How Sell Signal Works
Sell Signal scores approximately 1.1 million NYC residential properties on their likelihood of selling within the next 12 months. Six independent scoring layers surface the properties most likely to transact — so you know who to call and when.
The Base Rate Problem
In a typical year, only 30,000–50,000 of NYC's 1.1 million residential properties change hands — a base rate of just 3–4.5%. A model that predicts "no sale" for every property would be 96% accurate. Sell Signal has to dramatically beat this by concentrating high scores on the 3–4% that actually transact.
6-Layer Scoring Architecture
Each layer captures a different category of sell signals, scored independently, then combined into a composite probability score. This modular design means layers can be refined or added without redesigning the core engine.
L1: Ownership & Equity
How long the owner has held the property and how much unrealized equity they've accumulated. The median hold before sale in NYC is ~9 years.
L2: Property & Building Events
External pressure events that create sell motivation — expiring tax abatements, assessment spikes, distress indicators.
L3: Market Context
Is this a good time to sell in this specific location? These signals don't predict who will sell, but when selling is attractive.
L4: Listing & Transaction History
The most directly predictive signals — this owner has already demonstrated intent to sell.
L5: Life Events & Demographics
Life changes that commonly precede residential transactions — court filings, age, relocation, family changes.
L6: Behavioral & Digital
Digital behavior suggesting an owner is thinking about selling. Being built incrementally.
Score Tiers
| Tier | Score | P(Sale) |
|---|---|---|
| Very High | 80–100 | >40% |
| High | 60–79 | 15–40% |
| Moderate | 40–59 | 5–15% |
| Low | 20–39 | 2–5% |
| Very Low | 0–19 | <2% |
Why NYC-Specific Matters
National propensity models miss the signals that matter most in New York. Sell Signal is built on NYC-specific data that no national model can access:
Data Sources
| Data | Status |
|---|---|
| Ownership history & transfers | Live |
| Sale prices & dates | Live |
| Mortgage recordings | Live |
| Owner name & LLC detection | Live |
| Assessed value & tax class | Live |
| Listing history (expired, withdrawn) | Live |
| Building sales velocity | Live |
| 421a / J-51 abatement schedules | Live |
| Price reductions | Live |
| Failed contracts | Live |
| Rent stabilization status | Live |
| Comp pricing gap | Coming |
| Neighborhood price momentum | Coming |
| Days on market trends | Coming |
| Lis pendens / pre-foreclosure | Coming |
| Tax liens | Coming |
| DOB permits (nearby construction) | Coming |
| Buyer demand (saves / favorites) | Future |
| Divorce filings | Future |
| Probate / estate proceedings | Future |
| Owner estimated age | Future |
| Relocation signals | Future |
| Owner searches own address | Future |
| Valuation tool usage | Future |
| Agent inquiry signals | Future |
Early Validation
Tested on 26 real NYC properties across three cohorts — the algorithm correctly orders sold properties above expired listings above stable long-term holds:
Cohen's d: 0.73 (good effect size) · Label-Score Alignment: 95.9% · Ordering: SOLD > EXPIRED > STABLE ✓