In the 1980s, the Urban Institute developed a housing market simulation model to assess the relationships between public policies and household incomes, individual demographics, and housing affordability across a metropolitan region. This model assumed that housing property development is elastic, meaning if consumers can afford new housing, developers will provide that new housing, wherever they want it. The past few decades have shown the reality of housing development doesn’t run parallel to consumer demand — in fact, the United States faces a shortage of millions of housing units, with the cost of housing in some markets much higher than the marginal costs of new construction.
Land-use policy is one explanation for this discrepancy. Local governments across the US use regulations to shape how land is developed, notably through rules like zoning that specify the use and form of new construction. In combination with local real estate demand, demographics, and building costs, these policies influence where and how much housing is built. Localities sometimes harness restrictive zoning codes specifically to prevent housing construction — even when demand for new units is there — because of opposition to new neighbors. Comparative research shows restrictive zoning is associated with higher housing prices, as well as social and racial segregation.
As part of the development of Urban’s new Housing Market Forecaster, we decided to incorporate the reality of restrictive land-use regulations and the effects of the real estate market. We integrated a new “module” linking zoning rules with expected housing production by leveraging data from what occurred over the past decade to inform projections of housing production. The model then combines projected housing units with estimated future demographics to forecast how housing affordability and the distribution of people are likely to change in a metropolitan housing market.
In this post, I describe how we developed our approach to estimate the effects of local regulations on housing production for the Washington, DC, metropolitan region. I also explain how we developed alternative zoning approaches intended to reflect potential policy changes.
Collecting regional land-use policy data
We focused our work on the DC region, which we defined as the 36 jurisdictions (counties, cities, and towns) that are part of the Metropolitan Washington Council of Governments. These jurisdictions each hold land-use authority — meaning each has the power to zone land and specify zoning rules — and they collectively housed 5.7 million people in 2021.
For each jurisdiction, we scanned through existing zoning codes (as of summer 2022) to obtain detailed data on allowed housing development in each of the region’s 1,091 zoning districts. We considered only development that’s allowed by right, meaning projects that can be developed without requiring public hearings or votes by elected or appointed officials. The zoning codes included a variety of lot-feature regulations, such as maximum dwellings per parcel, minimum lot area per unit, and maximum number of dwellings per acre. We then developed a series of formulas to estimate the maximum theoretical housing density in each district (i.e., how many housing units would be present if developers built to the maximum allowed, including if that meant demolishing existing buildings).
To map the data, we collected shapefiles of zoning district boundaries from as many of the jurisdictions as possible. Several smaller towns in Maryland and Virginia didn’t provide such shapefiles, so we drew the districts ourselves using GIS tools based on images taken from static PDFs on their websites.
Developing a model for estimating housing supply growth
With our local zoning code dataset constructed, we needed to collect additional demographic and housing data to model future housing markets. In total, we collected data on neighborhood demographics at the census tract (neighborhood) level across several decades, constant-geography housing data from the Historical Housing Unit and Urbanization Database 2010, employment data from the US Census Bureau’s Longitudinal Employer-Household Dynamics dataset, subsidized housing data from the National Housing Preservation Database, and parks and roadway data from each of the relevant jurisdictions using information from their websites. We combined these data using areal interpolation at the zoning district level to provide a unified set of variables representing housing units, demographics, and other information for each of the tracts across the region.
One limitation in our work is that we were only able to collect zoning data as of 2022, which meant we had to assume that zoning rules and boundaries remained constant over our analysis period. In reality, localities change district boundaries and rules over time, but based on discussions with local stakeholders, we believe most zoning in the region remained relatively stable between 2005 and 2019, our study period.
We built our model on the assumption that only using zoning district characteristics to estimate changes in the number of housing units by area would be inadequate. Jurisdictions in different parts of the region could have identical zoning codes, but one could see a massive increase in construction and the other virtually none. These differences could result from the fact that housing development is the product of not only local regulations but also the real estate market.
