Urban Growth and Its Aggregate Implications

Summary

"Urban Growth and Its Aggregate Implications" is a paper by Gilles Duranton (Wharton School, University of Pennsylvania) and Diego Puga (CEMFI, Madrid), published in Econometrica, vol. 91, no. 6 (November 2023), pp. 2219–2259 (DOI: 10.3982/ECTA17936). It circulated earlier as NBER Working Paper No. 26591, issued December 2019 [VERIFY: BACKLOG.md's internal note dates the working paper to 2020; NBER's own listing and IDEAS/RePEc both give the issue date as December 2019 — this page follows the NBER date]. Both authors are among the most cited urban economists working today, with no Georgist affiliation, placing this alongside the Mirrlees Review and Hsieh & Moretti (2019) as mainstream, methodologically rigorous work that independently arrives at conclusions relevant to the Georgist case. Duranton and Puga build a dynamic, quantifiable model of a whole system of cities — not a single city — in which incumbent residents deliberately restrict city growth through planning regulations, and they use the calibrated model to estimate the aggregate output and growth costs of that restriction.

The Core Argument and Model

The paper starts from what Fujita and Thisse call the "fundamental tradeoff" of urban economics: bigger cities generate more output through agglomeration economies (human-capital spillovers that foster entrepreneurship and learning), but they also impose greater urban costs — longer, more congested commutes and higher housing costs as cities expand outward. Duranton and Puga's central modeling choice is to resolve this tradeoff through an explicit political-economy mechanism: incumbent residents in each city set planning regulations (a "housing permitting cost," what Glaeser, Gyourko and Saks call a "regulatory tax") to cap population growth at the level that maximizes their own welfare — not the level that would maximize aggregate output. This is deliberately inefficient by construction: incumbents limit entry to avoid seeing their productivity gains "dissipated" into higher commuting and housing costs, at the expense of potential newcomers who remain stuck in less productive locations.

The model is calibrated with a small number of parameters estimated from US microdata: an urban-cost elasticity with respect to city population of about 0.07 (rising to about 0.11 once congestion effects, elasticity ~0.04, are included), and an agglomeration elasticity of earnings with respect to city population of about 0.04 in the short run and about 0.08 in the long run once learning effects are incorporated. The authors report that three independent estimation strategies for urban costs — using commuting-cost data, house-price data, and a cross-city aggregate approach — converge on similar estimates, which they treat as validating the model's microfoundations.

Quantitative counterfactuals

Duranton and Puga use the calibrated model to run several counterfactual policy experiments:

1. Relaxing planning regulations in seven large, highly-restricted US cities (cities with population above three million and a wedge between periphery house prices and replacement cost exceeding $200,000), allowing permitting to rise to the 75th percentile level observed across all US cities over a 30-year period. This implies population increases of up to 38% in New York and 21% on average across the seven cities (roughly 18 million additional inhabitants combined). The authors find this raises aggregate output by 7.95% and aggregate consumption by 2.16%, with newcomers and rural residents seeing consumption gains of 6.55% and average consumption inequality between incumbents and newcomers narrowing substantially (from 63.6% to 67.7% of incumbent consumption). Sensitivity analysis across a plausible parameter range puts the output effect between roughly 5.7% and 8.2%, and the consumption effect between roughly 2.1% and 2.2%.
2. Growth-rate counterfactuals. Separately, the paper estimates that the ongoing spatial reallocation of a fixed population toward more productive cities — as underlying productivity, human capital, and transport costs evolve — raises the annual aggregate output growth rate by about 0.7 percentage points, and that accounting for the fact that aggregate US population is also growing (so urbanisation itself must proceed) "nearly doubles" this contribution.
3. Freezing city growth entirely (1950 populations held fixed). If no US city had been allowed to grow in population after 1950, the authors estimate annual output-per-person growth over 1950–2010 would have fallen from the actual 2.1% to about 0.9% — a 1.2 percentage-point annual gap that compounds to a 26.8% consumption loss over 60 years.

