How we score.
Five sub-scores, 26,295 ZIPs. No black box — the sub-score structure, sources, and normalization are all published. When we change something, we version it.
What goes in, and how it's weighted.
Every ZIP is evaluated across five sub-scores. Each one aggregates several public-data signals, normalizes them into a 0–100 percentile against every other US ZIP, and contributes to the composite with the weight shown below.
How accessible is ownership relative to local incomes, payments, and rents?
Is the underlying economy adding people, jobs, and income quality — or leaking them?
How stretched or fragile does the market look versus its own history and fundamentals? Includes a 10% Physical Risk wedge from FEMA's National Risk Index.
Which way are buyer-seller conditions moving right now, independent of absolute levels?
Place-quality signals — walkability, transit access, and job density, weighed alongside demand fundamentals. Higher Livability typically explains part of the price premium in dense, walkable, transit-rich areas.
A deliberate, ordered blend. The composite HavenScore weighs these five sub-scores in the order shown above. We don't lift the score for markets that feel desirable on top of their Livability contribution — where people are moving and paying up shows up separately in the Demand Dial, so cost and desirability don't get silently blended.
Place quality, weighed.
Livability blends walkability, transit access, and job density (70% of the sub-score) with demand fundamentals (30%). Higher Livability typically explains part of the price premium in dense, walkable, transit-rich areas — so the score gives those markets credit for the lifestyle they deliver instead of treating that premium as pure overvaluation.
Livability emphasizes density and access. Quiet college towns and walkable older suburbs can still score well via the demand-fundamentals weight, but car-only rural ZIPs will score low. Job density runs on Census LEHD LODES (2023 vintage); walkability and transit access run on EPA SLD (vintage frozen at June 2021 until EPA refreshes).
What a new buyer actually pays.
Affordability now uses state-level effective property tax rates from the Tax Foundation (2024) — the rate a typical new buyer at market price would owe. The prior release derived a per-ZIP rate by dividing the Census median property tax paid by Zillow's ZHVI. That ratio measures what existing owners pay on assessments that may be decades old, not what a new buyer faces — a methodologically different question.
Known limitation. State-level rates assume county effective rates track the state average. They do for roughly 45 states. In Cook County (IL), New York City's five boroughs, and the urban New Jersey counties (Hudson, Bergen, Essex), county effective rates diverge from the state mean by about 0.5 to 0.9 percentage points — a HavenScore impact of about one point in those counties. County overrides may ship in v1.2 if user feedback warrants.
Both rates appear on every ZIP detail: Property tax (state-level) is the primary figure and feeds Affordability, while Current owners pay shows the prior Census-derived rate as a transparency cell.
Three personas — and one signal we keep separately.
The composite above is the overall HavenScore. For real decisions, your perspective matters — a buyer and a seller looking at the same ZIP are asking opposite questions. Each persona uses different sub-score weights, and Demand Dial sits separately so cost and desirability never get blended.
Entry cost first. Downside risk second. Everything else, third.
Read methodology →Pricing power and how fast the market clears at today's prices.
Read methodology →Does rent cover carry? How clean is the exit?
Read methodology →Where people are moving and paying up — never baked into the score. Read it alongside the HavenScore, not through it.
Reading the score.
Bands are quintiles of the national ZIP distribution, recomputed each monthly release. A score of 72 means that ZIP performs better than 72% of scored US markets on the composite formula.
Entry costs overwhelm local incomes, or the market looks stretched enough that downside risk dominates.
Pressure is showing in affordability or risk signals. The market is tighter than balanced, not yet stretched.
Roughly middle of the national distribution — no clear structural edge, no glaring warning.
Prices and fundamentals line up reasonably well. A workable entry point for patient, well-underwritten buyers.
Strong on affordability and fundamentals; momentum and risk signals aren't flashing red.
Bands are a reading aid, not a hard threshold. Treat anything near a boundary as a signal to look at the pillar-level detail.
Not one dial. A level, a trajectory, and a liquidity check.
Current buyer-seller pressure from the same signals that feed the score — inventory, days-on-market, price-cut share, sale-to-list, and price acceleration.
The month-over-month change in those same inputs. A Balanced market cooling hard reads as Leaning Buyer — a lead indicator, not a lag.
Fires when inventory stays high, homes sit on market, and prices go nowhere — the signature of a market that isn't clearing at fair value.
The gauge has five states plus the Frozen overlay: Buyer's Market, Leaning Buyer, Balanced, Leaning Seller, Seller's Market. Leaning labels mark transitions inside the central range; once a market has moved into Buyer's or Seller's territory, the level label wins because the transition has already happened. The same data is narrated differently for each persona — a buyer in a Seller's Market reads “expect bidding wars”; a seller reads “maximum leverage.”
