The Houses Outlive Us
Why ageing, shrinking households, and durable housing could split national property markets for decades
Keywords: demographics, ageing, housing, house prices, regional economics, population decline, migration, household formation, vacancy, collateral, banking, long-term forecasting
There is a comfortable way to think about housing and demographics. A country grows, households multiply, more homes are needed, and prices rise. Later, the country ages, births fall, the population stops growing, and eventually there are fewer people. At that point, the story appears to run in reverse: fewer people should mean fewer buyers, less demand, and lower house prices.
That story is attractive because it is simple. It is also almost certainly too simple to describe what an ageing society actually does to property markets.
The question is not merely how many people exist. The question is how many households exist, where those households are located, what sort of housing they demand, how much housing has already been built, how quickly unwanted housing can disappear, and what happens to local credit when property values begin to fall. Once those pieces are put together, the likely future is not a smooth national decline in house prices. It is a widening divergence between regions.
Some cities may continue to gain population, capital, skilled workers, and housing demand even while their country shrinks. Other places may inherit a housing stock built for a population that no longer exists. Because houses are durable, the excess supply does not vanish when the people do. It remains in place, increasingly vacant, increasingly cheap, and increasingly difficult to finance. If property is also the principal collateral behind household and small-business borrowing, falling regional property values may then feed back into the local economy through weaker bank balance sheets and tighter credit.
The result is a different research question from the one usually asked.
Instead of asking, “Will ageing make house prices fall?”, the more useful question is:
When demographic decline finally reduces household numbers, why do some regions remain resilient while others enter persistent cycles of vacancy, falling prices, weaker collateral, credit contraction, and further population loss?
That is the question behind a research project I am developing in regional economics and finance. The eventual academic paper is being designed for Regional Studies, with other regional and urban economics journals as fallbacks. What follows is the economic logic of the project, the evidence already supporting the idea, and the empirical structure needed to determine whether the mechanism is real.
Population is not housing demand
The first mistake in demographic housing analysis is treating population as though it were synonymous with housing demand.
It is not.
A city of one million people divided among 500,000 households is not the same housing market as a city of one million people divided among 350,000 households. The first needs roughly one dwelling for every two residents before allowing for vacancies, second homes, and institutional accommodation. The second needs materially fewer dwellings even though the headline population is identical.
This distinction becomes more important as societies age because household size can continue falling after population growth has stopped. Children leave home. Couples separate. Widows and widowers remain in homes previously occupied by two people. Older adults may live independently for longer than earlier generations did. Marriage patterns change. Fertility falls. The number of one-person households rises.
The practical consequence is that a country can experience stagnant or even declining population while the number of households continues to increase for a considerable period.
The housing market therefore has at least two demographic turning points:-
the population peak; and
-
the household peak.
For housing, the second may matter more.
A useful first-order identity is simply:
Total housing demand = number of households × housing services demanded per household
This is elementary, but it changes the problem. A population projection by itself cannot tell us what will happen to housing unless we know how that projected population will be organised into households and what those households will demand.
The older demographic housing literature established that age structure can influence demand. Mankiw and Weil (1989), in one of the classic papers in this area, constructed age-specific housing-demand profiles and used demographic composition to explain movements in aggregate housing demand and real house prices. Their work helped establish the intuition that large generations moving through house-buying ages can alter property demand.
The striking feature of that paper, viewed from the present, is not simply its result. It is the difficulty of long-horizon extrapolation. The historical relationship between demographics and prices appeared sufficiently strong to motivate an extremely bearish long-term housing implication. Later experience demonstrated why projecting a cross-sectional age-demand relationship far into the future is hazardous. Housing prices are affected by income, interest rates, credit, supply constraints, migration, productivity, taxation, construction costs, and the geography of economic opportunity, not only by the age composition of the population (Mankiw & Weil, 1989).
Takáts (2010) revisited the demographic-asset-price problem using a panel of 22 advanced economies from 1970 to 2009. His estimates associated higher population with higher real house prices and higher old-age dependency with lower prices. When combined with demographic projections, the estimated demographic component implied significant future headwinds. For the United States, the paper estimated roughly 80 basis points a year of demographic headwind relative to neutral demographics over the subsequent forty years. Crucially, Takáts emphasised that this was a demographic component, not an unconditional house-price forecast. Italy and Korea, for example, had experienced substantial house-price growth despite estimated demographic headwinds (Takáts, 2010).
