Demand Analysis by Asset Class: Public-Data Models for 30+ Property Types

Every commercial property type sells the same thing: a unit of capacity that a definable population will pay to use for a definable period. A self-storage facility sells cubic feet to households in transition. A car wash sells minutes of tunnel time to vehicle owners within a drive-time radius. A cold storage warehouse sells pallet positions and temperature to a food system that has to move product before it spoils. The demand question is therefore the same question in thirty different costumes: how large is the population that uses this kind of capacity, how much of it do they use, how much capacity already exists, and what is the pipeline. What changes from one asset class to the next is not the logic but the data. This pillar sets out, for more than thirty property types, which federal, state and public series measure the demand driver, which series count the competing supply, and what ratio the two produce. Every figure carries a named source and a year, because a demand model that cannot show its provenance is an opinion with a spreadsheet attached.

The approach rests on a fact about the United States that is easy to overlook: the federal statistical system is organized by industry and by person, not by real estate, and that turns out to be exactly what demand analysis needs. The Census Bureau counts people by single year of age in every county. County Business Patterns counts establishments by six-digit industry code in every ZIP code. State highway departments count vehicles on named road segments. The Coast Guard compiles boat registrations from every state. The Department of Agriculture measures cubic feet of refrigerated space in odd-numbered years. None of these agencies set out to serve a lender looking at a marina loan in a coastal county, yet together they supply a more complete demand record for that marina than for the suburban office building the same lender would treat as ordinary collateral. That inversion, between how risky an asset class is treated and how observable its demand actually is, runs through everything that follows.

Demand is an asset-class problem, and the public record is organized by asset class

Generic market analysis treats demand as a function of population and income in a trade area, which is adequate for a grocery store and nearly useless for a travel center, where the customer is a combination truck that does not live anywhere nearby. The discipline that avoids this error starts by naming the demand unit for the asset class. For multifamily the unit is the renter household, and the driver series is household formation by age and income. For senior housing the unit is the age-qualified, income-qualified household, and the driver is the population aged 75 and older by county, read from the Census Bureau's Vintage 2025 estimates and the American Community Survey's income-by-age tables. For a wedding venue the unit is the marriage, and the driver is a provisional count that the National Center for Health Statistics publishes by state of occurrence: 2,041,926 marriages in 2023, a rate of 6.1 per 1,000 residents (CDC NCHS, FastStats, provisional 2023 data). For a data center the unit is a megawatt of deliverable power, and the driver is a utility's interconnection queue rather than any population at all.

Once the demand unit is named, the public record usually supplies three things. First, a driver series that measures the population of users, with a vintage and a geography. Second, a supply series that counts the existing competing capacity, almost always County Business Patterns at the six-digit NAICS level, supplemented by permits, licenses or registrations specific to the class. Third, a pipeline series that shows what is coming: the Building Permits Survey for residential classes, local permit records and public securities filings for the rest. The analytical product is the ratio of driver to supply, expressed in the natural units of the asset class: square feet of storage per capita, vehicles per car wash bay, registered vessels per wet slip, refrigerated cubic feet per resident, licensed childcare slots per hundred children under five, physicians per 100,000 residents, truck parking spaces per mile of Interstate. The ratio is what a credit committee can compare against other markets, against the same market five years earlier, and against the project in front of it.

The counterintuitive finding of this library's asset-class work is that the property types lenders classify as special purpose, and therefore treat as the riskiest collateral, are the ones whose demand is most completely observable in federal records. Car washes draw on vehicle counts from the American Community Survey and traffic counts from state highway departments. Marinas draw on state boat registrations compiled by the Coast Guard since 1989. Cold storage draws on a biennial federal census of refrigerated space down to the state and the warehouse type. Hotels draw on airport enplanements and county employment. RV parks draw on a federal satellite account that values RVing at $27.5 billion of 2024 value added (Bureau of Economic Analysis, Outdoor Recreation Economic Statistics, 2024 data released March 2026). The generic asset classes that lenders treat as ordinary collateral, suburban office and in-line retail above all, have the thinnest public demand record of any class in the table below, because nothing in the federal system registers a tenant's decision to lease 4,000 square feet of office. Risk in the collateral and observability of demand run in opposite directions. The practical consequence is that the public-data opportunity is largest exactly where credit committees ask the most questions, and the rest of this article is organized to exploit that.

The demand stack for thirty-plus property types

The table below is the working map for the pillar. Each row names the demand unit, the public driver series, the public supply series and the ratio that the analysis produces. Cluster articles in this pillar take one row at a time and build the full model; the reserve rows added in August 2026 extend the set to marinas, early education, behavioral health, grocery-anchored retail, data centers, manufactured housing communities and agricultural facilities. Where a row names County Business Patterns, the 2023 vintage released on 26 June 2025 is the current file, coded to NAICS 2017, and the establishment counts that appear later in this article come from its national file.

