How the housing map projections are built and scored
Beta, October 30, 2026 Updated October 31, 2026. Model version v2-2026-10, inputs as of September 2026. Model estimate, not a guarantee or advice.
What the map projects
Two things, each with its own track record. First, a 12-month projection of the typical home value (Zillow ZHVI) for every Florida county and for 942 Florida ZIP codes: an expected change, a range, and the odds that the value is higher in 12 months (September 2026 to September 2027). Second, 3-month direction odds for the median sale price, days on market, new contracts and price cuts, taken from Momentum Realty's monthly Market Pulse for the 13 MLS markets it covers, mapped to the 148 counties inside those markets.
The projection layer is in beta as of October 30, 2026: every call is logged and will be scored monthly against what Zillow later reports, and the method will change if the scores say it should. Both layers describe prices and market conditions only. They are not an appraisal, not an automated valuation of any specific property, and not advice on where to live. Markets and prices, nothing about neighborhoods or the people who live there.
The 12-month county model, in plain English
Beta (2026-10-30). A ridge regression fitted on Florida counties one month at a time: 17071 county-months from 2002-12 to 2025-09 (every month whose 12-month outcome is known). Each row uses only what was known in that month: the trailing 12-month home value change, the 12 months before that, the last 3 months, the share of listings with a price cut, inventory versus 12 months earlier, and how far the value sat above or below a log-linear trend fitted on that month and earlier only (expanding window from January 2000, so the 2006-2011 cycle is in the fit). The target is the true 12-month point-to-point change in the Zillow typical home value (ZHVI). Price cuts and inventory are only available from 2018; earlier rows carry the training mean plus a missing flag. Accuracy is measured leave-one-year-out by the forecast origin year: the model is refitted without a year, then asked to predict every month of that year. The range and the direction odds come from those out-of-sample misses. Live inputs outside the training range are pulled back to the nearest training value. The fitted intercept over the full sample is +4.6%. Out of sample the model is close to unbiased: average miss +0.2 points (predicted minus actual); no shift is applied. Sanity gate: where Momentum's Market Pulse covers the county and its 3-month median price odds point the other way (model above +5%% while the Pulse gives 60%%+ odds of flat or lower, or the reverse), the county is marked "model under review": its odds are shown, its expected change is not, its range is widened to 10-90%%, and the drivers say so. Known limits: adjacent years share overlapping 12-month windows, so the year-by-year test is a little less independent than it looks; mortgage rates, insurance, storms and migration are not inputs; the horizon is exactly 12 months but the live value is the latest month available.
Inputs and what the fitted model does with them
Coefficients are shown per standard deviation of the input (how much the expected 12-month change moves when that input is one standard deviation higher) and per unit (per percentage point). The fitted intercept over the full 2002-2025 sample is +4.6%. The last column is the range of each input seen in training; live inputs outside it are pulled back to the nearest edge. Missing flags are 1 for months before the price-cut and inventory series begin.
| Input | Key | Per std. dev. | Per point | Training range |
|---|---|---|---|---|
| Home value change, trailing 12 months | g12 | +1.21 | +0.106 | -33.21 to 44.89 |
| Home value change, months 12 to 24 back | glag | -2.31 | -0.204 | -33.21 to 44.89 |
| Home value change, trailing 3 months | mom3 | +7.62 | +2.508 | -11.37 to 11.72 |
| Value vs. log-linear trend fitted on months up to that one only (expanding window from 2000) | gap | +0.81 | +0.034 | -65.83 to 61.90 |
| Share of listings with a price cut (month) | cuts | -1.30 | -0.200 | 3.94 to 39.22 |
| Price-cut share missing (before 2018-03) flag | cuts_na | +0.36 | +0.764 | 0.00 to 1.00 |
| For-sale inventory vs. 12 months earlier | inv | -2.14 | -0.069 | -63.33 to 336.36 |
| Inventory missing (before 2019-03) flag | inv_na | -0.74 | -1.607 | 0.00 to 1.00 |
Reading it: recent momentum (the last 3 and 12 months) carries the most weight; a strong prior year tends to be followed by a weaker one; more price cuts and rising inventory pull the projection down. These are tendencies measured across 17071 county-months, not laws.