That distinction is central. Demography may push. It does not determine the entire path.
Ageing does not automatically create a nation of forced sellers
The second mistake is assuming that older households mechanically consume less housing.
It sounds plausible. People reach retirement, children leave, the family house becomes unnecessarily large, and the household downsizes. If every older household behaved that way rapidly, ageing would release a large quantity of housing back onto the market.
But the relationship is more complicated.
Eichholtz and Lindenthal’s 2010 preprint, based on a detailed cross-section of English households, is useful precisely because it distinguishes simple age effects from characteristics associated with cohorts and human capital. Their analysis finds that housing demand is strongly related to human capital and need not fall sharply with age. Education, health, income, wealth, and the characteristics of particular generations can matter substantially for how much housing older households continue to consume (Eichholtz & Lindenthal, 2010).
That matters for long-term forecasting because a 75-year-old household in 2070 will not necessarily resemble a 75-year-old household observed in a survey taken decades earlier.
Future older households may be richer. They may live longer. They may remain healthier for longer. They may have accumulated different levels of pension and housing wealth. They may have fewer children. They may have different preferences for location, apartment living, care provision, accessibility, and multigenerational living. Mortgage structures may differ. Inheritance patterns may differ. Home-equity products may be more common. Remote work may alter where retirement occurs.
A model that says “people aged 75 historically consumed X units of housing, therefore people aged 75 in 2070 will consume X” quietly assumes away almost everything that makes long-term economic forecasting difficult.
This suggests a second improvement over a simple demographic forecast: cohort composition must be allowed to change.
Age is not destiny. An age coefficient is not a law of nature.
The national average may be the least interesting number
The third mistake is assuming that a national demographic trend creates a national housing outcome.
Japan makes the problem obvious.
Hashimoto, Hong, and Zhang (2020) examine Japan’s demographic decline and housing market across regions. Their findings show an asymmetric relationship between population change and housing prices: the price decline associated with population loss is larger than the price increase associated with equivalent population gains. The explanation is economically intuitive. Housing is durable. When population expands, new housing can be built. When population contracts, the existing stock cannot be “unbuilt” with equal ease.
The same paper also highlights the spatial character of decline. Population losses are not distributed uniformly. Rural and peripheral regions can shrink while major urban areas continue attracting people. The national population can fall while a successful metropolitan region remains expensive and supply constrained (Hashimoto et al., 2020).
This produces a deceptively simple but important proposition:
National population decline does not imply uniform regional housing decline.
The national average may therefore become less informative precisely when demographic change becomes more severe.
Imagine two regions inside the same shrinking country.
Region A is a productive metropolitan core. It has universities, high-value employment, specialist healthcare, transport connectivity, deep labour markets, cultural amenities, and large pools of human capital. Young adults leave weaker regions and move into it. International migrants also disproportionately choose it. Even if the national population falls, Region A can continue gaining households.
Region B is a smaller peripheral area. Its young people leave for Region A. Births fall. Deaths exceed births. Shops and services close. The tax base weakens. Employers struggle to recruit. The homes left behind are not necessarily in the locations or of the types demanded by the remaining population.
These two places share the same national interest rate, national demographic trend, national tax system, and national monetary policy. Their housing equilibria can nevertheless move in opposite directions.
That is the regional economics problem at the heart of the project.
Figure 1. Proposed mechanism. Demographic change affects housing through household formation and migration. Because the inherited housing stock adjusts slowly downward, some regions can develop persistent overhang. If lower property values weaken collateral and credit, the regional decline can become self-reinforcing. This is a conceptual figure, not an empirical result.
Houses have memory
Most goods disappear relatively quickly when consumers stop wanting them. Housing does not.
This is the property that makes demographic decline economically different from demographic expansion.
If a city gains 50,000 households, it can respond through higher prices, higher rents, denser occupation, conversion of existing structures, or new construction. Supply may be slow, but positive construction is feasible.
If the same city loses 50,000 households, the adjustment cannot simply run in reverse. Builders cannot construct minus 20,000 houses. Existing dwellings remain standing until they are demolished, converted, abandoned, or allowed to decay. Demolition is costly. Ownership can be fragmented. Taxes can create perverse incentives. Local governments may resist recognising permanent decline. Land values may fall below the cost of clearing structures. Some properties may have no economically meaningful market at all.
