Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Feb 3.
Published in final edited form as: Int Econ Rev (Philadelphia). 2019 Nov 1;61(1):127–158. doi: 10.1111/iere.12419

Urban Population and Amenities: The Neoclassical Model of Location

DAVID ALBOUY 1,1, BRYAN A STUART 1,1
PMCID: PMC7857490  NIHMSID: NIHMS1657153  PMID: 33542584

Abstract

We develop a neoclassical general-equilibrium model to explain cross-metro variation in population and density. We provide new methods to estimate traded and non-traded productivities, and elasticities of housing and land supply, using density and land area data. From wage and housing cost indices, the model explains half of U.S. density and population variation and finds that quality of life determines location choices more than trade productivity; productivity and factor substitution in housing matter most, but are weak in nicer areas. Relaxing land use regulations would increase population in the West, raising both quality of life and productivity experienced by residents.

Keywords: Population, density, quality of life, productivity, amenities, housing supply, land supply, local labor markets

JEL Numbers: R23, R12, R31, H2

1. Introduction

Economists and policy makers are interested in knowing why incomes, prices, and populations vary across locations. Most research on this question uses a spatial equilibrium model in which individuals and firms choose their location in response to differences in economic fundamentals that vary across space (Rosen, 1979; Roback, 1988). This “neoclassical” model has been used to understand how amenities—broadly defined—determine wages and housing costs, but many of its core quantitative predictions, particularly about the cross-sectional distribution of population, have never been examined.2

This paper studies the implications of the canonical spatial equilibrium model for household location decisions. We make two methodological innovations that generate new evidence on the distribution of population across space. First, we develop predictions of the full competitive model featuring three factors—mobile labor and capital, and immobile land—and two outputs—a good tradable across cities, and a home good that is not. Cities vary in their amenities along three dimensions: quality of life for households, and trade productivity and home productivity for firms. We highlight the role of home (non-tradable) production, as the model predicts it plays a large role in location decisions. We develop a technique that estimates how observable variables influence both scale (productivity) and substitution parameters in local home production. This allows us to estimate housing production functions that differ for each metro in the United States using a simple cross section of data on population density, wages, and housing costs. In addition, we circumvent the pervasive problem of missing land price data using the structure of the model.

We derive the structural relationships between quantities, such as population, and the three amenity types implied by this popular model. Through log-linear analytical expressions, we show for the first time how these relationships depend on expenditure shares, land supply, tax rates, and—importantly—substitution responses in consumption and production. By imposing fewer restrictions, we nest many models in the literature, especially by letting elasticities of substitution differ from one. The model’s assumptions make a city’s prices invariant with its land supply, while its quantities rise in proportion with it. This simplifies matters, as population differences due to density may be examined separately from such differences due to land area.

Turning to data, we assess how well the neoclassical model explains observed population and density differences across 274 metropolitan areas in the United States. We first restrict the model to have constant home productivity across metro areas, as this generates population and density predictions using only wage and housing cost data. The parametrized model explains half of the observed variation in population density across cities. We then relax that assumption and use the density data and the full model for estimation purposes. This produces new trade productivity and home productivity differentials, as well as elasticities of housing and land supply, for all 274 metros.

We then examine what drives population and density differences across cities. Quality of life determines location size more than trade productivity, although home productivity may dominate both.3 Less home-productive areas have stricter land use regulations and rugged geography. Furthermore, we find that these variables reduce factor substitution in the home sector and the responsiveness of urban land supply. Thus, differences in the home sector and land supply critically shape where people live, and how much they can take advantage of places with higher quality of life and trade productivity.

Finally, we conduct counterfactual simulations that relax land use constraints. By increasing substitution possibilities in home production, this change allows households to take better advantage of local amenities. For example, the population of San Francisco would rise by 1.4 million people (20 percent); Los Angeles, by 1.9 million (12 percent). Overall, population would rise by 6 percent in the West, and fall in the Midwest and South. The total value of amenities experienced by households would rise by 0.6 percent of income.

This paper contributes to a broad literature on spatial equilibrium models that study the distribution of economic activity across space (e.g., Rappaport, 2008a,b; Glaeser and Gottlieb, 2009; Saiz, 2010; Lee and Li, 2013; Desmet and Rossi-Hansberg, 2013; Suárez Serrato and Zidar, 2016; Diamond, 2016; Hsieh and Moretti, 2019).4 Our technique of using a cross-section of easily observed data—wages, rents, population, and land area—is particularly novel and easy to apply. It generates new estimates of local home productivity differences as well as quality of life and (amended) trade productivity.5 It also provides a method of estimating housing and land supply elasticities from a single cross section of data. One may estimate how these vary across cities with additional geographic or regulatory data.

2. The Neoclassical Model of Location

2.1. System of Cities with Consumption and Production

The national economy contains many cities, indexed by j, which trade with each other and share a homogeneous population of mobile households. Cities differ in three attributes, each of which is an index summarizing the value of amenities; quality of life Qj raises household utility, trade productivity AXj lowers costs in the traded sector, and home productivity AYj lowers costs in the non-traded sector. Households supply a single unit of labor in their city of residence, earning local wage wj. They consume a numeraire traded good x and a non-traded “home” good y with local price pj. All input and output markets are perfectly competitive, and all prices and per-capita quantities are homogeneous within cities.

Firms produce traded and home goods out of land, capital, and labor. Land Lj is heterogeneous across cities, immobile, and receives a city-specific price rj. Each city’s land supply L0jL˜(rj) depends on an exogenous endowment L0j and a common supply function L˜(rj). The supply of capital in each city Kj is perfectly elastic at the price ı¯. Labor Nj is supplied by households who have identical size and tastes, and own diversified portfolios of land and capital, which pay an income R=∑jrjLj/NTOT from land and I=∑jı¯Kj/NTOT from capital, where NTOT=∑jNj is the total population. Total income mj = wj + R + I varies across cities only as wages vary. Out of this income households pay a linear federal income tax τmj which is redistributed in uniform lump-sum payments T.6 Household preferences are modeled by a utility function U(x, y; Qj) which is quasi-concave over x, y, and Qj. The expenditure function for a household in city j is e(pj, u; Qj) ≡ minx,y{x + pjy : U(x, y; Qj) ≥ u}. Quality of life Qj enters neutrally into the utility function and is normalized so that e(pj, u; Qj) = e(pj, u)/Qj, where e(pj, u) ≡ e(pj, u; 1).

Firms produce traded and home goods according to the function Xj=AXjFX(LXj,NXj,KXj) and Yj=AYjFY(LYj,NYj,KYj), where FX and FY are weakly concave and exhibit constant returns to scale, with Hicks-neutral productivity. Unit cost in the traded good sector is cX(rj,wj,ı¯;AXj)≡minL,N,K{rjL+wjN+ı¯K:AXjF(L,N,K)=1}. Let cX(rj,wj,ı¯;AXj)=cX(rj,wj,ı¯)/AXj, where cX(rj,wj,ı¯)≡cX(rj,wj,ı¯;1) is the uniform unit cost function. A symmetric definition holds for unit cost in the home good sector cY.

2.2. Equilibrium of Prices, Quantities, and Amenities

Each city is described by a block-recursive system of sixteen equations in sixteen endogenous variables: three prices (pj, wj, rj), two per-capita consumption quantities (xj, yj), and eleven city-level production quantities (Xj,Yj,Nj,NXj,NYj,Lj,LXj,LYj,Kj,KXj,KYj). The endogenous variables depend on three exogenous attributes Qj, AXj, AYj, and the land endowment L0j. As in the Hecksher-Ohlin model, the system first determines prices—where most researchers stop—then, per-capita consumption quantities and city-level production quantities.7 We adopt a “small open city” assumption and take nationally determined variables u¯, ı¯, I, R, T as given.

We log-linearize the system, as in Jones (1965), to obtain a model that can be solved analytically with linear methods. This allows us to analyze models in which the economy is not fully Cobb-Douglas. The full nonlinear system is explained in Appendix B. In Appendix C, we verify that the log-linearized model approximates the nonlinear model well for the values in our data.

The log-linearized model involves several economic parameters, evaluated at the national average. For households, denote the shares of gross expenditures spent on the traded and home good as sx ≡ x/m and sy ≡ py/m; the shares of income received from land, labor, and capital income as sR ≡ R/m, sw ≡ w/m, and sI ≡ I/m. For firms, denote the cost shares of land, labor, and capital in the traded good sector as θL ≡ rLX/X, θN ≡ wNX/X, and θK≡ı¯KX/X; the equivalents in the home good sector are ϕL, ϕN, and ϕK. Finally, denote the shares of land, labor, and capital used to produce traded goods as λL ≡ LX/L, λN ≡ NX/N, and λK ≡ KX/K. To fix ideas, assume the home good is more cost-intensive in land relative to labor than the traded good, both absolutely, ϕL ≥ θL, and relatively, ϕL/ϕN ≥ θL/θN, implying λL ≤ λN. For any variable z, we denote the log differential by z^j≡lnzj−lnz¯≈(zj−z¯)/z¯, where z¯ is the national average.

2.2.1. Equilibrium Price Conditions for Households and Firms

Since households are fully mobile, they receive the same utility u¯ across all inhabited cities. Firms earn zero profits in equilibrium. These conditions imply

−sw(1−τ)w^j+syp^j=Q^j (1a)
θLr^j+θNw^j=A^Xj (1b)
ϕLr^j+ϕNw^j−p^j=A^Yj. (1c)

Equations (1a)–(1c) simultaneously determine the city-level prices p^j, r^j, and w^j as functions of the three attributes Q^j, A^Xj, and A^Yj. The tax rate and cost and expenditure shares that determine the relative importance of these prices are evaluated at the national average. These conditions provide a one-to-one mapping between unobservable city attributes and potentially observable prices. Households pay more for housing and get paid less in nicer areas. Firms pay more to their factors in more trade-productive areas, and they do the same relative to output prices in more home-productive areas. Albouy (2009, 2016) examines these conditions in detail.

2.2.2. Consumption Conditions for Households

In their consumption x^j and y^j, households face a budget constraint and obey a tangency condition:

sxx^j+sy(p^j+y^j)=(1−τ)sww^j (2a)
x^j−y^j=σDp^j, (2b)

where w^j and p^j are determined by the price conditions. Equation (2b) depends on the elasticity of substitution in consumption, σD ≡ −e · (∂e/∂p2)/[∂e/∂p · (e − p · ∂e/∂p)] = −∂ ln(y/x)/∂ ln p. Substituting equation (1a) into equations (2a) and (2b) produces the consumption solutions x^j=syσDp^j−Q^j and  y^j=−sxσDp^j−Q^j. Because of homothetic preferences, in areas where Qj is higher, but pj is the same, households consume less of x and y in equal proportions, so the ratio y/x remains constant—similar to an income effect. Holding Qj constant, areas with higher pj induce households to reduce the ratio y/x through a substitution effect.

In a more general model, household types could vary in their taste for home goods. In such a model, households with stronger tastes for y sort to areas with a lower p, and higher values of σD correspond to cases where households vary more in their tastes for home goods.

