America’s Housing Affordability Crisis and the Decline of Housing Supply

Introduction

Real, constant-quality house prices for the nation are 15 percent above their pre-Global Financial (GFC) crisis peak. Concern about a growing housing affordability crisis has spread far beyond professional economists and beyond the traditionally expensive coastal markets. This paper will document and attempt to understand how changes in the nature of housing supply in America have helped lead to our current situation. We begin by reporting four key facts about the changing nature of American housing markets, primarily by using decennial census and American Community Survey (ACS) data that runs from 1950-2023.

The most important fact is that the intensity of housing production has declined substantially over the last half century or so. As Figure 1 in the next section shows, the 1950s and 1960s were a golden age of new construction, with extremely high rates of housing unit production in any market with growing demand for housing (which was virtually the entire country in those decades). Starting from a base of 36 million housing units in 1950, the national stock increased by 50 million homes over the next three decades to 1980. Housing unit growth rates then dropped by more than one third between the 1970s and the next two decades. In the 1980s and 1990s, the growth rate of housing was barely half the rates seen in the 1950s and 1960s. The first decade of the new century saw slightly less growth followed by significantly lower housing unit production in the 2010s. The most recent data indicate some recovery in housing production, but building levels remain far below their post-war heyday. Standard economic logic suggests that the combination of rising prices and slowing growth in production means that a tepid supply side plays an important role in explaining today’s high prices.

These national changes mask important heterogeneity across six metropolitan area housing markets that will be a particular focus of this paper: Atlanta, Dallas, Detroit, Los Angeles, Miami and Phoenix. Figure 2 below shows the differences across these markets in the rate of increase in new supply from 1950s through the 2010s and up to 2023. From the immediate postwar period to the 1970s, there was enormous heterogeneity in the rates of production, driven partially by the robust demand for sunbelt housing in the then relatively less populous markets of Atlanta, Dallas, Miami and Phoenix. The 1980s saw the beginnings of a convergence in these rates, and by the 2010s, all these six markets had similarly low rates of housing supply growth.

Twenty years ago (Glaeser and Gyourko, 2005), the prominent heterogeneity in U.S. housing supply was between extremely high rates of housing production in Sunbelt markets and extremely low rates of housing production in America’s large coastal markets (e.g., Boston, New York City, Los Angeles and San Francisco), as well as in markets in apparently long-term decline from deindustrialization (e.g., Cleveland, Detroit). By the 2010s, building levels in key Sunbelt markets such as Miami (FL) and Phoenix did not look much different from historically supply-constrained markets such as Los Angeles or even markets in secular economic decline such as Detroit. The limited data since the 2020 census do not suggest this situation is changing in a material way.

The decline in housing unit production intensity can help explain why housing prices are now so high. An important recent paper by Baum-Snow and Han (2024) estimates highly inelastic supply sides for many very local housing markets in urban areas across the country. This clearly is consistent with a slow rate of supply increase. Orlando and Redfearn (2024) posits that the nature of supply could be changing in previously elastically-supplied metropolitan areas such as Houston in ways that would lead to higher prices in equilibrium. More broadly, Baum-Snow (2023) highlights how housing supply constrains urban growth from the neighborhood level up. Baum-Snow and Duranton’s (2025) recent review also explores the link between changing supply conditions and housing affordability.

In the second part of our paper, we shift to estimating supply curves within metropolitan areas to understand the changes, especially in the sunbelt markets that had once been housing supply superstars. In Section III, we present a simple model of housing supply and demand. One central theme of that model is that over periods of twenty years of more, it is almost impossible to imagine a variable that would shift demand and not shift supply as well, because any variable that shifts demand for a location also will shift who lives in the area and residents then determine the permitting environment. Two core implications of our framework are that when supply limitations become more important, the positive correlation between price and construction will diminish and the negative correlation between price and density will move towards zero or even flip sign.