This suggests a central state variable for the research:
Housing-overhang ratio = usable housing stock / number of households
Call the ratio the housing-overhang measure. The exact empirical version will need refinement because “usable stock” is not identical to total recorded dwellings. Some vacant homes are seasonal properties. Some are temporarily vacant. Some are uninhabitable. Some are in locations where demand remains strong. Nevertheless, the basic ratio captures the mechanism better than population growth alone.
The central hypothesis is that adjustment becomes nonlinear once housing stock exceeds household demand by enough.
Below some level of excess supply, the market absorbs change through ordinary vacancy, turnover, rent adjustment, conversion, and moderate price movement.
Above a critical level, the system changes character. Vacancies become persistent. Transactions thin out. Comparable sales disappear. Maintenance is deferred. Properties become difficult to insure or finance. New construction falls toward zero, but that is not enough because the existing stock remains too large. Prices need to fall much further to attract marginal buyers, and even a low price cannot compensate for a location with deteriorating employment, schools, transport, or services.
The proposed empirical model therefore looks for a threshold rather than assuming one linear national elasticity.
In plain language:
If housing stock per household is already high, another loss of households should hurt prices more than the same loss would in a tight market.
That is a testable proposition.
It also gives demographic decline a mechanism. We are no longer saying merely that population falls and prices fall. We are specifying why, where, and under what market conditions the effect should become severe.
A demographic problem can become a financial problem
Housing is not merely a consumption good. It is an asset, a store of household wealth, a source of collateral, a tax base, and one of the principal assets supporting bank lending.
That creates the possibility of a financial feedback loop.
Suppose a shrinking region develops persistent housing overhang. Vacancy rises. House prices fall. The value of collateral behind existing mortgages falls with it. Loan-to-value ratios rise mechanically even if borrowers have made every scheduled payment. New buyers face weaker collateral values. Banks become more cautious about future recovery values in default. Developers cannot justify new projects. Small businesses whose owners use property as collateral may also face tighter borrowing conditions.
The mechanism can be stated without mathematical notation:
Demographic decline → housing overhang → lower property values → weaker collateral → tighter credit → weaker investment and employment → more out-migration.
The last arrow matters most.
If credit tightening merely follows demographic decline, finance is an outcome. If weaker credit then causes further economic decline and migration, finance becomes an amplifier.
That is what I mean by a regional collateral trap.
The term is deliberately conditional at this stage. It must be demonstrated, not asserted. If the banking evidence does not support the feedback mechanism, the academic paper should not use “collateral trap” as a headline claim. The paper would then remain a strong study of demographic shrinkage, housing-stock irreversibility, and regional housing divergence, with finance treated as an implication rather than a proven causal channel.
Fortunately, the data required to test the financial mechanism are increasingly available.
Wharton Research Data Services provides bank regulatory data drawn from mandatory filings, including commercial-bank Call Reports and bank-holding-company FR Y-9 reports. These contain balance-sheet, income, capital, risk, and off-balance-sheet information that can be used to examine how banks with different geographic exposures respond to deteriorating property markets (Wharton Research Data Services, n.d.).
The ambition is therefore not merely to observe falling regional house prices. It is to link local demographic and housing shocks to the financial institutions exposed to them.
A credible result would look something like this: banks more exposed to regions experiencing persistent household decline and housing overhang subsequently reduce property-related lending, increase provisions, experience higher losses, or contract local credit more than otherwise comparable banks exposed to demographically stronger regions.
That would convert a regional housing story into a regional finance story.
The real forecasting problem is not 2075. It is whether we can predict 2005 from 1990
Long-term demographic projections invite false precision.
It is easy to take the latest population projection, attach estimated coefficients, extend a line to 2050 or 2075, and produce a graph that looks scientific. The difficulty is knowing whether the model had genuine forecasting power or merely explained the past after seeing it.
The research design should therefore make forecasting credibility part of the contribution.
Before projecting anything several decades forward, the model should be forced to forecast periods for which we already know the outcome.
The idea is to reconstruct historical information sets.