2.2.3. Production Conditions for Traded and Home Good Sectors

Given prices and per-capita consumption, output X^j, Y^j, employmen N^j, N^Xj, N^Yj, capital K^j, K^Xj, K^Yj, and land L^j, L^Xj, L^Yj are determined by eleven equations describing production and market clearing. The first six are conditional factor demands describing how input demands depend on output, productivity, and relative input prices:

N^Xj=X^j−A^Xj+θLσXLN(r^j−w^j)−θKσXNKw^j (3a)
L^Xj=X^j−A^Xj+θNσXLN(w^j−r^j)−θKσXKLr^j (3b)
K^Xj=X^j−A^Xj+θLσXKLr^j+θNσXNKw^j (3c)
N^Yj=Y^j−A^Yj+ϕLσYLN(r^j−w^j)−ϕKσYNKw^j (3d)
L^Yj=Y^j−A^Yj+ϕNσYLN(w^j−r^j)−ϕKσYKLr^j (3e)
K^Yj=Y^j−A^Yj+ϕLσYKLr^j+ϕNσYNKw^j. (3f)

The dependence on input prices is determined by partial (Allen-Uzawa) elasticities of substitution in each sector for each pair of factors, e.g., σXLN≡cX⋅(∂2cX/∂w∂r)/(∂cX/∂w⋅∂cX/∂r). Our baseline model assumes that production technology does not differ across cities, implying constant elasticities; we relax this assumption for the home good sector below. To simplify, we also assume that partial elasticities within each sector are the same—i.e., σXNK=σXKL=σXLN≡σX, and similarly for σY —as with a constant elasticity of substitution (CES) production function.

Higher values of σX correspond to more flexible production of the traded good, as firms can vary the proportion of inputs they employ. In a generalization with multiple traded goods sold at fixed prices, firms specialize in producing goods for which their input costs are relatively low.8 A related argument exists for home goods. A higher value of σY means that housing producers can better combine labor and capital to build taller buildings in areas with expensive land. For non-housing home goods, retailers may use taller shelves and restaurants would hire extra servers to make better use of space.

Three conditions express the local resource constraints for labor, land, and capital under the assumption that factors are fully employed:

N^j=λNN^Xj+(1−λN)N^Yj (4a)
L^j=λLL^Xj+(1−λL)L^Yj (4b)
K^j=λKK^Xj+(1−λK)K^Yj. (4c)

Equations (4a)–(4c) imply that sector-specific factor changes affect overall changes in proportion to the factor share. Local land is determined by the supply function in log differences,

L^j=L^0j+εL,rr^j, (5)

with the endowment differential L^0j and the land supply elasticity εL,r≡(∂L˜j/∂r)⋅(rj/L˜j).

Finally, the market clearing condition for home goods that demand equals supply is

N^j+y^j=Y^j. (6)

Walras’ Law makes redundant the market clearing equation for traded output, which includes per-capita net transfers from the federal government.

2.2.4. Total Population, Density, and Land

The log-linearized model readily separates intensive margin population differences holding land supply constant, i.e., density, from extensive margin differences driven by land supply. If we define population density as, N∗j≡Nj/Lj then the total population differential is a linear function of differentials in density, the land endowment, and land supply determined by rent:

N^j=N^∗j+L^0j+εL,rr^j, (7)

where N^∗j and r^j, depend on amenities Q^j, A^Xj, A^Yj but the land endowment L^0j does not.9

2.3. Solving the Model for Relative Quantity Differences

We express solutions for the endogenous variables in terms of the amenity differentials Q^j, A^Xj, and A^Yj. Only equations (1a)–(1c) are needed to solve the price differentials:

r^j=1sRλNλN−τλL[Q^j+(1−τλN)sxA^Xj+syA^Yj] (8a)
w^j=1sw1λN−τλL[−λLQ^j+(1−λL)sxA^Xj−λLsyA^Yj] (8b)
p^j=1sy1λN−τλL[(λN−λL)Q^j+(1−τ)(1−λL)sxA^Xj−(1−τ)λLsyA^Yj]. (8c)

Higher quality of life leads to higher land and home good prices but lower wages. Higher trade productivity increases all three prices, while higher home productivity increases land prices but decreases wages and the home good price.

Putting solution (8c) in equations (2a) and (2b) yields the per-capita consumption differentials:

x^j=σD(1−τ)λN−τλL[σD(λN−λL)−(λN−τλL)σD(1−τ)Q^j+(1−λL)sxA^Xj−λLsyA^Yj] (9a)
y^j=−sxsyσD(1−τ)λN−τλL[sxσD(λN−λL)+sy(λN−τλL)sxσD(1−τ)Q^j+(1−λL)sxA^Xj−λLsyA^Yj]. (9b)

Households in home-productive areas substitute towards home goods and away from traded goods, while households in trade-productive areas do the opposite. In nicer areas, households consume fewer home goods; whether they consume fewer traded goods is ambiguous, as the substitution effect is positive, and the income effect is negative.

Solutions for the other quantities, which rely on equations (3a)–(6), are more complicated and harder to intuit. To simplify notation, we express the change in each quantity with respect to amenities using three reduced-form elasticities, each composed of structural parameters. For our central example, the population differential is written

N^j=εN,QQ^j+εN,AXA^Xj+εN,AYA^Yj+L^0j, (10)

where εN,Q is the elasticity of population with respect to quality of life; εN,AX and εN,AY are defined similarly. In terms of structural parameters, the first reduced-form elasticity, εN,Q, is

εN,Q=λN−λLλN+σD[sxsy(λN−λL)2λN(λN−λLτ)]+σX[λLλN−λLτ(λLsw+λNsR)]+σY[1λN−λLτ(λL2(1−λN)swλN+λN(1−λL)sR−(λN−λL)2syλN)]+εL,r[1sRλN(λN−τλL)]. (11)

We provide similar expressions for εN,AX and εN,AY in Appendix D. The full structural solution to (10) is obtained by substituting in these expressions.

Collecting terms for each structural elasticity in (11) highlights that nicer areas can have higher population via five behavioral responses. The first term reflects how households consume fewer goods from the income effect, and thus require less land per capita, e.g., by crowding into existing housing. The second term, with σD, captures how households substitute away from land-intensive goods, accepting additional crowding. The third, with σX, expresses how firms in the traded sector substitute away from land towards labor and capital, freeing up space for households. The fourth, with σY , reflects how home goods become less land intensive, e.g., buildings get taller. The fifth, with σL,r, provides the population gain on the extensive margin from more land being used.

Each reduced-form elasticity between a quantity and amenity has up to five similar structural effects. Unlike the price solutions, (8a)–(8c), the quantity solutions require more epistemically demanding knowledge of substitution elasticities, i.e., of behavioral responses to prices. Below we initially focus on quantity differences holding geography constant, focusing on density. This case sets L^j=0. In section 6, we consider how to estimate εL,r and L^0j.

2.4. Identification of Production Amenities and Land Values

While cross-metro data on wages and housing rents (which proxy for home good prices) are readily available, land values are not. As a result, we cannot identify trade and home productivity from (1b) and (1c).10 Our solution is to use widely available data on population density as a replacement for land values. Consider combining equations (1b) and (1c) to eliminate r^j:

 Inferred costs j≡θLϕLp^j+(θN−ϕNθLϕL)w^j=A^Xj−θLϕLA^Yj. (12)

The left-hand side of (12) equals traded producer costs inferred from wages and home good prices. Trade productivity raises these inferred costs, while home productivity lowers them. Albouy (2016) assumes that home productivity is constant, A^Yj=0, so that land values may be inferred from (1c), and A^Xj equals the inferred costs. The ensuing estimates are biased downwards in home-productive areas, although A^X is only slightly biased if ϕL << ϕL.

Combining equations (1a) and the analog of equation (10) for density yields the following expression, which says that “excess density” not explained by quality of life, on the left, must be explained by either trade or home productivity, on the right:

 Excess density j≡N^∗j−εN∗,Q[syp^j−sw(1−τ)w^j︸Q^j]=εN∗,AXA^Xj+εN∗,AYA^Yj. (13)

Equations (12) and (13) are exactly identified: the inferred amenities perfectly predict density. Solving these equations identifies each productivity from observable differentials N^∗j, w^j, and p^j:

A^Xj=θL[N^∗j−εN∗,Q(sypj−sw(1−τ)wj)]+ϕLεN∗,AY[θLϕLpj+(θN−ϕNθLϕL)wj]θLεN∗,AX+ϕLεN∗,AY (14a)
A^Yj=ϕL[N^∗j−εN∗,Q(sypj−sw(1−τ)wj)]−ϕLεN∗,AX[θLϕLpj+(θN−ϕNθLϕL)wj]θLεN∗,AX+ϕLεN∗,AY (14b)

High excess density and high inferred costs imply high trade productivity. High excess density with low inferred costs imply high home productivity. Home productivity is identified more from density than trade productivity, as ϕL > θL. The value of land is found by substituting (14a) into (1b), yielding:

r^j=N^∗j−εN∗,Q(syp^j−sw(1−τ)w^j)−εN∗,AXθNw^j−εN∗,AY(ϕNw^j−p^j)θLεN∗,AX+ϕLεN∗,AY. (14c)

As seen in the numerator of (14c), this rent measure depends on density not explained either by quality of life or productivity differences inferred from non-land prices.

The key to this approach is that an observed quantity, population density, replaces an unobserved price, land rents. In principle, one could use data on total metro population and land supply instead of density, but that would require a value for the land supply elasticity εL,r. There is no consensus on the appropriate value of this parameter, although we attempt to estimate it below.

2.5. Incorporating Preference Heterogeneity

Here we consider an extension in which individuals have idiosyncratic preferences over destinations. While such preferences are arguably less important for the long-run equilibrium captured by our model, this extension provides a clear connection to discrete choice models.

Suppose that the indirect utility of worker i in city j is money metric, with Vij=v(pj,wj;Qj)+ϵij, where v(pj, wj; Qj) is a function common to households and ϵij is an idiosyncratic preference shock. Following the literature—e.g., Kline and Moretti (2014) Suárez Serrato and Zidar (2016); and Hsieh and Moretti (2019)—we make the convenient assumption that ϵij is distributed i.i.d. Type 1 Extreme Value with scale parameter ψ ≥ 0. This implies that population in city j is

Nj=exp[v(pj,wj;Qj)/ψ]∑j′=1Jexp[v(pj′,wj′;Qj′)/ψ]. (15)

Log-linearizing equation (15) yields a modified mobility condition,

−sw(1−τ)w^j+syp^j+ψN^j=Q^Nj, (16)

where the subscript on Q^Nj distinguishes this from the quality of life differential in equation (1a). With idiosyncratic preferences, we observe willingness-to-pay of a resident on the margin of moving to another metro, i.e., Q^j=Q^Nj−ψN^j. The greater the population, the lower will be this willingness-to-pay for a given quality of life. Q^Nj then recovers this value, netting out changes in willingness-to-pay due to population differences. As seen in equation (16), identifying quality of life with idiosyncratic preferences requires not only data on wages and housing casts, but also data on population. As ψ approaches zero, there is no dispersion in idiosyncratic preferences, returning the original equation (1a).

