Methodology
Market divergence
The HousingGauge score rates current buying conditions in a market. Divergence measures something else: how differently the ZIP codes inside a region have appreciated. A third question, whether lagging ZIP codes later catch up, is a research question, tested separately and published with its evidence.
Divergence
For each eligible ZIP code, annualized appreciation over a window of L years is (V_t / V_(t−L))^(1/L) − 1, from the Zillow Home Value Index (middle tier, smoothed, seasonally adjusted). Divergence is the population standard deviation of those rates across a region's ZIP codes, in percentage points: √(Σ w (g − ḡ)² / Σ w), where ḡ is the weighted mean (the region benchmark). The default weight w is the ZIP's housing units (ACS); equal weights are shown alongside. Windows: 1, 3, 5, 10 years, computed at every quarter-end since 2001 and at the latest month.
Each region also reports the 10th, 50th and 90th percentile ZIP growth, the 90th–10th spread, the share of ZIP codes with falling values, the share beating the region, and the share beating it by five points or more.
A stable cohort
Historical divergence uses only ZIP codes with a value every month from January 2000 to the latest month, so changes over time reflect markets rather than ZIP codes entering the data. Each region page states how many ZIP codes that is and their share of the region's housing. ZIP tables and maps use every eligible ZIP code with values at both ends of a window.
Divergence Index
A 0–100 percentile: where a region's reading ranks among every region's readings at that date or earlier (same window and weighting). It uses no later data, and regions with fewer than 20 cohort ZIP codes get no index, so small samples don't produce extreme ranks. A high index means unusual geographic variation. It is not a buy or sell signal.
Which ZIP codes count
- ZIP Code Tabulation Areas (the Census approximation of postal ZIP codes) assigned to the county holding most of their land, and that county's metro (2023 delineations).
- At least 1,000 residents and 400 housing units, at least 50% of land inside the metro, and a current Zillow value.
- Everything else is shown as insufficient data and left out of statistics.
Convergence is not catch-up
- Growth convergence: ZIP growth rates become more similar. It can happen while every ZIP code falls together.
- Price-level convergence: cheaper ZIP codes close part of their price gap. Equal growth rates don't close gaps.
- Catch-up: earlier laggards beat their region afterward. A ZIP rising 4% a year while its region rises 8% did not catch up.
Turning points
A turning point asks a different question from divergence: not how far apart a region's ZIP codes are, but whether one ZIP code's long divergence from its region is reversing. Everything comes from one series per ZIP code, its relative price: its typical value divided by a benchmark index of its region (each month's benchmark growth is the housing-weighted average growth of the region's ZIP codes). Relative price falls while a ZIP code trails its region and rises while it outpaces it.
- Pent-up gap: annualized change in relative price over the 5 years ending a year ago (also shown over 10 years).
- Last 12 months and the 12 months before: change in relative price over each year, in percentage points.
- Lagged: bottom quarter of the region on the pent-up gap and at least 1 point a year behind. Led: top quarter and at least 1 point a year ahead. Regions need 12 ZIP codes with complete data.
- Turning up after a long lag: lagged, still losing ground the year before, and keeping pace or better over the last 12 months, so its relative price has stopped falling and started rising. The other phases: catching up (two years at pace or better), lag narrowing (still trailing, by less), still falling behind, leader turning down, leader losing ground, leading, and near the region.
- Context: 12-month growth of the region's and the nation's benchmark indexes, and the region's divergence cycle; divergence has settled after a peak when the 5-year dispersion is at or below 70% of its peak over the prior 10 years.
- Strength orders lagging ZIP codes for lists: depth of the lag, the swing from last year to this year, and regional and national growth (weights 0.35, 0.35, 0.15 and 0.15). It is an ordering, not a score or a forecast.
The research engine classifies every year end since the 1980s with the same code and measures what happened next, on both sources, against random picks. A turning point is shown as a description of the last six years unless that test validates it.
New businesses and incomes
Prices are only half of a neighborhood turn; the other half is the neighborhood itself. HousingGauge tracks, for every ZIP code it covers:
- Business establishments (Census ZIP Codes Business Patterns, every year since 1994) and restaurants, cafés and bars (NAICS 722, since 1998).
- Income per tax return and the share of returns over $100,000 (IRS Statistics of Income by ZIP code, tax years since 2011).
- Adults with a bachelor's degree and residents aged 25–34 (American Community Survey 5-year estimates, 2010–14 and the latest release).
Each is measured as growth over the latest 3 published years minus the region's growth (sums over the region's ZIP codes), so national trends, including inflation in nominal incomes, cancel out. Business data are published about 2 years after the year they describe and IRS data about 3; the research uses only years that had been published by each date. ZIP codes need at least 20 establishments (5 restaurants and bars) or 500 tax returns at the start of the window. A turning point is confirmed by new businesses when the ZIP is turning up and its establishments grew faster than its region's.
The research engine
Model convergence-v3 tests seven questions with two independent sources: FHFA's annual repeat-sales ZIP index (back to the 1980s for many ZIP codes) and Zillow's ZHVI (from 2000). For every start year it uses only information available then: prior appreciation, price levels, the region's own divergence history, lagged metro employment (BLS) and the mortgage rate (FRED). Outcomes are realized appreciation over the next 1, 3, 5 years; nothing is forecast, and a start year contributes only once its whole horizon has been observed.
- Rules were fixed before evaluation: quartiles of prior metro-relative growth; high divergence = at or above the 75th percentile of the region's own history; affluent = top 25% of ZIP prices; adjacency = ZIP centers within 5 miles; boom = metro growth of 8% or more; affordability shock = the 30-year rate up 1.5 points or more in a year.
- Each region-year counts once. 95% intervals come from a circular block bootstrap over start years (block = horizon) and are omitted below eight start years. A random-selection null shows what picking ZIP codes at random would have produced.
- Results are reported for start years up to and after 2010, so a pattern must hold in both halves.
- A predictive Catch-Up Potential label is published only if a predetermined rule beats the random null, has an interval above zero and holds in both halves, in both sources.
Sources and licenses
- Zillow Research: ZHVI and ZORI by ZIP. Free with attribution; commercial terms are being verified.
- FHFA: annual ZIP5 house price index, developmental, U.S. government work.
- U.S. Census Bureau: ACS 5-year ZCTA statistics, 2020 ZCTA–county relationships, gazetteer coordinates.
- NBER: county-to-CBSA crosswalk (OMB 2023 delineations). BLS LAUS metro employment. FRED 30-year mortgage rate. IRS Statistics of Income ZIP data (coverage research only).
Limitations
- Appreciation is not investor return: rent, vacancy, costs, taxes and financing are excluded.
- Zillow's ZHVI is smoothed and its history follows Zillow's current methodology; FHFA's ZIP index includes appraisals. Both revise history, so each research run records file hashes.
- Housing-unit weights come from the latest ACS, a small look-ahead for weighted history; equal-weighted figures avoid it.
- ZCTAs and county-based metro assignment approximate postal ZIP codes and their metros.
- No analysis uses race, ethnicity or any other protected characteristic.
Market divergence dashboard · Research report · HousingGauge score methodology
