Document Type

Working Paper

Publication Date



College of Business and Economics

Document Number





This paper develops a framework for addressing the omitted variable bias that plagues most real estate research. We incorporate qualitative information from text to control for property attributes that are generally unobserved. The textual information is entered by real estate agents for every property sold on a Multiple Listing Service (MLS). The agents, who arguably have the most local market and property specific knowledge, use the unstructured text to highlight important information that is not clearly conveyed in other areas of the listing. Although the framework can be applied universally in real estate research, we demonstrate its effectiveness in the estimation of agent-owned sales premiums. Similar to previous studies, we find agent-owned premiums between 2% to 6% when no textual information is included. When we include the textual information the agent-owned premiums dissipate. The results suggest that the market distortions reported in Rutherford et al. [2005] and Levitt and Syverson [2008] do not exist.