Preference heterogeneity dampens the relationship between population and amenities, as can be seen by combining equations (1a), (10), and (16):

N^j=11+ψεN,Q(εN,QQ^Nj+εN,AXA^Xj+εN,AYA^Yj+L^0j). (17)

This dampening occurs because firms in a city need to pay incoming migrants an increasing schedule in after-tax real wages to have them overcome their taste differences. Preference heterogeneity does not affect the estimates of trade and home-productivity in equations (14a) and (14b).

3. Parameter Choices and Reduced-Form Elasticities

3.1. Parameter Choices

The parametrization we use, shown in Table 1, was set in Albouy (2009). It is based on a literature review that does not refer to population data. We focus on the substitution elasticities, as they are rarely considered. The parametrization sets them to σD = σX = σY = 0.667. This is consistent with households having higher housing expenditures in high rent areas and housing having a higher cost-share of land in high value areas (Albouy et al., 2016a; Albouy and Ehrlich, 2018). When we incorporate preference heterogeneity, we use a value of ψ = 0.07 from Notowidigdo (Forthcoming) or ψ = 0.30 from Hsieh and Moretti (2019).11 Appendix E contains additional details on the parametrization. We report several sensitivity analyses below. Starting in section 5.4, σY is estimated and allowed to vary by city.

Table 1:

Notation and Constant Geography Parametrization

Parameter Name Notation Value
Cost and Expenditure Shares
Home good expenditure share sy 0.36
Income share to land sR 0.1
Income share to labor sw 0.75
Traded good cost share of land θL 0.025
Traded good cost share of labor θN 0.825
Home good cost share of land φL 0.233
Home good cost share of labor φN 0.617
Share of land used in traded good λL 0.16
Share of labor used in traded good λN 0.704
Tax Parameters
Average marginal tax rate τ 0.392
Average deduction level δ 0.284
Substitution Elasticities
Elasticity of substitution in consumption σD 0.667
Elasticity of traded good production σX 0.667
Elasticity of home good production σY 0.667 or city-specific
Miscellaneous Elasticities
Dispersion of idiosyncratic preferences ψ 0.07 or 0.30 (see text)
Elasticity of land supply εL,r 0 or see text
City-specific Attributes
Quality of life Qj city-specific
Trade productivity AXj city-specific
Home productivity AYj constant or city-specific

Parametrization for all but miscellaneous elasticities pre-set in Albouy (2009). Elasticity of land supply set to zero for determining reduced-form elasticities in Section 4, constant geography exercises in Table 3, and density exercises in Table 4. Substitution elasticities estimated by city in Table 5. See Appendix E for details on parametrization.

3.2. Parametrized Reduced-Form Elasticities

Panel A of Table 2 demonstrates how the three reduced-form elasticities for population depend on the structural elasticities. For example, the five responses in the elasticity of population to quality of life from (11) are given by the sum: εN,Q = 0.77 + 1.14σD + 1.95σX + 8.01σY + 11.85 εL,r. Numerically, the extensive margin of land supply is critical. On the intensive margin of population density, the substitution parameter in the housing sector stands out as the most important. The intuition is straightforward: increasing population density without building densely strains other substitution margins. Higher densities would be achieved only by increasing the occupancy of existing structures or releasing land from the traded good sector. When σD = σX = 0.667 and L^j=0, density and amenities are related through σY as:

N^∗j=(2.84+8.01σY)Q^j+(0.79+2.06σY)A^Xj+(1.15+2.61σY)A^Yj. (18)

Setting σY = 0.667 produces N^∗j=8.18Q^j+2.16A^Xj+2.88A^Yj. The elasticity of substitution in non-traded production accounts for about two-thirds of the reduced-form elasticities.

Table 2:

Relationship between Reduced-Form and Structural Elasticities, Population and Housing

A: Reduced-Form Population Elasticity with Respect to:
Quality of life
εN,Q
Trade productivity
εN,AX
Home productivity
εN,AY
σD 1.142 0.719 −0.077
σX 1.953 0.468 0.636
σY 8.012 2.055 2.608
εL,r 11.848 4.014 3.856
Constant 0.773 0.000 0.773
B: Reduced-Form Housing Elasticity with Respect to:
Quality of life
εY,Q
Trade productivity
εY,AX
Home productivity
εY,AY
σD −0.336 −0.212 0.023
σX 1.953 0.468 0.636
σY 8.012 2.055 2.608
εL,r 11.848 4.014 3.856
Constant −0.227 0.000 0.773

Table 2 decomposes reduced-form elasticities into substitution elasticities in consumption (σD), traded good production (σX), home good production (σY), and the elasticity of land supply (εL,r). For example, the reduced-form elasticity of population with respect to quality of life is εN,Q = 0.773 + 1.142σD + 1.953σX + 8.012σY + 11.848εL,r.

A one-point increase in Q^j is equal in value to a one-point increase in income, while one-point increases in A^Xj and A^Yj have values of sx and sy of income, due to their sector sizes. To make the productivity elasticities comparable in value, they can be normalized:

N^∗j=8.18Q^j+3.38sxA^Xj+8.01syA^Yj. (19)

Quality of life and home productivity have large impacts on local population density: raising their value by one-percent of income increases density by eight percentage points. Trade productivity’s impact is less than half as large. As a result, funds spent to attract households directly may be more effective at boosting population than funds spent to attract firms.

Setting the marginal tax rate τ to zero reveals that taxes cause much of these asymmetries: N^∗j=6.32Q^j+5.81sxA^Xj+7.55syA^Yj. Taxes push workers away from trade-productive areas towards high quality of life and home-productive areas (Albouy, 2009). Remaining asymmetries arise from other sources. The income effect from quality of life makes households more willing to crowd into existing residential space, while an output effect from home productivity, provides additional space. Trade productivity, on the other hand, raises labor costs in the home sector, putting a brake on growth.

In a Cobb-Douglas economy, σD = σX = σY = 1, the implied reduced-form elasticities are 37–50 percent higher than if σ = 0.667. If substitution margins are inelastic, then assuming a Cobb-Douglas economy—as many do–could inflate quantity predictions and associated welfare calculations.

Table 3 displays the reduced-form elasticities for all endogenous prices and quantities. Appendix Table A.1 contains results with preference heterogeneity (discussed above) and feedback effects between population density and amenities (discussed in Appendix D.7). While we focus on population and density here, many other quantities—such as capital stocks—could be investigated with additional data. A key challenge for these other quantities is that accurate data on them are often unavailable at the metro level.

Table 3:

Parametrized Relationship between Amenities, Prices, and Quantities

Quality of life Trade productivity Home productivity
Price/quantity Notation Q^ A^X A^Y
Land value r^ 11.848 4.014 3.856
Wage w^ −0.359 1.09 −0.117
Home price p^ 2.543 1.609 −0.172
Trade consumption x^ −0.446 0.349 −0.037
Home consumption y^ −1.986 −0.621 0.067
Population density N^ 8.181 2.162 2.885
Capital K^ 7.937 2.865 2.779
Land L^ 0 0 0
Trade production X^ 7.962 3.338 2.934
Home production Y^ 6.195 1.541 2.951
Trade labor N^X 8.202 2.278 3.012
Home labor N^Y 8.131 1.887 2.581
Trade capital K^X 7.962 3.005 2.934
Home capital K^Y 7.891 2.615 2.503
Trade land L^X 0.06 0.328 0.362
Home land L^Y −0.011 −0.063 −0.069

Each value in Table 3 represents the partial effect that a one-point increase in each amenity has on each price or quantity, e.g., N^j=8.181Q^j+2.162A^Xj+2.885A^Yj. All variables are measured in log differences from the national average.

Appendix E.2 discusses alternative parametrizations using different sets of share parameters from Haughwout (2002), Glaeser and Gottlieb (2009), and Rappaport (2008a,b), and AppendixE.3 describes results from these alternatives. After normalizing the elasticities by the size of the traded and non-traded sectors as in equation (19), the results from our preferred parametrization always lie within the range of values implied by these three alternatives. In the exercises that follow, these parametrizations generally imply a stronger role for trade productivity and a weaker role for quality of life, but this is largely due to how those attributes are estimated.

4. General Equilibrium Elasticities and Previous Estimates

Elasticities characterizing how population and housing respond to changes in prices are regularly estimated. The general equilibrium framework here models consumption, labor, and land markets simultaneously, complementing empirical work in two ways.12 First, it provides a useful lens for understanding existing empirical strategies. Price and quantity shifts depend ultimately on fundamental shifts in quality of life, trade productivity, and home productivity. As shown, a shift in one amenity—for example, quality of life—typically changes all prices and quantities. Second, similarities between elasticities in the parametrized model and prior empirical work help tell us if a parametrization is sensible.

4.1. Local Labor Supply and Demand

In partial equilibrium, increasing demand traces out a local labor supply curve. The immediate analogy of an increase in labor demand here is an increase in trade productivity, and so the following ratio provides a “local” general equilibrium elasticity of labor supply:

∂N^∗∂w^|Q^,A^Y=∂N^∗/∂A^X∂w^/∂A^X=0.66σD+0.43σX+1.88σY=1.98. (20)

The resulting labor supply curve slopes upwards as higher density raises demand for home goods and their prices, requiring higher wage compensation. A ceteris paribus increase in the wage, holding home good prices constant, does not identify a labor supply elasticity in this model. Since trade productivity increases both wages and home good prices, a constant home good price requires either a simultaneous decrease in quality of life, shifting in labor supply, or an increase in home productivity, shifting out housing supply.

Longer-run labor supply elasticity estimates are usually in the range of 1 to 4 (Bartik, 1991; Blanchard and Katz, 1992; Notowidigdo, Forthcoming; Albouy et al., 2019), close to the values predicted in (20), for a range of substitution elasticities.13

Increasing supply traces out a local labor demand curve. The closest analogy to a shift in supply is an increase in quality of life. The resulting labor demand curve slopes downward: holding productivity constant, a larger work force pushes down wages, as firms complement labor with ever scarcer land. The parametrized elasticity of labor demand is

∂N^∗∂w^|A^X,A^Y=∂N^∗/∂Q^∂w^/∂Q^=−2.15−3.18σD−5.44σX−22.31σY=−22.79

This large value is broadly consistent with evidence that immigration-induced increases in local labor supply have only weak effects on relative wages (e.g., Bartel, 1989; Card, 2001).14

Panel A of Figure 1 illustrates how general equilibrium elasticities of labor supply and demand vary with elasticities of substitution in consumption and production, assumed to be equal (σD = σX = σY ≡ σ). Without substitution responses, σ = 0, labor supply is perfectly inelastic, and labor demand has a smaller elasticity of −2.15, due only to income effects from Q. The size of the structural substitution parameters is instrumental in determining both demand and supply elasticities.