Calibrating the Housing Production Forecaster based on historic trends
The forecaster’s housing production module seeks to project future housing units by subregional area, which we define as public use microdata areas (PUMAs). The DC metropolitan area has 40 PUMAs. Our model’s primary dependent variable is the number of housing units by PUMA over time.
We began by assessing the correlations between the variables we collected to ensure our regressions didn’t suffer from multicollinearity. We then developed nine regression types and used Monte Carlo simulations to select the regression model with the lowest mean standard deviation and lowest maximum standard deviation among many tests.
Using census tracts as the unit of analysis, we ultimately deployed a regression with a series of independent variables:
- initial housing stock (American Community Survey 2005–09 data)
- a series of measures of allowed housing density (representing zoning restrictiveness, from our review of local zoning codes)
- share population non-Hispanic white
- population density per square mile
- share of workers travelling more than 30 minutes to work
- median gross rent
- Gini index of income inequality
- average distance from the White House (to represent the region’s center)
Our dependent variable was the number of housing units per tract in 2015–19 (American Community Survey). We also incorporated fixed effects for each jurisdiction with land-use authority into the model, to reflect local political variation in appetite for housing construction.
Our regression model shows statistically significant associations between higher housing unit growth and both higher population densities and higher local rent levels (indicating market demand for development). We also find that zoning capacity for new development (the difference between maximum allowed housing density and preexisting housing units) is associated strongly and significantly with more development.
Different jurisdictions are also more or less likely to accommodate new housing, once controlling for other characteristics. Virginia’s Arlington and Loudoun Counties are most likely to allow increased housing, whereas Maryland’s Montgomery and Prince George’s Counties are least likely. These differences could reflect local political differences in receptiveness to adding housing beyond market demand or demographics.
Projecting housing units by area in the coming decade
Using the coefficients produced by the regression model comparing changes from 2005–09 with 2015–19, we projected changes in housing stock by PUMA between 2015–19 and 2025–29.
Our model estimates that the region would add a total of about 159,000 housing units in the next decade, which would mark a decline from the roughly 184,000 added in the preceding decade. This result suggests the region is unlikely to meet the housing demands of the region’s growing population, with overly restrictive zoning restrictions in the neighborhoods where there’s a demand for housing development posing one key barrier.
Although our model demonstrates zoning capacity for additional housing units, that projection is not equivalent to actual housing construction. We estimate that in certain areas, such as Falls Church, Virginia, current zoning has no by-right capacity for additional housing units. However, theoretically, the region has the capacity to house millions more housing units, but many of those potential units would be on land far from the region’s center or in communities with low housing values, where new development is unmarketable. In both cases, new development is less likely. Given that a large share of land is taken up by nonhousing uses (such as commercial space, schools, and parks), it may be necessary to have a much larger capacity for future housing development than currently exists to encourage construction.
To acknowledge these constraints, we also modeled what would happen if localities throughout the region reformed their zoning codes to increase the capacity for new housing by 50 percent. When we did this, we found housing unit growth would increase substantially, with 252,000 new units over the next decade.
Under baseline zoning conditions, northwest DC and parts of southern Montgomery County, Maryland, would actually lose housing units in the coming decade because of smaller apartments being converted into larger units, which is a phenomenon that occurs in gentrifying communities. These areas have some of the most restrictive zoning codes but also the strongest local real estate markets. Zoning reforms in these areas would encourage housing growth and lead to more new construction.
Implications for land-use policy
With our Housing Market Forecaster, we have developed a new approach to estimating how land-use regulations may affect housing availability across a metropolitan area. Our model required integrating data across a broad array of resources and necessitated detailed work examining the zoning codes of dozens of jurisdictions.
Despite these difficulties, our results show promise for other researchers interested in understanding how zoning and land-use regulations affect housing availability. Our process demonstrates how to connect local zoning policies with other local demographics, while documenting how changes over time might influence outcomes. Policymakers considering reforms to local land-use laws may consider undertaking a similar method to identify key concerns for their respective regional housing markets.
