BasisCRE valuation · NYC public records
When a building’s price moves, everyone says the cap rate moved.
A cap rate is three forces wearing one number.
Debt cost. Income. Required return. Basis reads NYC’s public deed and mortgage record and splits each price move into the three — exactly, to the dollar.
01The hero — exact calibrated decomposition
One real trade, decomposed to the dollar.
118 2nd Avenue sold in 2016, then again 2024 — a −$1.65M move over 7.6 years. The engine prices each deed by the buyer’s own return logic, with the real CEMA-recovered loan, then attributes the move across the three forces. They sum back to the price move exactly.
- LTV
- 83.2% CEMA-recovered
- NOI
- $805K proxy: price × 5.00%
- Base rate
- 1.8% 10y UST
- Debt cost
- 8.8%
- Required return ρ
- 0.47% engine output
- LTV
- 63.0% CEMA-recovered
- NOI
- $1,007K grown 3.00%/yr over hold
- Base rate
- 4.2% 10y UST
- Debt cost
- 7.9%
- Required return ρ
- 15.75% engine output
The verified result is the footing: the three forces reproduce the −$1.65M move to the dollar — that’s the proof the wiring is right. Their split is what the engine’s return logic attributes, not a clean economic measurement: here it reads fundamentals pushing price up $3.59M while the required-return leg absorbs the rest. Debt lands slightly positive ($0.90M) even as base rates rose, because the deal deleveraged — LTV fell 83.2% → 63.0% over the hold. Read the required-return leg as illustrative: the entry ρ leans on a convexity proxy that overpenalizes high leverage.
02The hard read — CEMA recovery
The loan isn’t where the record says it is.
To dodge NYC’s mortgage-recording tax, most large loans are CEMAs: the recorded “mortgage” is a tiny gap note, and the real consolidated balance hides in a separate consolidation agreement (an M&CON) — sometimes beside a stale older one. A naive read of 300 West 44th Street sees no real debt at all.
On the $218.6M purchase, that’s a real 86.9% LTV senior — not all-cash. The parcel also carries a stale 2007 $104.0M M&CON; the engine selects the most-recent active-chain instrument, $190.0M, and reads it from the SODA document_amt field — metadata, not OCR.
03The self-grading loop
It grades its own data before the model sees it.
A gold set of verified extractions gates every scrape: amounts exact to the dollar, blanket agreements forced to refuse. The grader self-tests 17 / 17; the v1 scrape replays 12 attempted + 5 deferred. Bad data never reaches the engine — and the loop refuses to scale on it.
Same-BBL deed pairs surviving the gate, by rate-vintage exit cohort — every row of attrition shown, nothing hidden:
Four populated cohorts, every bucket ≥ 14 survivors. The waterfall is the honesty artifact — the model only decomposes what survived the gate.
04The signal — and where it goes
The rate force grows into the high-rate regime.
The rate force — real dated Treasury rates against the real recovered LTV, the one proxy-free leg — takes a larger share of higher-for-longer exit moves than of pre-2020 ones. The thesis direction, read straight off the public record.