Metropolitan Land Values

Summary

"Metropolitan Land Values" is a peer-reviewed article by David Albouy (University of Illinois at Urbana-Champaign and NBER), Gabriel Ehrlich (University of Michigan), and Minchul Shin (University of Illinois), published in The Review of Economics and Statistics, vol. 100, no. 3 (July 2018), pp. 454–466 (DOI: 10.1162/rest_a_00710).[1] The Review of Economics and Statistics is a long-established, peer-refereed MIT Press journal in mainstream empirical economics, and Albouy is a well-published urban/public economist whose related work on urban land rents and amenity values ("What Are Cities Worth?") already appears elsewhere on this wiki. The paper's contribution is explicitly empirical and methodological, not advocacy: it builds the first cross-sectional index of directly observed, transaction-based land values covering nearly every metropolitan statistical area (MSA) in the United States, using a large proprietary database of actual land sales rather than the "residual" method (total property value minus estimated structure value) that most prior national land-value estimates had relied on.[1] The paper is directly relevant to the Georgist case because it supplies an independently derived, non-advocacy estimate of the scale and distribution of US land value — precisely the empirical quantity Georgist revenue-sufficiency arguments depend on.

The Core Argument and Findings

The authors' central problem is that market data on land value (as opposed to combined land-plus-structure property value) have historically been "notoriously piecemeal," to the point that the Federal Reserve's Flow of Funds accounts stopped publishing a land-value series in 1995 after the residual method produced implausible negative land values in some periods.[1] To overcome this, Albouy, Ehrlich, and Shin use the CoStar COMPS database, a large commercial real-estate dataset of roughly 67,000 actual land-parcel transactions across 324 MSAs and consolidated metropolitan statistical areas (CMSAs) from 2005–2010, and combine these transaction prices with an econometric model grounded in the classic monocentric city model of urban land rent (Alonso 1964, Mills 1967, Muth 1969).[1] Because observed land sales are geographically nonrandom (they are concentrated near city centers and along transportation corridors, and many MSAs have very few recorded sales in a given year), the authors develop a hierarchical Bayesian ("empirical Bayes") shrinkage estimator: each city's estimated land-value gradient (how value declines with distance from downtown) is pulled toward a theory-informed prior that itself varies systematically with a city's urbanized area and coastal proximity, with the degree of shrinkage automatically larger for MSAs with fewer observations.[1] A cross-validation exercise (holding out subsets of observations and testing predictive accuracy against a naive average-price model) confirms the shrinkage specification substantially reduces prediction error relative to unshrunken, MSA-by-MSA estimates.[1]

Headline empirical findings, reported for the 2005–2010 sample period:

• Land values rise with city size, and rise faster at the center. A 10% larger urban footprint is associated with an 8% higher central land value; central land-value gradients (the ratio of central to peripheral value) are also steeper in larger cities, though highly variable — some very large cities have relatively flat gradients.[1] The highest central land values are found in New York, Chicago, Washington DC, San Francisco, and Los Angeles, where land near the center is on the order of 21 times more valuable than land ten miles away (versus an unweighted average ratio across all metros of only about 4).[1]
• Extreme concentration in a handful of coastal metros. New York has by far the highest estimated central land value, at roughly $123 million per acre; the next four highest-central-value metros are Chicago, Washington DC, San Francisco, and Los Angeles–Long Beach, in the $17–38 million-per-acre range.[1] Aggregated to the CMSA level, the authors find that five urban agglomerations — New York, Los Angeles, San Francisco, Washington DC, and Chicago — together account for 48% of the total value of all urban land in the United States, out of 324 metro areas covering 76,581 square miles of urban land nationally.[1] (By total value rather than central value, the ranking differs slightly: the New York PMSA alone is estimated at roughly $2.5 trillion, with Los Angeles–Long Beach close behind at roughly $2.3 trillion, even though Los Angeles's urban area is more than 1.8 times larger in square miles.)[1]
• Aggregate US urban land value is large relative to GDP, and fell sharply after 2006. The authors' national total climbs from $28.1 trillion in 2005 to a peak of $30.4 trillion (about 2.2 times nominal GDP) in 2006, then falls to $19.1 trillion (1.28 times GDP) by 2010 — a peak-to-trough decline of about 40%, smaller than the roughly 66% decline implied by the Flow-of-Funds/residual series over the same period.[1] The average value of urban land per acre likewise peaked in 2006 at $624,000 (an 8% rise over 2005) before falling to $373,000 by 2009, about 65% of its 2005 level.[1]
• The transaction-based index is systematically higher and more stable than residual estimates. Compared with the widely used residual-method estimates of Davis and Palumbo (2008) for 45 metros, the authors' transaction-based average value is $722,000 per acre versus $392,000 for the residual method, and the transaction-based index shows markedly less within-city volatility (an average coefficient of variation of 0.24 versus 0.44).[1] The authors argue the residual method mechanically misattributes costs from land-use regulation and construction inefficiency to land value, and can produce implausible results (they note a contemporaneous Flow-of-Funds-based estimate that implied negative $178 billion of corporate land value in 2009).[1]

