LE - The Value of Residential Land and Structures During the Great Housing Boom and Bust

8/20/2019 LE - The Value of Residential Land and Structures During the Great Housing Boom and Bust

1/30

Nicolai V Kuminoff, Jaren C Pope

Land Economics, Volume 89, Number 1, February 2013, pp. 1-29 (Article)

DOI: 10.1353/lde.2013.0003

For additional information about this article

Access provided by York University (14 Nov 2014 12:02 GMT)

http://muse.jhu.edu/journals/lde/summary/v089/89.1.kuminoff.html

http://muse.jhu.edu/journals/lde/summary/v089/89.1.kuminoff.htmlhttp://muse.jhu.edu/journals/lde/summary/v089/89.1.kuminoff.html


2/30

The Value of Residential Land and Structures duringthe Great Housing Boom and Bust

Nicolai V. Kuminoff and Jaren C. Pope

ABSTRACT. This study examines how the value of residential land and structures evolved during thegreat housing boom and bust, using data on morethan a million residential properties that were sold in10 metropolitan areas between 1998 and 2009. Weuse a hedonic estimator to disentangle the market value of land and structures at a local (Census tract)level. Our estimates reveal substantial heterogeneityin the evolution of the market value of land and struc-tures within metropolitan areas. Surprisingly, lower-value land at the urban fringes of metropolitan areaswas the most volatile during the boom-bust. (JELR14, R21)

I. INTRODUCTION

Housing is a major source of wealth in theUnited States. The Federal Reserve’s Flow of Funds Report documents that the asset valueof owner-occupied housing for the entireUnited States was approximately $23 trillionin 2006—more than the capitalized value of the NYSE, Amex, and Nasdaq exchangescombined. A house’s value can be decom-posed into two components: the value of theland on which the house is built, and thevalue of the structures that comprise thehouse itself. Decomposing property valueinto the value of land and structures is im-portant for several reasons. First, some citiesand counties tax land and structures at dif-ferent rates (Chapman, Johnston, and Tyrrell2009; Banzhaf and Lavery 2010; Cho, Lam-bert, and Roberts 2010). Successful imple-mentation of a split-rate tax requires accurateestimates for each component of value. Sec-ond, structures depreciate differently fromland. Documenting this difference is neces-sary for calculating tax code allowances for

Land Economics • February 2013 • 89 (1): 1–29

ISSN 0023-7639; E-ISSN 1543-83252013 by the Board of Regents of theUniversity of Wisconsin System

depreciation and for insurance companies toreimburse homeowners for damaged struc-tures. Third, understanding how the value of land has evolved relative to the value of structures may help households, banks, andlocal governments to manage risk withintheir nancial portfolios. Finally, tracking theevolution of land and structural values withinand across metropolitan areas may provideinsights into the forces that drive boom-bustcycles in real estate.

There are three primary techniques for de-composing property values into the value of land and structures. The “teardown” ap-proach derives land value from the salesprices (plus demolition costs) of propertiesthat were purchased with the intention to teardown the existing structures (e.g., Rosenthaland Helsley 1994; Dye and McMillen 2007).The “replacement cost” approach infers landvalue from the difference between propertyvalues and the depreciated costs of replacingthe structures on the property (e.g., Davis andHeathcoat 2007; Davis and Palumbo 2008).Finally, the “hedonic” approach estimatesland value by regressing the sales prices of properties on the characteristics of the landand structures. In this case, the land value iscommonly dened as the marginal implicitprice per square foot of a lot (e.g., Clapp

1980; Glaeser, Gyourko, and Saks 2005).Bell, Bowman, and German (2009) providean overview of the three methods, describingtheir strengths and weaknesses. The difcultywith the teardown approach is the sparsenessof the data. There are too few teardowns toapply the methodology at a high level of spa-tial resolution in large geographic areas. Bell,Bowman, and German (2009, 177) describe

The authors are, respectively, assistant professor, De-

partment of Economics, Arizona State University,Tempe; and assistant professor, Department of Eco-nomics, Brigham Young University, Provo, Utah.


3/30

February 2013 Land Economics2

FIGURE 1Standard and Poor’s Case-Shiller National Housing

Price Index

a more nuanced set of trade-offs between thereplacement cost and hedonic approaches butconclude that “the contribution [hedonic]principle of value seems more consistent with

the notion of market value.”1

Despite thisconceptual advantage, the replacement costapproach is more commonly used. This ispartly due to its apparent simplicity and rela-tively low data requirements. However, italso reects concerns that omitted variablesmay confound hedonic estimates of landvalue. 2 If the houses located in neighbor-hoods with higher land values also tend tohave nicer physical characteristics that arenot readily observed by the analyst (e.g.,granite countertops, hardwood oors) thenthe value of land will be confounded with thevalue of unobserved physical characteristics.

The objective of this paper is twofold: (1)to rene the hedonic approach to estimatingland value in a way that mitigates omittedvariable bias, and (2) to use our rened meth-odology to analyze how the market value of land and structures evolved within and acrossmajor metropolitan areas during the 2000s.Our data describe the sale prices, physical at-tributes, and geographic locations of morethan a million houses that were sold in 12 met-ropolitan areas between 1998 and 2009: Bos-ton, Charlotte, Cincinnati, Detroit, LosAngeles, Miami, Oakland, Philadelphia, Pitts-burg, San Francisco, San Jose, and Tampa.According to the Standard & Poor’s Case-Shiller repeat sales index, residential propertyvalues in major metropolitan areas more thandoubled between 1998 and 2006 and then de-clined by approximately 40% between 2006and the end of 2009 (Figure 1). 3 Our data span

this remarkable boom-bust cycle.

1 In the literature on land valuation, the hedonic ap-proach is also referred to as the “contribution” approach.

2 Indeed, some prior hedonic estimates for land valueseem strikingly low. For example, the estimates by Glaeser,Gyourko, and Saks (2005) suggest that the value of a fullacre of land in Boston in 1998 was less than $30,000 and inSan Francisco it was less than $200,000.

3 This gure shows that housing prices more than dou-bled from 1998 to 2006 but then declined substantially from2006 to 2009 for 20 metropolitan areas included in the index.

The data for this gure comes from the the Standard andPoor’s Case-Shiller U.S. National Values Home Price Index.For documentation, see www.standardandpoors.com.

