When a Deed Is Not a Market Sale: Foreclosure Transfers, Statutory Consideration, and Repeat-Sales House-Price Measurement

New York City's public sales file can keep foreclosure-related lender acquisitions at amounts set by a legal rule. A rule based on the parties to each deed flags 16,655 potentially affected deeds among 1.47 million records from 2003 to 2025. Conventional price screens keep most of them, and 94–95% of the screened lender acquisitions recorded since 2016 appear in the public sales file at the same amount. Dropping the flagged endpoints changes a house repeat-sales index by 3.69 log points in 2010Q3 (simultaneous 95% band 1.59 to 5.78), and by 2.63 under a narrower definition. The estimated peak-to-trough contrast is −0.64 points, with an interval from −3.03 to 1.25. A hedonic comparison on the common attribute sample is consistent with repeat-sales pairing contributing to the later shortfall. Whole-path inference, other endpoint rules and the assignment of unresolved cases qualify the result. Of 400 AI-reviewed deeds, 84 remain unresolved, and agreement among the resolved cases is not independent human validation. Files: the full paper, a short note on the same study, the code archive (MIT licence) and the data archive. See revision_notes.md for earlier changes.