Relation to the Georgist Case

This paper's relevance to Georgism is evidentiary rather than argumentative: the authors do not discuss land value taxation, Henry George, or fiscal policy anywhere in the paper except in a one-line historical framing note that "urban land values have been central to questions of wealth, income, and taxation since the seminal works of Ricardo (1821) and George (1884)," which they cite only to motivate why land values matter as an object of study — not to argue for any tax policy.[1] What the paper directly supports is the empirical premise behind the land rent could fund government outcome: independent, market-transaction-based evidence that aggregate US urban land value is very large in absolute terms and relative to GDP (over twice GDP at the 2006 peak), giving a mainstream, non-Georgist anchor figure for arguments about how much revenue a land value tax base could plausibly support. It also complicates any simple, uniform-rate vision of a national LVT: the extreme concentration of value in a handful of coastal metro areas means the practical revenue capacity and administrative stakes of land taxation are enormously uneven across the country, with most US urban land — by area, not value — worth relatively little per acre.

Nuances and Limits

• Coverage is urban land only, and structures/agricultural land are excluded. The $19–30 trillion aggregate figures cover only the roughly 76,581 square miles the authors classify as urban across 324 metro areas; they explicitly do not include rural, agricultural, or non-metropolitan land, nor do they include the value of structures. The paper's own back-of-envelope comparison suggests that if 40% of total metro-area land is publicly owned (roads, parks, civic buildings), only about $18.2 trillion of the total would be privately held — a reminder that the headline aggregate is not directly a "taxable private land value" figure without further adjustment.[1]
• The sample period (2005–2010) spans the peak and crash of the mid-2000s housing bubble. The dramatic decline in aggregate land value the paper documents (a roughly 40% peak-to-trough fall) is specific to this boom-bust window and should not be read as a stable, general estimate of "US land value" for other periods; the authors themselves present it as a case study in how their index behaves relative to residual measures during a volatile episode, not as a steady-state figure.
• The shrinkage methodology assumes a specific theoretical prior (the monocentric city model). For MSAs with very few or no recorded transactions, estimated values lean heavily on this theory-based prior rather than direct observation; the authors are transparent that "meta-city" predictions for data-thin metros are model-derived rather than directly measured, and that the shrinkage estimator, while reducing mean-squared error in cross-validation, may understate the true uncertainty in these estimates because it does not fully account for uncertainty in the estimated hyperparameters.[1]
• CoStar COMPS is a commercial real-estate database, not a full census of land sales. It disproportionately captures larger, more commercial-oriented transactions and excludes extreme outliers (parcels priced under $100/acre or over 5,000 acres, or more than 60 miles from a city center); the authors' own median lot size in the sample is 3.5 acres, which is not necessarily representative of small residential lots.[1]
• This is a measurement paper, not a study of assessment or tax administration. It says nothing about whether governments could feasibly assess land value at this granularity for tax purposes using existing administrative capacity — a separate and contested question addressed on this wiki's land cannot be assessed objection page.

Bears On

• Outcome: Land rent could fund a large share of government — supplies an independent, transaction-based estimate that aggregate US urban land value exceeded twice GDP at its 2006 peak, a mainstream anchor figure for arguments about the scale of a potential land-value tax base (though the paper covers urban land only, not the larger total-US-land estimates found in other sources on this page).
• Concept: Land Value Tax — establishes, via a rigorous non-Georgist methodology, that the land-value base such a tax would target is large and highly unevenly distributed across metro areas.
• Objection: Land value can't be assessed accurately — the paper's own shrinkage/imputation approach for data-thin metros is a sophisticated illustration of how much modeling is required to estimate land value where direct transaction data is sparse, a consideration relevant to (though not decisive on) that objection.
• Concept: FIRE Sector — the paper's finding that land value is overwhelmingly concentrated in a handful of finance- and government-heavy coastal metros (New York, Washington DC, San Francisco) is consistent with accounts linking land-rent concentration to the growth of finance, insurance, and real estate.

See Also

• Land Value Tax
• Objection: Land value can't be assessed accurately
• What Are Cities Worth?
• FIRE Sector
• New Estimates of Value of Land of the United States

Sources

1. David Albouy, Gabriel Ehrlich & Minchul Shin, "Metropolitan Land Values," The Review of Economics and Statistics, vol. 100, no. 3 (July 2018), pp. 454–466. DOI: 10.1162/rest_a_00710; working-paper draft (October 25, 2017) freely available via David Albouy's site — used throughout for the abstract, methodology (CoStar COMPS data, monocentric-city prior, empirical Bayes shrinkage estimator), all headline findings (central land values, the 48% concentration figure, aggregate land value and its ratio to GDP by year), and the comparison to the Davis-Palumbo residual method.
2. William Larson, "New Estimates of Value of Land of the United States," BEA Working Paper WP2015-3, 2015 — cited within Albouy, Ehrlich & Shin as a comparable official land-value estimate; see this wiki's dedicated page for Larson's BEA-based estimate of roughly $23 trillion in total US land value (2009), a broader, all-land figure that differs in scope and method from this paper's urban-only, transaction-based estimate.
3. Morris A. Davis & Michael G. Palumbo, "The Price of Residential Land in Large US Cities," Journal of Urban Economics, 63(2), 2008, pp. 352–384 — used for the residual-method comparison estimates cited in this page's Core Findings and Nuances sections.