Direct Comparison to Hsieh & Moretti (2019)

The paper explicitly engages with Hsieh & Moretti's (2019) "Housing Constraints and Spatial Misallocation", which this wiki already treats as Core-tier evidence for the same outcome. Duranton and Puga note that Hsieh and Moretti's static model — which finds an 8.9% aggregate US output gain if three of the most productive cities raised their housing-supply elasticity to the median city's level — is "similarly motivated" but mechanically different: in Hsieh and Moretti's framework, an increase in city population is always detrimental to existing residents (the welfare-optimal city size for an incumbent is zero, so planning regulations are modeled as exogenous), whereas in Duranton and Puga's framework the tradeoff between agglomeration benefits and urban costs is explicit, planning regulations are endogenous, and relaxing regulation in a few cities causes regulation to loosen elsewhere too (as pressure on other housing markets eases), which is itself an added source of aggregate gains not present in the Hsieh-Moretti mechanism. That two independently derived, differently specified models — one static and treating regulation as exogenous, one dynamic and endogenizing it — converge on similar-magnitude aggregate output effects (7.95% vs. 8.9%) from relaxing land-use restriction in a handful of major cities is a meaningful robustness signal, though the two papers are not measuring identical counterfactuals and should not be read as replications of one another.

The paper also situates itself relative to Davis, Fisher & Whited (2014), who find that housing/infrastructure investment and agglomeration feedbacks together boost aggregate growth by "about 10%" in a neoclassical growth model, and to Desmet & Rossi-Hansberg (2013), a static framework finding that reducing cross-city differences in productivity, amenities, or frictions has large effects on city population but only minor welfare effects — a useful counterpoint suggesting the literature does not uniformly find large output effects from every reallocation channel.

Relation to the Georgist Case

This paper's subject is land-use/planning regulation, not land value taxation, and the authors do not propose or model an LVT. Its relevance to the wiki's land rents suppress productivity outcome is therefore the same kind of indirect but substantively strong relevance already established for Hsieh & Moretti: the paper's mechanism is that restricting the supply of space in high-productivity locations funnels the resulting scarcity value into higher land/housing costs that price out the workers and firms who would gain the most from being there, misallocating population and depressing aggregate output relative to a less-restricted counterfactual. That is precisely the land-scarcity-rent channel this wiki's outcome page describes, now modeled with endogenous regulation and validated against a second, independently specified paper (Hsieh-Moretti) with a similar-magnitude result. It complements Bakker (2023), which measures the aggregate-TFP cost of privately captured urban land rents directly, and Fiorentino & Moogan (2025), which models how taxing land value specifically (rather than deregulating land use) could ease the same misallocation.

Honest limits on this connection: Duranton and Puga's policy lever is relaxing planning regulation (allowing more construction), not taxing land value. The paper offers no evidence on LVT incidence, capitalization, administrability, or on whether a land value tax would by itself induce incumbent residents to relax the regulations the model identifies as the actual binding constraint — indeed, since the model treats planning regulation as a political choice by incumbents to protect their own welfare, it raises the (unaddressed by this paper) question of whether LVT would need to be paired with land-use liberalization to have the aggregate effects estimated here, echoing the "capture without supply reform doesn't lower prices" caution already documented at Land capture didn't make housing cheap.

Nuances and Limits

• Land-use deregulation, not land value taxation. As above: this is evidence that restricting the quantity of urban land available for development is aggregate-output-costly, not evidence for any particular tax policy.
• Model-dependent estimates. Like Hsieh & Moretti, the headline 7.95%/2.16% figures come from a calibrated structural model with specific functional-form and parameter choices, not a natural experiment. The authors report the output effect is sensitive to parameter choice (ranging roughly 5.7%–8.2% across plausible alternatives) though the consumption effect is more stable (roughly 2.1%–2.2%).
• Two acknowledged modeling limitations. The authors note their overlapping-generations setup does not have migrants internalizing benefits to future generations (though they argue this would, if anything, strengthen the link between city size and regulation strictness they identify), and — more significantly by their own account — the model does not incorporate the durability of housing structures, meaning it cannot capture the asymmetry between city growth and decline (cities build in response to positive shocks but do not "un-build" in response to negative ones) documented by Glaeser & Gyourko (2005).
• US-specific calibration. All parameter estimates and counterfactuals use US data (1950–2010 aggregates; US metro microdata for parameters); the authors explicitly flag extending the framework to other countries, including developing countries with different urbanisation-growth relationships, as future work.
• Innovation is exogenous. The model treats the evolution of local total-factor productivity as an exogenous stochastic process; it does not model cities as endogenously generating innovation, which the authors note a richer model might incorporate.
• Convergent but not identical estimates with Hsieh-Moretti. The similarity between this paper's 7.95% and Hsieh & Moretti's 8.9% is worth treating as suggestive corroboration of the general magnitude of housing/land-use-restriction costs, not as two measurements of the same quantity — the counterfactuals, city samples, and mechanisms differ (see above), and (per this wiki's Hsieh-Moretti page) the Hsieh-Moretti figure itself has since become the subject of an active, unresolved post-publication replication dispute (Greaney, 2026) that this Duranton-Puga paper predates and does not address.

