Methodology
HavenScore computes a 0–100 composite score for every US ZIP code using free, publicly available data. This page documents exactly how scores are calculated, what data we use, and how often it updates.
Data Sources
All data is sourced from free, authoritative public sources with no proprietary or paid data feeds.
Zillow ZHVI (Zillow Home Value Index)
Median home values and year-over-year price growth by ZIP code. Published monthly.
Zillow ZORI (Zillow Observed Rent Index)
Median asking rents and year-over-year rent growth by ZIP code. Published monthly.
Realtor.com Research Data
Active inventory, days on market, and price cut share by ZIP code. Published monthly.
Redfin Market Tracker
Median sale prices, sale-to-list ratios, and transaction volume by ZIP code. Published monthly.
US Census Bureau ACS (American Community Survey)
Median household income, population, demographics, vacancy, cost burden, and education levels. Updated annually.
Freddie Mac PMMS (Primary Mortgage Market Survey)
Current 30-year fixed mortgage rates. Updated weekly via FRED API.
FRED (Federal Reserve Economic Data)
Federal funds rate, yield curve, consumer sentiment, housing starts, CPI, delinquency rates, and other macro indicators. Updated monthly.
BLS (Bureau of Labor Statistics)
County-level unemployment rates and employment growth. Updated monthly.
FHFA (Federal Housing Finance Agency)
House Price Index (HPI) — 1-year and 5-year price changes by 3-digit ZIP area. Updated quarterly.
BEA (Bureau of Economic Analysis)
Per-capita personal income and income growth by county. Updated annually.
IRS SOI Migration Data
County-level net migration rates and income quality of movers. Updated annually.
The HavenScore
The HavenScore is a weighted composite of four sub-scores, each normalized to a 0–100 percentile scale against all scored ZIP codes nationally. A score of 75 means that ZIP performs better than 75% of ZIP codes across all four dimensions.
HavenScore = (Affordability × 0.40) + (Growth × 0.25) + (Risk × 0.20) + (Momentum × 0.15)
Affordability
40%- •Price-to-income ratio relative to national distribution
- •Monthly payment at current 30yr rate as % of income
- •Property tax rate
- •Rent-to-price ratio — higher = better value for buyers
- •Poverty rate and housing cost burden
- •Sale-to-list ratio — are buyers paying above or below asking?
Growth
25%- •YoY home price change — Zillow + FHFA confirmation
- •Employment growth — BLS
- •Net migration rate — IRS migration data
- •Population growth — Census ACS
- •Building permits per capita — FRED
- •Income quality — BEA income growth + IRS income inflow/outflow
Risk
20%- •Overvaluation vs ZIP historical avg price-to-income
- •Housing cost burden — severely burdened owners and renters
- •Price cut share — signal of seller distress
- •Vacancy rate, poverty rate, unemployment rate
- •HPI vs ZHVI divergence — when government and private data disagree
Momentum
15%- •Inventory change — acceleration or deceleration
- •Price acceleration — are gains speeding up or slowing?
- •DOM velocity — rate of change in days on market
- •Migration trend — demand momentum signal
How Sub-Scores Work
Each sub-score input is normalized to a percentile rank against all other US ZIP codes with sufficient data. A ZIP code with a price-to-income ratio in the 80th percentile nationally (meaning 80% of ZIPs are less affordable) receives a low Affordability component score. Inputs are inverted where needed so that higher always means better.
Sub-scores are then weighted and summed to produce the final HavenScore. ZIPs with insufficient data for a sub-score have that sub-score excluded and the remaining weights are proportionally redistributed. ZIPs are only scored when at least three of four sub-scores can be computed.
Risk sub-score is inverted: a ZIP with very low risk (few price cuts, inventory near historical average, DOM stable) receives a high Risk sub-score. Higher is always better across all four sub-scores.
Buyer/Seller Market Gauge
The Buyer/Seller Gauge is a separate indicator from the HavenScore. It measures current market conditions on a spectrum from −100 (Strong Buyer's Market) to +100 (Strong Seller's Market). Zero is a balanced market.
| Input Signal | Buyer Signal | Seller Signal | Weight |
|---|---|---|---|
| Price trend (YoY) | Declining or flat | Rising fast | 20% |
| Active inventory | Above average | Below average | 17% |
| Days on market | Longer than usual | Shorter than usual | 13% |
| Price cut share % | High (many cuts) | Low (few cuts) | 13% |
| Price acceleration | Decelerating / negative | Accelerating | 12% |
| Employment conditions | High unemployment | Low unemployment | 9% |
| Income quality of movers | Outflow of wealth | Inflow of wealth | 8% |
| Sale-to-list ratio | Below asking | Above asking | 8% |
Update Schedule
- Monthly: HavenScore, all sub-scores, Buyer/Seller Gauge, home values, rents, inventory, DOM, price cuts, unemployment rates, employment growth
- Weekly: Mortgage rates (Freddie Mac PMMS via FRED API)
- Quarterly: FHFA House Price Index (HPI 1yr/5yr changes)
- Annually: Census ACS (income, population, tax rates, permits, vacancy, cost burden), FEMA risk scores, IRS migration, HMDA denial rates, BEA income growth
Data freshness timestamps are displayed on every ZIP detail page. Scores are never shown without indicating when they were last computed.
Price Forecast
HavenScore generates 3-month, 6-month, and 12-month price forecasts for every scored ZIP code using machine learning trained on 26 years of historical market data across 33,000+ ZIP codes.
The forecast model considers 80+ features per ZIP including local market conditions, regional economic indicators, national macro signals (interest rates, yield curve, consumer sentiment), demographic trends, and spatial relationships between neighboring markets.
Each forecast includes a confidence tier (High, Medium, Low) based on model certainty and data availability for that ZIP, plus prediction intervals showing the expected range of outcomes.
Why this forecast?Every prediction is accompanied by a breakdown of the top factors driving it — which features are pushing the price up or down. This uses SHAP (SHapley Additive exPlanations), a technique from game theory that measures each feature's individual contribution to the prediction.
The model is validated through walk-forward backtesting across multiple market cycles including the 2008 financial crisis and the 2022 rate correction. Accuracy metrics are computed on true out-of-sample predictions — the model never sees future data during validation.
Limitations
Forecasts are not financial advice. While our ML model is validated across 26 years of market history, past performance does not guarantee future accuracy. Real estate markets can be affected by events no model can predict.
Coverage gaps. Some rural ZIP codes have insufficient transaction volume for Zillow or Realtor.com to publish reliable estimates. These ZIPs are marked as having limited data and may show fewer metrics or no score.
Overvaluation % is relative. It compares the current price-to-income ratio to the long-term average for that specific ZIP — not to some national ideal. A high overvaluation % in a desirable city does not automatically mean prices will fall.
Income data lags. Census ACS income data is typically 12–18 months old. We use the most recent available data and label it accordingly.