The location of each house conveys accessto a specic bundle of local public goods andalso denes the commuting opportunities thatwould be faced by a working household.These localized amenities may be in limitedsupply due to zoning regulations and otherforms of development restrictions (Glaeser,Gyourko, and Saks 2005). As a result, it isimportant to recognize that land values mayvary across neighborhoods within a metro-politan area. Equally important is the need torecognize that the market values of land andstructures may evolve differently over time.The relative price of land may increase overtime as developable land becomes scarcer.Likewise, changes in credit constraints orwealth may alter the relative demands for thepublic and private attributes of housing inways that differ across time and space.

To characterize the spatiotemporal varia-tion in the value of land and structures, we

estimate housing price functions for each met-ropolitan area, while allowing the shape of theprice function to change from year to year.Our estimator extends the conventional he-donic approach in two ways. First, we usexed effects for U.S. Census tracts to capturespatial variation in localized amenities thatcontribute to land value through a parcel’s lo-cation, rather than its size. Second, we addinteractions between the xed effects andsquare footage of living space to capture spa-

tial variation in latent attributes of structures.We then generate estimates for annual averagevalues of land and structures at the level of an


4/30

89(1) Kuminoff and Pope: Value of Residential Land and Structures 3

individual Census tract. Our spatially explicitestimates are typically an order of magnitudelarger than estimates based on the conven-tional hedonic approach.

When we aggregate our tract-level esti-mates up to the level of a metropolitan area,they are generally consistent with the metro-politan area averages reported prior to theboom by past studies using the replacementcost method (e.g., Davis and Palumbo 2008).The same is true after the bust. However, thetwo sets of estimates diverge during theboom-bust period. Our estimates for land val-ues do not rise as fast during the boom or fallas quickly during the bust. Our estimates im-ply that the market value of structures ex-ceeded their replacement cost during theheight of the boom. The differences can belarge—up to 100% for San Francisco. One po-tential explanation is that local housing mar-kets are less than perfectly competitive. Witha small share of houses on the market at anyone time, the unique bundle of amenities pro-vided by a desirable neighborhood may allowhome sellers to command a markup on thestructural characteristics of their houses, asTaylor and Smith (2000) rst observed. In-deed, we nd that neighborhoods with higherpreboom land values (presumably the higher-amenity neighborhoods) had larger markupson structures during the boom. Over time, wewould expect these markups to stimulate newconstruction, following the general logic of Tobin’s q-theory.

Consistent with Davis and Palumbo’s(2008) analysis of the variation between met-ropolitan areas, we nd that land values aremore volatile in metropolitan areas where the

supply of housing is less elastic. Interestingly,we nd the opposite pattern within metropol-itan areas. Neighborhoods at the urban fringe,where we would expect the supply of housingto be most elastic, were the neighborhoodsthat experienced the most volatility in housingprices and land values during the boom andbust. This general pattern can be seen in theCase-Shiller index. Figure 2 4 displays indices

4 This gure shows the substantial heterogeneity in price

changes both across and within metropolitan areas. Thethree lines show the evolution of prices for a metropolitanarea’s bottom “tier,” middle “tier,” and top “tier”of the price

for the lowest, middle, and highest tier of houses (ranked by 2010 value) for Miami, SanFrancisco, Boston, and Atlanta. Within eachmetropolitan area, the lowest-value houses

were the most volatile and the highest-valuehouses were the least volatile. 5 We nd thatthe higher-value houses tend to be locatedcloser to the city, where the supply of land isleast elastic, and the lower-value houses tendto be located at the outskirts of the surround-ing suburbs, where most of the new housingis built. This suggests that factors other thansupply elasticity of housing are playing an im-portant role in the evolution of land and struc-tural values.

Overall, this research makes three contri-butions to the literature on land valuation.First, we use high-resolution spatial xed ef-fects to rene the conventional hedonic ap-proach to decomposing housing prices intothe implicit value of land and structures. Sec-ond, we use a consistent estimation strategyto provide new estimates for how land valuesevolved within and across several metropoli-tan areas during the remarkable boom-bustcycle of the 2000s. Finally, while prior studieshave reported that land values vary acrossspace and time, we document two novel fea-tures of this variation that deserve more atten-tion in future research: (1) the least valuableland at the urban fringes of metropolitan areaswas the most volatile during the recent boom-bust cycle; and (2) the market value of struc-tures exceeded construction costs during theboom, with the largest markups occurring inthe most afuent neighborhoods.

II. THE MARKET VALUE OF LAND ANDSTRUCTURES IN A METROPOLITAN

AREA

We begin from a standard description of residential sorting. Heterogeneous householdsare assumed to choose from a stock of houseswith different lot sizes and structural charac-

distribution. Breakpoints are dened by metropolitan area asof August 2010. The data for this gure also comes fromthe the Standard & Poor’s Case-Shiller Home Price Index.

*Supply elasticities are based on Saiz (2010).5 One can nd the same pattern in the other 16 majormetropolitan areas tracked by the Case-Shiller index.


5/30


FIGURE 2Hetergogeneity in the Evolution of Housing Prices across and within Metropolitan Areas

teristics (e.g., bedrooms, bathrooms, squarefeet). Their collective location choices will inturn inuence the supply of neighborhoodamenities (e.g., public school quality, com-mute time to the city center, preservation of open space) through a combination of voting,social interactions, and feedback effects. 6 For-mally, an individual household’s utility max-imization problem is

6 The new empirical literature on Tiebout sortingstresses the need to recognize that neighborhood amenitiesare typically endogenous to the collective location choicesmade by the households in a metropolitan area (Kuminoff,Smith, and Timmins 2010). For example, urban develop-ment may provide opportunities for dining and nightlife,while increasing trafc congestion and degrading air andwater quality. Homeowners may be asked to vote on as-sessments to fund open space preservation and publicschools. Academic performance among students in thoseschools may depend on the distribution of income and edu-

cation among parents in the school district. While we do notmodel these mechanisms, our framework is consistent withtheir presence.

max U (g ,l , x ,b;α ) subject to y = b + P . [1]i kt jk jkt it it jkt j,k

In period t , household selects one of i j =houses located in one of 1 , . . . , J k = 1, . . . ,K k

neighborhoods. Their utility depends on thelot size of the parcel ( ), the structural char-lacteristics of the house ( ), the amenities pro- x

vided by the neighborhood ( ), and the in-gcome the household has left over to spend onthe numeraire good ( ) after paying the an-bnualized after-tax price of housing ( ). TheP jk t household’s idiosyncratic preferences are rep-resented by .α it

Sellers in this market may include a mix of developers and individuals selling theirhouses. There is no need to be more specicabout the supply side of the market. Under apair of weak restrictions on consumer prefer-

ences, any market outcome consistent withutility maximizing behavior can be describedby a hedonic price function.