Take a year such as 1990. Use only demographic projections, housing information, economic variables, and institutional data that were genuinely available at that time. Estimate what the model would have predicted for later regional housing outcomes. Then repeat the exercise using an information set from 2000, then 2010, and so on.
This creates pseudo-real-time forecasting.
The model ladder can be written simply:
Model 0: house prices + standard macroeconomic controls
Model 1: Model 0 + population
Model 2: Model 1 + age structure
Model 3: Model 2 + household formation
Model 4: Model 3 + migration
Model 5: Model 4 + housing stock, vacancies, construction, and demolition
Model 6: Model 5 + mortgage and bank-credit conditions
The question is not which model best fits historical data in sample. The question is whether adding household structure, migration, housing-stock irreversibility, and financial exposure produces better forecasts that survive outside the sample used to build the model.
This approach is also a direct response to the history of demographic housing forecasting. Mankiw and Weil (1989) showed how powerful the demographic signal could appear when historical age-demand relationships were extrapolated. Takáts (2010) was more explicit about the distinction between demographic headwinds and actual price forecasts and about the possibility that other forces can dominate demographic pressure.
A modern paper should go further. It should ask whether the demographic mechanism would have warned us correctly about regional housing divergence using information available before that divergence occurred.
If it cannot do that, a forecast to 2075 deserves little confidence.
What evidence would actually convince me?
A good research project should state in advance what evidence would support the thesis and what evidence would weaken it.
The demographic block requires regional population by age, births, deaths, fertility, and longevity. But that is merely the beginning.
The household block requires household counts, average household size, household composition, and preferably information about the age and tenure of household heads. This lets us identify the date at which household formation stops offsetting population decline.
The migration block requires internal migration, ideally origin-destination flows rather than just net migration. A region losing 20,000 people because young graduates leave is economically different from a region losing 20,000 retirees through mortality. Both reduce headcount. They do not have the same implications for future income, labour supply, housing demand, or the tax base.
The housing-stock block requires total dwellings, occupied units, vacant units, structure type, age of stock, permits, completions, conversions, demolitions, and abandonment where measurable.
The price block requires quality-adjusted house-price measures rather than simple median transaction values. The Federal Housing Finance Agency’s House Price Index is useful because it provides repeat-sales measures across national, state, metropolitan, county, ZIP-code, and census-tract geographies, with data extending back to the mid-1970s (Federal Housing Finance Agency, n.d.).
The U.S. Census Bureau’s American Community Survey provides detailed regional information on housing units, occupancy, vacancy, tenure, household characteristics, housing costs, and the structure of the housing stock. These measures make it possible to distinguish an ordinary level of transactional vacancy from persistent excess supply (U.S. Census Bureau, 2024).
The financial block requires mortgage originations, loan-to-value measures, credit performance, and bank exposures. WRDS bank regulatory data make at least part of this feasible because balance-sheet and regulatory measures can be matched to depository institutions over time (Wharton Research Data Services, n.d.).
The project therefore has a deliberately modular structure. If the banking linkage proves weak, the housing and regional mechanism can still stand. If the banking linkage is strong, it becomes an additional contribution rather than an assumption built into the story.
Why the United States is a useful laboratory even though the problem is global
The most severe demographic shrinkage today is not necessarily in the United States. Japan, parts of Europe, and parts of East Asia provide more advanced examples of ageing and population decline.
So why begin the deep empirical work with U.S. data?
Because identification often depends more on data architecture than on choosing the country with the most dramatic headline trend.
The United States combines unusually granular house-price measures, detailed household and housing-stock data, extensive mortgage information, bank regulatory data, and large differences in regional demographic experience. Some counties and metropolitan areas have experienced persistent population loss; others have attracted large inflows. Some regions have elastic housing supply; others are tightly constrained. Some local banking markets are exposed to property decline; others are not.
That variation can be used to identify the mechanism carefully.
The broader strategy is therefore:
United States: deep identification and financial transmission.
Japan, Europe, and OECD regions: external validation, demographic stress cases, and long-horizon projection.
This is preferable to constructing a giant international panel too early and pretending that every country measures house prices, vacancies, demolition, household formation, and regional boundaries in exactly the same way.
A smaller number of well-harmonised regional panels may yield more credible economics than a larger but internally inconsistent dataset.