Property type	Demand unit	Public driver series	Public supply series	Working ratio
Self-storage	Household in transition	ACS households, tenure, movers in the past year; Vintage 2025 population	CBP 531130; Nonemployer Statistics; local permits	Rentable square feet per capita in a 3-mile ring
Car wash	Vehicle within drive time	ACS table B25044 vehicles available; state DOT AADT	CBP 811192	Vehicles per wash site; AADT capture rate
Small-bay flex industrial	Service firm with 5 to 20 employees	QCEW and CBP employment by sector; CBP size classes	CBP 531120, 493110; local permits	Small establishments per existing flex square foot
General warehousing	Pallet throughput	QCEW 493110 employment; Freight Analysis Framework	CBP 493110	Warehouse employment per resident
RV park and campground	RV-owning household and park visitor	BEA outdoor recreation satellite account; National Park Service visitation	CBP 721211	Sites per thousand regional visitors; seasonal occupancy
Travel center	Combination truck	HPMS combination-truck AADT; FAF5 corridor flows; Jason's Law parking inventory	CBP 447110 and 447190; FHWA parking survey	Truck AADT per fueling position; parking spaces per corridor mile
Gasoline station with convenience store	Vehicle trip	State DOT AADT; ACS vehicles	CBP 447110	AADT per station within the interchange
Hotel and motel	Room night	Airport enplanements (BTS, FAA); QCEW employment; BEA tourism	CBP 721110; SEC EDGAR EX-102 loan tapes; local permits	Enplanements per room; employment per room
Casino hotel	Gaming visit	State gaming commission revenue reports	CBP 721120	Gaming revenue per room
Wedding and event venue	Marriage	NCHS marriage rates by state; ACS B12501 marriages in the past year	CBP 812990, 722320; assessor use codes	Marriages per venue per year
Independent living, 55-plus	Age-qualified household	Vintage 2025 population 55 and older; ACS B19037 income by age of householder	CBP 623311, 531110; HUD and state registries	Units per thousand income-qualified households
Assisted living	Resident aged 80 and older needing help with daily activities	Vintage 2025 single year of age; ACS income by age	CBP 623312; state licensing rolls	Licensed beds per hundred residents aged 80 and older
Memory care	Resident with cognitive impairment	Vintage 2025 85-plus cohort; CDC prevalence studies	State licensing rolls (memory care endorsement)	Beds per hundred residents aged 85 and older
Skilled nursing	Medicare or Medicaid resident day	CMS utilization files; Vintage 2025 85-plus	CBP 623110; CMS Care Compare	Certified beds per thousand residents aged 75 and older
Continuing care retirement community	Entry-fee household	ACS income and home value by age	CBP 623311	Units per thousand qualified households
Child day care	Child under 5 with all parents working	Vintage 2025 under-5 population; ACS B23008; DOL National Database of Childcare Prices	CBP 624410; state licensing rolls	Licensed slots per hundred children under 5
Preschool and early education	Child aged 3 to 4	Vintage 2025 single year of age; state pre-K enrollment	CBP 624410, 611110; state licensing	Seats per hundred children aged 3 to 4
Medical office	Patient visit	HRSA Area Health Resources Files; ACS B27010 insurance by age; CMS NPPES	CBP 621111; NPPES practice locations	Physicians per 100,000 residents; square feet per physician
Dental office	Insured patient	HRSA dentists by county; ACS insurance coverage	CBP 621210	Dentists per 100,000 residents
Outpatient behavioral health	Client episode	SAMHSA N-SUMHSS; HRSA mental health shortage areas	CBP 621420	Facilities per 100,000 residents
Psychiatric and substance use hospital	Inpatient day	SAMHSA N-SUMHSS; CMS utilization	CBP 622210	Beds per 100,000 residents
Residential treatment	Resident stay	SAMHSA N-SUMHSS	CBP 623220; state licensing	Beds per 100,000 residents
Cold storage	Pallet position under refrigeration	USDA NASS cold storage stocks; Census of Agriculture; port and processing volumes	USDA NASS Capacity of Refrigerated Warehouses; CBP 493120	Refrigerated cubic feet per resident and per acre of production
Multifamily rental	Renter household	ACS households by age and tenure; Vintage 2025; HVS vacancy	Building Permits Survey 5-plus units; NRC starts and completions; SOMA absorption	Units in pipeline per hundred renter households; 3-month absorption share
Marina	Registered vessel	USCG registrations by state; BEA boating and fishing value added	CBP 713930; Army Corps permits	Registered vessels per wet slip
Grocery store	Food-at-home dollar	ACS income; BLS Consumer Expenditure Survey; USDA SNAP retailer data; ERS Food Access Research Atlas	CBP 445110; SNAP authorized-store roll	Food-at-home spending per grocery square foot
Convenience store	Vehicle trip and resident	AADT; ACS population and vehicles	CBP 445120, 447110	Residents and AADT per store
Data center	Megawatt of deliverable power	LBNL electricity use series; utility interconnection queues; EIA Form 861	CBP 518210; utility large-load filings	Queued megawatts against substation capacity
Manufactured housing community	Pad-renting household	Census Manufactured Housing Survey shipments and prices; ACS B25024 mobile homes; HVS rents	CBP 531190; state park registries	Pads per thousand households below the area median income
Agricultural storage and processing	Bushel, head or carcass	Census of Agriculture; NASS crop and livestock reports	CBP 493130, 311; USDA licensed warehouses	Storage capacity per acre harvested
Rural commercial (USDA B&I scope)	Guaranteed-loan project	Census of Agriculture; ACS rural population	CBP by county; USDA program data	Project scale against county demand
Suburban office	Office-using job	QCEW employment in office-using sectors; LEHD LODES	Assessor rolls; local permits	Office-using jobs per square foot
In-line retail	Household retail dollar	ACS income; BLS Consumer Expenditure Survey	Assessor rolls; CBP retail sectors	Retail spending per square foot

Two rows are deliberately included to make the inversion visible. Suburban office and in-line retail, the classes most lenders treat as ordinary, have no federal series that records their demand unit directly; the analyst has to infer office-using employment from the Quarterly Census of Employment and Wages and retail spending from household income, and the supply count comes from assessor rolls that vary in quality by county, as the library's state-by-state review of parcel records documents. The special-purpose rows above them carry a direct federal count of the user population. That asymmetry is not a curiosity; it is the reason an analytics platform built on public data can be more useful on a marina or a cold storage loan than on an office loan.