6/30


Assumption 1.

a. is strictly increasing inU (g ,l , x ,b ;α )i kt jk jkt it for all .b b∈(0, y )it

b. Let represent household i’s preference≥ iordering over all potential location choicesthat satisfy the household’s budget constraint.

is invariant to i’s actual location choice.≥ i

The rst condition is self-explanatory. Thesecond condition simply limits the scope forany one household to inuence prices or thesupply of neighborhood amenities. For ex-ample, suppose household i has exceptionallybright children. If i were to move from its cur-rent house in school district R to a new housein school district S , then school quality mayincrease marginally in S due to peer effects,and decrease marginally in R. These adjust-ments may be followed by changes in housingprices. Condition b implies that these changesmust be sufciently small to leave i’s prefer-ence ordering over the two houses un-changed. 7 The need for this restriction be-comes apparent in the proof of Theorem 1,which is simply a variation on results derivedby Bajari and Benkard (2005). 8

Theorem 1. Suppose that Assumption 1 holdsfor every household. Then for any two houses,

and , it must be true that if j,k r ,s P = P jk t rst , , and .g = g l = l x = x kt st jk rs jkt rst

Proof . Suppose i chooses given P jkt > P rst . j,k Then U i(gkt ,l jk , x jkt , yit − P jkt ;α it ) < U i(gst ,lrs ,

because utility is strictly x , y − P ;α )rst it rst it increasing in the numeraire. This preference

ordering is invariant to whether i locates inor . Therefore, cannot be a utility- j,k r ,s j,k

7 Theorem 1 can also be proven under an alternative as-sumption that households ignore their own contributions tothe supply of neighborhood amenities.

8 Our theorem recognizes that neighborhood amenitiesmay be determined endogenously through a Tiebout sortingprocess. In contrast, Bajari and Benkard (2005) characterizemarkets where product attributes (other than price) are de-termined exogenously. They also model unobserved productattributes and restrict utility to be Lipschitz continuous inorder to guarantee Lipschitz continuity of the price function.

While it is straightforward to add these elements to ourmodel, they are unnecessary to guarantee the existence of aprice function.

maximizing location for i in period t , whichis a contradiction. QED

Theorem 1 states that property values will

be functionally related to neighborhoodamenities, lot sizes, and structural housingcharacteristics during a single period. Relativeto the majority of the empirical literature thatmaintains the assumptions of Rosen’s (1974)hedonic model as a basis for measuring thewillingness to pay for urban amenities, The-orem 1 is notable for what it does not assume.We do not require the market to be perfectlycompetitive. Nor do we require households tobe free to choose continuous quantities of every housing characteristic in every neigh-borhood. Some characteristics may be ap-proximately continuous; others may be dis-crete. For example, households may be free tochoose square footage continuously oversome range, whereas public school qualitymay change discretely as one crosses the bor-der between two adjacent school attendancezones (Black 1999).

Relaxing continuity and perfect competi-tion means that we lose the ability to interpretequilibrium marginal implicit prices for char-acteristics as measures that identify consum-ers’ marginal willingness to pay for thosecharacteristics or rms’ marginal costs (Feen-stra 1995; Taylor and Smith 2000; Kuminoff,Smith, and Timmins 2010). If discreteness inthe choice set prevents buyers from settingtheir marginal rates of substitution betweencharacteristics equal to the corresponding im-plicit price ratios, then the market clearingprices for individual characteristics may un-derstate or overstate individual marginal will-

ingness to pay. Likewise, if sellers can exertsome degree of market power, then implicitprices for specic characteristics may notequal their marginal costs.

The reason for relaxing continuity and per-fect competition in our conceptual model isthat in reality some neighborhoods are con-strained by geographic features such as waterbodies, steep terrain, and public land that canproduce discreteness in the supply of ameni-ties, whereas other neighborhoods use zoning

regulations to explicitly constrain further ur-ban development. If a constrained neighbor-hood provides access to a unique bundle of


7/30


amenities, then the amenity bundle may con-vey market power to property owners in theneighborhood (e.g., see Taylor and Smith2000). We return to this point in Section VI,

where we consider local market power as apotential explanation for some of our results.Our specication for the hedonic price

function, , describes aP = P (g ,l , x ) jk t kt jk jkt spatial landscape at a single point in timewhere prices, amenities, and location choicesare all dened such that no household wouldprefer to move, given its income and prefer-ences. This is a single-period snapshot of mar-ket outcomes; it may or may not be a long-run steady state. Current period incomes andpreferences may reect temporary macroec-onomic factors. Credit may be unusually easy(or difcult) to obtain relative to a long-runequilibrium. The average household may beunusually optimistic (or pessimistic) about thefuture asset value of housing. Budget con-straints may reect other temporary macro-economic shocks. As all of these factorschange over time, households may adjust theirbehavior in ways that alter the shape of theprice function and generate boom-bust cycles.

During a boom-bust cycle, the evolution of the price function can be decomposed intochanges in the market value of land and struc-tures. To illustrate this, we rst dene the mar-ket value of a property at a single point in timeas its current annualized price.

Denition 1. is the market P ≡P (g ,l , x ) jkt kt jk jk t value of property j,k in period t .

The value of the underlying land is then de-ned by the thought experiment where we re-

move all of the structural characteristics fromthe property.

Denition 2. is the land LV ≡P (g ,l ,0) jkt kt jk value of property j,k in period t .

measures what a vacant (but otherwise LV jk t identical) parcel to j would sell for in the sameneighborhood. 9 This denition of land valuecaptures the spatial trade-off between com-muting costs and accessibility to the city cen-

9 This assumes the undeveloped parcel is also zoned forresidential development.

ter (Alonso 1964; Muth 1969; Mills 1967), aswell as the value of local public goods andurban amenities conveyed by the neighbor-hood (Tiebout 1956). 10

Finally, subtracting land value from totalmarket value yields the value associated witha property’s structural characteristics, . x jk t

Denition 3. is the struc-SV = P − LV jk t jk t jk t tural value of property j,k in period t .

While it is conceptually straightforward to de-compose property value into the value of landand structures, empirical implementationpresents several challenges.

III. ESTIMATING THE MARKET VALUEOF LAND AND STRUCTURES

Background

If life were more like a laboratory experi-ment, there would be no need to estimate landvalues. Sales of vacant parcels would be ran-domly distributed throughout metropolitan ar-eas, and we would simply measure their trans-action prices. The problem, of course, is thatvacant land sales typically occur at the fringesof urban areas. We rarely observe such trans-actions occurring in built-up neighborhoods.In an established neighborhood, the closestsubstitute for a vacant land sale is likely to bea “teardown.”