Bears On

• Outcome: High land rents suppress productivity — supplies a second, independently derived quantitative model (endogenous planning regulation rather than exogenous housing-supply elasticity) reaching a similar-magnitude estimate of the aggregate output cost of restricting growth in high-productivity cities, strengthening the case that this is a real and non-trivial effect rather than an artifact of one paper's modeling choices.
• Research: Hsieh & Moretti (2019), Housing Constraints and Spatial Misallocation — directly engaged and compared in this paper; the two papers' convergent 7.95%/8.9% estimates are mutually reinforcing but derived from different mechanisms and should be cited together with that caveat.
• Research: Bakker (2023), Urban Land Rents and TFP — a parallel measurement of the aggregate-TFP cost of privately captured land rent; this paper supplies a structural, regulation-based account of the same broad phenomenon.
• Research: Fiorentino & Moogan (2025), LVT and Urban Agglomeration Dynamics — models the tax-based policy response (LVT) to the same land-scarcity/agglomeration tradeoff that Duranton-Puga model via land-use deregulation; read together they suggest LVT and land-use liberalization may be complementary rather than substitute levers.
• Objection: Land capture didn't make housing cheap (Singapore/Hong Kong) — this paper's finding that incumbents use planning regulation to protect their own welfare at newcomers' expense helps explain why capturing land value without addressing land-use restriction might not lower prices: the regulatory constraint, not tax design, is this paper's binding friction.
• Concept: Deadweight Loss — the paper frames incumbent-imposed planning regulation explicitly as "a source of deadweight loss for society."

See Also

• Hsieh & Moretti (2019): Housing Constraints and Spatial Misallocation
• Bakker (2023): Urban Land Rents and TFP
• Fiorentino & Moogan (2025): LVT and Urban Agglomeration Dynamics
• High land rents suppress productivity
• Deadweight Loss

Sources

1. Gilles Duranton & Diego Puga (2023), "Urban Growth and Its Aggregate Implications," Econometrica, 91(6): 2219–2259. DOI: 10.3982/ECTA17936 · Econometric Society (open PDF) · author copy, Diego Puga — used for the abstract, model description, all quantitative counterfactual figures (7.95%/2.16%/0.7pp/26.8%), parameter estimates, and the direct comparison to Hsieh & Moretti (2019). Directly fetched and read in full (final version, 30 June 2023) in this session.
2. Gilles Duranton & Diego Puga (2019), "Urban Growth and its Aggregate Implications," NBER Working Paper No. 26591, issued December 2019. NBER — used for confirming the working-paper number, issue date, and abstract, and for the earlier circulation history ahead of the 2023 Econometrica publication.
3. Chang-Tai Hsieh & Enrico Moretti (2019), "Housing Constraints and Spatial Misallocation," American Economic Journal: Macroeconomics, 11(2): 1–39. Wiki summary — used for the comparison of the two papers' mechanisms and magnitudes, cross-checked against Duranton & Puga's own discussion of Hsieh-Moretti in section 8 of their paper.

[VERIFY: the exact issue date of NBER WP 26591 — this page follows NBER's and IDEAS/RePEc's own listing of December 2019; BACKLOG.md's internal task note says "2020," which may reflect a revised-version date or simply be imprecise. The published Econometrica version (2023) is confirmed with high confidence via the DOI, the Econometric Society's own publication page, and the final-version PDF's dateline of 30 June 2023.]