Figure 1:

Figure 1:

Relationship between Labor and Housing General Equilibrium Elasticities and Substitution Possibilities

Panel (a) displays ∂N^/∂w^, where the change in both density and wages is due to a change in the indicated amenity, as a function of the substitution elasticity σD = σX = σY ≡ σ. Panel (b) displays similar results for the elasticity of housing with respect to housing prices.

4.2. Local Housing Supply and Demand

A city’s housing stock is closely tied to its population. The difference between the two depends only on substitution and income effects in consumption:

Y^j=N^j−sxσDp^j−Q^j=6.20Q^j+2.41sxA^Xj+8.20syA^Yj. (21)

Relative to population, housing responds less to quality of life and trade productivity and more to home productivity. This relationship holds both for total numbers and density.

Two potential demand shifts may trace out a housing supply curve: quality of life or trade productivity. The elasticity is generally greater for the former than the latter:

∂Y^∂p^|A^X,A^Y=∂Y^/∂Q^∂p^/∂Q^=−0.09−0.13σD+0.77σX+3.15σY+4.66εL,r (22a)
∂Y^∂p^|Q^,A^Y=∂Y^/∂A^X∂p^/∂A^X=−0.13σD+0.29σX+1.28σY+2.49εL,r. (22b)

These equal 2.44 and 0.96 when σD = σX = σY = 0.667 and εL,r = 0. The two formulae point out that the magnitude of a housing supply elasticity depends on the demand shock. Both elasticities have large coefficients on σY, from housing being produced more densely on the intensive margin. Supply also depends strongly on the extensive margin, through εL,r Subtly, it depends on the intensive margin through σX: as land released from the traded sector moves to the home sector, workers demand more land. The elasticities also incorporate small reductions in demand from rising prices, seen in the negative constant and coefficient on σD.

The predicted values of the supply elasticities are consistent with the range of estimates seen in Green et al. (2005) and Saiz (2010), for different cities. However, those estimates vary considerably across metros, from what their authors argue are differences in geography or regulatory constraints. This suggests that these constraints affect either the intensive or extensive margin of housing supply by altering the parameter values of σY or ε L,r across metros.15

A shift in supply due to home productivity could arguably identify a metro-level housing demand curve. In the neoclassical model, home productivity increases the amount of housing much more than it lowers prices:

∂Y^∂p^|Q^,A^X=∂Y^/∂A^Y∂p^/∂A^Y=−4.48−0.13σD−3.69σX−15.12σY−22.37εL,r

When σD = σX = σY = 0.667 and εL,r = 0, this is −17.12. This large magnitude rests heavily on households having homogeneous tastes for locations. If so, improvements to housing productivity increase the quantity of housing much more than they lower prices.

Panel B of Figure 1 illustrates how general equilibrium elasticities of housing supply and demand vary with elasticities of substitution in consumption and production. As elasticities of substitution increase, the difference between housing supply elasticities identified by quality of life and trade productivity grows.

5. The Relationship between Density, Prices, and Amenities

5.1. Data

We now apply the neoclassical model empirically to see if it can predict population levels. We examine cities in the continental U.S., defined at the Metropolitan Statistical Area (MSA) level using 1999 Office of Management and Budget consolidated definitions (e.g., San Francisco is combined with Oakland and San Jose), of which there are 274. We use the 5-percent sample of the 2000 United States Census from Ruggles et al. (2004) to calculate wage and housing price differentials, controlling for relevant covariates (see Appendix F for details). Population density is from the 2000 Census Summary Tract Files. For each census tract, we take the ratio of population to land area and then population-average these densities to form metro-level densities, shown in Figure 2. We use MSA population weights throughout.

Figure 2:

Figure 2:

Metropolitan Population Density, Thousands per Square Mile, 2000

Figure 3 displays kernel density estimates of wages, housing prices, and population density across MSAs. Population density varies by an order of magnitude more than wages and prices. This basic fact is reflected in our amenity estimates below.

Figure 3:

Figure 3:

Distribution of Wages, House Prices, and Population Density, 2000

Predicted population density, calculated under the assumption of equal home-productivity across metros, depends only on wages and housing prices.

5.2. Predicting and Explaining Population Density

We first consider how well the model predicts population density using price information alone. As in Albouy (2016), we use estimates of Q^j and A^Xj based on w^j and p^j, from equations (1a) and (12), and assume A^Yj=0. Thus, predicted population density is simply εN∗,QQ^j+εN∗,AXA^Xj, approximated by 8.18Q^j−2.16A^Xj≈2.89p^j−2.37w^j. The prediction error for each city is then ξj=N^∗j−εN∗,QQ^j−εN∗,AXA^Xj.16

Figure 4 plots actual and predicted density against each other for each of the 274 metros, along with a 45-degree line. Overall, 49 percent of density variation is explained by this restricted neoclassical model, which to reiterate, does not estimate a single parameter.17 This model does underpredict density for a number of large, relatively old cities—such as New York, Chicago, and Philadelphia—as well as large Texan metros—including Houston, Dallas, and Austin. It overpredicts density for a number of metros in California and Florida, including San Francisco and Naples. Figure 3 shows that in general, the model predicts a more-compressed density distribution than is observed, even though the reduced-form elasticities are quite high.

Figure 4:

Figure 4:

Actual and Predicted Population Density, 2000

See text for estimation details. High density metros have population density which exceeds the national average by 80 percent, medium density metros are between the national average and 80 percent. Low density and very low density metros are defined symmetrically.

To see if other elasticities of substitution provide better fits, we consider how well other combinations of σD, σX, and σY predict density under certain restrictions. Figure 5 graphs the variance of the prediction error, V ar(ξ j), as a function of these elasticities. If σD = σX = σY = σ, V ar(ξ j) is minimized at σ = 0.7, close to the value of 0.667 from the pre-set parametrization. The common Cobb-Douglas case σ = 1 fits notably worse. Fixing σX = 0.667 reduces V ar(Ξj) for all other values of σD = σY . Fixing both σD = σX = 0.667, as in the lowest curve, reduces V ar(ξj) by roughly the same amount. V ar(ξj) is reduced the most by setting σY = 0.667. This underlines how responses in the home sector are key for understanding how population density varies.18

Figure 5:

Figure 5:

Variance of Error in Fitting Population Density using Quality of Life and Trade Productivity, as Function of Substitution Elasticities

We evaluate the variance for σ equal to 0, 0.1, 0.2, 0.3, 0.333, 0.4, 0.5, 0.6, 0.667, 0.7, 0.8, 0.9, 1, 1.25, and 1.5.

5.3. Using Density to Estimate Trade and Home Productivity

We next relax the restriction that home productivity is constant (A^Yj=0) and use density data to separately identify trade and home productivity, as described in Section 2.4. Panel A of Figure 6 displays our measures of inferred cost and excess density for MSAs. They are estimated from the left-hand sides of equations (12) and (13) under the parametrization with σD = σX = σY = 0.667. The figure includes iso-productivity lines for each of the traded and home sectors at average levels of productivity, A^Xj=0 and A^Yj=0.

Figure 6:

Figure 6:

Results of Parametrized Model, 2000

See note to figure 4 for metro density definitions.

To visualize the productivity estimates, first consider the downward-sloping iso-trade-productivity line, along which cities have average trade productivity. Above and to the right of this line, cities have high inferred costs, together with considerable excess density. This indicates they have above-average trade productivity. Second, consider the upward-sloping iso-home-productivity line. Above and to the left of it, cities have high excess density, with low inferred costs. This indicates high home productivity. Vertical deviations from this second line are equal to the prediction error ξj in section 5.2. Since the first line is almost vertical, and the second almost horizontal, excess density—or prediction error for N∗j—affects home productivity measures much more than trade productivity ones. Note that the slopes of either line rise with the substitution elasticities, as productivity has a greater impact on excess density.

Panel B of Figure 6 graphs trade and home productivity directly, through a change in coordinates of Panel A. Examining each quadrant in turn, Chicago and Philadelphia have high levels of both trade and home productivity, while New York is the most productive overall. San Francisco has the highest trade productivity but low home productivity. San Antonio has low trade productivity and high home productivity. Santa Fe and Myrtle Beach are unproductive in both sectors.19

These home productivity estimates deserve several comments. First, they reflect density much more than prices.20 Second, higher values of σY raise their sensitivity to prices relative to density. Third, home productivity is highest in large, older cities.

This third phenomenon may be due to the static assumptions of the neoclassical model, which can introduce various specification errors. Indeed, the cores of these cities were built prior to World War I, when most land-use regulations were absent, intra-city transport costs were higher, and European immigration was at its peak. These factors caused housing to be built at high densities, which have largely endured.21 The high excess density in these cities may also be an enduring legacy of these cities having relatively higher trade productivity and quality of life in earlier times. These relative advantages fell with the decline of manufacturing and the rise of air conditioning (Glaeser and Tobio, 2008). Yet durable housing and moving costs have prevented populations from adjusting fully to the neoclassical baseline. These speculations warrant further investigation in models incorporating such dynamic mechanisms.22

To summarize the data and findings, Table 4 presents estimates of population density, wages, housing costs, inferred land values, and amenity differentials for a selected sample of metropolitan areas. Appendix Table A.3 contains a full list of metropolitan and non-metropolitan areas and compares trade productivity estimates that make use of density with those that do not (i.e., based on inferred costs only).

Table 4:

List of Selected Metropolitan Areas, Ranked by Population Density

Population density Wage Home price Land value Quality of life Trade productivity Home productivity
Name of metropolitan area N^j w^j p^j r^j Q^j A^Xj A^Yj
New York, Northern New Jersey, Long Island, NY-NJ-CT-PA 2.295 0.217 0.432 3.406 0.032 0.272 0.502
Los Angeles-Riverside-Orange County, CA 1.259 0.134 0.455 1.947 0.080 0.164 0.086
San Francisco-Oakland-San Jose, CA 1.219 0.259 0.818 2.051 0.138 0.273 −0.173
Chicago-Gary-Kenosha, IL-IN-WI 1.200 0.134 0.230 1.791 0.008 0.160 0.274
Miami-Fort Lauderdale, FL 0.973 0.010 0.127 1.373 0.037 0.043 0.200
Philadelphia-Wilmington-Atlantic City, PA-NJ-DE-MD 0.967 0.115 0.062 1.411 −0.038 0.134 0.341
San Diego, CA 0.881 0.061 0.485 1.440 0.123 0.088 −0.110
Salinas (Monterey-Carmel), CA 0.847 0.102 0.602 1.444 0.142 0.124 −0.200
Boston-Worcester-Lawrence, MA-NH-ME-CT 0.806 0.121 0.343 1.279 0.051 0.136 0.033
Santa Barbara-Santa Maria-Lompoc, CA 0.722 0.058 0.666 1.299 0.182 0.082 −0.326
…
Florence, AL −1.388 −0.132 −0.352 −2.099 −0.047 −0.166 −0.223
Myrtle Beach, SC −1.392 −0.188 −0.125 −2.025 0.051 −0.212 −0.468
Florence, SC −1.397 −0.140 −0.336 −2.114 −0.038 −0.173 −0.247
Johnson City-Kingsport-Bristol, TN-VA −1.408 −0.190 −0.352 −2.148 −0.019 −0.217 −0.272
Gadsden, AL −1.436 −0.136 −0.422 −2.192 −0.068 −0.172 −0.177
Goldsboro, NC −1.499 −0.197 −0.288 −2.228 0.003 −0.225 −0.358
Dothan, AL −1.523 −0.189 −0.404 −2.322 −0.036 −0.220 −0.259
Anniston, AL −1.569 −0.202 −0.425 −2.398 −0.037 −0.234 −0.264
Ocala, FL −1.572 −0.170 −0.295 −2.363 −0.009 −0.205 −0.365
Hickory-Morganton-Lenoir, NC −1.614 −0.135 −0.220 −2.357 −0.004 −0.174 −0.416
Rocky Mount, NC −1.631 −0.122 −0.241 −2.384 −0.017 −0.165 −0.394
Standard deviation 0.871 0.116 0.282 1.323 0.050 0.129 0.200

Table 4 includes the top and bottom ten metropolitan areas ranked by population density. The first three columns are estimated from Census data, while the last four columns come from the parametrized model. See text for estimation procedure. Standard deviations are calculated among the 274 continental metropolitan areas using metro population weights. All variables are measured in log differences from the national average.

