REIT Supplemental Database

Client: Internal • Industry: Capital Markets / REIT Analytics • Completed: May 25, 2026
SEC EDGAR XBRL REIT Analytics Python ETL DuckDB Capital Markets

Challenge

Rigorous cap-rate analysis of public multifamily REITs runs into a data wall. SEC EDGAR’s XBRL feed exposes OperatingIncomeLoss (post-D&A, post-G&A) but not Property NOI — and same-store NOI margin, recurring capex, and regional breakouts live only inside the quarterly supplemental PDFs each REIT files as 8-K exhibits. That leaves analysts choosing between un-rigorous rule-of-thumb margins or expensive paid feeds (NCREIF, Yardi, Green Street). The gap surfaced while drafting the first Compound essay: without structured supplementals, the cap-rate math couldn’t be grounded in primary sources.

Solution

Built a free, transparent, auditable supplemental-disclosure database — built to the standard a multifamily pod at a quant fund would expect: reproducible, queryable, and traceable to source on every record.

  • PDF discovery + extraction pipeline that locates each REIT’s quarterly supplemental 8-K exhibit on EDGAR, archives an immutable copy, and parses tables via pdfplumber with an LLM-vision fallback for non-standard layouts
  • Pod-grade metric set per REIT per quarter: same-store revenue / expenses / NOI, the all-important same-store NOI margin, occupancy, lease trade-outs, recurring vs. total capex ($/unit), regional NOI breakouts, debt structure (WAvg rate, maturity, net-debt/EBITDA), and FFO/AFFO per share
  • Source-citation contract — every metric record carries source URL, local filename + SHA-256, page number, section name, verbatim quoted text, and a confidence level (high / medium / human-verified / absent), so any figure in a Compound essay traces back to the exact page of the exact filing
  • XBRL cross-check reconciling each quarter’s extracted total_revenue against the company’s XBRL Revenues concept (within ±1%), with handling for tagging quirks (e.g. CPT tags property revenue as OperatingLeaseLeaseIncome, not Revenues)
  • Sanity bounds + anomaly log — derived NOI margins validated against a Sun Belt multifamily band (~55–70%); out-of-range or mis-extracted quarters are flagged to a cross-check anomaly log rather than silently passed
  • DuckDB store + per-REIT trackers with parquet exports, sector and entity reference pages, and a query cookbook

Impact

  • Coverage spanning 8 public REITs across multifamily and single-family rental (MAA, UDR, CPT, EQR, AVB, ESS, INVH, AMH) — roughly 280 quarterly periods backfilled from 2010 to present, several thousand source-traced metric records, with zero extraction failures in the latest backfill phases
  • Surfaces the one number XBRL hides — same-store NOI margin — enabling implied-cap-rate analysis grounded in disclosed property income (e.g. MAA implied cap rate of ~6.7% reconciled within the 6–7% range of professional commentary)
  • An adversarial red-team review of the pipeline raised the dataset’s trustworthiness rating from 6.5 to 8.5/10 after every flagged extraction bug was resolved against verbatim XBRL and source-HTML cross-checks
  • Replaces a four-to-five-figure paid-data dependency with a free, inspectable, locally controllable alternative that directly grounds the public Compound research