015). Given the very limited number of economic evaluation studies of residents' opposition and support for green energy facilities and the inconclusive results, it is important to obtain more evidence on the effects of biogas projects on the hypothetical decrease in housing prices and its influence on social acceptability.
Based on a database of 61,942 sales between 2015 and 2021 on real estate sales in the Great West (Normandy and Brittany), a difference-in-difference (DiD) counterfactual method is used to evaluate the effects of biogas units on the price of real estate in the areas near these units. This study aims to continue the work initiated by ZEMO et AL. (2019) by applying a new field of study (Normandy and Brittany) and making changes to the method (DiD). We will test what role the size of the biogas plants plays and how the existence of opposition from the local population to the establishment of such facilities influences local real estate prices.

Alon Sagi (p), Dani Broitman, Avigdor Gal

Martin Faulques

Dwelling units are both consumer and investment products. This double facet view, together with the urban spatial development and the dynamics of consumer’s preferences, are among the factors that cause a constant change in urban neighborhoods demand. But sometimes, the demand for some neighborhoods rises more than can be explained as rational behavior or in traditional supply and demand schemes. In some cases, this rise in demand might turn into a price bubble: a situation in which a price of a certain asset (a stock a house or a virtual currency) increases over its real value and eventually ‘burst’ in rapid and dramatic price decrease. According to the ‘efficient market hypothesis’ (EMH), the most exactable economic theory to explain assets pricing, in an efficient market, all prices are fully reflect the assets values and expectations for future returns are already included in it. Therefore, since future value expectations are already embedded in current prices, predicting price bubbles in advanced (or any other future price change) is impossible. That unless one can find a new, better and faster, method of market analyzing. This is the main issue of the current research: Most past research analyzed housing bubbles post-mortem or tried to predict it locally and during a short period. The current research aims to find a universal long time prediction tool for change in demand for urban neighborhoods. More specifically, we aim to predict housing price changes in a neighborhood relative to its region. For this task we use millions of housing transactions in 7,195 neighborhoods in England and Wales since 1995, together with hundreds of socio-economic features taken from the UK censuses. We aim to predict future market scenarios, using innovative machine and deep learning methods. We use the data from 1995 to 2011 as a training set, and test whether the demand for a neighborhood rises, declines, or remains unchanged between 2011 to 2016, and then again between 2016 to 2021. This way we get 9 possible tags to classify. Although the obtained results are preliminary and only partially accurate, our conclusion is that we can be optimistic regarding the capability of machine learning methods to forecast future prices in general, and price bubbles in particular.

Maike Metz-Peeters, Sven Werenbeck-Ueding (p)

Alon Sagi

A rich literature on house price indices looks into the development of housing prices or rents over time, when controlling for a variety of housing characteristics. The most commonly used approaches typically assume time trends to be constant or allow only for limited spatial heterogeneity. Evidence on trend heterogeneity with respect to housing characteristics is, however, sparse, yet crucial to understand the dynamics behind the emerging developments.

In the hedonic regression approach, analyzing time trend heterogeneity would require the explicit and precise modeling of interaction terms, thereby imposing strict, unrealistic formal assumptions on the data generating process (DGP). In contrast, the repeat sales method only considers housing units for which prices are observed in more than one period, thereby being inefficient in its use of data. To overcome these issues, Longford (2009) and McMillen (2012) view time trends in the light of the Potential Outcomes Model (POM) and suggest the use of matching estimators that essentially compare units from different periods that are similar in their covariates. This approach provides a nonparametric way to control for housing characteristics, establishes a simple framework for heterogeneous time trend estimation and removes effects that arise from changes in the spatial and structural composition of observed housing markets by excluding units with a lack of common support. However, similar to hedonic regressions, the consideration of the spatial dimension into the model is based on the inclusion of unflexible and arbitrary fixed effects. In addition, matching estimators are well known to suffer from the curse of dimensionality.

We adopt the view on house prices as potential outcomes and apply causal forests to estimate heterogeneous time trends of house and apartment sales prices and rents, respectively. This method from the field of causal machine learning can in essence be interpreted as an data-adaptive form of k-nearest neighbor matching and overcomes the curse of dimensionality by letting neighborhoods in the covariate space get wide along irrelevant dimensions, enabling it to identify spatial neighborhoods with relatively constant price levels. Thus, we explore its capability to flexibly consider the spatial dimension in regions with high observation density, by directly including the observation‘s coordinates as regressors. Our analysis uses data from ImmoScout24, Germany’s leading online platform for real estate advertisement, which contains detailed geocoded information on dwellings in Germany advertised for sale or rent between 2010 and 2020.

Paloma Taltavull De La Paz (p)

Sven Werenbeck-Ueding

The paper builds the cycles of transactions and prices in the short term rental market from 2015 to 2021 for 46 European and Asian cities to understand how short-term tenants move across the cities. Having the cycles at the city level, the paper builds a panel and estimates a supply-demand model for the short-term rental market and analyses the endogenous relationship and ripple effect among cities using the VECM framework. Results suggest the existence of links between cities, both in short-term rental contracts and prices, supporting the hypothesis that short-term rental market visitors choose cities clustered in networks and that those networks compete with each other as alternative destinations. The evidence suggests that, through the technological platforms and individually, a particular city network is chosen as an alternative to others, revealing tastes changes and inducing segmentation in rental price growth and investment.