People: cohorts, households and income

The demographic base for every consumer-facing class is the Census Bureau's population estimates program, and the current vintage is the one to cite. Vintage 2025, whose county detail by single year of age, sex, race and Hispanic origin was released on 25 June 2026, is the first vintage to carry full demographic detail from the 2020 Census in its base. Its headline for asset-class work is the age structure. Residents aged 65 and older numbered 64,617,088 on 1 July 2025, an increase of 8,988,997 or 16.2% since 1 April 2020, against growth of 3.1% for the population as a whole. The South accounted for 3,694,167 of that gain, 41% of the national increase in the cohort, and grew its 65-and-older population by 17.5% in five years. The South was also the only region that grew in all five age groups the Bureau reports; the Northeast grew 0.7% overall, the Midwest 1.1%, the West 1.9%. The national median age reached 39.4 years, with Maine at 44.9 and Utah at 32.6 marking the extremes (U.S. Census Bureau, Vintage 2025 Population Estimates, 2026).

These numbers are the raw material of cohort arithmetic. Senior housing demand is not a function of the 65-plus population but of the 80-plus population, and within it the share that can pay: the American Community Survey's table B19037 gives household income by age of householder, and the 2020-2024 five-year file released on 8 January 2026 carries it for every county, while the 2024 one-year file released on 11 September 2025 carries it for areas of 65,000 residents or more. The senior housing demand cohort math article in this pillar works the chain from single year of age to income-qualified households to penetration rate, and the same chain, with the ages reversed, produces childcare demand: the population under five from Vintage 2025, the share with all parents in the labor force from ACS table B23008, the licensed supply from state rolls, and the price environment from the Department of Labor's National Database of Childcare Prices, which carries county-level prices from 2008 through 2022 and was last updated on 9 December 2025.

Household formation drives multifamily, self-storage and manufactured housing demand, and the ACS supplies it by age of householder and tenure. Self-storage is the class most sensitive to transitions rather than levels: households that moved in the past year, divorced in the past year, or added a member are the users, which is why the self-storage demand model treats mover rates as a driver alongside population and why the supply-per-capita benchmark in the benchmarks pillar has to be built from rentable square feet rather than establishment counts. Vehicles available by household, ACS table B25044, is the driver for car washes and the second driver for convenience retail. Health insurance coverage by age, table B27010, is the closest public proxy for payor mix in medical and dental office demand, and marriages in the past year, table B12501, put a county-level number behind the marriage series that the wedding venue article uses at the state level.

The discipline with all of these is vintage and geography. A demand model that mixes a 2025 population estimate with a 2019-2023 ACS income distribution is mixing reference dates five years apart; the model should state both dates and, where the difference matters, carry the estimate forward with the population series rather than silently assume the income distribution has not moved. Block-group ACS estimates carry margins of error that can exceed the estimate itself for small cohorts; the library's trade-area demographics article shows where ring apportionment breaks, and the remedy is to aggregate to the tract or county before computing a ratio that will appear in a credit memo.

Movement: traffic, vehicles and freight

For the classes whose customer arrives by road, the driver series is a traffic count, and traffic counts are a state function with a federal aggregation. Every state department of transportation maintains annual average daily traffic on its counted segments, and the Federal Highway Administration assembles the counted network into the Highway Performance Monitoring System. The HPMS public release carries, for every segment of the Federal-Aid system in the 50 states, the District of Columbia and Puerto Rico, the all-vehicle AADT together with separate fields for combination-truck and single-unit-truck AADT; the public geospatial release on the FHWA site is the 2018 data year, with the page last updated on 13 September 2022, so current counts come from the state portals, whose coverage and update cadence the library's AADT sources article reviews state by state and its national traffic atlas maps. FHWA's own caveat belongs in every model that uses the file: national summaries built from the public release can differ from the Highway Statistics tables, and the data are not suited to navigation.

The car wash is the purest traffic-capture class. Its demand unit is the vehicle, its driver is AADT on the fronting road combined with vehicles available in the surrounding households, and its supply count is County Business Patterns code 811192, which counted 19,807 employer car wash establishments in 2023 employing 172,644 people; 52% of those establishments had fewer than five employees, which is the signature of the older self-serve and in-bay formats rather than the express tunnel. The car wash demand article builds the capture-rate model and the membership sizing that express operators and their lenders need. Gasoline stations with convenience stores, code 447110, numbered 96,002 establishments, the largest count in the mobility group, and other gasoline stations added 13,172; under the 2022 revision of NAICS these codes become 457110 and 457120, a change the Quarterly Census of Employment and Wages has already adopted for 2024 and County Business Patterns will adopt with a later vintage.