Back-test by year (out-of-sample)
For each origin year the model was refitted without that year and then asked to predict every county-month in it. "Direction right" is the share of county-months where the sign of the projection matched what happened 12 months later. "Inside the range" uses a range built only from the other years' misses, so it is a fair test. The miss column is the median error in percentage points. Rows under 60% direction are shaded. Out-of-sample average miss (predicted minus actual): +0.2% points; close enough to unbiased that no shift is applied to the point estimate.
| Year predicted | Counties | Direction right | Inside the range | Median miss | Actual avg | Predicted avg |
|---|---|---|---|---|---|---|
| 2002 | 51 | 100% | 69% | 5.3 pts | +11.4% | +6.6% |
| 2003 | 607 | 99% | 31% | 6.9 pts | +14.4% | +7.6% |
| 2004 | 602 | 99% | 4% | 12.2 pts | +22.8% | +10.6% |
| 2005 | 614 | 100% | 36% | 8.1 pts | +19.4% | +16.7% |
| 2006 | 623 | 48% | 10% | 11.2 pts | -2.6% | +9.1% |
| 2007 | 714 | 86% | 10% | 10.2 pts | -14.7% | -4.7% |
| 2008 | 732 | 100% | 46% | 6.4 pts | -16.7% | -10.3% |
| 2009 | 732 | 97% | 92% | 2.5 pts | -8.7% | -7.6% |
| 2010 | 739 | 29% | 16% | 8.8 pts | -7.7% | +1.4% |
| 2011 | 743 | 67% | 80% | 2.6 pts | -1.4% | -3.0% |
| 2012 | 759 | 73% | 55% | 5.1 pts | +7.0% | +3.0% |
| 2013 | 769 | 93% | 92% | 2.3 pts | +8.1% | +8.3% |
| 2014 | 791 | 90% | 86% | 2.9 pts | +6.6% | +5.3% |
| 2015 | 792 | 97% | 84% | 3.0 pts | +8.0% | +6.1% |
| 2016 | 792 | 97% | 95% | 1.9 pts | +7.0% | +6.3% |
| 2017 | 792 | 95% | 87% | 2.1 pts | +7.7% | +5.8% |
| 2018 | 792 | 99% | 92% | 1.8 pts | +6.4% | +6.0% |
| 2019 | 804 | 98% | 92% | 2.1 pts | +7.6% | +6.5% |
| 2020 | 804 | 100% | 26% | 8.9 pts | +18.2% | +8.9% |
| 2021 | 804 | 100% | 51% | 5.6 pts | +23.1% | +18.6% |
| 2022 | 804 | 72% | 24% | 14.5 pts | +3.9% | +16.7% |
| 2023 | 804 | 56% | 59% | 4.3 pts | +2.4% | -2.3% |
| 2024 | 804 | 40% | 57% | 5.6 pts | -2.9% | +3.1% |
| 2025 | 603 | 72% | 90% | 2.4 pts | -2.4% | -2.1% |
The honest reading of that table: the model is good at the ordinary year and bad at the turning points. It did not see the 2006 top coming (48% direction right) nor the 2010 relapse (29%), it under-called the 2004-2005 and 2020-2021 surges, over-called 2022 (it expected the run to continue), and it was wrong-footed again in 2024 (40%). It called 2007-2009 and 2025 correctly. Direction was right in nearly every county in the years when nearly every county moved the same way, which flatters the headline number.
Live calls and the public scoreboard
Every projection the map publishes is written to an append-only ledger on the day it is made (level, area, expected change, range, odds, review flag, model version). Live calls will be scored monthly: once 12 months have passed, each call is scored against the realized Zillow value and the result is published here and on the map. Back-test numbers and live numbers are always labeled separately.
Sanity gate against the near-term layer
Where a Momentum Market Pulse market covers the county, the 12-month call is checked against the Pulse 3-month median price odds. If the model says more than +5% while the Pulse gives 60% or better odds of flat or lower (or the reverse), the county is marked "model under review": the map shows its odds but not its expected change, the range is widened to the 10-90% band, and the drivers say "Near-term indicators disagree with the 12-month model." Counties under review in this run: Miami-Dade County, Indian River County, De Soto County, Martin County, Collier County, Monroe County, Broward County, Charlotte County, Manatee County, Palm Beach County, Lee County, Sarasota County, Pinellas County.
