← Back to Model Lab
The evidence behind the lab

Useful questions. Traceable data. Testable ideas.

Each view has a different job. Current market conditions help you compare listings. Ownership scenarios help you examine costs. Forecasts need a record of predictions and later outcomes.

How research informs what you can use

These studies motivate questions and methods. Their results are evidence from their own samples; we have not independently replicated them in these eight cities.

Market conditions · available

Look beyond the last sale

Listing behavior can add information about changing demand. The lab now shows inventory and time to contract, with a comparison to the same season a year earlier.

The Anenberg–Laufer listing study uses richer listing information. Research on buyer search also motivates a separate demand indicator. Our two-measure description has not been tested as a forecast.

Explore buying conditions →
Rent or buy · available

Connect the price to living costs

Rents and the cost of owning help put a purchase price in context. Explore loan payments, upkeep, fees and the effect of a different price path.

Himmelberg, Mayer and Sinai explain why ownership costs matter for valuation. Gallin’s rent research motivates longer-term adjustment tests. A scenario calculator does not establish what prices will do.

Build an ownership scenario →
Price forecasts · experimental

Check the simplest rule first

Our original price-trend, relative-price and tax-base models remain available, with their past errors and personal weights. None earned selection over matching San Mateo County in the original validation.

Research on combining regional housing forecasts motivates comparison with simple blends. Adding models or tuning weights to familiar history does not establish better predictions.

Inspect the forecasts and misses →

What history is available?

Reviewed release retrieved 2026-09-16. These are revised historical series downloaded now, not an archive of what readers could have known in each past year. Coverage is eight cities in San Mateo County plus county context where available.

2,784City / home-type market windows
991City and county rent observations
162Housing-stock snapshots
Retained coverage. Market windows end monthly and overlap. The rental count excludes missing months.
PlaceHousing stockCity-wide rentsMarket conditions
BurlingameApr 2010–Jan 2026 · 18 datesMay 2017–Jul 2026 · 111 monthsMar 2012–Aug 2026 · 174 windows per home type
Daly CityApr 2010–Jan 2026 · 18 datesNov 2017–Jul 2026 · 105 monthsMar 2012–Aug 2026 · 174 windows per home type
MillbraeApr 2010–Jan 2026 · 18 datesJul 2019–Jul 2026 · 84 monthsMar 2012–Aug 2026 · 174 windows per home type
PacificaApr 2010–Jan 2026 · 18 datesJan 2019–Jul 2026 · 90 monthsMar 2012–Aug 2026 · 174 windows per home type
Redwood CityApr 2010–Jan 2026 · 18 datesMar 2015–Jul 2026 · 136 monthsMar 2012–Aug 2026 · 174 windows per home type
San BrunoApr 2010–Jan 2026 · 18 datesJan 2019–Jul 2026 · 91 monthsMar 2012–Aug 2026 · 174 windows per home type
San MateoApr 2010–Jan 2026 · 18 datesApr 2015–Jul 2026 · 136 monthsMar 2012–Aug 2026 · 174 windows per home type
San Mateo CountyApr 2010–Jan 2026 · 18 datesJan 2015–Jul 2026 · 139 monthsCounty series not retained
South San FranciscoApr 2010–Jan 2026 · 18 datesMay 2018–Jul 2026 · 99 monthsMar 2012–Aug 2026 · 174 windows per home type

Single-family and condo/co-op market series stay separate. Zillow’s retained city rent series covers all rental housing, including multifamily; it does not match either sale population. The lab asks you to supply a comparable property’s rent for a personal cost comparison.

Jobs: 108 Q2 observations across 2012–2023. Completions: 62 accepted city-years across 2018–2025. San Bruno 2020 and South San Francisco 2024 remain unavailable. Jobs count workplaces; completions include replacement construction.

What would make a new forecast credible?