5.4. City-Specific Elasticities of Substitution

Because geographic and regulatory environments are heterogeneous, housing producers’ ability to substitute between land, labor and capital may vary considerably. This heterogeneity is interesting in its own right, and it can impact the model’s predictions. We model heterogeneity across metros by assuming that σYj is a linear function of Ij, the Wharton Residential Land Use Regulatory Index (WRLURI) from Gyourko et al. (2008); Sj, the average slope of land from Albouy et al. (2016b); and vj, a residual: σYj=σY0+σYIIj+σYSSj+vj. We normalize Ij and Sj to have mean zero and standard deviation one.23 Similarly, we assume that home productivity is given by a linear function: A^Yj=aIIj+aSSj+uj As shown in Appendix G, these assumptions yield the following equation:

N^ej=σY0G^j+σYIIjG^j+σYSSjG^j+aI(k1+σY0k2)Ij+aS(k1+σY0k2)Sj+σYIaIk2(Ij)2+σYSaSk2(Sj)2+(σYIaS+σYSaI)k2IjSj+ej, (23)

where N^ej≡N^∗j−1.00p^j+0.79w^j is density explained by all but σYj anzd A^Yj, G^j≡2.82p^j−2.37w^j captures observable demand shifts from Q^j and A^Xj, k1 and k2 are known positive constants, and ej is a residual. Non-linear least squares provides the parameter estimates under orthogonality conditions for uj and vj discussed in Appendix G.

The estimator here differs from alternative approaches in several ways. First, it is identified by level differences in population, not changes. Second, as implied by (22a) and (22b), it handles demand shifts asymmetrically, putting more weight on quality of life than trade productivity. Third, it accounts for all demand shifts, absent specification error. This eliminates the need for instrumenting demand shifts (e.g., with January temperature or Bartik employment shocks) that are ostensibly exogenous to supply.

An important concern is that Ij and Sj are correlated with unobserved supply shifters in uj or vj. As a way of testing the specification, the model provides three over-identifying restrictions. The linear reduced-form equation of (23) has eight terms {G^j,Ij,Sj,IjG^j,SjG^j,IjSj,(Ij)2,(Sj)2}, with coefficients that depend non-linearly on the five structural parameters, {σY 0, σY I, σY S, aI, aS}.

The data fail to reject the implied structural restrictions of the model (p = 0.62), providing support for the estimates of equation (23), shown in Table 5. The baseline estimate of σY 0 in column 1 is 0.69, again close to the pre-set value. The results in column 2, which restricts A^Yj=0, imply (intuitively) that σYj is negatively related to both regulations and average slope: a one-standard-deviation increase reduces the elasticity by 0.16 and 0.30. The predicted elasticities σYj have a population-weighted mean of 0.89—higher than without the interactions—with a standard deviation of 0.37.24 This model now explains 61 percent of the variation in density, as opposed to the 49 percent from before with a uniform σY. Column 3 holds σYj constant and lets A^Yj vary: it falls by about 7 percent with a one standard deviation increase in slope. Column 4 presents the full model estimates which resemble those in columns 2 and 3.

Table 5:

The Determinants of Substitution Possibilities and Productivity in the Home Sector

Dependent variable: Population density not explained by home sector
(1) (2) (3) (4)
Elasticity of Substitution in Home Sector
 Baseline σY 0 0.693 (0.247) 0.887 (0.272) 0.865 (0.325) 1.106 (0.348)
 Wharton Land Use Regulatory Index (s.d.) σI −0.157 (0.127) −0.184 (0.112)
 Average slope of land (s.d.) σS −0.296 (0.188) −0.306 (0.168)
Housing Productivity
 Wharton Land Use Regulatory Index (s.d.) aI 0.014 (0.038) −0.005 (0.015)
 Average slope of land (s.d.) aS −0.071 (0.018) −0.068 (0.023)
Observations 274 274 274 274

Table 5 presents results of estimating equation (23) by nonlinear least squares. Explanatory variables are normalized to have mean zero and standard deviation one. Regressions are weighted by population. Robust standard errors are in parentheses.

The estimates of s σYj imply city-specific elasticities of housing supply according to the formulae from Section 4. We calculate these housing supply elasticities for when they come from shifts in trade productivity, under constant geography (εL,r = 0), setting σD = σX = 0.667 and σYj to the predicted value from column 2 of Table 5. A regression of the supply elasticities from Saiz (2010) on these calculated elasticities yields a slope of 1.04 (s.e. 0.19), an intercept of 0.45 (s.e. 0.22), and a correlation coefficient of 0.45. This slope is indistinguishable from one. The intercept is close to the value predicted in equation (21) from the consumption response, sxσD = 0.43, as Saiz estimates a population response, rather than a housing one.25 The similarity is remarkable, given how differently the elasticities are estimated.

6. Land Area and the Total Population of Cities

The neoclassical model explains density levels across metros fairly well. To explain entire population levels, it must also model land area, which varies tremendously. The original Rosen-Roback model ignores the internal structure of cities and takes its land as homogeneous and fixed. We supplement it with a simple land supply function from equation (5). Now, this function may vary both in its endowment, L^0j, and supply elasticity, εL,rj.

To our knowledge, our framework provides the first to estimate both the intensive (density) and extensive (land) margins of urban growth separately. The neoclassical model makes this separation clean, as it uses a unified land market of homogeneous land, with constant returns in urban production. For estimation purposes, take the land supply parameters as linear functions of covariates W j, with L^0j=WjβL0+uj and εL,rj=ε¯+Wjβε+vj. W j includes Ij, Sj, and also the log land share (i.e., the share which is not water) from Saiz (2010).

We measure land using the square miles in the Census urban area; metropolitan areas, defined by counties, contain a considerable amount of land for non-urban use, which we exclude. Panel A of Figure 7 plots this land area against the inferred land rent from (1c) when A^Yj=0, i.e., r^j=(p^j−ϕNw^j)/ϕL. Since cities are small and open to mobile labor and capital, the demand for land is perfectly elastic at each city’s price r^j. With no other covariates, the regression line traces out a supply curve. The slope equals the supply elasticity, given by ε¯=0.85 the weighted intercept provides a base land endowment of 773 square miles.

Figure 7:

Figure 7:

Urban Land Area and Inferred Land Rents

Notes: Inferred land rent comes from equation (1c) assuming A^Yj=0. See note to figure 4 for metro density definitions.

Table 6 reports results from specifications with covariates.26 A one standard deviation increase in slope lowers the land endowment by over half, while a one standard deviation increase in land not covered by water increases it by almost a quarter. In the fully-interacted model, the population-weighted average elasticity of land is 1.58 (unweighted mean is 1.47) but is reduced by about 0.25 from a one standard deviation increase in slope or regulation. While these results are not as well identified as those in Table 5, they do accord with intuition, suggesting that the land measure and inferred land rents contain valuable information.

Table 6:

The Determinants of Land Supply

Dependent variable: Log urban area, square miles
(1) (2) (3)
Inferred land rent 0.852 (0.190) 1.202 (0.210) 1.468 (0.142)
Wharton Land Use Regulatory Index (s.d.) 0.073 (0.094) −0.086 (0.086)
Average slope of land (s.d.) −0.610 (0.099) −0.546 (0.078)
Log land share (s.d.) 0.240 (0.098) 0.279 (0.078)
Interaction between inferred land rent and
 Wharton Land Use Regulatory Index (s.d.) −0.219 (0.089)
 Average slope of land (s.d.) −0.247 (0.087)
 Log land share (s.d.) 0.069 (0.117)
Constant 6.650 (0.143) 6.650 (0.112) 6.898 (0.108)
Observations 274 274 274
R-squared 0.377 0.526 0.594

Inferred land rent is constructed without using density data. All explanatory variables are normalized to have mean zero and standard deviation one. Regressions are weighted by population. Robust standard errors are in parentheses.

To examine how well the model explains cross-metro population differences, we use equation (7) to predict the total population differential as the sum of the predicted land differential –equal to WjβL0+ε¯r^j, from column 2 of Table 6—and the simple predicted density differential—conditional on p^j and w^j. Figure 8 plots actual metro population against this prediction. The simple prediction explains 49 percent of cross-metro population variation, without using data on either density or population. This fit is reasonably successful, as it is based on a model built to value amenities from prices alone. Nevertheless, the model could clearly be improved to incorporate other economic factors.

Figure 8:

Figure 8:

Actual and Predicted Population

See note to figure 4 for metro density definitions.

7. Population Determinants and Counterfactual Exercises

7.1. Why Do People Live Where They Do?

Here we answer the question of whether people follow jobs, jobs follow people, or both follow housing. To do this, we decompose the variance of cross-metro differences in population due to quality of life, trade productivity, and home productivity. Column 1 of Table 7 begins with a restricted model of density with constant home productivity A^Y=0, and substitution elasticities set to σ = 0.667. Quality of life accounts for nearly half of the explained variance, dominating trade productivity (i.e., inferred costs), even though the latter shows greater cross-sectional variation in value (see Appendix Figure A.2). Quality of life and trade productivity are positively correlated.

Table 7:

Fraction of Density and Population Explained by Quality of Life, Trade Productivity, Home Productivity, and Land

Variance/Covariance Component Notation Density Population
(1) (2) (3) (4)
Quality of life Var(εN,QQ^) 0.488 0.225 0.585 0.223
Trade productivity Var(εN,AXA^X) 0.189 0.103 0.315 0.142
Home productivity Var(εN,AYA^Y) - 0.438 - 0.393
Land Var(L^0) - - 0.167 0.064
Quality of life and trade productivity Cov(εN,QQ^,εN,AXA^X) 0.323 0.138 0.457 0.161
Quality of life and home productivity Cov(εN,QQ^,εN,AYA^Y) - −0.141 - −0.133
Quality of life and land Cov(εN,QQ^,L^0) - - −0.41 −0.157
Trade and home productivity Cov(εN,AXA^X,εN,AYA^Y) - 0.237 - 0.263
Trade productivity and land Cov(εN,AXA^X,L^0) - - −0.114 −0.036
Home productivity and land Cov(εN,AYA^Y,L^0) - - - 0.078
Total variance of prediction 0.350 0.758 2.190 5.740
Data used to construct attributes
Wages and housing prices Yes Yes Yes Yes
Density No Yes No Yes
Predicted land intercept No No Yes Yes

Predicted land intercepts come from column 2 of Table 6.