Live calls scored so far: 0. 1009 projections are on the ledger waiting for their horizon (first scorable: 2027-09). Until then the scoreboard shows the back-test above, labeled as such.
The 3-month odds (Momentum Market Pulse)
Momentum Realty's Market Pulse scores each MLS market it covers from 0 to 100 using four parts: supply (months of inventory), speed (days on market), over-list sales (share closing above list) and price cuts (share of active listings with a cut). 0 is a strong buyer advantage, 100 a strong seller advantage; the map calls this the market balance score. The score has been computed back to 2005 where the MLS history allows, and the next-12-month median price change that followed each score tier is reported as the tier average. The 3-month odds for median price, days on market, new contracts and price cuts are set from the current score, its direction and the contract pipeline, and are published monthly in each market's Pulse report. Counties outside a Pulse market show "not covered". MLS data is deemed reliable but not guaranteed.
| Market | Month | Score | Label | Back-test |
|---|---|---|---|---|
| BeachesMLS | 2026-09 | 48 | Balanced, soft | 256 months |
| Daytona Beach Area MLS | 2026-09 | 23 | Moderate buyer advantage | 256 months |
| Emerald Coast MLS | 2026-09 | 34 | Moderate buyer advantage | 256 months |
| Florida Keys MLS | 2026-09 | 11 | Strong buyer advantage | 256 months |
| Georgia MLS | 2026-09 | 58 | Balanced, firm | 37 months |
| ScoutMLS (Golden Isles) | 2026-09 | 23 | Moderate buyer advantage | 256 months |
| Florida Gulf Coast MLS | 2026-09 | 30 | Moderate buyer advantage | none yet |
| MIAMI MLS (SEFMLS) | 2026-09 | 38 | Balanced, soft | none yet |
| NEFAR MLS (realMLS) | 2026-09 | 44 | Balanced, soft | 256 months |
| Central Panhandle MLS | 2026-09 | 34 | Moderate buyer advantage | none yet |
| Pensacola MLS | 2026-09 | 52 | Balanced, firm | 253 months |
| Space Coast MLS | 2026-09 | 51 | Balanced, firm | 256 months |
| Stellar MLS | 2026-09 | 52 | Balanced, firm | 70 months |
The other scores on the panel
12-month projection (0-100): the model's odds that the typical home value is higher in 12 months, computed from the out-of-sample misses (the share of past misses that would still leave the expected change above zero). The expected change and the range sit under the number; the price direction composite and its parts (recent appreciation, days on market, inventory, price cuts) are shown inside the parts.
Affordability (0-100, higher is more affordable): the average of two clamped linear parts. Price-to-income: a typical value of 2.5 times area median income scores 100 and 7.5 times scores 0. Payment burden: a monthly cost of 15% of area median income scores 100 and 55% scores 0 (principal and interest at 20% down and the current mortgage rate, plus insurance and tax estimates where available). Income is from the Census ACS 5-year estimate and is used only inside the ratio.
Momentum, long-term growth and investor scores are described in the map's "About the data" panel; they are unchanged.
Valuation, two lenses. Trend lens (the "Overvaluation vs. own long-run trend" bar and map metric): a log-linear trend is fitted to the area's own monthly Zillow ZHVI from January 2000 (or its first month) through the latest month, and the gap is the latest value divided by the fitted trend value, minus one. It is the same fit the projection model uses for its trend-gap feature at the latest month and the same line drawn on the panel's value chart, so the three agree. Within 5% of trend reads "fairly valued", 5% to 15% "moderately over- or undervalued", beyond 15% "well above or below trend". ZIPs with under 60 months of history show no trend read. Income lens (the "Price-to-income vs. state median" map metric): the area's typical value divided by its Census ACS median household income, compared with the state's median ratio; the verdict text reads it as comfortable (under 3.5x income), stretched (3.5x to 5x) or very stretched (over 5x). The two lenses can disagree: a retiree, second-home or high-wealth market can sit near its own trend while looking very stretched against local paychecks, which is exactly why both are shown.