Despite the difficulties of measuring the true underlying housing supply curve over longer time horizons, we believe that the actual empirical link between prices and construction, which we call the “empirical housing supply curve,” is an inherently important object, even if we cannot be sure what parameters that curve actually represents. The empirical housing supply curve informs us about whether building is happening in places where there is more demand. A positive slope means that the market seems to be responding to demand signals, and that space is being produced where people value it most. A flat empirical housing supply curve suggests that we should expect housing supply to remain fixed even if price rise. The empirical housing supply curve is also relevant for telling us about the size of the social losses due to limited housing construction, since expensive areas are likely to have the biggest gap between how much consumers value housing units and how much housing costs to build.

We first focus on our six core metropolitan areas, and we follow the results of four different estimation strategies: ordinary least squares, instrumenting with lagged price, instrumenting with geography, and instrumenting with both geography and lagged price. As implied by the discussion above, none of these approaches will solve the problem that unobserved factors which influence demand are correlated with unobserved factors that reduce supply. Consequently, while they will not estimate the actual parametric supply curve, the instrumentation strategies help reduce the direct feedback from construction to price.

The declining relationship between house price and new construction is the dominant fact shown by this analysis. In the overwhelming number of America’s largest metropolitan areas, the tractlevel correlation between price and housing growth was lower over the 2000 to 2020 period than it had been during the 1970 to 1990 period. Housing construction used to supply high demand areas within metropolitan areas, but that is much less true more recently than in the past. This complements the well-documented fact that, across metropolitan areas, places that build a lot aren’t expensive and places that are expensive don’t build a lot (Glaeser and Gyourko, 2018).

Where did the empirical housing supply curve flatten most sharply? Variables that are typically thought to correlate with land use regulation reliably predict the shift. The Wharton Residential Land Use Regulatory Index (WRLURI) is reliably correlated with the downward shift. The lagged overall growth in the housing stock in the market is associated with a smaller and less precisely estimated shift. The share of educated workers in the metropolitan area, which is also thought to be one factor that drives land use regulation, is also correlated with the change.

We find that the estimated relationship between density and the growth of housing was initially negative, but became closer to zero over time, which is also compatible with changing supply conditions gaining importance. The shift is weaker for single family housing, and stronger when multi-family units are included in our outcome measures. While more work remains to be done on this issue, the evidence thus far does not suggest that housing supply growth is slowing primarily because particular neighborhoods have become “built out.”

When we look at the link between density and the growth in overall housing, there are many more positive relationships, especially during the 2000 to 2020 time period. A natural interpretation of a positive link between density and overall housing growth is that permitting multifamily housing projects is easier in some places than in others. Areas that have density may also have the ability to add more density, perhaps because they have been zoned for multifamily housing. This fact again pushes back on the idea that a lack of land is responsible for the slowdown in America’s housing growth.

We also look at interactions between density and price in our six core areas. In the 1970s and 1980s, the sunbelt areas typically built in more expensive, lower density areas. However, that pattern changed over time. Growth became increasingly important in higher density, higher price areas. In Miami and Los Angeles, growth also became concentrated in higher density, lower price areas, probably because it was easier to build high density, multifamily projects in those areas. Again, this work suggests that building has become far more difficult in the higher value parts of sunbelt metropolises, even or especially when those areas have relatively low density levels.

In sum, this paper documents the decline in the rate of new housing production in major metropolitan areas, and fundamental shifts in the empirical housing supply curve. America once responded to demand by delivering more density, but it does no longer. This suggests that not only is America failing to deliver housing in its most production metropolitan areas (Hsieh and Moretti, 2019; Duranton and Puga, 2023), but we are also failing to deliver housing in our most desirable neighborhoods.

The plan of paper is as follows. The next section documents key facts about the current state of American housing markets, especially how they have changed over time. This is followed in Section III with a model that outlines our framework for understanding changes in housing supply. Section IV then outlines our empirical strategy and presents and discusses key results for a select set of markets. The analysis is expanded to a much broader set of markets in Section V, with Section VI concluding the paper.

Read a PDF of the conference draft here.