Travel centers are a freight class, and their driver is the combination truck. HPMS truck AADT on the Interstate segment, corridor flows from the Freight Analysis Framework, and the one federal survey that measures the constraint directly: the Jason's Law truck parking surveys that MAP-21 required of FHWA. The 2019 survey, presented to the National Coalition on Truck Parking on 1 December 2020, counted about 313,000 truck parking spaces nationally, 40,000 at public rest areas and 273,000 at private truck stops, up 6% and 11% respectively from the 2014 survey, which had counted about 36,000 public and more than 272,000 private spaces. Of the 11,696 drivers who responded in 2019, 43% more than in 2014, 98% reported difficulty finding safe parking and 75% said the problem arose at least weekly; 79% of truck stop operators said they did not plan to add parking. Drivers named New York, New Jersey, Pennsylvania, Illinois and Georgia most often, along with the whole of the I-95 corridor, I-5 in California, the Chicago region and the Pacific corridors (FHWA, Jason's Law Truck Parking Survey, 2015 report and 2019 results, 2020). For a lender on a travel center loan, that is a demand statement of unusual clarity: the constraint is measured, its growth is measured, and the incumbents have declared that they will not relieve it. The travel center demand article turns it into a fuel and parking model.

Records: marriages, registrations and authorizations

A third family of drivers is administrative rather than statistical. Governments record marriages, register boats, authorize retailers, license providers and enroll patients, and each of those records is a direct count of a demand unit. The wedding venue is the cleanest case. The National Center for Health Statistics compiles provisional counts of marriages from state health departments and publishes rates by state of occurrence for 1990, 1995 and 2000 through 2023; the national figures for 2023 are 2,041,926 marriages and a rate of 6.1 per 1,000 residents, with 672,502 divorces reported by 45 states and the District of Columbia (CDC NCHS, 2025). The Census Bureau measures the same event from the other side: the American Community Survey asks whether a person married in the past twelve months, and its 2022 estimate was 16.7 marriages per 1,000 women aged 15 and older, with the District of Columbia at 24.4, Colorado at 20.9 and Nebraska at 20.8 at the top and New Mexico at 12.1 at the bottom of the states (U.S. Census Bureau, America Counts, 2024). The difference between the two series is itself a demand signal: NCHS counts the marriage where it happens, the ACS counts it where the couple lives, and a state whose occurrence rate far exceeds its residence rate is importing weddings. The wedding venue article and the benchmarks pillar's marriage records series develop that reading.

Marinas have a registration series with a 36-year history. The Coast Guard's Recreational Boating Statistics report compiles the vessels registered by each state: 10,777,370 in 1989, a peak of 12,942,414 in 2005, 11,546,512 in 2023 and 11,674,073 in 2024, of which 10,852,992 were mechanically propelled and 6,480,886 were between 16 and 26 feet long, the size band that fills wet slips and dry stacks. Florida registered 1,171,432 vessels in 2024, followed by Minnesota at 865,379, Michigan at 795,494, Ohio at 630,288, Wisconsin at 602,849, California at 596,703 and Texas at 552,138 (U.S. Coast Guard, Recreational Boating Statistics 2024, published 2025). The report's own warning matters for any per-capita comparison: states differ in what they register, with Minnesota, Ohio, Arkansas and South Carolina registering all watercraft with exceptions while California and New York register motorized vessels, so the scope column in the state table travels with the number. The marina demand article pairs the registration series with water access and slip inventories.

Retail authorization rolls serve the grocery class. The Department of Agriculture's SNAP retailer data lists every store authorized to accept benefits at any point in the past twenty years, with store type, county, coordinates and authorization dates, and because authorization requires meeting federal stocking standards under 7 CFR 278.1, the roll is a verified inventory of food retail by format. Combined with the Economic Research Service's Food Access Research Atlas, which flags low-income, low-access tracts, it lets the grocery-anchored retail demand analysis identify trade areas where food-at-home spending exceeds the supply that serves it. Provider registries serve the health classes: the CMS National Plan and Provider Enumeration System gives practice locations for every billing provider, and the Health Resources and Services Administration's Area Health Resources Files, whose 2024 release is dated 29 January 2026, carry more than 6,000 county, state and national variables from more than 60 sources, including physicians and dentists per county. Those are the inputs to the medical and dental office demand model and to the behavioral health analysis, which adds the Substance Abuse and Mental Health Services Administration's annual facility survey.

Counting supply: County Business Patterns and its blind spots

Supply is counted before it is measured, and the count that spans every asset class is County Business Patterns. The 2023 vintage, released on 26 June 2025, reports establishments, employment during the week of 12 March, first-quarter payroll and annual payroll for nearly 1,000 industries at national, state, metropolitan, county, congressional district and ZIP code levels, coded to NAICS 2017. Its national file counted 8,361,342 establishments with paid employees and 139,831,742 employees. The figures that matter for this pillar sit at the six-digit level. Hotels and motels other than casino hotels numbered 55,895 establishments with 1,497,840 employees; child day care services 82,162 establishments and 1,045,052 employees; offices of physicians 204,617 and offices of dentists 135,665; assisted living facilities for the elderly 20,884, nursing care facilities 18,126 and continuing care retirement communities 5,639; supermarkets and grocery stores 62,947; self-storage lessors 18,564; RV parks and campgrounds 5,018; marinas 3,739; refrigerated warehousing 1,376; data processing and hosting 18,544; lessors of other real estate property, the code that carries manufactured home sites, 8,957 (U.S. Census Bureau, County Business Patterns 2023, released 2025). The County Business Patterns article in the methodology pillar explains the file structure and the suppression flags.