The decomposition in column 2 lets AY vary across metros, so that the entire population density is explained by all three attributes. As before, quality of life dominates trade productivity, yet both are dominated by home productivity. While all three attributes are important in explaining density, it appears that people and jobs follow housing more than anything else. Because the home productivity estimates are heavily based on density, this conclusion should be treated cautiously. Nonetheless, it complements the finding that home sector substitution is key in how population responds to amenities.

The decompositions in columns 3 and 4 bring in land supply to account for total population. To keep the accounting tractable, we use the specification from column 2 of Table 6, with a uniform price elasticity of 1.20 for land, and vary land endowments. In column 3, we see quality of life continues to dominate trade productivity, while both dominate the land endowment. Finally, column 4 considers the full model for population. As in column 2, home productivity dominates quality of life and trade productivity. The largest interaction is the positive one between home and trade productivity.

One stimulating result from this table is that quality of life is negatively related to land and home productivity. Unfortunately, the most attractive areas of the U.S. are often the most difficult to build on. This appears to stem mainly from two causes. First, coastlines and rugged terrain are associated with higher quality of life (Albouy, 2008), but lower land supply and ability to build densely, as shown above. Second, higher quality of life areas tend to have more land use restrictions, although these do not seem to improve quality of life (Albouy and Ehrlich, 2018).27

Appendix Table A.5 explores how the results are affected by preference heterogeneity, endogenous amenity feedback, and geographically-neutral federal taxation. Incorporating preference heterogeneity places makes quality of life relatively more important than trade and home productivity. To some extent, this is mechanical, as equation (16) infers that quality of life rises with population if there is preference heterogeneity. This also explains why quality of life is positively correlated with home productivity in this model. Incorporating endogenous amenity feedback—where population density decreases quality of life and home productivity while increasing trade productivity—yields similar results, as these forces cancel each other out to some degree. As discussed in more detail in Appendix D.7, endogenous amenity feedback and preference heterogeneity have similar (and so, hard to separate) effects in a log-linearized model. On the policy side, if federal taxes were made geographically neutral, trade productivity would determine locations more than quality of life; people would follow jobs more than the opposite.28

7.2. What if Chicago was as Nice as San Diego?

As quality of life is key in determining where people live, consider what would happen if the city with the largest growth potential, Chicago, were given the quality of life of one of America’s nicest cities, San Diego. In this counterfactual, Chicago receives none of the attributes that lower San Diego’s population. As seen in Table 8, Chicago has an elastic home good sector, with σY = 1.23, which from (18) implies N^∗j=12.69Q^j. The difference in quality of life between San Diego and Chicago is 0.12, explained entirely by climate and geography (Albouy, 2008), making it effectively exogenous. Therefore, the model predicts that the population of Chicago would expand by 3.59(= exp(12.69 × 0.12) − 1) times. Based on the 2000 numbers, this implies a population of 42 million, double that of New York City!29 A sunny and beautiful city with such great home productivity would likely be full of gleaming skyscrapers, packed with residents. On the other hand, if San Diego’s quality of life fell to that of Chicago’s, the long-run effect would be less dramatic, as its home good sector is less responsive. With σY = 0.30, N^∗j=5.24Q^j, so San Diego’s population would fall by 47 percent, from 2.8 to 1.5 million.

Table 8:

Changes in Population from Relaxing Land Use Regulations

Panel A: Metro-Level
Home Subs. Elasticity (σYj) 2000 Population, Mil.
Main city in MSA Estimated Counterfactual Observed Counterfactual
San Francisco 0.29 0.42 7.0 8.4
San Diego 0.30 0.47 2.8 3.2
Los Angeles 0.45 0.57 16.4 18.3
Boston 0.71 0.96 5.8 6.2
Seattle 0.11 0.31 3.5 3.7
New York 0.94 0.97 21.2 21.8
Philadelphia 0.89 1.04 6.2 6.3
Miami 1.07 1.21 3.9 3.9
Denver 0.49 0.74 2.6 2.6
Phoenix 0.67 0.84 3.2 3.2
Washington-Baltimore 0.87 0.94 7.6 7.5
Detroit 1.15 1.15 5.5 5.3
St. Louis 1.27 1.27 2.6 2.5
Chicago 1.23 1.23 9.2 8.9
Houston 1.22 1.22 4.7 4.6
Dallas 1.15 1.15 5.2 5.1
Cleveland 1.07 1.07 3.0 2.9
Tampa 1.17 1.17 2.4 2.3
Pittsburgh 0.70 0.71 2.4 2.3
Portland 0.48 0.50 2.3 2.2
Minneapolis 1.02 1.05 3.0 2.9
Atlanta 0.96 0.98 4.1 4
Panel B: Change in Regional Dist. Panel C: Change in Amenity Dist.
Northeast 1.01 Quality of life 0.002
Midwest 0.98 Trade productivity 0.005
South 0.97 Home productivity 0.003
West 1.06 Total value 0.006

The counterfactual decreases the WRLURI to the population-weighted mean in metro areas with an above-average WRLURI, thus increasing the home substitution elasticity based on the estimates in column 2 of Table 5. Land supply and amenities are held fixed in the counterfactual.

A related thought experiment is to make home productivity in San Diego as high as in Chicago. From (18), the elasticity of population density in San Diego with respect to home productivity is N^∗j=1.93A^Yj. As a result, a 0.38 log point increase in home productivity (see Table 4) would increase population in San Diego by 8 percent. The small response stems from the relatively low elasticity in San Diego’s home good sector of 0.3. If this elasticity and home productivity both rise to Chicago’s, San Diego’s population would then grow by 4.24(= exp(4.36 × 0.38) − 1) times.

7.3. The Effect of Relaxing Land Use Regulations

The parametrized model readily permits nationwide counterfactual policy exercises. Given the importance of the home good sector, we focus on the effects of lowering land use regulations in cities for inhabitants with above-average regulation. This exercise is similar to Hsieh and Moretti (2019)—who lower regulations more dramatically—although we examine levels instead of growth.30 We hold amenities fixed in these counterfactuals.

Table 8 presents results from these counterfactual exercises. In Panel A, column 1 shows the estimated elasticity of substitution in housing, and column 2 shows the counterfactual elasticity when cities with above-average regulation are lowered to the average.31 Column 3 shows each metro’s population in 2000, and column 4 shows the population under the counterfactual. The elasticities in several coastal cities, notably San Francisco, Los Angeles, and San Diego, increase substantially. This permits many more people to take advantage of their amenities. Because population must balance, less attractive cities, such as Detroit, Atlanta, and Dallas, lose residents even without changes in their elasticity. As seen in Panels B and C, the West would gain population from the South and Midwest, and people would live in more amenable and productive places. The increase in the total value of amenities experienced by households would equal 0.6 percent of GDP.

As discussed in Section 2.5, population will respond less to changes in amenities or regulations in the presence of imperfect mobility. The predicted populations presented here most sensibly describe a long-run equilibrium.

8. Conclusion

This paper studies what the canonical spatial equilibrium model of household location decisions implies for population numbers. Our full characterization flexibly captures interactions between labor and housing markets, workers’ preferences, and firms’ production technology. Location decisions are determined by quality of life, trade productivity, home productivity, and land. We allow land use regulations and geography to affect both home productivity and substitution possibilities in home good production. We also introduce a novel approach to estimating home productivity, elasticities of substitution in the home good sector, and land supply elasticities.

We find that a restricted model—in which home productivity does not vary across metros—explains half of the observed variation in density and population across metro areas. When we use the structure of the model to jointly estimate quality of life, trade productivity, and home productivity, we find that quality of life is more important than trade productivity in determining household location decisions, and home productivity may be more important than either. We use the model to quantify the effects of relaxing land use regulations, which would increase substitution possibilities in housing production. Households would move to nicer and more productive cities. Population would increase by 6 percent in the West—driven by increases of 20 percent in San Francisco, 14 percent in San Diego, and 12 percent in Los Angeles—while falling in the Midwest and South.

The neoclassical framework accounts for the most basic factors that affect urban life. The version we develop here requires relatively simple data, and thus it could be used to understand location decisions in other periods and countries. It is also remarkably versatile for adding features, such as agglomeration, multiple types of workers, and preference heterogeneity. We hope that this version serves as a unifying framework for future work on the determinants of urban population.

Supplementary Material

1

Acknowledgments

For their help and input we thank David Agrawal, Nate Baum-Snow, Rebecca Diamond, Jesse Gregory, Andrew Haughwout, Jordan Rappaport, Will Strange, Juan Carlos Suárez Serrato, and Owen Zidar; conference participants at the 2012 Urban Economics Association annual meeting, 2012 National Tax Association annual meeting, 2013 American Real Estate and Urban Economics Association annual meeting, 2013 Canadian Economics Association annual meeting, 2013 National Bureau of Economic Research Summer Institute meeting in Urban Economics, 2013 Housing Urban Labor Macro meeting in Atlanta; and seminar participants at Calgary, Cornell, the Cleveland Federal Reserve, Georgia State University, IEB Barcelona, the Kansas City Federal Reserve, Maryland, Michigan, Minnesota (Applied Economics), NYU Abu Dhabi, the Paris School of Economics, Purdue, Sciences Politiques, Toronto, and the Toulouse School of Economics. During work on this project, Albouy was supported by the NSF grant SES-0922340 and Stuart was supported by an NICHD training grant (T32 HD007339) and an NICHD center grant (R24 HD041028) to the Population Studies Center at the University of Michigan.

Footnotes

This paper was previously presented as “Urban Quantities and Amenities.” Any mistakes are our own.

2

We use the term “neoclassical” as it relies on standard elements in consumer and producer theory, resembling the two-sector models of Heckscher (1919) and Ohlin (1924) on trade, Uzawa (1961) and Stiglitz (1967) on growth, and Harberger (1962) on tax incidence.

3

The first two attributes concern the problem of whether “jobs follow people” or “people follow jobs” (e.g., Carlino and Mills, 1987; Hoogstra et al., 2017) while the third addresses whether both jobs and people follow housing (e.g., Glaeser et al., 2005). The cross-sectional method we employ assesses the relative importance of these dimensions without relying on timing assumptions necessary in studies based on time-series evidence.

