Methodology
Every number on the explorer, where it comes from, how stale it is, and what it cannot tell you. If you only read one section, read the one on the yield basis: it is the choice that decides what the map is comparing.
Sample data
The data pipeline has not run yet, so these figures are generated rather than measured. The coverage counts on this page are therefore counts of generated figures, and the formulas and sources described below are what the pipeline will produce once it has run.
Read this first
ACS median home values are self-reported by homeowners, they run high, and they lag about eighteen months. The American Community Survey asks owners what they think their home is worth. People overstate, consistently, and the five year estimate describes a window that closed a year and a half before it was published. That figure is the denominator of the default gross yield on this map, so the default yield is biased low on the value side.
ACS median gross rent has the mirror problem in the numerator: it INCLUDES utilities, and it is a median over the occupied stock rather than over units coming to market, so it is weighted toward long-tenured leases and reads below what a unit lists for today.
The two errors push the ratio in opposite directions and partially cancel. That is a coincidence of sign, not a justification, and how much they cancel is unknown and varies by metro. The default map is a reliable RANKING and an unreliable LEVEL. Anybody underwriting a deal should switch to the market basis and then check the number against a real rent quote and a broker opinion of value.
Every yield here is GROSS. Nothing is netted out for taxes, insurance, management, vacancy loss, maintenance or capital reserves. A six percent gross yield is not a six percent return, and after those costs it is usually closer to half of one.
The yield basis
A metro can carry three rents and two home values, and they do not measure the same thing. A gross yield is a ratio of one of the rents to one of the values, so it is not one number with measurement noise on it: ACS rent over ACS value and Zillow rent over Zillow value can differ by more than a percentage point on the same metro, in either direction, because both the numerator and the denominator are biased and the biases do not move together.
So a basis is chosen for the whole map, never per metro. The tempting design is to let each metro use the best pair it happens to carry and fall back to ACS where Zillow is absent. That produces the highest coverage and it is wrong. Every metro would get a plausible number, the map would have no holes, and the colour ramp would be comparing a Zillow yield in Phoenix against an ACS yield in Wheeling. The ranking that produced would be an artifact of which metros Zillow covers rather than of the housing market. There is no fallback path anywhere in the engine and no function that blends two bases.
A map with gaps is honest about what is not known. A map that fills the gaps with a different measurement is not.
| Basis | What the two series are | Metros covered |
|---|---|---|
| Census | ACS median gross rent, includes utilities; ACS median home value, self-reported | 393 of 393 |
| Market | Zillow ZORI asking rent, excludes utilities; Zillow ZHVI, current market estimate | 220 of 393 |
| Official | HUD Fair Market Rent, two bedroom, 40th percentile; ACS median home value, self-reported | 391 of 393 |
That coverage difference is the single most important thing to understand about this tool. The census basis is the default because it is the only pair that exists for effectively every metro, so the default map's holes are real coverage gaps rather than gaps the basis choice introduced, and because both figures come from one survey with one sample frame, so the level is biased but the bias is applied the same way everywhere. It is also public domain end to end, which matters because the plan commits to Zillow being one deletable file.
Every measure
Gross rent yield
monthly rent x 12 / home value x 100
How much rent a market produces per dollar of price. The one measure that exists on a consistent basis for effectively every metro, which is why it is the default layer.
- Inputs
- The rent and the home value of the selected basis
- Publishers and lag
- Census American Community Survey, 5 year estimates
- Zillow Home Value Index and Observed Rent Index
- HUD Fair Market Rents
What it cannot tell you
- GROSS. Nothing is netted out for property taxes, insurance, management, vacancy loss, maintenance or capital reserves. A six percent gross yield is not a six percent return and is usually closer to half of one.
- It is a ratio of two medians over different populations. The median rented unit and the median owned home are not the same dwelling, so this is not the yield on any particular property.
- On the census basis the rent includes utilities, which inflates the numerator, and the home value is self-reported, which inflates the denominator. The two errors push the ratio in opposite directions by amounts nobody has measured per metro.
Price to rent
home value / (monthly rent x 12)
How many years of gross rent the price represents. Exactly 100 divided by the gross yield, so it carries no extra information, but it spreads the expensive coastal metros apart where the yield compresses them into one band.
- Inputs
- The rent and the home value of the selected basis
- Publishers and lag
- Census American Community Survey, 5 year estimates
- Zillow Home Value Index and Observed Rent Index
- HUD Fair Market Rents
What it cannot tell you
- Carries every limitation of the gross yield, because it is the same two numbers the other way up.
- Sensitive to the rent in the denominator, so a metro with a thin or noisy rent estimate swings further here than it does on the yield.
Rent to income
monthly rent x 12 / median household income x 100
An affordability indicator for the metro. Useful as a ceiling check: a market where rent already takes a large share of income has less room to raise it.