Named, attributed, traceable.
Every number on the site traces back to one of the 16 sources below — each named, with its refresh cadence and how it feeds the score. The list is split into the inputs that drive the composite and the datasets we surface for context only.
Composite-score sources
Inputs that feed the five sub-score HavenScore composite. 12 sources, refreshed on documented cadences.
Home values, rents, inventory, transaction indicators across price tiers.
Home value (ZHVI) and rent (ZORI) indices courtesy of Zillow Research. HavenScore is not affiliated with, endorsed by, or sponsored by Zillow Group, Inc.
Active inventory, days on market, price-cut share at the ZIP level.
Listing and inventory data from Realtor.com Economic Research. HavenScore is not affiliated with Move, Inc. or News Corp.
Sale prices, sale-to-list ratio, and transaction volume at the ZIP level.
Data provided by Redfin, a national real estate brokerage. HavenScore is not affiliated with Redfin Corporation.
Household income, population, vacancy, owner-occupied share, and housing cost burden from the 5-year ACS.
Public-domain U.S. Census Bureau data (17 U.S.C. §105). No endorsement implied. Surfaced as the transparency cell labeled “Current owners pay” — median property tax paid divided by median home value.
Workplace Area Characteristics (WAC) job counts at the Census-block level, aggregated to ZIP. 2023 vintage.
Public-domain Census Bureau LEHD LODES Workplace Area Characteristics data (17 U.S.C. §105). No endorsement implied.
Current 30-year fixed mortgage rate used in payment math.
Mortgage rate data from Freddie Mac's Primary Mortgage Market Survey, used with attribution per Freddie Mac's terms.
House Price Index for longer-term price trends.
Public-domain FHFA House Price Index (17 U.S.C. §105). No endorsement implied.
Personal income and growth at the county level.
Public-domain Bureau of Economic Analysis data (17 U.S.C. §105). No endorsement implied.
County unemployment and labor force.
Public-domain Bureau of Labor Statistics data (17 U.S.C. §105). No endorsement implied.
Macro indicators (rates, employment, inflation).
Macro indicators republished by FRED (Federal Reserve Bank of St. Louis). Series-level citations preserved.
Per-ZIP Small Area Fair Market Rents used in rent-vs-buy math.
Public-domain HUD User SAFMR / FMR data (17 U.S.C. §105). No endorsement implied.
State-level effective property tax rates (2024 vintage) — the rate a new buyer would pay, used in the Affordability sub-score.
State-level effective property tax rates from Tax Foundation (taxfoundation.org). Used under fair use with attribution; HavenScore is not affiliated with Tax Foundation.
Enrichment data — context and cross-cutting inputs
These datasets surface as context alongside the score. EPA Smart Location signals (walkability + transit access) and Census LEHD LODES job counts feed the Livability sub-score; the FEMA NRI composite feeds the 10% Physical Risk wedge inside Risk; and IRS migration feeds Livability's demand-fundamentals block. HMDA lending data and the per-hazard FEMA cells (flood, wildfire, hurricane) remain Snapshot-only context and are not blended into the score.
County-to-county migration patterns and average inflowing income (surfaced on the ZIP Context sliver, not blended into the score).
Public-domain IRS SOI migration data (17 U.S.C. §105). No endorsement implied.
Mortgage application volume and denial rate at the ZIP level (surfaced on the ZIP Context sliver, not blended into the score).
Public-domain CFPB / FFIEC HMDA data (17 U.S.C. §105). No endorsement implied.
National Walkability Index from the Smart Location Database (vintage frozen at June 2021).
Public-domain EPA Smart Location Database walkability index (17 U.S.C. §105). No endorsement implied.
National Risk Index — flood, wildfire, hurricane, and composite risk per ZIP.
Public-domain FEMA National Risk Index data (17 U.S.C. §105). No endorsement implied.
EPA Walkability Index vintage is frozen at June 2021 (Census 2019 block groups) — the SLD release cadence is irregular. Job density runs on Census LEHD LODES, refreshed annually (2023 vintage).
How rent is blended across ZIPs.
We use Zillow's rent index where coverage is high and fall back to HUD Small Area Fair Market Rents for smaller or rural ZIPs, so every scored ZIP carries a rent estimate.
No score is better than a misleading one.
We publish a HavenScore only when we have enough signal across pillars to trust the output. Some rural and low-transaction ZIPs don't have sufficient coverage in public data, and we label them rather than paper over the gap.