4

We discuss the relationship between our model and previous work in detail in Appendix A. Haughwout and Inman (2001) develop a similar model for one city without home production. Rappaport (2008b,a) derives quantitative implications of roughly the same model in a setting with two cities, one large and one small. Glaeser and Gottlieb (2009) assume unitary elasticities of substitution, fix separate land supplies in home and traded production, and derive analytical expressions, without applying them to data. Lee and Li (2013) use a similar model to explain Zipf’s law of city sizes, without identifying particular cities or amenities. Saiz (2010) and Desmet and Rossi-Hansberg (2013) apply their models directly to city-level data using monocentric models with inelastic housing demand and constant density, and without land in trade production or labor in home production. Modeling decadal changes, Suárez Serrato and Zidar (2016) and Diamond (2016) exclude labor from non-traded production and land from traded production, and assume at least two unitary elasticities of substitution while introducing heterogeneous tastes for cities. Allen and Arkolakis (2014), Bartelme (2015), Caliendo et al. (Forthcoming), and Fajgelbaum et al. (2015) consider trade costs and monopolistic competition in models that start from, yet restrict the benchmark neoclassical model. Our results on the role of land use regulations complement recent work by Hsieh and Moretti (2019), who focus on how these regulations affected economic growth between 1964 and 2009.

5

Previous work estimating quality of life and trade-productivity differences includes Beeson (1991), Gabriel and Rosenthal (2004), Shapiro (2006), Albouy (2016).

6

The model can be generalized to allow nonlinear income taxes. Our empirical implementation adjusts for state taxes and tax benefits to owner-occupied housing.

7

The recursive structure vanishes if workers have idiosyncratic preferences for cities or amenities depend endogenously on quantities, as described below.

8

For example, areas with high land costs and low labor costs would produce goods that use labor intensively. A representative zero-profit condition is formed by an envelope of the zero-profit conditions for each good, with a greater variety of goods reflected in greater substitution possibilities.

9

In principle, land supply can vary on two different margins. At the extensive margin, an increase in land supply corresponds to a growing city boundary. Extensive margin differences can be driven by the land endowment L^0j or the supply function εL,rr^j. At the intensive margin, an increase in land supply takes the form of employing previously unused land within a city’s border. The assumption of full utilization in (4b) and (5) rules out intensive changes.

10

Albouy et al. (2018) estimate r^j using transaction purchase data, which is only available for recent years. Their analysis discusses several conceptual and empirical challenges from this approach. Moreover, land value data is generally not available in most years in most countries.

11

The parameter σ in Notowidigdo’s moving cost function is analogous to − ψ in equation (16). Notowidigdo estimates σH = −0.066 for high-skill workers and σL = −0.065 for low-skill workers. The relationship to Hsieh and Moretti (2019) is straightforward (see their discussion on p. 18).

12

The model features a “local” general equilibrium, in the sense that there are interactions between input and output markets within each metro, while each metro is assumed to be small enough not to affect the nationwide variables u¯ or ı¯.

13

Most researchers from Bartik (1991) onwards do not find that these elasticities differ much between 10-year and longer horizons. Notowidigdo (Forthcoming) estimates a preference dispersion parameter of ψ = 0.07, in a model that holds quality of life constant. In our model, this reduces the elasticity of labor to nominal wages to 1.22 (see Table A.1).

14

A number of papers estimate the relationship between immigration-induced (total) labor supply changes and (average) wage changes, which is the closest empirical analog to (∂N^/∂w^)|A^X,A^Y. Results from such regressions vary widely, as discussed by Borjas (1999).

If demand for the traded good is not perfectly elastic, as in a model with heterogeneous traded output, then the elasticity of labor demand will be lower. To illustrate this in a partial equilibrium setting, let demand for the local traded good be X^j=−ηp^Xj where pXj is its price, formerly fixed. Let land supply for traded-good firms be provided in a segmented market by L^Xj=L^0Xj+εXr^Xj. From the equations governing the firm—(1b), (3a), and (3b)—one can derive a general form of Marshall’s Rule for labor demand in the trade sector. It includes additional terms of trade productivity and the land endowment:
N^Xj=−[σX(η+εX)+θNεX(η−σX)−θK(η−σX)σX]w^j+(η−1)(σX+εX)A^Xj+(η−σX)θLL^0XjηθL+σX(1−θL)+εX.
The coefficient on wages increases with η, meaning labor demand is more elastic when product demand is elastic. If we take εX = 1 (see Table 6 below), then a commonly estimated value of η = 4 (e.g. from Suárez Serrato and Zidar (2016)) produces a labor demand elasticity of −3.2, while η = ∞ produces a an elasticity of −22. We also see that wages here rise with the endowment of land, comparable to a fixed capital, as in Glaeser and Gottlieb (2008) or Hsieh and Moretti (2019).
15
Consider a typical partial equilibrium setting of home good supply with exogenous wages, but where we let land values be set endogenously by the zero-profit condition (1c). For simplicity, also assume that home and traded good land markets are segmented. Thus, residential land is given by Yj=L^0Yj+εYr^Yj. Then the supply of housing increases with prices, productivity, and land endowments, while it falls with wages:
Y^j=σY(1−ϕL)+εYϕLp^j−(σY+εY)ϕNϕLw^j+[1+σY(1−ϕL)+εYϕL]A^Yj+L^0Yj
Parametrized, Y^=(2.2+4.3εY)p^j−(1.8+2.6εY)w^j+(3.2+4.3εY)A^Yj. The base coefficient of 2.2 on p^j is similar to estimates in the literature. The formula also highlights the separate role of an extensive margin through land supply in εY and L^0Yj, as well as productivity in A^Yj in determining supply. Local wages, w^j, play a particular role as Q^j and A^Yj lower wages, while A^Xj raises them. Thus, omitting wages from an elasticity regression could create an omitted variable bias in estimating the structural parameters, either up or down, depending on the origin of the price change. Furthermore, with heterogeneous preferences (section 2.5), the total elasticity, net of demand, is lower.

Note that even without modeling local labor inputs, ϕN = 0, we cannot reduce the supply relationship between home-good prices and population as Q^j is present in (21).

16

Slight differences stem from state taxes. Note, an unrestricted regression of log density on wages and housing costs (naturally) produces a higher R-squared of 0.72 > 0.47, with N^∗j=4.40w^j+0.90p^j+ej=0.63Q^j+6.26A^Xj+ej. The value of 0.63 is much lower than the parametrized value of εN,Q, 8.18, while the value of 6.26 is larger than εN,AX. The two parameters in this regression cannot identify three separate substitution elasticities. But if the elasticities are constrained equal, σD = σX = σY = σ, then the constrained regression produces values similar to the parameterization: N^∗j=8.57Q^j+2.30A^Xj+ej, implying σ = 0.68.

17

We assess model fit by reporting the square of a linear correlation coefficient, from a linear fit with an imposed slope of one.

18

As described in Appendix E.3, our parameterization of the share parameters explains a significantly higher fraction of the cross-metro variance in population density than alternatives.

19

Panel B of Figure 6 also includes isoclines for excess density and inferred costs, which correspond to the axes in Panel A. If quality of life is held constant, trade productivity and home productivity must move in opposite directions to keep population density constant. Similarly, home productivity must rise faster than trade productivity to keep inferred costs constant.

20

According to the parametrization in Appendix Table A.2, A^Yj=0.32N^∗j+0.73w^j−0.93p^j, which largely reflects density since density varies so greatly and housing prices and wages are positively correlated. Trade productivity is A^Xj=0.03N^∗j+0.84w^j+0.01p^j. Quality of life depends only on the price measures Q^j=−0.48w^j+0.32p^j. Land values reflect all three measures, r^j=1.37N^∗j+0.49w^j+0.32p^j, although density is key.

21

Consistent with this explanation, Table A.9 shows that 59 percent of the variation in home productivity is explained by four variables: the Wharton Residential Land Use Regulatory Index (WRLURI), average slope of land, log infrastructure stock as of 1980 from Albouy and Farahani (2017), and the share of housing units in 1980 that were built before 1940.

22

Albouy and Ehrlich (2018) use data on land values to infer productivity in the housing sector, which comprises most of the non-traded sector. While the two approaches generally agree on which large areas have high home productivity, the land values approach suggests that larger, denser cities generally have lower, rather than higher housing productivity. This apparent contradiction actually highlights what the two methodologies infer differently. Productivity measures based on current land values provide a better insight into the marginal cost of increasing the housing supply, by essentially inferring the replacement cost. Productivity measures based on density are more strongly related to the average cost of the housing supply, thereby reflecting the whole history of building in a city. The distinction matters particularly for older cities where older housing was built on the easiest terrain, and in decades prior to strict residential land use regulations, which typically grandfather pre-existing buildings.

23

When these variables are missing, we impute them with the population-weighted mean. The results are robust to just focusing cities with non-missing variables.

24

The unweighted mean and standard deviation are 0.94 and 0.45.

25

Saiz’s empirical strategy examines temporal variation using industrial composition, immigrant enclaves, and sunshine as sources of exogenous variation in demand. By combining quality of life and productivity shifters, the estimates may not be directly comparable, although we suspect that productivity shifters are more important in his analysis.

26

Summary statistics are in Appendix Table A.4.

27

The equilibrium model ignores that people are gradually moving to areas with nicer weather (Rappaport, 2007; Glaeser and Tobio, 2008).

28

We use our amenity estimates and parametrized model to predict prices and quantities (including population density) for each city in the absence of location-distorting federal income taxes. Because we estimate amenities using observed density, wage, and housing price data, we cannot estimate amenities in the absence of distortionary federal taxes.

29

A change of this kind would increase the welfare of the country by reducing the density outside of Chicago. This would create a feedback effect that would lower the population increase by some minor amount given that this population change is around 10 percent of the national population.

30

We lower regulations to the population-weighted average of the WRLURI, while Hsieh and Moretti (2019) lower regulations to the median city in their sample. The reduction in WRLURI considered by Hsieh and Moretti is over half a standard deviation lower than our baseline counterfactual.

31

These rely on the estimates in column 2 of Table 5.

