Experience Stern | Faculty & Research

Overview: In the paper titled, “Unequal Impact of Zestimate on the Housing Market,” NYU Stern Professor Runshan Fu, along with co-authors Yan Huang (Carnegie Mellon University), Nitin Mehta (University of Toronto), Param Vir Singh (Carnegie Mellon University), and Kannan Srinivasan (Carnegie Mellon University), examines the impact of Zillow’s Zestimate—a machine learning (ML)-generated home value estimate—on housing market outcomes, particularly across different socioeconomic segments. The study explores how Zestimate influences buyer and seller decisions, market efficiency, and socioeconomic inequality.

Why study this now: Housing is a crucial component of household wealth and financial stability. However, property valuation is complex due to factors like significant item-specific heterogeneity, dynamic market conditions, and market frictions. The introduction of ML-based estimates like Zestimate aims to reduce these inefficiencies by offering property value signals. Understanding its impact is critical because: it can influence pricing strategies, negotiation dynamics, and market liquidity; it affects both buyer surplus and seller profits by altering expectations and decision-making; and it raises concerns about whether ML algorithms exacerbate or mitigate socioeconomic inequality in real estate.

What the authors found: Using a sample of 4,027 properties listed in Pittsburgh, Pennsylvania between February and October 2019, the researchers found that:

• On average, Zestimate increases buyer surplus by 5.94% and seller profits by 4.36%, and enhances buyer-seller match quality by reducing uncertainty about a property’s value, enabling sellers to hold out for better offers without excessive time on the market
• Despite being less accurate in poor neighborhoods, Zestimate reduces greater initial uncertainty, leading to larger relative gains for these areas
• Overvalued Zestimates can lead to longer selling times, while undervalued ones may result in lower sale prices. However, the uncertainty reduction effect dominates, meaning the overall impact of Zestimate remains positive despite occasional valuation errors

What does this change: Many people assume that AI tools like Zestimate only help the rich. This study shows that is not necessarily true. While Zestimate is more accurate in wealthier areas, it helps poorer neighborhoods even more because they start with less information. This research shows that when judging AI, it is important to look beyond just accuracy—there should be consideration of how it changes people’s decisions and improves their options.

Key insight: “We show that Zestimate overall benefits the housing market, as on average it increases both buyer surplus and seller profit,” note the authors. “Moreover, Zestimate actually reduces socio-economic inequality, as our results reveal that both rich and poor neighborhoods benefit from Zestimate but the poor neighborhoods benefit more.”

This research is forthcoming in Marketing Science.