What the long-term future might actually look like
If the mechanism is correct, several predictions follow.
1. The household peak should matter more than the population peak
A country can pass peak population without immediately developing excess housing because falling household size continues supporting dwelling demand. Housing-market weakness should intensify when household numbers themselves begin to decline.
This implies that commentators watching only total population may call the housing turning point too early.
2. Decline should be geographically concentrated
National shrinkage should increase the importance of migration within the country. Productive cores may retain or gain young households while weaker regions lose them. The distribution of population becomes more important than the national total.
The same national demographic trend can therefore support rising prices in one market and severe decline in another.
Figure 2. Stylized illustration only, not a forecast. If households and capital concentrate geographically while peripheral regions retain an inherited housing stock, national demographic decline can coexist with appreciation in core regions and substantial real-price decline elsewhere.
3. Vacancy should become an early-warning variable
Prices alone may adjust slowly, particularly where transaction volumes become thin. Persistent vacancy can reveal that the physical stock is beginning to exceed effective household demand before a conventional price index fully captures the deterioration.
A useful warning system may therefore combine household decline, vacancy persistence, construction collapse, and the age of the housing stock rather than relying on prices alone.
4. Downward adjustment should be more violent than upward adjustment
If housing is durable, demographic contraction should have an asymmetric effect. A growing city can respond through construction. A shrinking city cannot respond through negative construction. The inherited stock remains.
Hashimoto et al. (2020) document precisely this asymmetry in Japan: housing-price declines associated with population losses are larger than price increases associated with population gains of comparable magnitude.
That is one of the strongest empirical clues supporting the broader mechanism.
5. The quality of housing demand should change before aggregate demand collapses
Ageing affects not only the quantity of housing but the type demanded. Accessibility, proximity to healthcare, transport, maintenance requirements, apartment versus detached housing, and neighbourhood services can all change with age.
This means a region may experience simultaneous scarcity and surplus: too many large older houses in poorly connected locations and too few accessible dwellings near services.
The aggregate count of “housing units” can therefore hide a serious mismatch in housing characteristics.
6. Property decline can become an economic-development problem
Once the collateral mechanism is added, the consequences extend beyond homeowners.
A region with declining collateral values may experience weaker small-business borrowing, lower construction activity, lower municipal revenues, weaker local-bank performance, and less investment. If those effects encourage further out-migration, the process becomes cumulative.
The central policy question then changes from “How do we prop up house prices?” to “How do we manage the orderly contraction and reallocation of capital in places where the inherited built environment no longer matches the population?”
That is a much harder question.
The policy temptation will be to preserve every price. That may be a mistake.
Governments dislike falling house prices for understandable reasons. Housing is household wealth. Mortgages sit on bank balance sheets. Property taxes fund local government. Construction employs large numbers of workers. A falling market is politically painful.
But if the underlying problem is structural demographic decline, policies designed simply to prevent prices from adjusting can make the physical mismatch worse.
Subsidising new construction in a region with persistent excess stock would be particularly perverse. It would increase the quantity of housing precisely where the long-run problem is too much housing relative to households.
Likewise, tax systems that discourage demolition or make it costly to consolidate abandoned parcels can leave obsolete housing in place. The correct policy may involve accepting that some structures should disappear, some neighbourhoods should contract, and some public infrastructure should be consolidated.
This does not imply abandoning declining regions. It implies distinguishing between policies that improve productivity and quality of life and policies that merely prevent market clearing.
There are also distributional complications. A cheap house is not automatically affordable in an economically meaningful sense. A property that costs very little but is located far from employment, healthcare, schools, transport, or social networks may have low value because it provides poor access to the things households need.
Similarly, an older homeowner in a shrinking town can be “asset rich” on paper yet unable to sell at the expected price. If transaction volumes collapse, quoted valuations can overstate realisable wealth. Demographic housing decline may therefore affect retirement planning as well as regional development.
What would falsify the thesis?
The project should fail if the evidence fails.
Several findings would weaken the proposed mechanism.
First, if household decline adds no explanatory or predictive power once population change is known, then the distinction between population peak and household peak is less important than proposed.
Second, if housing-stock-per-household measures do not predict nonlinear price or vacancy responses, the stock-overhang mechanism is weak.
