Essays on local housing and real estate brokerage markets.

As noted by the late housing economist John Quigley (1979), housing markets are characterized by a peculiar combination of features, such as durability, spatial fixity and multidimensional heterogeneity. The combination of these features differentiate it from other product markets and it has been shown that they have a non-negligible impact upon market outcomes in housing and real estate brokerage markets. The inherent spatial fixation and the multidimensional heterogeneity of housing units, for example, suggest that search costs in housing markets are substantial. Potential buyers frequently visit multiple housing units before a transaction takes place, which is costly in terms of time and money. These high search costs, but frequently also the high transaction costs (registration duties, notary fees, and so on), suggest that housing markets only slowly adjust to a new equilibrium after an external shock takes place. It has been observed, for example, that listing prices only slowly decrease during housing market busts. The different chapters presented in this dissertation are all, in some sense, related to the observation made by Quigley (1979). In the first two chapters we study the effects of borders. Since houses are fixed in space and housing market policies are frequently organized at the national level, border effects may appear. In the third chapter, we acknowledge that not only houses are heterogeneous, but neighborhoods are also heterogeneous in terms of their housing stock composition. We investigate whether different bundles of houses at the neighborhood level have an impact upon the prices of the properties sold. In the fourth chapter, finally, we investigate focus on the real estate brokerage industry. We examine the both market structure and market size and apply the structural model to the Belgian real estate brokerage market.

The first chapter investigates convergence and spatial and temporal diffusion patterns of regional housing prices with a special interest for the linguistic border that splits Belgium into two distinct linguistic regions. Especially in the UK, a large literature has emerged that investigates the so-called “ripple effect hypothesis” (Meen, 1999). Many authors have found that regional house price movements exhibit a distinct spatial pattern over time, rising first in a cyclical upswing in the south-east and then spreading out over the rest of the country. By analyzing the joint evolution of housing prices from 21 predefined Belgian regions on both sides of the linguistic border, we examine whether there are certain distinguishable systematic recurring patterns. A preliminary analysis using the pairwise approach to test for convergence that was developed by Pesaran (2007) shows the different regional housing price series are interdependent, which suggests that the general evolution of housing prices is driven by macroeconomic developments. The results from this preliminary analyses, however, also show that the degree of interdependency among regional house price series is larger within each linguistic region, which suggests that the linguistic border might play a role. As a second step, we test whether we find evidence for a (single) dominant region in Belgium. Therefore, we estimate bivariate VAR(4) models with error correcting coefficients, which allow us to assess whether prices in certain regions are long-run forcing, in the sense of Granger and Lin (1995), upon prices in the remaining regions. After having established that the region of Antwerp is an eligible candidate, we estimate an extended version of a model that was recently proposed by Holly et al. (2011). Our empirical model allows for a dominant region, error-correcting mechanisms and takes into account the unique federal setting of Belgium. The results show that Antwerp is indeed a suitable dominant region. We furthermore find that the role of the linguistic border is ambiguous. While housing prices in regions located along the central Antwerp (Flanders)-Namur (Wallonia) are highly interdependent, housing prices on both sides of the linguistic border in peripheral regions in the eastern and western parts of Belgium move relatively independent. Using Generalized spatio-temporal Impulse Response Functions,  we show that there is evidence for a ripple effect in Belgium.

In the first chapter, spatial and temporal diffusion patterns of housing price series within a single country were examined. It was shown that the role of the linguistic border is ambiguous. Where the first chapter thus examines the effects of a linguistic border, in the second chapter the focus is shifted towards a national border. Since the seminal contributions of McCallum (1995) and Engel and Rogers (1996), border effects have been studied extensively. A prominent example in the context of housing markets is Miceli et al. (2014), who examine the price gap “at the border” using listings from both Dutch and German borders near the border. While many of these studies use data from both sides and examine the gap at the border, in the current study, we only have data for Belgian dwellings. Therefore, we examine the determinants of housing prices of houses located near the Belgian-Dutch border and examine whether houses that are located more closely to the Netherlands are more expensive as a result of spatial arbitrage. Housing prices in the Netherlands have traditionally exceeded those in Belgium, which has provided an incentive for Dutch households to move across the border. Since these Dutch households typically want to live close to the Netherlands, the prices of Belgian dwellings close to the border are expected to increase as a result of an increased demand. Using a large dataset with  detailed information concerning the prices, characteristics and exact locations of approximately 6.500 transactions, we are able to examine whether dwellings located more closely to the Belgian-Dutch border are more expensive as a result of spatial arbitrage. The baseline results from a spatial hedonic pricing model show that the price difference between a property located at the border and a property located 15 kilometers from the border was approximately 13% in 2003. The results, however, also indicate that this arbitrage effect has been declining over time. While properties located 15 kilometers from the border were approximately 13% less expensive in 2003, this effect has decreased to only 2% in 2015. The boom in Belgian housing markets since 2005, combined with the subsequent collapse of the Dutch housing market in the aftermath of the financial crisis, suggests that the incentive for spatial arbitrage has declined over time, which is in line with our findings. In a series of robustness checks and extensions it is also shown that the results are robust with respect to alternative specifications of the model and an alternative measures of distance to the border. It is furthermore shown that the estimated arbitrage effect is not constant across space. More specifically, our results show that the estimated spatial arbitrage effect is especially strong in the eastern parts of Flanders, where sparsely populated Flemish regions border Dutch cities. In a final extension of the baseline model we limit the analysis to a single geographical region (the districts Turnhout and Maaseik) and only use those transaction for which we know the previous address of the buyer. Descriptive statistics reveal that Dutch buyers that is, on average, 16% more expensive, but also differs in other dimensions. The results from a spatial hedonic model, however, reveal that even after controlling for an extensive list of control variables Dutch buyers pay a premium. The results, however, also reveal that this premium has been declining over time. We interpret this finding as evidence for “anchoring bias”. 