Momentum MLS (live)
The "Momentum MLS (live)" metric group and the panel block of the same name come straight from the MLS feeds Momentum Realty subscribes to (BeachesMLS, Central Panhandle MLS, Daytona Beach MLS, Emerald Coast MLS, Florida Gulf Coast MLS, Florida Keys MLS, Georgia MLS, MIAMI MLS, Pensacola MLS, Scout MLS, Space Coast MLS, Stellar MLS, Tallahassee MLS, realMLS (NEFAR)). Three sources are combined for each county: the daily county truth file (active listings, pending or under-contract listings, median list price, list price per square foot, days on market of current listings, closings in the last 12 months), the Market Pulse cache (closed sales by month: count, median sold price, days on market of sold homes, sale-to-list ratio, share sold over list, cash share; the 90-day figures blend the latest three monthly medians weighted by closings, and "vs. 1 year ago" compares the same three months a year earlier) and the hourly feed events (new listings and new pendings in the last 7 days, price cuts in the last 30 days, each listing counted once per event kind). Months of supply is active listings divided by average monthly closings over the last 90 days. City dots show active, pending and median list price for cities with 25 or more active listings in our feeds.
Thin-data rule. A county shows live figures only when our feeds carry at least 25 active listings and at least 6 closings in the last 90 days (or 12 months where the 90-day count is missing). Below that the panel says "thin data" and names the feed and the counts rather than showing figures that would describe a handful of homes. Momentum's coverage is concentrated where its agents work, so many counties outside Northeast Florida are thin; the Realtor.com figures above the block cover every county. Dates: counts updated October 30, 2026, closed sales from the Pulse cache updated October 29, 2026 (October 2026 is month to date), feed events through October 31, 2026. All MLS figures are deemed reliable but not guaranteed.
True cost to own
An estimate of the monthly cost of owning the typical home in an area, at the Zillow typical value (ZHVI), built from four parts. Principal and interest: 20% down at the current Freddie Mac 30-year average rate (the same basis as the affordability section). Property tax: the county's typical FY2025-26 adopted millage with the Florida homestead exemption applied ($25,000 on all levies plus $25,000 more on non-school levies for value above $50,000); Georgia counties use the DOR 2025 digest at the 40% assessment ratio with the $2,000 state homestead exemption. City and district levies vary, so this is the county's typical bill, not a quote. Insurance: the average premium on Citizens Property Insurance policies in force in the county (Citizens Property Insurance, Detail by County, April 2026), a market reference rather than a quote; private-market premiums differ. HOA: the listing-weighted median monthly dues across the communities in Momentum's MLS fee records (fee records through October 31, 2026), included in the total only where the county has at least 50 listings with fee fields; the panel says when HOA is excluded. Also shown: the share of listings with a CDD, and the share of homes inside a FEMA special flood hazard or coastal zone across our community flood records (through October 2026; counties with fewer than 30 records show no value).
ZIP rule. ZIP figures use the county's millage and Citizens premium at the ZIP's own typical value; HOA, CDD and flood share are county-level only. Every figure in this group is an estimate at the typical value, not a bill or a quote for any specific property.
Supply and flows
A county-level Florida group added October 2026. Each metric names its source and date in the panel. The figures count homes, movers and dollars; they say nothing about who the movers are, and the map never rates an area by its residents.