The file has four blind spots that an asset-class model must name. First, it counts employer establishments only. A self-storage facility run by a remote manager, an RV park operated by its owner, a manufactured housing community with no on-site payroll: none of these appear, and the 91% of self-storage establishments that report fewer than five employees is a warning that the class sits at the edge of the file. The Nonemployer Statistics series fills part of the gap, and for self-storage the honest supply count is rentable square feet from local permits and assessor rolls, not establishments. Second, it counts establishments, not properties: a hotel operator with three properties in a county is three establishments, a dental group with one address and twelve chairs is one. Third, county cells for thin industries are suppressed and carry an employment range flag rather than a number, so a county-level ratio for a class with four establishments is built on a flag, and the model should say so. Fourth, the NAICS vintage lags: CBP 2023 is coded to NAICS 2017, QCEW 2024 to NAICS 2022, and codes that moved between vintages, gasoline stations above all, cannot be joined across the two series without a crosswalk.

The size-class columns are the under-used part of the file. The share of establishments with fewer than five employees separates formats within a code: 31% for hotels and motels, which says most of the count is full-service or select-service properties with staff, against 82% for all other personal services, the code that holds wedding chapels, and 78% for lessors of nonresidential buildings, the code that holds most small-bay flex landlords. For the small-bay flex industrial model, the demand side is the same file read from the other direction: the number of establishments with five to nineteen employees in the sectors that lease small bays, contractors, wholesalers, repair shops and light producers, is the tenant population, and its growth against the flex inventory is the demand ratio.

Pipeline and absorption: the multifamily case

Multifamily is the asset class with the most complete public pipeline record, and it is the template for reading pipeline anywhere. The Census Bureau and HUD publish New Residential Construction monthly, and the December release carries the annual totals by structure type. In 2024 permit-issuing places authorized 1,478,000 housing units, of which 441,600 were in buildings with five or more units; starts of such units were 336,200 and completions 591,700, the highest multifamily completion total in decades. In 2025, on preliminary figures, permits for five-plus units rose 4.3% to 460,400, starts rose 18.0% to 396,600, and completions fell 20.3% to 471,800, while single-family permits fell 7.4% to 909,600 and total permits fell 3.6% to 1,425,200. In December 2025 the seasonally adjusted annual rate of five-plus permits was 515,000, starts 402,000 (U.S. Census Bureau and HUD, New Residential Construction, December 2025, release CB26-28, 2026). The sequence matters more than any single number: the completion wave of 2024 has passed, starts turned up in 2025, and the units started in 2025 will complete in 2026 and 2027 into whatever demand exists then.

Absorption is measured by a survey most lenders have never heard of. The Survey of Market Absorption of New Multifamily Units follows privately financed, unsubsidized, unfurnished units in buildings of five or more units from completion and reports the share rented within three months. The series peaked at 75% for units completed in the third quarter of 2021 and has since fallen below 50% for four consecutive quarters, a run the survey had never recorded before; in the latest release discussed by the National Association of Home Builders in March 2026, 47% of the 93,680 units completed in the quarter were rented within three months, completions stayed above 90,000 for a seventh consecutive quarter, and the median asking rent of new units was $1,860, up 5.3% from $1,766 twelve months earlier (NAHB analysis of Census SOMA, 2026; quarter labels to be read from the Census table). The Housing Vacancies and Homeownership survey supplies the stock-level context: in the second quarter of 2026 the rental vacancy rate was 7.3%, against 7.0% a year earlier, the homeowner vacancy rate 1.2%, the homeownership rate 65.0%, and the median asking rent for vacant units $1,531; rental vacancy was 9.5% in the South and 6.9% in the Midwest, and the Bureau notes that its fourth-quarter 2025 estimates rest on November and December data only because of the federal funding lapse (U.S. Census Bureau, HVS, release CB26-116, 28 July 2026).

The multifamily supply pipeline from permits article builds the county-level version of this reading, and the methodology pillar's Building Permits Survey article explains the difference between permit-issuing places and the county totals that the annual file carries. For the other asset classes, the pipeline record is thinner and has to be assembled: local permit records by use code, which the platform's parcel and permit layers surface, and for hotels, self-storage and retail the securitized loan filings in SEC EDGAR, whose EX-102 schedules disclose property type, occupancy and net operating income for every loan in a public conduit deal and which the library's hotel supply pipeline article reads for lodging.

Food-system and industrial demand

Cold storage is the asset class where the federal record measures supply most precisely, because the Department of Agriculture has surveyed refrigerated warehouses in odd-numbered years since the 1920s. The 54th survey, with a reference date of 1 October 2025 and published in February 2026, found gross refrigerated capacity of 3.99 billion cubic feet, up 7.9% from 3.70 billion two years earlier; usable capacity, which strips out aisles, posts, coils and blowers, was 3.25 billion cubic feet or 82% of gross, and freezer space made up 79% of usable space. The 931 warehouses divided into 476 public warehouses holding 2.46 billion gross cubic feet, 62% of the total, and 455 private and semiprivate warehouses holding 1.53 billion. California led with 400 million cubic feet, followed by Georgia at 304, Washington at 301, Wisconsin at 297 and Texas at 254 million (USDA NASS, Capacity of Refrigerated Warehouses, 2026). County Business Patterns adds the establishment view: 1,376 employer establishments in refrigerated warehousing with 66,236 employees in 2023, against 16,753 general warehousing establishments.