Every scored ZIP has cleared our coverage threshold across enough sub-scores to support a trustworthy composite.
New scores publish as new data arrives. Historical values stay immutable so trends stay honest.
Sparse ZIPs are marked “insufficient data” instead of given a potentially misleading score.
How we handle missing data. When a single ZIP-level signal is missing, we substitute the containing county or metro average so the pillar can still compute. Pillars that lean meaningfully on those substitutes are labeled with a small Blended badge on the ZIP detail page so the source is visible at a glance. When a whole pillar is too sparse to score we exclude it and recompute from the rest; when coverage is too thin across pillars to produce a trustworthy composite we withhold the score entirely and show the available raw metrics instead.
Forecasts are separate from the score.
Alongside the score, HavenScore publishes 3-, 6-, and 12-month price forecasts from a separate machine-learning model — the HavenScore Forecast Model (v1). Validated against held-out history, shipped with calibrated uncertainty bands.
A tool for informed decisions. Not a guarantee.
HavenScore helps homebuyers read markets with more rigor than gut feel. It is still a starting point, not a conclusion.
Not financial advice
Nothing here constitutes investment advice, a buy/sell recommendation, or a prediction that any specific outcome will occur. Consult a licensed real estate professional and financial advisor.
Not a price prediction
The HavenScore reflects current and recent conditions. A high score today means a market looks relatively strong given what the data shows now — not that prices will rise or risks won't change.
Not a recommendation to act
A low score isn't a vote against a ZIP, and a high score isn't a green light. Personal finances, life stage, and local knowledge sit outside our data.
Not neighborhood quality
HavenScore evaluates market conditions — affordability, growth, risk, momentum, and a Livability sub-score capturing walkability, transit access, and job density. It does not score schools, crime, or community character.
Questions we get a lot.
Still can't find the answer you're looking for? The sub-methodology pages linked above go deeper on each persona.
How often does the HavenScore refresh?
Monthly for price, rent, inventory, and mortgage inputs. Annual for Census demographics, BEA income, and IRS migration. Every ZIP page shows the freshness timestamp for each pillar so you always know how current the underlying data is.
Why isn't desirability baked into the score?
Place-quality fundamentals — walkability, transit, job density — enter the score through the Livability sub-score at 8%. But raw desirability (where people are paying up to live right now) still sits outside the composite in the Demand Dial. Lifting the score for high-demand markets would hide the cost-versus-quality trade-off most buyers are trying to evaluate.
How do you handle ZIPs with thin public data?
If a single pillar is too sparse to score, we drop that pillar and recompute the composite from the remaining pillars with sufficient signal. If coverage is too thin across pillars to produce a trustworthy composite, we withhold the score entirely and surface the raw metrics we do have.
Do the forecasts use the same data as the score?
The forecast model is separate from the HavenScore. It's trained on decades of ZIP-level price history and publishes 3-, 6-, and 12-month price paths with 80% prediction intervals. We validate those intervals on held-out history across every major US region and rate environment — see the forecast methodology page for details.
Can I download the underlying numbers?
Pro plans include CSV export aligned with the public site release. Free readers can inspect every ZIP page, every pillar's sub-score, and the raw inputs that drive each one.
When we change the math, we version it.
- v1.2May 2026Swapped the job-density input from the EPA SLD intersection-density proxy (2021 vintage) to Census LEHD LODES actual workplace job counts (2023 vintage). Walkability and transit access stay on EPA SLD until EPA ships a new release.
- v1.1May 2026Added Livability sub-score at 8% of HavenScore. Added a 10% Physical Risk wedge inside Risk using FEMA's National Risk Index composite. Rebuilt Property Tax to state-level statutory effective rates from Tax Foundation 2024; the prior Census-derived rate now appears as a transparency cell labeled “Current owners pay.”
- v1April 2026First calibrated release. Four sub-scores — Affordability, Growth, Risk, Momentum. 80% prediction-interval coverage verified across every horizon on held-out history, with accuracy broken out by region and rate environment.
Before v1
The math went through several unpublished drafts while we built the validation and calibration infrastructure. v1 is the first release we've verified against real outcomes and stand behind. Earlier work included:
- Iterated on sub-score weights before settling on the original Affordability / Growth / Risk / Momentum balance.
- Separated Demand Dial from the composite score so raw desirability doesn't mask cost.
- Upgraded the forecasting engine to improve accuracy across horizons and markets.
Housing sites sell urgency. We publish the math — one score per ZIP, with the reasoning shown, so you can argue with us instead of guessing at a market.
The HavenScore editorial team