As stated in the introduction, housing is a highly heterogeneous good, which complicates its valuation. Fortunately, (housing) economists have developed methods, such as the hedonic pricing method (Rosen, 1974), to deal with the inherent heterogeneity that characterizes the housing commodity. The hedonic pricing method, which considers houses as “bundles of characteristics”, has since become one of the main workhorses of housing economists studying the determinants of housing prices. At a slightly more aggregate level neighborhoods can be defined as “bundles of housing units”. While numerous papers using the hedonic pricing method have shown that different bundles of characteristics are valued differently by buyers, there have been virtually no studies that investigate the housing stock composition of the neighborhood influences the price of an otherwise equivalent dwelling. In the third chapter, we examine how buyers value the characteristics of neighboring properties using a sample of approximately 6.100 dwellings from the Flemish part of the Brussels Metropolitan Area. More specifically, we examine how the average values and diversity measures of several characteristics, such as interior space, plot size, age and type of construction, influence the price of the property that is up for sale. We therefore combine detailed individual transaction data with administrative and geospatial data and employ spatial hedonic models. Our results reveal that the price of an observationally equivalent property is increasing in the average size, year of construction, maximum roof height and distance to the road of neighboring properties, but decreases in the average size and shape of the plots of neighboring properties. Our finding that the price of a property is increasing in the average size of neighboring properties provides evidence for the “tax capitalization hypothesis” of Hamilton (1976), who argues that relatively smaller houses benefit from the presence of larger houses. The results for the diversity measures reveal that home values are higher in neighborhoods that are characterized by low levels of diversity in building types, year of construction, interior space, building shapes and distance to the road of neighboring properties. Our findings furthermore indicate that differences in the composition of the neighborhood’s housing stock are responsible for price differentials as large as 12%. While the current chapter provides just one example of how geospatial data can be used in econometric analysis, our findings have important and potentially far-reaching consequences for real estate professionals, policy makers and urban planners. We furthermore show that the results presented in this chapter are robust with respect to alternative specifications and alternative levels of aggregation.

The inherent spatial fixation and multidimensional heterogeneity of the housing commodity also result in high search costs for potential buyers and sellers. Buyers, for example, frequently visit multiple housing units before they proceed to a purchase, which is costly both in terms of money and time. In order to reduce these search costs, real estate agents have traditionally operated as intermediaries between buyers and sellers. Real estate agents provide promotional activities, select eligible buyers and help with the bargaining process and administrative duties. In return, they typically charge a commission fee that is frequently expressed as a percentage of the sales price (e.g. 6% in the US (Hsieh & Moretti, 2003)). The fact that these commission fees are relatively stable across time, thus during busts and booms of the housing market, and the fact that there are low barriers to entry in the real estate brokerage market, has led many to believe and conclude that there is a lack of price competition in the real estate brokerage market. As shown by Hsieh and Moretti (2003), a lack of price competition in combination with low barriers to entry can result in large welfare losses as a result of the excessive entry real estate agents. The main intuition for this result is that these brokers merely “steal business” from incumbent brokers and can be productively employed elsewhere. In recent decades, however, developments in ICT have made it less costly for sellers (buyers) to search for an eligible buyer (seller/home), which has put a downward pressure on the commission rates charged by real estate agents. In this chapter, we examine whether lower commission rates are necessarily beneficial from a social point of view. We do this by developing a model of the real estate brokerage industry, where we make assumptions about the behavior of buyers, sellers and real estate agents. Our results indicate that it is socially optimal for intermediaries to have some market power. That is, it is socially optimal that intermediaries charge a service fee that is above the per-match cost as this excludes some low valuation buyers, which are disliked by sellers, as well as some high reservation price sellers, which are disliked by buyers. Secondly, we show that matching markets are generally characterized by an excessive number of intermediaries that operate in the market compared to what is socially optimal. When calibrating the model using data from the Belgian real estate brokerage industry, which provides an adequate setting to test our model, the welfare counterfactuals suggest that the observed commission rate of 4.3% is below the socially optimal commission rate, which is estimated to be in the range between 5.1% and 24%. A welfare gain of 1% to 11% could be established when regulating broker service fees, given the number of brokers that currently operate in the market. When also regulating broker entry, a further welfare gain of 7% to 69% could be realized. Various other policy relevant welfare counterfactuals are constructed and discussed. In general, the results presented in this chapter show that the traditional paradigm “more competition is always better” does not per definition hold in matching markets.