The demand side of cold storage is the food system itself, and it is measured in the same department. The 2022 Census of Agriculture, released on 13 February 2024, counted 1.9 million farms and ranches, 7% fewer than in 2017, on 880 million acres, 2% fewer; the average farm grew 5% to 463 acres, family operations ran 95% of farms and 84% of the land, and farms sold $543 billion of agricultural products against $389 billion in 2017, with production expenses of $424 billion and net cash income of $152 billion. The average producer was 58.1 years old, 0.6 years older than in 2017; 296,480 producers, 9% of the total, were under 35 (USDA NASS, 2022 Census of Agriculture, 2024). Read against the capacity survey, the two series give the cold storage demand article a ratio that no private source publishes: refrigerated cubic feet per dollar of regional agricultural output, tracked state by state across two surveys. The agricultural facilities article extends the same sources to grain storage, packing and processing, where the Department of Agriculture's own lending rules define the market evidence a guaranteed loan must carry: the OneRD regulation at 7 CFR 5001.3 defines a feasibility study as a report by an independent qualified consultant evaluating the economic, market, technical, financial and management feasibility of a project, and the legacy Business and Industry rule at 7 CFR 4279.150 states that the income approach of an appraisal is not an acceptable substitute. This library stays on the demand measurement side of that line; the requirement itself belongs to the lender's counsel and the program's guidance.

Infrastructure-constrained classes: power, water, fiber and parking

Some asset classes are constrained not by the population of users but by an input the site must deliver, and for those the demand model inverts: the question is not how many users exist but how much of the constrained input the site can obtain. Data centers are the defining case. Lawrence Berkeley National Laboratory's report for the Department of Energy, published on 20 December 2024, estimated that U.S. data centers used 58 terawatt-hours of electricity in 2014 and 176 terawatt-hours in 2023, about 4.4% of national consumption, with load growth tripling over the decade; its scenarios for 2028 run from 325 to 580 terawatt-hours, 6.7% to 12.0% of projected national use, a doubling or tripling in five years (LBNL, 2024 United States Data Center Energy Usage Report, 2024). A population model tells a lender nothing useful about a site in that market; what matters is the utility's interconnection queue under the Federal Energy Regulatory Commission's Order No. 2023, the substation capacity in the Energy Information Administration's Form 860 and Form 861 files, the fiber routes in the FCC's National Broadband Map, and the water rights and withdrawals in state and Geological Survey records. The data center constraints in public records article builds the screen from those sources, and its demand ratio is queued megawatts against deliverable capacity rather than anything per capita.

Travel centers are constrained by parking in the same way that data centers are constrained by power, and the Jason's Law surveys measure that constraint as directly as the Berkeley report measures load. The point of pairing the two in one emblem is methodological: both series are published by a federal agency, both are updated on a known cadence, both measure the binding constraint rather than a proxy for it, and neither appears in a conventional market study. A demand model that cites them is making a statement a credit committee can check. Marinas sit in the same family, with the constraint being water access and federal permits under section 10 of the Rivers and Harbors Act and section 404 of the Clean Water Act; so do RV parks, where the RV park demand article reads seasonality from National Park Service visitation and the outdoor recreation satellite account, which valued the whole outdoor recreation economy at $696.7 billion of value added in 2024, 2.4% of GDP, with boating and fishing at $38.4 billion, RVing at $27.5 billion, hunting, shooting and trapping at $16.5 billion and snow activities at $7.6 billion (BEA, 2026).

What the lender reads: special-purpose properties and the SBA record

Lenders already sort asset classes by observability; they just call it something else. The Small Business Administration's standard operating procedure for its 7(a) and 504 programs, SOP 50 10 8, effective 1 June 2025, distinguishes limited or special purpose properties from general-purpose real estate and requires a larger equity injection on 504 projects that finance them. The standard contribution on a 504 project is 10%; it rises to 15% for a special purpose property or a new business and to 20% when both conditions hold. The examples that lender guidance on the SOP cites include hotels and motels, gasoline stations, car washes, nursing homes and assisted living facilities, marinas, and cold storage facilities where more than half of the square footage is equipped for refrigeration; the full list sits in the SOP itself. The logic is collateral logic: a special purpose building has few alternative users if the operator fails. The logic of this pillar is the mirror image: a special purpose building has a demand unit the public record counts, so the probability that the operator fails can be read from the data before the loan is made. The library's special-purpose property risk article in the lender workflow pillar takes up what the loan data shows.