| Metric | Definition | Source and date | Coverage |
|---|---|---|---|
| Net migration per 1k residents | People on inbound tax returns minus people on outbound returns, per 1,000 county residents (population is used only as the denominator). Also shown: movers in and out, and the net adjusted gross income that moved with them, in dollars and per resident. | IRS SOI county-to-county migration, tax year 2022-2023 | 67 counties |
| Where movers come from | The top three out-of-state origins by inbound returns. The IRS in-state row mixes moves between Florida counties with non-mover rows, so the panel states in words that most in-state moves come from other Florida counties and shows no count for it. | same | 67 counties |
| Single-family permits per 1k homes | Single-family units permitted in the calendar year per 1,000 existing housing units (ACS). Also shown: the unit count, all units versus the prior year, and a five-year bar of all permitted units. | U.S. Census Bureau, Building Permits Survey, calendar 2025; statewide monthly context from the same survey through June 2026 | 67 counties |
| Distressed share of active listings | Share of active for-sale MLS inventory carrying a short-sale, foreclosure, REO, auction or pre-foreclosure signal (status plus remarks, negation-guarded), over true active inventory. The change column compares the current monthly release with the release two months earlier (the series began August 2026, so a three-month comparison is not yet possible). | Momentum Florida Distress Index, October 2026 release (published October 30, 2026) | 47 counties with 200+ active listings |
| Resales at a net loss | Share of homes bought in 2022-23 and resold since that netted less than the prior purchase price after about 8%% selling costs. Also shown: the same share for homes built 2020 or later, and its gap in points against the county's older stock (an association with buying new, not a cause). | Momentum Realty analysis of MLS resale pairs, as of August 29, 2026 | 35 counties (24 with the built-2020+ figure) |
Thin-data rule. Distress is published only for counties with at least 200 active listings; the resale figures only for counties with enough resale pairs to call a share; permits and migration cover all 67 Florida counties. Georgia counties do not have this group yet. Nothing in the group feeds the projection model.
Compare and share
"+ Compare" on any county or ZIP panel adds it to a tray of up to three areas; "Compare" opens a side-by-side table of typical value, 12-month change, the model's expected change and odds, market balance, cost to own, days on market, months of supply, affordability, price-to-income, price cuts, net migration and distress. The strongest reading in each row is shaded (higher growth, odds and affordability; lower cost, days on market, supply, price cuts and distress); typical value and market balance are shown without a highlight because neither is "better" in one direction. The table links to each area's page and carries a copyable link (?compare=). "Share" copies a link that restores the metric, level, selected area and compare set.
National projections (model version 3)
The projection model was refitted on every US county: 552,570 county-months across 3061 counties, forecast origins January 2002 to 2025, the same six inputs as the Florida model (trailing 12-month change, the 12 months before that, the last 3 months, price-cut share, inventory versus a year earlier, and the gap to each county's own expanding log-linear trend); state indicators for Florida and Georgia were tested and did not improve the Florida back-test, so they are not used. The target is the true 12-month change in the Zillow typical home value. Accuracy is measured leave-one-year-out by forecast origin year.
All US counties, out of sample: direction right 81.1% of the time, 59.4% of outcomes inside the 20-80% range, median miss 3.0 points, average lean +0.0 points. Florida counties only, same national model: direction 83.7%, inside the range 58.3%, median miss 5.1 points (version 2, Florida-only fit: 82.9%, 58.5%, 4.6). Florida projections on the map stay on version 2 until the national model matches it on all three measures; version 3 is used for every other state.
Most counties are projected from the Zillow July 2026 snapshot (origin May 2026, horizon May 2027); Florida and Georgia use the site's newer months (origin 2026-09). The range for a county outside Florida is the 20-80% band of all national misses; Florida uses its own misses. The Momentum Pulse sanity gate applies only where a Pulse market exists (Florida and Georgia).
