Properties likely to list — before the “For Sale” sign goes up.
Cox Proportional Hazards Model (Cox 1972) blending probate filings, utility disconnects, lis pendens, cosmetic permits, mortgage payoffs, tax-assessment jumps, and tenure into a per-property hazard ratio. Each property gets a 0-100 composite + a predicted listing-window bucket (0-30d / 30-90d / 90-180d / 180-365d). Knock the door, send the postcard, or call the listing agent — before the competitor does.
| Rank | Address | Contacts | Window | Top driver | Tenure | Last sale | HR | PL | Actions |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Folly Beach · 29425 · Folly Beach | 30-90d | Active lis pendens | 6.4y | $889k | 4.31 | 73 | ||
| 2 | Sullivan's Island · 29420 · Sullivan's Island | 30-90d | Active lis pendens | 9.5y | $2.59M | 4.22 | 72 |
We use the canonical industry algorithm for time-to-event prediction: Cox, D.R. (1972), “Regression Models and Life-Tables”, Journal of the Royal Statistical Society, Series B, 34(2), 187-220. The hazard h(t) — the instantaneous probability of listing at time t given the property has not yet listed — is modeled as a baseline rate multiplied by the exponential of a linear combination of covariates:
h(t | x) = h₀(t) · exp(β₁·x₁ + β₂·x₂ + ... + βₖ·xₖ) HR = exp(Σ βᵢ·xᵢ) ← “hazard ratio”
Each signal we collect contributes a multiplicative hazard ratio (HR > 1 = above baseline). Composite HR is the product of active signals; it's mapped to a 0-100 score for UI convenience and bucketed into a predicted-listing window via discrete-time survival approximation (Allison 1982).
- probate · HR 3.5 — NAR Profile of Home Buyers and Sellers (annual). Estate sales cluster within 12-18mo of probate filing.
- distress · HR 2.8 — Foote-Gerardi-Willen 2008 (FRB Boston): foreclosure-pre-sale listing rate.
- utility_disconnect · HR 2.4 — Federal Reserve Survey of Consumer Finances mobility correlate.
- cosmetic_permit · HR 2.0 — Sirmans-Macpherson 2006 listing-prep / staging research.
- insurance_pressure · HR 1.8 — Post-2022 SC carrier-pause and non-renewal forced-sale data.
- payoff_signal · HR 1.7 — clear-title-pre-sale RMC pattern (broker observation, post-2024).
- tenure (7-15yr peak) · HR 1.5 — JCHS Harvard tenure-conditional hazard estimates.
- assessment_jump · HR 1.4 — NAR seller-survey "equity-tap" motivation.
- title_event · HR 1.4 — probate-adjacent recording flurry (broker observation).
- comparable_sales · HR 1.3 — Genesove-Mayer 2001 prospect-theory FOMO neighborhood effect.
- hood_velocity · HR 1.2 — zip-level listing density (broker calibration).
Baseline 90-day hazard h₀ = 1.5%, derived from US Census ACS tenure distributions + NAR annual listing rate. Recipe + weights: apps/worker/processing/pre_listing.py. Coefficients will be refit via supervised learning once we have ≥6mo of MLS-listing labels.
ECOA-clean: no demographic data, no school-district scores, no credit-score inputs. Property events only — same hygiene as the rest of Pulse.