The loan data is public. The SBA's FOIA releases of the 7(a) and 504 portfolios carry more than one million loan records with the industry code of every borrower, and MMCG Analytics' SBA layer is derived from them. MMCG's analysis of loans approved from fiscal 2014 through fiscal 2021, counted by loan rather than by dollar, found an all-industry adverse-resolution rate of 7.6%, the share of loans that ended in charge-off, liquidation or guaranty purchase, with child day care services at 3.9%, assisted living facilities at 3.8%, offices of dentists at 2.9% and gasoline stations with convenience stores at 2.4% among the lowest-risk industries and fitness and recreational centers at 11.0% among the highest (MMCG Research, SBA 7(a) Performance Series, 2026). Several of the asset classes in this pillar therefore carry a measured credit history by industry alongside their demand series, which is a combination no paid dataset offers. Two rules travel with that history. The first is suppression: no performance rate is shown, in this library or on the platform, for any cohort of fewer than ten loans, so a county-level rate for a thin asset class is aggregated up to the state or omitted. The second is role: MMCG Analytics supplies the data and the analysis, and the credit decision rests with the lender and its own policy. The SBA FOIA loan datasets article in the methodology pillar documents the file structure and the suppression standard.

Method: the provenance standard and ten failure modes

Every figure in this pillar carries a named source and a year, and every cluster article inherits that rule. The provenance standard article sets out why; this section sets out how it fails in practice. Ten failure modes account for most of the bad demand models that reach a credit file. One, the silent vintage: a population figure from 2025 divided by a supply count from 2019 with neither date stated. Two, the wrong cohort: senior housing demand built on the 65-plus population when the resident is 84. Three, the establishment count used as a capacity count, which overstates supply where facilities are small and understates it where they are large. Four, the employer-only count used for a class that lives outside the employer universe, self-storage and RV parks above all. Five, the residence-based series used for an occurrence-based demand unit, which is how a wedding venue in a destination county gets modeled on its own residents' marriages. Six, the ring drawn without regard to the road network, which the library's trade-area article treats at length. Seven, the suppressed cell treated as zero. Eight, the NAICS join across vintages without a crosswalk. Nine, the pipeline read from permits alone, without completions and absorption, which in 2025 would have missed a 20% fall in multifamily completions. Ten, the constraint ignored: a data center or a travel center modeled on population when the binding variable is megawatts or parking spaces.

The remedy for all ten is the same discipline: name the demand unit, name the driver series with its vintage and geography, name the supply series with its vintage and its known blind spots, compute the ratio in the natural units of the class, and carry every source and date into the document the lender will read. The public-data stack pillar describes the series in depth; this pillar describes how they combine by asset class; the cluster articles that follow it, from hotel demand and childcare deserts to behavioral health, early education and manufactured housing, each work one row of the table above to the point where a ratio can be defended line by line. That is what demand analysis by asset class means when it is done from public data: not a lower-cost imitation of a subscription report, but a different product, built on records that anyone can check, for the property types where the records are richest and the credit questions are hardest.

Frequently asked questions

What is demand analysis by asset class in commercial real estate?

It is the practice of measuring the population that uses a specific kind of property, counting the capacity that already serves it, and expressing the two as a ratio in the natural units of that property type: registered vessels per slip, licensed childcare slots per hundred children under five, refrigerated cubic feet per resident. The driver series, the supply series and the ratio differ for each asset class, which is why a generic population-and-income study fails for travel centers, data centers and marinas.

Which public data sources measure demand for commercial property types?

The Census Bureau's population estimates and American Community Survey measure people, households, income, vehicles and marriages; County Business Patterns counts competing establishments by six-digit industry; the Building Permits Survey and New Residential Construction measure the pipeline; state departments of transportation and the FHWA's Highway Performance Monitoring System measure traffic; the Coast Guard compiles boat registrations; the National Center for Health Statistics counts marriages; the Department of Agriculture measures refrigerated capacity and farm output; the Bureau of Economic Analysis values outdoor recreation; and Lawrence Berkeley National Laboratory estimates data center electricity use. Each carries a vintage that the model must state.

How current are the public series used in demand analysis?

Vintage 2025 population estimates carry a reference date of 1 July 2025 and were released in June 2026. The 2024 one-year American Community Survey was released in September 2025 and the 2020-2024 five-year file in January 2026. County Business Patterns 2023 was released in June 2025. New Residential Construction is monthly, the Housing Vacancies survey quarterly, and the USDA refrigerated warehouse survey biennial, with the October 2025 reference date published in February 2026. A demand model should cite the reference date of every series, not the release date alone.

Why are special purpose properties easier to model from public data?

Because their demand unit is something a government already counts. Vehicles, registered boats, marriages, truck traffic, refrigerated capacity and interconnection queues are recorded by federal or state agencies on a known cadence. Generic office and in-line retail demand has to be inferred from employment and income because no agency records a tenant's leasing decision. The asset classes lenders treat as the riskiest collateral are therefore the ones whose demand is most directly observable.

Can public data replace paid market data for demand analysis?

For the driver side of the model, public series are usually the primary source that paid products repackage, and they carry the provenance a credit committee can check. For supply, public establishment counts and permits need the caveats described above, and rentable capacity for classes such as self-storage has to be built from permits and assessor rolls. The honest answer is that public data produces a different and more auditable product, strongest for the special purpose classes and weakest for generic office and retail.

How do lenders use demand ratios from public data?

As a comparable: the ratio for the subject market against other markets, against the same market in an earlier vintage, and against the project's own capacity. The SBA's public loan datasets add a measured credit history by industry, subject to the rule that no rate is shown for fewer than ten loans. The analytics inform the credit file; the lending decision remains the lender's.