Each proposed family needs a defined outcome, a fixed method and a test on later, uninspected data. The existing annual appreciation weights apply only to models with that same target. Cost scenarios and a multi-year forecast cannot be averaged into that annual prediction.

Forecast readiness after this data release
FamilyCandidate questionNext required evidence
Market pressure · pilot implementedHow many days will homes take to reach a contract next calendar quarter?Four rules and an equal blend now have a reproducible replay. Neither home type passed the full selection gate. Thirteen October–December 2026 forecasts are frozen for a live check, with weekly capture and immutable scoring. Apply the models and inspect their errors.
Rent fundamentalsDo prices adjust toward rents over several years?Match housing populations, test whether a stable relationship exists, and obtain enough mature forecast origins. The current mixed-population benchmarks do not pass that gate.
Jobs × housing responseDoes stronger demand with a slower supply response precede relative appreciation?Stock denominators are now available. Short shared job/completion history and unknown historical release dates still block the proposed multi-year test. Resolve availability and broaden coverage first.

2025 is useful for exploration, but we have already inspected it. We cannot relabel it as an untouched test after selecting a new rule. Every future issued forecast should retain its inputs, date, target and eventual error; the original record and planned future review remain intact.

The market pilot retains 343,154 California source rows. Coverage and sales-count rules using 2013–2018 inputs selected 299 single-family and 168 condo peer cities. Our eight cities are excluded from coefficient fitting, but other California cities can share their regional shocks. Local selection uses 2019–2021 decision dates; later replay uses 2022–2026. These quarterly outcomes do not overlap. Source histories are revised, and the assumed one-month reporting allowance does not establish historical release availability.

The original 174-period evidence snapshot retains its original “not tested” readiness assessment as history. This separate experiment records the later test and its failure to select an added model. Read the protocol, results and operations.

Sources, definitions and missing values

Housing-stock reconciliation

California DOF E-8 and E-5 agree on all nine overlapping April 2020 total and occupied housing benchmarks. We retain those census dates and use January 1 snapshots for annual comparisons. Total stock includes occupied and vacant units. Completions are divided by the January stock at the start of their reporting year; that ratio is not net stock growth.

The E-8 workbook declares May 1, 2025 while the current release page says May 2026. We record this discrepancy and do not infer when the exact historical values first became available. E-5 2026 estimates are provisional.

Market comparisons and small samples

Inventory is the count at period end. Closed sales, pending activity and median prices span three months. Days to contract describes homes going under contract. Year-earlier changes are calculated from retained levels, not the source’s precomputed change columns.

384 retained windows have fewer than ten closed sales or an unavailable count; 6 have unavailable sales. The two-measure summary is withheld if either comparison window has fewer than ten sales or a required measure is missing. Ten sales is a display policy, not a calibrated confidence level. Missing values stay unavailable; observed zeros stay zero. Price-cut history is not included.

Ownership calculations

The fixed-rate loan amortizes monthly. Cash flow includes principal repayments; the remaining balance is subtracted when modeling a sale, so principal is not counted twice as a cost. Purchase outlays, ongoing payments and sale proceeds are discounted using your chosen alternative return on cash. These are present values, not an inflation forecast.

The starting mortgage rate is the dated 6.76% national 30-year PMMS observation for 2026-09-10. Property tax, maintenance, insurance, transaction fees, future growth and the alternative return are editable examples. Mortgage insurance remains at the entered amount until the loan is paid off; the calculator does not model automatic cancellation, tax deductions, or every household expense.

Budget limits and mortgage history

The lab now retains 2,894 weekly national mortgage observations from 1971 through 2026-09-10. Freddie Mac PMMS, retrieved from FRED, Federal Reserve Bank of St. Louis supplies this history; the method changed on November 17, 2022. Monthly chart averages use observations dated within each month and the latest month may be partial.

The purchase-budget tool solves the monthly cash and upfront-cash constraints with your financing and cost assumptions. It does not assess loan qualification. The break-even appreciation tool finds the price-growth assumption that equalizes discounted costs; it is not a market prediction. Inputs stay in page memory and explicit copy actions are required to change a price or rate.

