Why these records need linking
New York City publishes its property sales and its recorded deeds and mortgages as separate open datasets, and the two do not join directly. The Department of Finance sales file has the price and date but no document number. ACRIS, the city register, has the deeds and mortgages but not the sales. Condominiums are the hardest case: the sales file usually identifies a unit by the building's billing lot, while ACRIS records its deed and mortgage under the unit's own lot, so a join on the parcel number misses most of them.
This dataset links the records using fixed, documented rules. The studies on this site and other researchers can reuse the same matches. The public files contain no party names.
In numbers
- 729,047house, condominium and co-operative sales, 2016–2025, all five boroughs
- 277,298house and condominium sales linked to a unique deed (97% of houses and 74% of condominiums attempted)
- 188,076sales with a mortgage recorded within 90 days, each with its loan amount
- 276,255linked deeds with flags for company buyers, shared surnames, shared addresses and lender take-title events
Show as a table
| Year | Houses | Condominiums |
|---|---|---|
| 2016 | 67.9% | 53.8% |
| 2017 | 69.6% | 55.3% |
| 2018 | 70.7% | 55.9% |
| 2019 | 72.3% | 57.6% |
| 2020 | 75.1% | 62.8% |
| 2021 | 78.3% | 66.0% |
| 2022 | 75.8% | 63.1% |
| 2023 | 68.6% | 48.7% |
| 2024 | 69.0% | 46.0% |
| 2025 | 68.8% | 46.5% |
Files in the dataset
Three tables that share keys: sales joins to links on sale_id, and links joins to deed_flags on deed_document_id. A blank cell means not applicable or not measured, never zero.
Files and main columns
| File | Rows | Main columns |
|---|---|---|
sales.csv.gz | 729,047 | Borough, block and lot; property group (house, condo, co-op); building class; address; price; sale date; geography; flags for ACRIS coverage and the $300,000–$6 million band |
links.csv.gz | 729,047 | Deed match (unique, none, ambiguous); ACRIS deed id; recovered unit lot; financed in three windows (30, 90 and 180 days); number of mortgages; largest loan amount and its document id; CEMA flag |
deed_flags.csv.gz | 276,255 | Party counts; company, trust or estate buyer; same surname (and a strict version); shared mailing address; lender name; deed event from the foreclosure paper's classifier; legal-rule amount |
The download also includes the raw extracts and their retrieval records, Python code using the standard library, and CODEBOOK.md, which defines every column and rule.
How the link works
- Deed. A deed matches a sale when its amount is within 0.5% of the price and its date within 45 days. Houses are matched on their own parcel. Condominiums are matched within the block among unit lots, which recovers the unit; it agrees with the sales file's own unit lot in 99.3% of unique matches. If deeds on two parcels qualify, the sale is left unlinked rather than guessed.
- Mortgage. A sale receives a recorded-mortgage flag if a mortgage is recorded on its parcel (for condominiums, the recovered unit) from 15 days before to 30, 90 or 180 days after the sale. A recorded mortgage is not verified purchase finance.
- Flags. Computed from the ACRIS party records of each deed, with the papers' own rules. Party names were used to compute flags and are excluded from the public release. The flags are a fixed snapshot.
Agreement with earlier studies
The financing flags agree with the archived panel of The Shape of the Tax for every one of 166,736 houses and all but 4 of 82,348 condominiums. Recomputing that paper's mortgage-recording-tax table from this dataset reproduces every reported statistic. Since that paper's version 2.5 takes its condominium loans from this dataset, that row is no longer an independent comparison. These checks show that the computations agree with earlier code, not that the links are accurate. Against the repeat-sale pairs of Who Gets the House?, the linked deed is the same at 15,902 endpoints and different at 1.
Limits
- A financed flag of 0 means no mortgage was recorded on the parcel in the window. It is not proof of a cash purchase.
- Co-operative and Staten Island sales are listed but never linked. ACRIS real-property mortgages do not measure financing secured by co-op shares, and Staten Island deeds are outside ACRIS.
- 10,835 condominium sales match deeds on more than one unit and 22,030 match none, so condominium coverage is lower than house coverage.
- The company, surname, address and deed-event flags are screens built from names. Existing review agreement does not establish independently verified population accuracy. The party records are not deposited, so these flags are supplied as a fixed snapshot.
Condominium coverage
A unique deed is linked for 74.0% of eligible condominium sales, from 78.9% in Manhattan to 44.7% in the Bronx. At prices from $100,000 to under $500,000, the corresponding figures are 77.0% and 40.4%. A borough comparison of financing therefore needs its coverage denominators.
Read the comparison and download the counts. Coverage describes how many records link. Agreement describes consistency with an earlier computation. Accuracy requires verified source evidence.
Quick start
For a runnable example without extra Python packages, extract the dataset and run python examples/recorded_mortgage_example.py build. The archive also includes a condominium selection example; see its README for arguments and outputs.
import pandas as pd
ids = {"deed_document_id": str, "mortgage_document_id_base": str}
sales = pd.read_csv("build/sales.csv.gz", low_memory=False)
links = pd.read_csv("build/links.csv.gz", dtype=ids, low_memory=False)
flags = pd.read_csv("build/deed_flags.csv.gz", dtype=ids)
d = sales.merge(links, on="sale_id").merge(flags, on="deed_document_id", how="left")
condos = d[(d.property_group == "condo") & (d.deed_match == "unique")]
print(condos.financed_base.mean()) # share with a mortgage recorded near the saleHow to cite
Loschi, P. (2026). NYC residential sales linked to ACRIS deeds and mortgages, 2016–2025 (Version 1.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22977532
@misc{loschi2026linked,
author = {Loschi, Pablo},
title = {NYC residential sales linked to ACRIS deeds and
mortgages, 2016--2025},
year = {2026},
version = {1.0.1},
publisher = {Zenodo},
howpublished = {Data set},
doi = {10.5281/zenodo.22977532}
}The concept DOI always resolves to the latest version. Please also cite the NYC Open Data sources: DOF annualized sales (w2pb-icbu) and ACRIS Master, Legals and Parties (bnx9-e6tj, 8h5j-fqxa, 636b-3b5g).