References

  1. Albouy D, “Are Big Cities Really Bad Places to Live? Improving Quality-of-Life Estimates across Cities,” NBER Working Paper No 14981 (2008).
  2. Albouy D, “The Unequal Geographic Burden of Federal Taxation,” Journal of Political Economy 117 (2009), 635–667. [Google Scholar]
  3. Albouy D, “What are Cities worth? Land Rents, Local Productivity, and the Total Value of Amenities,” Review of Economics and Statistics 98 (2016), 477–487. [Google Scholar]
  4. Albouy D, Chernoff A, Lutz C and Warman C, “Local Labor Markets in the U.S. and Canada,” Journal of Labor Economics 37 (2019), S533–S594. [Google Scholar]
  5. Albouy D and Ehrlich G, “Housing Productivity and the Social Cost of Land-Use Restrictions,” Journal of Urban Economics 107 (2018), 101–120. [Google Scholar]
  6. Albouy D, Ehrlich G and Liu Y, “Housing Demand, Cost-of-Living Inequality, and the Affordability Crisis,” NBER Working Paper No 22816 (2016a).
  7. Albouy D, Ehrlich G and Shin M, “Metropolitan Land Values,” Review of Economics and Statistics 100 (2018), 454–466. [Google Scholar]
  8. Albouy D and Farahani A, “Valuing Public Goods More Generally: The Case of Infrastructure,” Upjohn Institute Working Paper 17–272 (2017).
  9. Albouy D, Graf W, Kellogg R and Wolff H, “Climate Amenities, Climate Change, and American Quality of Life,” Journal of the Association of Environmental and Resource Economists 3 (2016b), 205–246. [Google Scholar]
  10. Allen T and Arkolakis C, “Trade and the Topography of the Spatial Economy,” Quarterly Journal of Economics 129 (2014), 1085–1140. [Google Scholar]
  11. Antràs P, “Is the U.S. Aggregate Production Function Cobb-Douglas? New Estimates of the Elasticity of Substitution,” The BE Journal of Macroeconomics 4 (2004), 1–36. [Google Scholar]
  12. Arzaghi M and Henderson JV, “Networking off Madison Avenue,” The Review of Economic Studies 75 (2008), 1011–1038. [Google Scholar]
  13. Bartel AP, “Where Do the New US Immigrants Live?” Journal of Labor Economics 7 (1989), 371–391. [DOI] [PubMed] [Google Scholar]
  14. Bartelme D, “Trade Costs and Economic Geography: Evidence from the US,” Mimeo, University of Michigan (2015).
  15. Bartik TJ, Who Benefits from State and Local Economic Development Policies? (Kalamazoo: Upjohn Institute, 1991). [Google Scholar]
  16. Baum-Snow N and Pavan R, “Understanding the City Size Wage Gap,” The Review of Economic Studies 79 (2012), 88–127. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Beeson PE, “Amenities and Regional Differences in Returns to Worker Characteristics,” Journal of Urban Economics 30 (1991), 224–241. [Google Scholar]
  18. Berndt ER, “Reconciling Alternative Estimates of the Elasticity of Substitution,” The Review of Economics and Statistics 58 (1976), 59–68. [Google Scholar]
  19. Blanchard OJ and Katz LF, “Regional Evolutions,” Brookings Papers on Economic Activity 1992 (1992), 1–75. [Google Scholar]
  20. Borjas GJ, “The Economic Analysis of Immigration,” Handbook of Labor Economics 3 (1999), 1697–1760. [Google Scholar]
  21. Caliendo L, Parro F, Rossi-Hansberg E and Sarte P-D, “The Impact of Regional and Sectoral Productivity Changes on the US Economy,” Review of Economic Studies (Forthcoming).
  22. Card D, “Immigrant Inflows, Native Outflows, and the Local Labor Market Impacts of Higher Immigration,” Journal of Labor Economics 19 (2001), 22–64. [Google Scholar]
  23. Carlino GA and Mills ES, “The Determinants of County Growth,” Journal of Regional Science 27 (1987), 39–54. [DOI] [PubMed] [Google Scholar]
  24. Chay KY and Greenstone M, “Does Air Quality Matter? Evidence from the Housing Market,” Journal of Political Economy 113 (2005), 376–424. [Google Scholar]
  25. Ciccone A and Hall R, “Productivity and the Density of Economic Activity,” American Economic Review 86 (1996), 54–70. [Google Scholar]
  26. Combes P-P, Duranton G, Gobillon L, Puga D and Roux S, “The Productivity Advantages of Large Cities: Distinguishing Agglomeration From Firm Selection,” Econometrica 80 (2012), 2543–2594. [Google Scholar]
  27. Davis DR and Dingel JI, “A Spatial Knowledge Economy,” NBER Working Paper No 18188 (2012).
  28. Desmet K and Rossi-Hansberg E, “Urban Accounting and Welfare,” American Economic Review 103 (2013), 2296–2327. [Google Scholar]
  29. Diamond R, “The Determinants and Welfare Implications of US Workers’ Diverging Location Choices by Skill: 1980–2000,” American Economic Review 106 (2016), 479–524. [Google Scholar]
  30. Epple D, Gordon B and Sieg H, “A New Approach to Estimating the Production Function for Housing,” American Economic Review 100 (2010), 905–924. [Google Scholar]
  31. Fajgelbaum PD, Morales E, Serrato JCSS and Zidar OM, “State Taxes and Spatial Misallocation,” NBER Working Paper No 21760 (2015).
  32. Gabriel SA and Rosenthal SS, “Quality of the Business Environment Versus Quality of Life: Do Firms and Households Like the Same Cities?” Review of Economics and Statistics 86 (2004), 438–444. [Google Scholar]
  33. Glaeser EL, “Learning in Cities,” Journal of Urban Economics 46 (1999), 254–277. [Google Scholar]
  34. Glaeser EL and Gottlieb JD, “The Economics of Place-Making Policies,” Brookings Papers on Economic Activity 39 (2008), 155–253. [Google Scholar]
  35. Glaeser EL and Gottlieb JD, “The Wealth of Cities: Agglomeration Economies and Spatial Equilibrium in the United States,” Journal of Economic Literature 47 (2009), 983–1028. [Google Scholar]
  36. Glaeser EL, Gyourko J and Saks RE, “Urban Growth and Housing Supply,” Journal of Economic Geography 6 (2005), 71–89. [Google Scholar]
  37. Glaeser EL and Tobio K, “The Rise of the Sunbelt,” Southern Economic Journal 74 (2008), 609–643. [Google Scholar]
  38. Goodman AC, “An Econometric Model of Housing Price, Permanent Income, Tenure Choice, and Housing Demand,” Journal of Urban Economics 23 (1988), 327–353. [Google Scholar]
  39. Green RK, Malpezzi S and Mayo SK, “Metropolitan-Specific Estimates of the Price Elasticity of Supply of Housing, and Their Sources,” American Economic Review 95 (2005), 334–339. [Google Scholar]
  40. Gyourko J, Saiz A and Summers A, “A New Measure of the Local Regulatory Environment for Housing Markets: The Wharton Residential Land Use Regulatory Index,” Urban Studies 45 (2008), 693–729. [Google Scholar]
  41. Harberger AC, “The Incidence of the Corporation Income Tax,” Journal of Political Economy 70 (1962), 215–240. [Google Scholar]
  42. Haughwout A, “Public Infrastructure Investments, Productivity and Welfare in Fixed Geographic Areas,” Journal of Public Economics 83 (2002), 405–428. [Google Scholar]
  43. Haughwout AF and Inman RP, “Fiscal Policies in Open Cities with Firms and Households,” Regional Science and Urban Economics 31 (2001), 147–180. [Google Scholar]
  44. Heckscher EF, The Effect of Foreign Trade on the Distribution of Income (Ekonomisk Tidskrift, 1919).
  45. Hoogstra GJ, Van Dijk J and Florax RJ, “Do Jobs Follow People or People Follow Jobs? A Meta-Analysis of Carlino–Mills Studies,” Spatial Economic Analysis (2017), 1–22. [Google Scholar]
  46. Hsieh C-T and Moretti E, “Housing Constraints and Spatial Misallocation,” American Economic Journal: Macroeconomics 11 (2019), 1–39. [Google Scholar]
  47. Ioannides YM and Zabel JE, “Neighbourhood Effects and Housing Demand,” Journal of Applied Econometrics 18 (2003), 563–584. [Google Scholar]
  48. Jaffe AB, Trajtenberg M and Henderson R, “Geographic Localization of Knowledge Spillovers as Evidenced by Patent Citations,” Quarterly Journal of Economics 108 (1993), 577–598. [Google Scholar]
  49. Jones RW, “The Structure of Simple General Equilibrium Models,” Journal of Political Economy 73 (1965), 557–572. [Google Scholar]
  50. Kline P and Moretti E, “People, Places, and Public Policy: Some Simple Welfare Economics of Local Economic Development Programs,” Annual Review of Economics 6 (2014), 629–662. [Google Scholar]
  51. Lee S and Li Q, “Uneven Landscapes and City Size Distributions,” Journal of Urban Economics 78 (2013), 19–29. [Google Scholar]
  52. Lucas RE, “Labor-Capital Substitution in U.S. Manufacturing,” in Harberger A and Bailey MJ, eds., The Taxation of Income from Capital (Brookings Institution, 1969), 223–274. [Google Scholar]
  53. McDonald J, “Capital-Land Substitution in Urban Housing: A Survey of Empirical Estimates,” Journal of Urban Economics 9 (1981), 190–211. [Google Scholar]
  54. Notowidigdo MJ, “The Incidence of Local Labor Demand Shocks,” Journal of Labor Economics (Forthcoming).
  55. Ohlin BG, The Theory of Trade (Handelns Teory, 1924).
  56. Rappaport J, “Moving to Nice Weather,” Regional Science and Urban Economics 37 (2007), 375–398. [Google Scholar]
  57. Rappaport J, “Consumption Amenities and City Population Density,” Regional Science and Urban Economics 38 (2008a), 533–552. [Google Scholar]
  58. Rappaport J, “A Productivity Model of City Crowdedness,” Journal of Urban Economics 63 (2008b), 715–722. [Google Scholar]
  59. Roback J, “Wages, Rents, and Amenities: Differences among Workers and Regions,” Economic Inquiry 26 (1988), 23–41. [Google Scholar]
  60. Rosen HS, “Housing Subsidies: Effects on Housing Decisions, Efficiency, and Equity,” in Auerbach AJ and Feldstein M, eds., Handbook of Public Economics (Elsevier, 1985), volume 1 of Handbook of Public Economics, chapter 7, 375–420. [Google Scholar]
  61. Rosen S, “Wage-Based Indexes of Urban Quality of Life,” Current Issues in Urban Economics 3 (1979). [Google Scholar]
  62. Rosenthal SS and Strange WC, “Evidence on the Nature and Sources of Agglomeration Economies,” in Henderson JV and Thisse J-F, eds., Cities and Geography (Elsevier, 2004), volume 4 of Handbook of Regional and Urban Economics, 2119–2171. [Google Scholar]
  63. Ruggles S, Sobek M, Alexander T, Fitch CA, Goeken R, Hall PK, King M, and Ronnander C, “Integrated Public Use Microdata Series: Version 3.0,” Minneapolis: Minnesota Population Center (2004).
  64. Saiz A, “The Geographic Determinants of Housing Supply,” Quarterly Journal of Economics 125 (2010), 1253–1296. [Google Scholar]
  65. Shapiro JM, “Smart Cities: Quality of Life, Productivity, and the Growth Effects of Human Capital,” The Review of Economics and Statistics 88 (2006), 324–335. [Google Scholar]
  66. Stiglitz JE, “A Two-Sector Two Class Model of Economic Growth,” Review of Economic Studies 34 (1967), 227–238. [Google Scholar]
  67. Su C-L and Judd KL, “Constrained Optimization Approaches to Estimation of Structural Models,” Econometrica 80 (2012), 2213–2230. [Google Scholar]
  68. Suárez Serrato JC and Zidar O, “Who Benefits from State Corporate Tax Cuts? A Local Labor Markets Approach with Heterogeneous Firms,” American Economic Review 106 (2016), 2582–2624. [Google Scholar]
  69. Uzawa H, “On a Two-Sector Model of Economic Growth,” Review of Economic Studies 29 (1961), 40–47. [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1

RESOURCES