When an existing structure is purchasedwith a plan to demolish it and build new hous-ing, the value of the underlying land shouldequal the sale price of the developed parcelless demolition costs. Rosenthal and Helsley

(1994) were the rst to apply this idea to inferland values from teardown properties in Van-couver, British Columbia. In subsequentwork, Dye and McMillen (2007) and Mc-Millen (2008) rened the econometrics tocontrol for the nonrandom selection of whichparcels are torn down and provided new evi-dence on land values in Chicago. While tear-downs can support a convincing quasi- ex-perimental approach to measuring land value,

10

Cheshire and Sheppard (1995) distinguish betweenthese two components of land value. While we could cer-tainly do the same, it is not essential to our analysis.


8/30


the active markets are too few and too thin toapply the method broadly across the UnitedStates or at a high level of spatial resolutionthroughout a single metropolitan area.

Since the lack of data makes it difcult tomeasure the market value of land directly, an-alysts have sought to estimate it indirectlyfrom hedonic regressions or replacement costequations. Both strategies begin by rearrang-ing the decomposition in Denition 3,

P = LV + SV . [2] jk t jk t jk t

Given data on the structural characteristics of houses and their transaction prices, equation[2] can be used to estimate land values. In thereplacement cost framework, two maintainedassumptions are sufcient to guarantee the es-timator will be consistent. First, the market forhousing is assumed to be sufciently compet-itive that the market value of a structure willequal the cost of rebuilding that structure inits current condition: replacement cost jkt ≡

. Second, the replacement RC ( x ) = SV t jkt jkt cost function is assumed to be known. Underthese assumptions, one can obtain a consistentestimate for land value as the residual ob-

tained by subtracting replacement cost fromthe price of housing,

LV = P − RC ( x ). [3] jk t jk t t jk t

Davis and Heathcoat (2007) applied this logicat the national level to develop the rst mac-roeconomic index of residential land value inthe United States. Davis and Palumbo (2008)rened their methodology to control for vari-ation in property values and construction costsacross major metropolitan areas. They devel-oped a database describing the value of landin 46 major metropolitan areas between 1984and present. 11

During boom-bust cycles, the replacementcost framework tends to attribute most of thechanges in property values to speculation onland. This follows from the mechanics of [2]–

11 While the published version of Davis and Palumbo’spaper presents estimates for 1984 to 2004, the Lincoln In-stitute of Land Policy maintains a web page where their

estimates are updated as new data become available: http:// www.lincolninst.edu/subcenters/land-values/metro-area-land-prices.asp.

[3]. If residential construction costs are rela-tively stable during a period when propertyvalues are rising rapidly, then observedchanges in property values will be interpreted

as changes in land value. This was exactlywhat happened during the recent boom. Thereplacement cost model indicates that the ratioof land value to property value on the WestCoast increased from 61% in 1998 to 74% in2004, for example (Davis and Palumbo 2008).We have no doubt that the market value of land did increase during the boom. However,the replacement cost estimates for the mag-nitude of the change may be too high if hous-ing markets are less than perfectly competi-tive or if zoning restrictions and permittingrequirements drive a wedge between con-struction costs and effective replacement costsin the short run.

The hedonic approach to estimating landvalues avoids the need to specify replacementcosts or assume that marketsareperfectlycom-petitive. Instead, the key maintained assump-tion is a parametric specication for the rela-tionship between the sale price of a house andits characteristics. Equation [4] presents a lin-ear example reecting the spatiotemporalstructureof past hedonic land value estimators.P = ξ + δ ⋅l + x β + ε . [4] jk t jk jk t jk t

In this case, provides an estimate of the im-plicit marginal price of land, and provides⋅ l jk an estimate for the property’s land value. Ef-forts to estimate from data on individualδ housing sales date back at least to Clapp’s(1980) study of land values in Chicago. 12 Overthe years, the methodology has been rened toallow more exible parametric specicationsfor the hedonic price surface (Cheshire andSheppard 1995) and extended to compare es-timates across 21 metropolitan areas (Glaeser,Gyourko, and Saks 2005).

There are two key challenges to developingcredible hedonic estimates for land values.The rst challenge—omitted variable bias—is widely recognized. For example, one might

12 In earlier work, Jackson (1979) used aggregate censustract data to estimate a coarse approximation to a hedonic

price surface in Milwaukee. In principle, his results couldalso be used to develop an approximation to the value of land.


9/30


expect that houses built on larger lots will alsotend to be built using higher-quality materials.Because data on building materials are typi-cally unavailable, their effect on sale prices

will be confounded with the value of land(McMillen 2008). Another concern is that anestimate for the depreciation of structures(from the coefcient on age) may be con-founded with unobserved neighborhoodamenities, because all of the houses in a sub-division tend to be built at about the same time(Davis and Palumbo 2008). More generally,there is always likely to be some degree of spatial correlation between observed parcelcharacteristics and unobserved neighborhoodamenities that will ultimately bias the esti-mator (Kuminoff, Parmeter, and Pope 2010).

The second challenge is to choose a spec-ication for the price function that is suf-ciently exible to capture the key features of spatial and temporal variation in land values.Past studies have focused on allowing the perunit price of land ( ) to vary exibly withinδ a metropolitan area (e.g., see Cheshire andSheppard 1995). While this is an importantdimension of heterogeneity, we hypothesizethat it is equally important to distinguish be-tween the variable (i.e., quantity-based) andxed (i.e., access-based) components of landvalue, while recognizing that marginal im-plicit prices for both may change over timedue to changing market conditions. 13

Access matters. This is a central theme of public, urban, and environmental economics.Commuters value access to the central busi-ness district (CBD).Homeowners valueaccessto local public goods and amenities that con-tribute to their quality of life. These values are

fundamental to the models of urban spatialstructure and neighborhood formation thatbuild on the work of Tiebout (1956), Alonso(1964), Mills (1967), and Muth (1969). Withina neighborhood, the value of access will be ap-proximately xed, independent of parcel size.As onemoves to a differentneighborhoodwithhigher crime rates, lower-quality schools, and/

13 In general, the shape of the equilibrium hedonic pricefunction will vary with changes in tastes, wealth, regula-

tions, and spatially delineated amenities. See Kuminoff,Smith, and Timmins (2010) and Kuminoff, Parmeter, andPope (2010) for details.

or a longer commute to the CBD, the value of access may drop sharply. To identify spatialvariation in access value separately from spa-tial variation in the per unit price of land, the

analyst must observe several housing transac-tions within each neighborhood during an in-terval over which land valuesarerelatively sta-ble. 14 To identify temporal variation in each of these components, the analyst must observe alarge number of observations in each period.Our econometric model is specially designedtoaccomplish these tasks using dataonthe uni-verse of housing sales within a metropolitanarea together with controls for omitted vari-ables.