California Department of Finance: California DOF E-5 population and housing estimates, May 2026

Original source · Definitions / release documentation

Current revised estimates. Total housing includes occupied and vacant units. Use January snapshots for annual comparisons; retain April census benchmarks separately. Workbook declares release May 1, 2026; 2026 estimates are provisional. This is not a historical-vintage archive.

Retrieved: 2026-09-16T01:17:56.946683+00:00. Historical first publication of these exact values is not established.

Retained artifact: dof-e5.xlsx

Retained artifact SHA-256: 4e10c3eb3e6635e1591e1b264e7037b6ac9b77ec3af9e2914848ca357851f2ba

California Department of Finance: California DOF E-8 historical population and housing estimates

Original source · Definitions / release documentation

Current revised estimates. Total housing includes occupied and vacant units. Use January snapshots for annual comparisons; retain April census benchmarks separately. Workbook declares May 1, 2025; the current release page says May 2026. Historical first publication of these exact bytes is unknown.

Retrieved: 2026-09-16T01:17:55.295927+00:00. Historical first publication of these exact values is not established.

Retained artifact: dof-e8.xlsx

Retained artifact SHA-256: e79e2a4b673e69b78914d597f5d9e55f54335ccece198cc38aa1cf8ad2b4181f

Zillow Research: Zillow Observed Rent Index — city, all rental homes and multifamily

Original source · Definitions / release documentation

Smoothed, not seasonally adjusted, nominal monthly asking-rent index. Selected original field values retained; full-download hash is recorded. Revised history retrieved in 2026. Rental stock differs from the for-sale populations.

Retrieved: 2026-09-16T01:20:54.803722+00:00. Historical first publication of these exact values is not established.

Retained artifact: zori-city.csv.gz

Retained artifact SHA-256: 059c44bd2bad324b3d607138bcc4f60d8d9040d7c6778ee676ae0113de4fc9c5

Zillow Research: Zillow Observed Rent Index — county, all rental homes and multifamily

Original source · Definitions / release documentation

Smoothed, not seasonally adjusted, nominal monthly asking-rent index. Selected original field values retained; full-download hash is recorded. Revised history retrieved in 2026. Rental stock differs from the for-sale populations.

Retrieved: 2026-09-16T01:20:54.570482+00:00. Historical first publication of these exact values is not established.

Retained artifact: zori-county.csv.gz

Retained artifact SHA-256: b32599dad6dd4aab97f7d548ec5be36b43ac02041f54d973ccfec7b4f7479470

Redfin Data Center: City housing-market history by property type

Original source · Definitions / release documentation

Exact selected original rows from a complete scan of the upstream object. Three-month rolling windows overlap; never sum them as monthly sales. Not seasonally adjusted. Recent estimates and history may be revised. Source precomputed change columns are not used; displayed year-over-year changes are derived from retained levels. Price-cut share is unavailable.

Retrieved: 2026-09-16T01:22:33.928835+00:00. Historical first publication of these exact values is not established.

Retained artifact: redfin-history.csv.gz

Retained artifact SHA-256: 122e4d2ce32009a5bbad65d075249ae3ee4ebac89b018f2128bcd1dee41ccc96

Freddie Mac: PMMS 30-year fixed mortgage benchmark

Original source · Definitions / release documentation

Manually reviewed dated statistical observation, checked against FRED MORTGAGE30US. Hash covers this retained extraction, not the upstream webpage. National conforming purchase-loan average, not a quote or a Bay Area jumbo-loan rate.

Retrieved: 2026-09-16. Dated observation: 2026-09-10.

Retained artifact: mortgage-benchmark.json

Retained artifact SHA-256: 1de6a6ded354442e07aa06a8f37bc57688728c4d73408829bcd873f6da71bb09