Housing often forms a substantial component of household wealth. Despite its critical importance, accurately measuring property value has historically posed significant challenges. Recent research highlights the influence of a well-regarded machine learning algorithm designed for pricing properties within the U.S. housing market. The findings indicate that the algorithm generally provided benefits across the market.

A collaborative team of researchers from Carnegie Mellon University, New York University, and the University of Toronto conducted this investigation. Their findings are documented in an article published in the journal Marketing Science. The research represents a pioneering effort to elucidate the ramifications of algorithm-generated property value predictions on the housing market, including economic and social outcomes. Param Vir Singh, a professor at Carnegie Mellon’s Tepper School of Business and one of the study’s co-authors, elucidated the significance of their findings, highlighting the groundbreaking nature of their work in assessing the impact of machine-generated property valuations.

Many online real estate platforms utilise proprietary algorithms that leverage vast datasets to estimate property values. These calculated values are then displayed on the platforms’ websites, providing potential buyers and sellers with critical information influencing their market decisions. This particular study focused on ‘Zestimate’, a machine learning algorithm devised by Zillow, among the foremost real estate marketplace companies in the U.S. and a pioneer in publishing algorithm-generated property valuations nationwide. The algorithm is noted for its heightened accuracy in affluent neighbourhoods compared to poorer ones, which has sparked concerns about potential increases in socioeconomic disparities.

The study examined over 4,000 property listings across 140 neighbourhoods in Pittsburgh, Pennsylvania, spanning from February to October 2019. The researchers employed a structural model of the housing market to analyse the effects of Zestimate. In this model, buyers and sellers faced uncertainties regarding property values, with Zestimate providing an unbiased indication.

The results revealed that Zestimate positively impacted both buyers and sellers, leading to an average increase in buyers’ surplus by 5.4 per cent and sellers’ profits by 4.2 per cent. This benefit was primarily attributed to the algorithm’s role in diminishing uncertainty, which allowed sellers to adopt a more patient approach, setting higher reservation prices while waiting for buyers who genuinely valued the properties. According to the study authors, this improved the quality of matches between sellers and buyers.

Moreover, the research found that Zestimate played a role in reducing socioeconomic inequality within the housing market. Both affluent and economically disadvantaged neighbourhoods gained from the algorithm, but the benefits were notably more substantial in poorer communities. For instance, the average increase in seller profits in these neighbourhoods was nearly 5 per cent, significantly higher than in mid-range and affluent areas. Similarly, the buyer surplus in poorer neighbourhoods saw an 8.3 per cent rise compared to lower percentages in other socioeconomic segments. This disparity was primarily due to more significant initial uncertainty in poorer areas, which stood to gain more from new, reliable information about property values.

However, the study also acknowledged certain limitations, including the absence of modelling scenarios where multiple buyers might engage in a bidding war, typically driving up the selling price. Furthermore, the model used in the study assumed buyers considered properties individually rather than exploring multiple listings simultaneously, which could otherwise heighten competition among sellers. Yan Huang, another co-author and associate professor at Carnegie Mellon, underscored the potential of property value prediction algorithms to mitigate uncertainty in real estate markets. Meanwhile, Kannan Srinivasan, also a professor at Carnegie Mellon, emphasised the importance of assessing the differential impacts of such algorithms, advocating for a comparative analysis of different demographic groups under scenarios with and without the algorithm’s intervention to appreciate its effects fully.

More information: Runshan Fu et al, Unequal Impact of Zestimate on the Housing Market, Marketing Science. DOI: 10.2139/ssrn.4480469

Journal information: Marketing Science Provided by Carnegie Mellon University