- Inputs
- The rent of the selected basis
- ACS median household income, all households
- Publishers and lag
- Census American Community Survey, 5 year estimates
- Zillow Home Value Index and Observed Rent Index
- HUD Fair Market Rents
What it cannot tell you
- NOT a household budget share. The numerator is a median over RENTED units and the denominator is a median over ALL households including owners, who out-earn renters in almost every metro. The true burden on the median renter is higher, by a margin that moves with the ownership rate.
- The conventional thirty percent affordability threshold is a threshold on the other ratio, rent against RENTER income, so do not draw a line at thirty on this one.
Permits per 1,000 units
units authorized in the trailing twelve months / housing units x 1000
How fast the stock is growing relative to its own size. The real answer to whether a market is about to be oversupplied, and the layer nobody presents well to retail investors. Per thousand existing units rather than per capita, because ten thousand permits is a boom in Boise and a rounding error in Chicago.
- Inputs
- Census Building Permits Survey, units authorized by CBSA
- ACS total housing units
- Publishers and lag
- Census Building Permits Survey
- Census American Community Survey, 5 year estimates
What it cannot tell you
- Authorized is not started and not completed. In a credit squeeze this overstates what actually gets built, sometimes by a lot.
- Single family and multifamily units are counted together. One large apartment complex reads the same as the same unit count spread across twenty subdivisions, and those are very different events for a rental investor.
- The numerator is monthly with roughly a five week lag and the denominator is an annual ACS figure, so the two sides of the ratio are not contemporaneous.
Rental vacancy
reported directly
The market's own verdict on whether the supply already delivered is being absorbed. Read alongside permits: high permits with low vacancy is a market keeping up, high permits with high vacancy is one that is not.
- Inputs
- ACS rental vacancy rate
- Publishers and lag
- Census American Community Survey, 5 year estimates
What it cannot tell you
- Backward looking by construction. It describes what has already happened to occupancy, not what the units under construction will do to it.
- The ACS metro level estimate carries a wide margin of error on small metros, where a swing of a point or two can be sampling rather than the market.
Home value change, 1 year
reported directly
One year change in the typical home value for the metro.
- Inputs
- FHFA metro house price index where it covers the metro, Zillow ZHVI where it does not
- Publishers and lag
- FHFA House Price Index
- Zillow Home Value Index and Observed Rent Index
What it cannot tell you
- The source varies by metro, and the two indices are built differently: FHFA is a repeat sales index over conforming mortgages, ZHVI is a valuation model over the whole stock. The panel names which one produced the figure for a given metro; the map does not, so a national map of this layer is two indices side by side.
- A price index change is not a return. It excludes transaction costs and says nothing about the rent the asset produced meanwhile.
Home value change, 3 year
reported directly
Three year total change, which is long enough to see through a single quarter and short enough to still describe the current cycle.
- Inputs
- The same index as the one year change, over three years
- Publishers and lag
- FHFA House Price Index
- Zillow Home Value Index and Observed Rent Index
What it cannot tell you
- A TOTAL, not an annual rate. Dividing it by three overstates the annual rate by a widening margin as growth rises, which is why the annualised version exists.
- Carries the same mixed-source limitation as the one year figure.
Home value change, 3 year annualised
(1 + three year change) ^ (1/3) - 1
The three year change as a compound annual rate, which is the form that compares against a yield or a discount rate.
- Inputs
- The three year home value change
- Publishers and lag
- FHFA House Price Index
- Zillow Home Value Index and Observed Rent Index
What it cannot tell you
- Compounding is not linear. Thirty three percent over three years is ten percent a year, not eleven, and the gap widens as growth rises.
- An average annual rate hides the path. Two metros with the same three year figure can have had very different years.
Rent change, 1 year
reported directly
Whether rents are moving, on the only series here that is current to the month.
- Inputs
- Zillow ZORI, one year change in observed asking rent
- Publishers and lag
- Zillow Home Value Index and Observed Rent Index
What it cannot tell you
- ZORI only. Absent for well over a hundred metros, which means this layer, and the supply pressure composite that leans on it, are thinner outside the larger markets than everything else on this tool.
- An asking rent index describes units coming to market, not the rent roll of the existing stock, which moves more slowly.
Supply pressure
0.45 x percentile(permits per 1,000 units) + 0.35 x percentile(rental vacancy) + 0.20 x (100 - percentile(rent change, 1 year))
A single reading of whether a market is crowded, with every component and every weight exposed so the arithmetic can be redone by hand.