Rening the Hedonic Approach toEstimation

Our approach to estimating land values re-lies on micro data that are sufciently rich toallow us to estimate annual price functions formetropolitan areas, while simultaneously us-ing spatial xed effects to capture the marketvalue of latent attributes of land and struc-tures. In the case of land, the issue is that noexisting database provides comprehensivecoverage of spatial variation in access-basedamenities below the level of a county. This isimportant because amenities often vary sig-nicantly within a county. To measure thisvariation we use spatial xed effects forneighborhoods, which we dene to be Censustracts. 15 Within a tract, access to amenitieswill be approximately xed. Children will beassigned to public schools in the same schooldistrict, their parents will face the same com-muting opportunities, and there will be littleor no variation in crime rates or air quality.Thus, we would expect tract xed effects toabsorb the composite value of access to theseand other neighborhood amenities.

14 Abbott and Klaiber (2011) make a similar point in thecontext of identifying what occupants are willing to pay fora particular amenity.

15 The U.S. Census Bureau denes Census tracts to be“as homogeneous as possible with respect to populationcharacteristics, economic status, and living conditions.” See“Chapter 10: Census Tracts and Block Numbering Areas,

U.S. Census Bureau,” Geographic Areas Reference Manual,which can be found at www.census.gov/geo/www/garm.html.


10/30


In the case of structures, micro-level dataare typically limited to the attributes recordedby the county assessor. Some houses havehardwood oors, granite countertops, sky-

lights, solar panels, and spas. Unfortunately,these improvements are rarely noted in thecounty records. If the quality of building ma-terials varies systematically across neighbor-hoods in ways we do not observe, then theiraverage effect on property values may be con-founded with our estimates for the xed com-ponent of neighborhood land value. To miti-gate this potential source of confounding, weadd a set of interactions between the xed ef-fects for neighborhoods and the square foot-age of the house. The resulting terms are in-tended to capture systematic variation acrossneighborhoods in the average value of asquare foot of structural improvements.

We adopt a semilog form for the estima-tion, regressing the log of transaction pricesfor all of the single-family residential prop-erties sold in a metropolitan area during yeart on their lot sizes, their structural character-istics, and two sets of xed effects,ln(P ) = ξ +δ l + γ sqft + x β +ε jk t kt kt jk t kt jk t jk t t jkt

123 14243 . [5] LV SV jk t jk t

The rst two terms after the equality corre-spond to the property’s land value. denotesξ kt the neighborhood xed effects. They willmeasure the component of land value that isconstant across all the houses sold within tractk during year t , regardless of lot size. Theneighborhood amenities that enter mayξ kt also interact with the size of the lot to inu-ence the variable component of land value.For example, the marginal value of yard size

may be larger in quieter neighborhoods withlower crime rates. Therefore, we allow the co-efcient on lot size, , to vary over neigh-δ kt borhoods as well.

The third and fourth terms after the equal-ity correspond to the value of structural im-provements. measures the component x β jk t t of property value that can be explained by thehousing characteristics that are observed.While we allow the implicit prices of char-acteristics to change over time, we restrict

them to be xed within a metropolitan areaduring the course of a year. mea-γ sqft kt jkt sures systematic variation in the average value

of a square foot of living space that variesacross neighborhoods due to unobservedstructural improvements.

Finally, we interpret the error term asε jkt

the composite of three effects. It will reect(1) unobserved idiosyncratic structural im-provements that differ from the tract average,(2) idiosyncratic access to amenities within aneighborhood, 16 and (3) misspecication inthe shape of the price function. To mitigatethe rst two effects, we aggregate our micro-level estimates for the value of land and struc-tures to report averages for Census tracts,counties, and metropolitan areas. This also al-lows us to compare our results to estimatesfrom the prior literature. While our resultingpoint estimates surely contain some error, andtheir standard errors may be affected by spa-tial autocorrelation, we expect the magnitudeof the bias in our point estimates to be smallerthan in previous hedonic studies because of the ways in which our model enhances spatialand temporal resolution and controls for omit-ted variables. 17 To evaluate the impact of these renements, we use the fact that ourmodel nests the conventional hedonic speci-cation as a special case. Equation [5] reducesto [4] if we omit spatial xed effects ( γ =kt

16 For example, all of the houses in a Census tract maybe located near public open space, but the handful of lotsthat are adjacent to the public lands may sell at an additionalpremium.

17 A referee notes that we could have used an explicitspatial model to attempt to mitigate the bias from spatialcorrelation between housing attributes and omitted vari-ables. Examples of explicit spatial models include the spatialerror, spatial lag, general spatial, and spatial Durbin models.We chose not to use these models for three reasons. First,

there is prior evidence that spatial xed effects outperformexplicit spatial models in terms of recovering accurate esti-mates for the implicit prices of housing attributes (Kuminoff,Parmeter, and Pope 2010). Second, our xed effects modelproduces estimates that are very similar to Davis and Pal-umbo’s (2008) replacement cost calculations prior to theboom, as we discuss in Section V. We interpret this evidenceas validating our model, since we expect the replacementcost methodology to provide good estimates for land valuesduring periods of market stability. The third reason is thatour sample sizes and covariate matrices are too large to im-plement the maximum likelihood routines for estimatingexible spatial autocorrelation models. As computing powercontinues to improve, an interesting avenue for future re-

search would be to develop new models that combine spatialxed effects with strategies for addressing more localizedforms of spatial autocorrelation within neighborhoods.


11/30


) and restrict the implicit price per acreξ = 0kt of land to be constant within a metropolitanarea ( ).δ = δ kt

IV. DATA AND SUMMARY STATISTICSOur analysis is based on more than one

million observations on the sales of single-family residential properties across the UnitedStates. We purchased the data from Data-Quick. This widely used commercial vendorof real estate data assembled the data fromassessor’s ofces in individual towns andcounties. The data include the transactionprice of each house, the sale date, and a con-sistent set of structural characteristics, includ-ing square feet of living area, number of bath-rooms, number of bedrooms, year built, andlot size. Using these characteristics, we per-formed some standard cleaning of the data,removing outlying observations, removinghouses built prior to 1900, and removinghouses built on lots larger than 5 acres.

