Methodology

Base Method

The basis for the methodology is 2024's Horizontal holdings: untangling the networks of corporate landlords by Taylor Shelton and Eric Seymour. They were neither consulted about nor involved in this extension of their wonderful research.

Data sources

Parcel records come from Fulton County and DeKalb County tax assessment data, obtained as GIS exports. Each record includes the parcel geometry, owner name and mailing address, land use code, acreage, and living units. The two datasets are unified into a single table of 600k+ parcels.

Georgia Secretary of State business registration data (bulk download) is matched against owner names to identify registered corporations, LLCs, and other entities. Matching uses a hybrid SQL + fuzzy approach, with manual review of ambiguous cases. 50k+ owner entities have been linked to SOS records.

Building complaints come from the City of Atlanta's Accela permitting system via the public records API. Records cover Building Code Complaints from January 2020 forward. Complaints are linked to parcels via spatial join against the city's address point and Tax Parcel datasets.

Corporate and institutional flags

Every owner name is tested against a set of patterns to assign one or both flags:

CORPORATE — owner name contains indicators of a business entity: LLC, LP, Inc., Corp., Co., Trust, Holdings, Properties, Realty, Investments, Fund, Partners, Ventures, and others. Also matches known patterns for single-family rental aggregators and out-of-state investment vehicles.

INSTITUTIONAL — owner is a government body, university, hospital, religious organization, HOA, or similar non-profit entity. Matched by name pattern and, where available, confirmed via SOS business type.

These flags have known gaps:

• Some abbreviations without punctuation are not caught (e.g. GEORGIA POWER CO without a period).
• Subdivision and condo association names sometimes appear as owner names in the raw data — these look like institutional owners but are data artifacts.
• Individual landlords with large portfolios are not flagged unless they hold property through a named entity.

Ownership clustering

Many landlords hold property across dozens of separately-named LLCs. To surface these networks, owner entities are linked into clusters when they share:

• A normalized mailing address (after stripping PO box numbers and standardizing via libpostal)
• A Georgia SOS registered agent, where the agent is a private individual or small firm rather than a national corporate agent
• A SOS principal office address

Linked entities form a graph; connected components become clusters. The algorithm uses a multi-pass approach to prevent large national agents (CT Corporation, Cogency Global, etc.) from collapsing unrelated owners into a single mega-cluster.

Which is to say - since clustering is imperfect, this implementation leans toward not over-linking owners.

Address normalization

Owner mailing addresses are initially normalized using libpostal, an open-source address parser. Additional work is done to support PO Boxes because libpostal typically strips them, which would incorrectly connect all owners using the same post office.

Limitations

This data reflects public records as of the last pipeline run. It does not reflect ownership changes that have occurred since then. Parcel boundaries and owner assignments come from county assessor records, which may lag behind actual deed transfers. This tool is for informational purposes; it is not a substitute for a title search.