- Inputs
- Permits per 1,000 units
- Rental vacancy
- Rent change over one year, inverted
- Publishers and lag
- Census Building Permits Survey
- Census American Community Survey, 5 year estimates
- Zillow Home Value Index and Observed Rent Index
What it cannot tell you
- A JUDGEMENT, not a measurement. The weights were chosen by reasoning and no regression established that 45/35/20 predicts anything. Three thoughtful people would pick three different weightings and draw three different maps, and none of them would be wrong.
- RELATIVE, not absolute. Every component is a percentile rank within the metro set supplied, so a score of 80 means toward the crowded end compared with other US metros. Nothing here has an absolute scale.
- Scored on two of three components where the third is missing, with the weights rescaled over what remained. A two component score and a three component score are not quite the same measure, and the metro panel says which one it is showing.
- Population change is deliberately outside it, even though the artifact carries it: it is a five year measure against three one year components, and it is a demand signal rather than a supply one.
The figures behind the measures
| Figure | What to know |
|---|---|
| Population | Reported directly, with its own source and publication date. |
| Households | Reported directly, with its own source and publication date. |
| Housing units | Total stock, occupied and vacant. The denominator of the permits measure. |
| Median household income | All households, owners and renters together. |
| Median home value, ACS | Self-reported by owners, runs high, and lags about eighteen months. |
| Median home value, Zillow | Current market estimate of the typical home value. Monthly. |
| Median gross rent, ACS | What renters pay INCLUDING utilities, across the occupied stock rather than units coming to market. |
| Observed rent, Zillow | Asking rent on units coming to market, EXCLUDING utilities. Absent for well over a hundred metros. |
| Fair market rent, 2 bed | HUD's administratively set 40th percentile gross rent. Annual, effective each October. |
| Owner occupied | Reported directly, with its own source and publication date. |
| Rental vacancy | Wide margin of error on small metros, where a point or two can be sampling rather than the market. |
| Homeowner vacancy | Reported directly, with its own source and publication date. |
| Permits, trailing 12 months | Units AUTHORIZED, which is neither started nor completed, and single family and multifamily are not distinguished. |
| Home value change, 1 year | Reported directly, with its own source and publication date. |
| Home value change, 3 year | Reported directly, with its own source and publication date. |
| Rent change, 1 year | From Zillow ZORI, so absent wherever ZORI is absent. |
| Population change, 5 year | A five year ACS change, a different horizon from every other figure here. Deliberately outside the supply pressure composite for that reason. |
Sources and how often they publish
| Source | Publishes | Lag |
|---|---|---|
| Census American Community Survey, 5 year estimatesThe federal backbone of this tool: population, households, housing units, income, home value, gross rent, vacancy and tenure all come from one survey with one sample frame, which is what makes the census basis internally consistent. Its home value is SELF-REPORTED by owners and runs high. | Annually, each December | About eighteen months behind the period it describes |
| Zillow Home Value Index and Observed Rent IndexThe current, market facing figures: ZHVI for value and ZORI for asking rent, which excludes utilities. Coverage is the catch. ZORI reaches a little over half the metro set, so the market basis has real holes and the tool shows them rather than filling them in. | Monthly, around the 16th | About three weeks |
| HUD Fair Market RentsAn administratively set 40th percentile gross rent by bedroom count. Official in the sense that federal programmes pay against it, which is not the same as being what the market charges. | Annually, effective each October | Set ahead of the year it applies to, from ACS data two to three years old |
| Census Building Permits SurveyUnits authorized by building permit, by metro. The only forward looking series here, and the reason the supply question can be answered at all. | Monthly | About five weeks |
| FHFA House Price IndexA repeat sales index over conforming mortgages, with a long history. Covers most but not all metros, which is why the home value change layers fall back to Zillow where it does not reach. | Quarterly | About two months after the quarter closes |
| Census TIGERweb and cartographic boundariesMetro names, codes, state memberships and geometry. Names drift with every revision while the five digit CBSA code persists, which is why nothing in this system joins on a name. | With each OMB delineation revision | Current vintage |
The ZIP level
Clicking a metro drills to its ZIPs. There are 33,791 ZIP Code Tabulation Areas in the country, so they are not a committed national layer: the boundaries are built ahead of time into one static file per metro and the browser fetches the single file it needs when you drill in. The national payload is exactly what it was before this level existed.
Which ZIPs are in which metro
Each ZCTA is assigned to the one metro whose polygon contains the ZCTA's Census published internal point, so every ZCTA is drawn at most once. A ZCTA straddling a metro boundary belongs wholly to the metro holding that point. A metro has no statement of its ZIP membership anywhere, because a tabulation area is not an administrative unit that anyone assigns to anything, so this is a geometric assignment rather than a published crosswalk. It is decided by the Census Bureau's own internal point rather than a centroid we computed, and it guarantees every area is drawn exactly once. What it gets wrong is at the edges: an area straddling a metro boundary belongs wholly to the metro holding that point.