The data also include the physical addressof each house, which we translated into lati-tude and longitude coordinates using GISstreet maps and a geocoding routine. The lat-long coordinates were then used to assigneach house to its corresponding census tract.The tract-level assignment provides theneeded spatial resolution to analyze trends inland values within and across metropolitan ar-eas during the boom-bust cycle. Furthermore,it allows us to use spatial xed effects to con-trol for the average effect of latent variableswithin each tract.

While we conducted the econometric anal-ysis for 10 metropolitan areas, we focus on

four of them in greater detail in order to illus-trate our main results: Miami, San Francisco,Boston, and Charlotte. 18 We selected thesefour because all have complete data between1998 and 2008, they provide geographic vari-ation on populous areas in the United States,they provide variation in the supply elasticityof land, and they differ in the intensity of theirboom-bust cycles. Figure 2 illustrates the dif-ferences in the sizes of their booms and bustsusing the Case-Shiller Home Price Index.

18 The other eight are Cincinnati, Detroit, Los Angeles,Oakland, Philadelphia, Pittsburg, San Jose, and Tampa.

Each panel also reports Saiz’s (2010) esti-mates for the supply elasticity of housing. 19

Table 1 provides summary statistics for thehousing transactions that we observe in Mi-

ami, San Francisco, Boston, and Charlotte.The rst two rows of each panel illustrate thatthe average sale price rose in all four areasbetween 1998 and 2006. The size of the in-crease was most striking in Miami ($162,000to $410,000) and San Francisco ($343,000 to$809,000), where prices more than doubled innominal terms. These increases do not reectany obvious changes in the composition of houses on the market. The structural charac-teristics of the average sale property are es-sentially constant over the study period. Ineach area, the median transaction was a sin-gle-family house with three bedrooms, twobaths, and between 1,600 and 1,900 squarefeet of living area. Naturally, Charlotte andMiami have newer housing stocks than SanFrancisco and Boston. Lot sizes also tend tobe larger in Charlotte and Boston than in Mi-ami and San Francisco, reecting variation inthe balance between sales from the cities andsuburbs. 20

V. RESULTS

Comparison to Preboom Estimates from theExisting Literature (1998–1999)

We begin by comparing our estimates forland values to previous gures generated bythe conventional hedonic estimator by Glae-ser, Gyourko, and Saks (2005) (henceforthGGS) and the replacement cost estimator de-veloped by Davis and Palumbo (2008)(henceforth DP). Neither study had the benetof our spatially delineated micro data on ac-tual housing sales. Instead, they combineddata from the American Housing Survey withother sources to generate estimates for aver-

19 His estimates are generated using information on geo-graphic constraints, regulatory constraints, and predeter-mined population levels in each metropolitan area.

20 A related literature on gentrication investigates howthe demographic composition of neighborhoods within ametropolitan area changes as people migrate from the sub-

urbs to the cities, and vice versa. See McKinnish, Walsh,and White (2010) for an interesting example of this line of research, as well as citations to the broader literature.


12/30


TABLE 1Summary Statistics for Miami, San Francisco, Boston, and Charlotte

Variable Mean Median Std. Dev. Min. Max. Obs.

MiamiHousing Price (1998) 161,679 129,000 144,276 17,500 3,500,000 9,799Housing Price (2006) 409,931 335,000 333,682 23,000 5,000,000 21,730Square Feet 1,948.49 1,740 861.07 251 9,997 194,242Bathrooms 2.11 2 0.80 0.5 10 194,242Bedrooms 3.18 3 0.83 1 10 194,242Year Built 1977 1983 20 1901 2008 194,242Lot Size (acres) 0.21 0.17 0.26 0.00 5 194,242

San FranciscoHousing Price (1998) 343,034 285,000 235,225 10,000 4,772,727 31,656Housing Price (2006) 809,485 710,000 430,730 15,000 4,900,000 42,193Square Feet 18,02.24 1,640 769.58 260 9,984 517,295Bathrooms 2.19 2 0.80 0.5 10 517,295Bedrooms 3.34 3 0.88 1 10 517,295

Year Built 1968 1968 25 1900 2008 517,295Lot Size (acres) 0.21 0.14 0.31 0.00 5 517,295

BostonHousing Price (1998) 275,770 234,000 177,842 16,321 4,150,000 14,399Housing Price (2006) 445,970 367,900 302,919 50,000 4,750,000 29,369Square Feet 1,875.02 1,662 880.97 252 9,989 281,920Bathrooms 1.93 2 0.83 0.5 9.5 281,920Bedrooms 3.25 3 0.82 1 10 281,920Year Built 1960 1960 30 1900 2008 281,920Lot Size (acres) 0.58 0.34 0.66 0.00 5 281,920

CharlotteHousing Price (1998) 176,186 137,000 143,959 13,000 2,500,000 4,909Housing Price (2006) 231,841 175,000 205,418 7,900 3,700,000 22,552Square Feet 2,090.49 1,876 937.11 412 9,968 129,596Bathrooms 2.44 2 0.86 0.5 10 129,596Bedrooms 3.31 3 0.70 1 10 129,596Year Built 1986 1994 21 1900 2009 129,596Lot Size (acres) 0.36 0.28 0.36 0.01 5 129,596

Note: Summary statistics for housing characteristics and lot size using the micro-level assessor data for single-family residential propertiesin Miami, San Francisco, Boston, and Charlotte.

age land values within several metropolitanareas. Fortunately, some of their estimatesoverlap with the spatial dimensions of ourdata prior to the onset of the housing market

boom, providing an opportunity for compari-son. The purpose of the comparison is to in-vestigate how our renements to the hedonicland value estimator inuence the accuracy of our results. We would expect DP’s replace-ment cost calculations to generate reasonableestimates for land values prior to the boom.The relative stability of the market in the midto late 1990s would have allowed developersconsiderable time to meet demand, mitigatingany wedge between construction costs and ef-

fective replacement costs. Thus, we viewDP’s estimates as the most reliable baselinefor comparison.

The task of estimating land values is a rela-tively small component of the overall analysisby GGS. Their main objective is to test thehypothesis that land use regulations impose an

effective tax that explains the rise in housingprices in major metropolitan areas. To illus-trate their point and to compare housing pricesto construction costs, GGS estimate the “free-market cost of land” using a conventional he-donic model (similar to equation [4] above)for 21 metropolitan areas based on data fromthe 1998 and 1999 installments of the Amer-ican Housing Survey. 21 We have the requisite

21 The results (reported in their Table 4) support their

hypothesis that the areas that we would expect to be morehighly regulated have larger differences between construc-tion costs and housing prices.