Sources

1. U.S. Census Bureau, County Business Patterns 2023, released 26 June 2025; national file cbp23us. https://www.census.gov/newsroom/press-releases/2025/2023-county-business-patterns.html
2. U.S. Census Bureau, Vintage 2025 Population Estimates, county characteristics, press release of 25 June 2026. https://www.census.gov/newsroom/press-releases/2026/vintage-2025-pop-estimates.html
3. U.S. Census Bureau, American Community Survey data releases: 2024 1-year estimates (11 September 2025) and 2020-2024 5-year estimates (8 January 2026). https://www.census.gov/programs-surveys/acs/news/data-releases.html
4. U.S. Census Bureau and U.S. Department of Housing and Urban Development, New Residential Construction, December 2025, release CB26-28 (2026). https://www.census.gov/construction/nrc/pdf/newresconst_202512.pdf
5. U.S. Census Bureau, Quarterly Residential Vacancies and Homeownership, second quarter 2026, release CB26-116, 28 July 2026. https://www.census.gov/housing/hvs/files/currenthvspress.pdf
6. U.S. Census Bureau, Survey of Market Absorption of New Multifamily Units (SOMA), program page, 2026; figures as reported by NAHB Eye on Housing, "Multifamily Absorption Rate Remains Below 50%", 3 March 2026. https://www.census.gov/programs-surveys/soma.html and https://eyeonhousing.org/2026/03/multifamily-absorption-rate-remains-below-50/
7. U.S. Department of Agriculture, National Agricultural Statistics Service, Capacity of Refrigerated Warehouses, February 2026 (reference date 1 October 2025). https://www.nass.usda.gov/Publications/Todays_Reports/reports/rfwh0126.pdf
8. U.S. Department of Agriculture, National Agricultural Statistics Service, 2022 Census of Agriculture, news release of 13 February 2024. https://www.nass.usda.gov/Newsroom/2024/02-13-2024.php
9. Lawrence Berkeley National Laboratory for the U.S. Department of Energy, 2024 United States Data Center Energy Usage Report, 20 December 2024. https://www.energy.gov/articles/doe-releases-new-report-evaluating-increase-electricity-demand-data-centers
10. Centers for Disease Control and Prevention, National Center for Health Statistics, FastStats: Marriage and Divorce, provisional 2023 data, page reviewed 17 March 2025. https://www.cdc.gov/nchs/fastats/marriage-divorce.htm
11. U.S. Census Bureau, America Counts, "How Does Your State Compare With National Marriage and Divorce Trends?", October 2024 (ACS 2022). https://www.census.gov/library/stories/2024/10/marriage-and-divorce.html
12. U.S. Coast Guard, Recreational Boating Statistics 2024, COMDTPUB P16754.38, 2025, Tables 29, 36, 37 and 38. https://www.uscgboating.org/library/accident-statistics/Recreational-Boating-Statistics-2024.pdf
13. Federal Highway Administration, Jason's Law Truck Parking Survey Results and Comparative Analysis, 2015; 2019 survey results presented to the National Coalition on Truck Parking, 1 December 2020. https://ops.fhwa.dot.gov/freight/infrastructure/truck_parking/jasons_law/truckparkingsurvey/es.htm and https://ops.fhwa.dot.gov/Freight/infrastructure/truck_parking/coalition/2020/mtg/mtg12012020_jasons_law.htm
14. Federal Highway Administration, HPMS Public Release of Geospatial Data in Shapefile Format, 2018 data year, page updated 13 September 2022. https://www.fhwa.dot.gov/policyinformation/hpms/shapefiles.cfm
15. U.S. Bureau of Economic Analysis, Outdoor Recreation Economic Statistics: U.S. and States, 2024, released 5 March 2026. https://www.bea.gov/news/2026/outdoor-recreation-economic-statistics-us-and-states-2024
16. Health Resources and Services Administration, Area Health Resources Files, 2024 release, 29 January 2026. https://data.hrsa.gov/topics/health-workforce/ahrf/
17. U.S. Department of Labor, Women's Bureau, National Database of Childcare Prices, 2008 to 2022, updated 9 December 2025. https://catalog.data.gov/dataset/national-database-of-childcare-prices
18. 7 CFR 5001.3 (definitions, OneRD Guaranteed Loan regulation) and 7 CFR 4279.150 (feasibility studies), Legal Information Institute, read 22 August 2026. https://www.law.cornell.edu/cfr/text/7/5001.3 and https://www.law.cornell.edu/cfr/text/7/4279.150
19. U.S. Small Business Administration, SOP 50 10 8, Lender and Development Company Loan Programs, effective 1 June 2025, and SBA Procedural Notice 5000-872764, Revisions to SOP 50 10 8, effective 30 September 2025, as posted by the National Association of Government Guaranteed Lenders, 2025; effective date and 7(a) minimums as summarized by Live Oak Bank, 2025; 504 special purpose property list and equity structure as summarized by 504 Capital Corporation, 2025. https://www.naggl.org/sba-notice-revising-sop-50-10-8/ ; https://www.liveoak.bank/blog/navigating-the-sbas-new-sop-50-10-8 ; https://504capital.com/blog/financing-special-purpose-properties-sba-504-loans/
20. MMCG Research, SBA 7(a) Performance Series: MMCG analysis of the public SBA 7(a) loan register, loans approved FY2014 to FY2021, 2026. https://mmcganalytics.com/sba-default-rates/