| Origin year | Rows | Direction | Inside range | Median miss | Lean |
|---|---|---|---|---|---|
| 2002 | 1,018 | 98% | 73% | 1.5 | -2.1 |
| 2003 | 12,335 | 99% | 65% | 2.2 | -3.3 |
| 2004 | 12,490 | 96% | 54% | 3.3 | -4.2 |
| 2005 | 12,802 | 93% | 68% | 2.4 | -1.8 |
| 2006 | 13,563 | 76% | 71% | 2.2 | +2.4 |
| 2007 | 14,168 | 48% | 47% | 4.0 | +5.1 |
| 2008 | 14,859 | 36% | 22% | 5.8 | +6.4 |
| 2009 | 15,329 | 56% | 58% | 3.3 | +2.9 |
| 2010 | 15,915 | 30% | 35% | 4.7 | +5.1 |
| 2011 | 16,759 | 65% | 74% | 2.1 | +0.1 |
| 2012 | 22,542 | 77% | 68% | 2.5 | -1.5 |
| 2013 | 24,563 | 86% | 76% | 2.0 | -0.6 |
| 2014 | 25,707 | 87% | 78% | 2.0 | -0.6 |
| 2015 | 29,015 | 90% | 71% | 2.3 | -1.0 |
| 2016 | 29,510 | 86% | 78% | 2.0 | -0.2 |
| 2017 | 30,035 | 92% | 72% | 2.3 | -2.0 |
| 2018 | 30,431 | 93% | 77% | 2.0 | -0.9 |
| 2019 | 35,603 | 96% | 67% | 2.6 | -1.9 |
| 2020 | 35,699 | 98% | 17% | 8.0 | -7.8 |
| 2021 | 35,791 | 96% | 53% | 3.5 | -1.7 |
| 2022 | 36,017 | 72% | 21% | 7.8 | +8.0 |
| 2023 | 36,109 | 80% | 71% | 2.4 | +0.2 |
| 2024 | 36,261 | 74% | 61% | 3.0 | +2.1 |
| 2025 | 16,049 | 70% | 64% | 2.8 | +1.8 |
Data sources and dates
- Zillow Research ZHVI, county, monthly 2000-01 to 2026-05 (snapshot July 2026, extended to 2026-09 with the site's latest months) (data/external/zillow/2026-07/County_zhvi_*.csv)
- Zillow Research share of listings with a price cut, county, monthly from 2018-03 (data/external/zillow/2026-07/County_perc_listings_price_cut_*.csv)
- Zillow Research for-sale inventory, county, monthly from 2018-03 (data/external/zillow/2026-07/County_invt_fs_*.csv)
- Zillow Research ZHVI, ZIP, monthly 2000-01 to 2026-05 (Florida rows) (data/external/zillow/2026-07/Zip_zhvi_*.csv)
- Current county and ZIP metrics (latest 13 months) (data/fl-market/counties.json, zips.json)
- Momentum Realty Market Pulse (MLS), latest month per market (reports/pulse/<market>/<YYYY-MM>.json)
- Momentum Realty MLS feeds: county truth, city truth, Market Pulse cache and hourly feed events (data/intel/county-truth.json, city-truth.json, pulse-cache/, .hot-events-*.json)
- Millage: county FY2025-26 adopted rates; Georgia DOR 2025 digest (data/millage-rates.json, data/ga/millage.json)
- Citizens Property Insurance average premium by county (data/fl-market/insurance.json)
- Momentum MLS community fee records and FEMA flood-zone records (data/api/community-fees.json, community-flood-all.json)
- IRS SOI county-to-county migration (data/fl-market/migration.json)
- U.S. Census Bureau Building Permits Survey, county and statewide monthly (data/fl-market/permits.json, data/external/fl_sf_permits_monthly.csv)
- Momentum Florida Distress Index monthly releases (data/fl-market/distress-release-*.json)
- Momentum Realty resale-at-a-loss analysis (data/api/resale-underwater.json)
Market data files updated 2026-10-16. Zillow data through September 2026. The model is refitted and the ledger scored on each monthly refresh.
Limitations
- Monthly rows overlap: adjacent origin years share part of their 12-month windows, so the year-by-year test is a little less independent than it looks.
- Price cuts and inventory are only available from 2018, so earlier training rows carry the training mean plus a missing flag; the model leans on momentum and trend more than on supply.
- The Zillow monthly snapshot on file ends in May 2026 and is extended to September 2026 with the site's latest months (same series, newer vintage). Vintage differences are small but not zero.
- ZIP projections apply the county model to the ZIP's own value history; price cuts and inventory are the county figures. ZIPs without a value history inherit the county projection and are labeled.
- Mortgage rates, insurance costs, storms, migration and the national economy are not inputs. The model is a market-conditions lean, not a forecast of events.
- Georgia counties get the 3-month Pulse layer where covered; the 12-month model is Florida only in this version.
- Beta: this is version v2-2026-10, published October 30, 2026. Expect the method to change as live calls are scored; changes are listed in the model version and this page.