13/30


TABLE 2Comparing Traditional Hedonic Estimates with Estimates Generated by the Replacement-Cost Method and

Our New Hedonic Method

Metropolitan Area Year

1

GGS Hedonic LandValues ($/acre)

2

Our Approximationto GGS ($/acre)

3

DP ReplacementCost Land Values

($/acre)

4

Our New HedonicLand Values

($/acre)

Boston 1998 29,621 20,038 237,063 212,523Cincinnati 1999 17,424 25,700 131,220 217,927Detroit 1999 16,117 5,227 96,927 238,939Los Angeles 1999 112,820 67,954 804,555 857,309Oakland 1998 101,930 94,525 967,995 908,507Philadelphia 1999 35,284 16,988 104,087 198,530Pittsburgh 1998 30,492 21,780 42,007 212,020San Francisco 1998 178,596 192,971 2,421,461 1,716,395San Jose 1998 170,755 125,017 1,533,329 1,227,703Tampa 1998 16,117 871 122,822 176,039

Note: Column 1 reports selected land values from Table 4 of Glaeser, Gyourko and Saks (2005), or “GGS,” converted to a per acre basis.Column 2 reports our replication of the GGS estimates, using a similar (but not identical) set of housing characteristics from assessor data.Column 3 reports replacement-cost estimates from Davis and Palumbo (2008), or “DP.” Finally, Column 4 reports results from our new hedonicxed effects estimator.

information to develop comparable estimatesfor 10 of their 21 metropolitan areas. Conven-iently, DP also report estimates for all 10 areas.

To provide the best possible comparison,we focus on the subset of our data that over-laps with the information used by GGS. Spe-

cically, we limit our data to the year thatmatches the year in which each metropolitanarea was covered by the AHS (either 1998 or1999). Then we subdivide metropolitan areasto match the disaggregate denitions used inthe AHS. This means subdividing the SanFrancisco consolidated metropolitan statisti-cal area into the San Francisco, Oakland, andSan Jose primary metropolitan areas, for ex-ample. While our micro data still differ fromthe AHS in terms of the number of observa-

tions and the richness of information on struc-tural characteristics, their spatial and temporaldimensions are the same.

The rst column in Table 2 simply repro-duces the estimates of land value (on a per acrebasis) from Table 4 of GGS. In Column 2, wereport the results from our attempt to come asclose as possible to replicating their estimatingequation, given the differences between thevariables inour dataand the AHS micro data. 22A quick comparison between Columns 1 and

22 Our results are generated using a simple linear modelestimated according to equation [4], using the combination

2 conrms that the two sets of estimates arequite similar (with Tampa as the exception).Overall, the estimates line up with our generalintuition for which metropolitan areas ought tohave more expensive land. San Francisco, SanJose, Oakland, and Los Angeles have the high-

est measures of land value, whereas Detroitand Tampa have the lowest. However, all of the estimates seem implausibly low for the late1990s. Could you really buy an acre of land inSan Francisco for under $200,000 or in Bostonfor under $30,000? A likely explanation is thatthe conventional hedonic estimator does notfully capture thexed component of landvalueassociated with access to thelocalpublicgoodsand amenities ( ).ξ kt

Column 3 reports the corresponding re-

placement cost estimates for land value fromDP. They used information published by R.S.Means Company (2004) to develop metro-politan-level estimates for replacement cost.Their measures for housing prices were de-veloped by combining data on price levels ineach metropolitan area during AHS surveyyears with time-series data on the percentagechange in housing prices from Freddie Mac’sConventional Mortgage Housing Price Index

of control variables that comes as close as possible to thespecication from GGS. Complete results will be providedupon request.


14/30


(CMHPI). The rank order in Column 3 is simi-lar to the rst two columns, but the replace-ment cost estimates are typically an order of magnitude larger! While there are some slight

variations in the datasets used to develop theestimates in Columns 1 and 3, none seem ca-pable of generating order of magnitude dif-ferences. 23 It seems more likely that the dif-ferences are due to estimation procedures. Inparticular, the access-based component of land value associated with local public goodswould be included in the replacement cost es-timates and excluded from the estimates gen-erated by conventional hedonic regressions.

Column 4 reports the estimates from ourrenement to the hedonic estimator, using thespecication in equation [5]. 24 Generalizingthe conventional hedonic model to allow foraccess-based amenities and latent housingcharacteristics increases our estimates by anorder of magnitude (moving from Column 2to Column 4). The resulting estimates alignmuch more closely with the estimates fromDP’s replacement-cost model.

Finally, it is important to reiterate that thesimilarity between our estimates and DP’s oc-curs during a two-year period prior to theboom. As we track the two sets of estimatesover the course of the boom-bust cycle, wesee some interesting differences.

The Evolution of Average Land Valuesduring the Boom-Bust Cycle (1998–2009)

We estimated equation [5] for each (met-ropolitan area, year) combination from 1998and 2009. Table 3 summarizes results for thefour metropolitan areas where we have a com-

plete set of data: Miami, San Francisco, Bos-23 For example, the CMHPI uses a slightly different def-

inition for metropolitan areas than the AHS. Also, since DPdo not report average lot size, we use the average lot size inour data to convert the DP estimates to a $ per acre measure.So, there are certainly some differences in the estimates thatare caused by differences in spatial-temporal components of the underlying datasets, but we think that it is highly unlikelythat these drive the large differences we document betweenGGS and DP.

24 We dropped tracts with fewer than 15 observationsper year to better ensure accurate estimates of tract-specic

land values. This drop (approximately 33.5% of our obser-vations) was necessary given our Census tract xed effectsidentication strategy.

ton, and Charlotte. It reports our measures forthe evolution of land values and the share of property value attributed to land (“landshare”), alongside the replacement cost esti-

mates from DP.25

The hedonic measures weregenerated by averaging our parcel-specic es-timates for land values and improved valuesover all of the housing transactions in eachmetropolitan area. There are some obviousdifferences between the two sets of estimatesat the market’s peak.

Figure 3 26 illustrates the differences graph-ically. Focusing on the rst column in the g-ure, it is clear that land values estimated byboth methods rise and fall during the boomand bust. Prior to the boom, the two sets of estimates are similar. The same is true follow-ing the bust. However, the peak amplitude ismuch larger in the replacement cost estimates.

The second column of Figure 3 illustrateshow estimates for the value of structuresevolved over the same period. The hedonicmodel suggests that the market value of struc-tures rose and fell in tandem with the marketvalue of land during the boom-bust cycle. Thereplacement cost measures rose steadily, fol-lowing a similar trend in every metropolitanarea. Once again, the differences between thehedonic estimates and the replacement costmeasures are largest at the height of the boom.