A ZCTA is not a postal ZIP
- A postal ZIP code is a set of delivery routes and has no area. The Census Bureau builds a ZIP Code Tabulation Area by giving each census block the ZIP most common among its addresses and dissolving blocks that agree, so a ZCTA approximates a delivery ZIP rather than equalling it.
- There are roughly 41,700 postal ZIP codes and 33,791 ZCTAs. The missing ones are mostly ZIPs with no residential addresses to tabulate, so an address in a PO box only, single building or unique ZIP will find nothing on this map.
- Where both exist the boundaries still differ, because a ZCTA follows census block edges and a delivery route does not. A property near the edge of its ZIP can be counted in a neighbouring ZCTA's figures.
- These boundaries are the 2020 census definitions and are fixed until 2030, while USPS changes routes continually.
A ZIP figure is softer than a metro figure
- The American Community Survey is a sample. A metro median rests on tens of thousands of interviews; a single ZCTA median can rest on a few dozen, and the margin of error scales with the inverse square root of the sample. Margins of 15 to 40 percent of the estimate are ordinary at this grain.
- Where the sample is too thin the Bureau publishes no estimate at all. Those ZCTAs render hatched here, the same as any other missing figure, because that is what they are.
- Suppression is not random. It removes small and thinly sampled areas, which skew rural, lower income and higher vacancy, so the ZCTAs that do carry a figure are a biased subset of the metro rather than a random sample of it.
- Hatched ZIPs are not a random scatter: ACS suppresses the thinnest samples and Zillow publishes only where it has transactions, so a ranking of the ZIPs that do carry a figure is not a ranking of the metro.
- Where ACS publishes a margin of error, the ZIP panel prints it beside the figure and states how much weight it will bear, using the Census Bureau's own thresholds: a coefficient of variation at or under 15 percent is reliable, up to 30 percent should be used with caution, and above that the estimate should not stand alone. Zillow and HUD publish no margin, so their rows say so rather than implying precision.
Two measures do not exist at this level
Permits per 1,000 units and the supply pressure composite are absent at ZIP grain, and not because of a coverage gap. The Census Building Permits Survey publishes by place, county and metro and by no ZIP geography at all, and the composite weights permits most heavily, so rescaling it over its two remaining components would produce a different measure under the same name. Choosing either layer while drilled into a metro colours the ZIPs by gross rent yield instead and says so above the map.
The shading means something else here
A metro is shaded against every metro in the country, so a dark metro in Ohio and a dark metro in Texas mean the same thing. ZIPs are shaded against the other ZIPs IN THAT METRO. That is deliberate and it is the opposite rule: a metro's ZIPs cluster tightly around the metro's own level, so a national ZIP ramp would make every ZIP in Cleveland one shade and every ZIP in San Jose another, and the map you drilled twice to reach would carry no information. The legend prints its numeric edges, which is what keeps that readable.
Two dates, which are not the same
The pipeline rebuilds the artifact every day. Nothing it reads publishes daily, so on most days the rebuild produces a byte-identical file and commits nothing. That is the design working: the daily cadence buys detection of a source that renamed a column or moved a URL, not fresher numbers.
So there are two dates and they answer different questions. The pipeline run date says when the file was assembled. The vintage on each figure says when its publisher dated it. Every figure on this tool prints its own vintage, and nothing anywhere prints the run date as if it were a data date. The data status page shows both side by side.
Model this market
The metro panel carries a link into the rental pro forma. Exactly two numbers travel and both are data: the metro's median home value becomes the purchase price, and its median rent times twelve becomes the gross potential rent, both on whichever basis the map is showing. One dwelling, not a building, because a metro median describes one home.
Everything else is the pro forma's own default, and one of those defaults is rescaled rather than carried, which is worth stating plainly. The pro forma's default expense lines total $710,000 a year against a $1.8M rent. Carried unchanged onto a single house they would open a model with a deeply negative NOI, which reads as a broken tool rather than as an invitation to fill the lines in. So each line is multiplied by the ratio of the new rent to the default rent, which holds the pro forma's own expense ratio. Those seven numbers are not a claim about what taxes or insurance cost in that metro, and nothing in the artifact would support such a claim. Replace them.
Floor area is assumed at 1,500 square feet, because the artifact carries none. It moves the price per square foot line and feeds nothing else: no income, no expense, no cap rate and no return.
One more thing about the census and official bases. Both of their rents are GROSS rents, meaning utilities are already inside the figure. The pro forma also carries a utilities expense line, so a model prefilled from either basis double counts them and understates NOI by whatever utilities actually cost. The panel says so at the point of the link, and the fix is to zero that line or to switch to the market basis, whose rent excludes utilities.