Third, if regions shed unwanted housing sufficiently rapidly through demolition, conversion, or alternative use, then durability may not create the persistent disequilibrium hypothesised.
Fourth, if migration does not systematically concentrate younger or higher-income households in stronger regions, the spatial-sorting channel is overstated.
Fifth, if banks exposed to shrinking property markets do not alter credit supply after controlling for borrower demand and local economic conditions, the collateral-trap extension should be rejected.
Finally, if the full model does not forecast historical regional outcomes better than simple macroeconomic or house-price models, then it should not be used to make confident projections to 2050 or 2075.
These are not inconvenient possibilities to be explained away after estimation. They are the tests that make the research useful.
Why this matters now
Demography moves slowly enough that economists can ignore it for years and then suddenly discover that the supposedly distant future has become the present.
Housing moves even more slowly. A dwelling built today may still exist when the demographic assumptions used to justify it have completely changed.
That interaction between slow demographic change and slower physical-capital adjustment is the reason the subject matters.
The popular version of the ageing debate asks whether there will be enough workers to support retirees. The fiscal version asks what happens to pensions, healthcare, and public debt. The asset-market version asks whether retirees will sell assets to smaller subsequent generations.
The regional housing version adds a physical fact that financial models can easily overlook:
the houses are still there.
A share certificate can change hands instantly. A bond can mature. A factory can eventually be repurposed. A detached house in a shrinking town occupies a particular plot of land in a particular place. Its value depends on whether someone wants to live there, whether that person can finance the purchase, whether nearby employment exists, whether services remain, and whether thousands of similar properties are also looking for buyers.
This is why the likely outcome of demographic decline is not one national price effect.
It is a geography of winners, survivors, and stranded housing capital.
The strongest cities may become even more dominant as people and capital concentrate. Middle regions may age without immediately shrinking because smaller households sustain demand. Peripheral areas may cross a threshold where the housing stock becomes too large for the household base, vacancy becomes persistent, prices fall asymmetrically, and collateral weakens.
The long-run projection problem is therefore not “What will the national house-price index be in 2075?”
A more useful set of questions is:-
Which regions will still be gaining households even when national population is falling?
-
Which regions will cross from ordinary slack into structural housing overhang?
-
How quickly can obsolete housing be removed or repurposed?
-
Which banks and borrowers are exposed to those regions?
-
Does weaker collateral amplify out-migration and regional decline?
-
Can historical projection vintages identify these transitions before they happen?
Those are questions that can be answered empirically.
If the research works, the result will not be a prophecy about the death of housing. It will be a framework for understanding why the same demographic future can produce radically different property markets within the same country—and why the physical durability of houses may turn a slow demographic shift into a long regional economic adjustment.
The population can shrink gradually.
The housing stock may not.
That difference is where the economics begins.
References
Eichholtz, P., & Lindenthal, T. (2010). Demographics, human capital, and the demand for housing [Preprint]. Maastricht University. https://maastrichtrealestate.com/upload/researches/Eichholtz-et-al_Demographics-Human-Capital-and-the-Demand-for-Housing.pdf
Federal Housing Finance Agency. (n.d.). FHFA House Price Index. Retrieved August 31, 2026, from https://www.fhfa.gov/data/hpi
Hashimoto, Y., Hong, G. H., & Zhang, X. (2020). Demographics and the housing market: Japan’s disappearing cities (IMF Working Paper No. 20/200). International Monetary Fund. https://doi.org/10.5089/9781513557700.001
Mankiw, N. G., & Weil, D. N. (1989). The baby boom, the baby bust, and the housing market. Regional Science and Urban Economics, 19(2), 235–258. https://doi.org/10.1016/0166-0462(89)90005-790005-7)
Takáts, E. (2010). Ageing and asset prices (BIS Working Papers No. 318). Bank for International Settlements. https://www.bis.org/publ/work318.htm
U.S. Census Bureau. (2024). Selected housing characteristics: American Community Survey 5-year estimates, Table DP04. https://data.census.gov/table/ACSDP5Y2024.DP04
Wharton Research Data Services. (n.d.). Bank Regulatory. The Wharton School, University of Pennsylvania. Retrieved August 31, 2026, from https://wrds-www.wharton.upenn.edu/pages/about/data-vendors/vendor-partner-bank-regulatory/