Does the difference between the two setsof estimates reveal something interestingabout the behavior of housing markets duringthe boom-bust cycle? Or does it merely reectdifferences in the underlying data? While ourcomparisons were made along a consistent setof spatial and temporal dimensions, the un-derlying micro data are not the same. DP’s

replacement cost estimates are based on in-tegrating the AHS data with Freddie Mac’sCMHPI, whereas our hedonic estimates come

25 The replacement cost results are the Davis and Pal-umbo (2008) estimates that have been updated and providedat www.lincolninst.edu/subcenters/land-values/. Also notethat we were unable to obtain assessor data for the year 2009in Boston. Therefore all results reported for Boston are forthe 1998–2008 time frame.

26 Column 1 shows the evolution of total land value, andColumn 2 shows the evolution of total structural value by

Primary Metropolitan Statistical Area. Both columns showestimates derived using the hedonic method and the replace-ment cost method.


15/30


T A B L E 3

E v o l u t i o n o f L a n d V a l u e s a n d

L a n d S h a r e s b e t w e e n 1 9 9 8 a n d 2 0 0 9

M i a m i

S a n F r a n c i s c o

B o s t o n

C h a r l o t t e

H e d o n i c

L a n d

V a l u e

R e p l a c e m e n t

C o s t L a n d

V a l u e

H e d o n i c

L a n d

S h a r e


C o s t L a n d

S h a r e

H e d o n i c

L a n d

V a l u e


C o s t L a n d

V a l u e

H e d o n i c

L a n d

S h a r e


C o s t L a n d

S h a r e

H e d o n i c

L a n d

V a l u e


C o s t L a n d

V a l u e

H e d o n i c

L a n d

S h a r e


C o s t L a n d

S h a r e

H e d o n i c

L a n d

V a l u e


C o s t L a n d

V a l u e

H e d o n i c

L a n d

S h a r e


C o s t L a n d

S h a r e

1 9 9 8 1 0 0

, 8 3 3

1 0 0 , 1 9 2

0 . 6 7

0 . 5 7

2 8 6 , 6 3 8

3 8 4 , 8 9 5

0 . 7 3

0 . 8 0

1 4 4 , 1 5 6

1 5 0 , 6 7 6

0 . 5 6

0 . 5 9

7 9 , 7

0 0

9 7 , 2

4 9

0 . 5 2

0 . 5 9

1 9 9 9 1 0 0

, 8 3 7

1 0 6 , 1 8 5

0 . 6 4

0 . 5 7

3 3 2 , 6 0 5

4 3 9 , 4 4 5

0 . 7 2

0 . 8 2

1 5 8 , 9 7 9

1 7 8 , 1 3 4

0 . 5 9

0 . 6 3

7 7 , 3

9 5

1 0 2 , 1 1 2

0 . 5 0

0 . 5 9

2 0 0 0 1 0 5

, 3 3 9

1 1 6 , 5 8 5

0 . 6 2

0 . 5 9

4 0 1 , 3 9 3

5 7 7 , 0 5 8

0 . 7 2

0 . 8 5

1 7 9 , 9 2 0

2 1 6 , 6 7 8

0 . 5 9

0 . 6 7

8 2 , 1

2 7

1 0 6 , 4 5 7

0 . 4 9

0 . 5 9

2 0 0 1 1 2 2

, 3 4 1

1 3 4 , 6 5 3

0 . 6 5

0 . 6 1

4 4 4 , 1 3 3

6 7 5 , 4 0 4

0 . 7 3

0 . 8 5

1 9 6 , 3 2 9

2 6 4 , 3 2 6

0 . 6 0

0 . 6 9

8 8 , 7

9 1

1 0 5 , 2 4 6

0 . 5 3

0 . 5 8

2 0 0 2 1 3 8

, 4 1 0

1 5 9 , 4 8 2

0 . 6 4

0 . 6 4

4 3 8 , 2 6 4

6 8 5 , 5 3 1

0 . 6 9

0 . 8 5

2 2 1 , 0 0 5

3 0 0 , 3 0 7

0 . 6 1

0 . 7 1

8 3 , 1

1 5

1 0 5 , 3 8 8

0 . 4 8

0 . 5 7

2 0 0 3 1 5 7

, 8 9 1

1 9 3 , 0 0 3

0 . 6 4

0 . 6 7

4 4 0 , 7 8 5

7 4 0 , 9 5 2

0 . 6 6

0 . 8 6

2 3 7 , 6 2 2

3 4 4 , 8 1 0

0 . 6 2

0 . 7 3

8 4 , 3

6 1

1 0 6 , 6 3 4

0 . 4 7

0 . 5 6

2 0 0 4 1 8 7

, 7 2 6

2 3 4 , 7 3 9

0 . 6 4

0 . 6 9

5 0 5 , 0 2 2

8 6 8 , 2 7 7

0 . 6 7

0 . 8 7

2 6 4 , 1 4 4

3 8 1 , 5 9 4

0 . 6 3

0 . 7 4

8 9 , 8

4 7

1 0 4 , 2 3 9

0 . 4 7

0 . 5 3

2 0 0 5 2 3 1

, 5 3 3

3 2 1 , 3 6 0

0 . 6 5

0 . 7 4

5 8 7 , 1 0 5

1 , 0 7 8 , 1 0 2

0 . 6 5

0 . 8 9

2 7 2 , 8 3 0

4 1 3 , 2 4 3

0 . 6 2

0 . 7 4

8 9 , 8

6 9

9 8 , 6

1 0

0 . 4 6

0 . 4 8

2 0 0 6 2 6 7

, 1 0 7

4 0 7 , 1 4 4

0 . 6 8

0 . 7 6

6 0 7 , 5 5 8

1 , 1 4 3 , 4 8 1

0 . 6 6

0 . 8 8

2 6 3 , 6 2 7

3 9 5 , 6 7 5

0 . 6 2

0 . 7 1

9 5 , 7

4 8

1 0 0 , 0 6 0

0 . 4 7

0 . 4 5

2 0 0 7 2 7 6

, 1 7 3

3 8 3 , 4 7 8

0 . 6 7

0 . 7 3

6 2 3 , 9 5 8

1 , 0 8 5 , 3 0 4

0 . 6 4

0 . 8 6

2 5 3 , 5 5 2

3 6 3 , 1 9 7

0 . 6 0

0 . 6 8

9 7 , 4

7 9

1 0 6 , 0 2 4

0 . 4 6

0 . 4 6

2 0 0 8 2 1 8

, 3 2 0

2 3 0 , 1 8 0

0 . 6 7

0 . 6 0

5 1 7 , 2 7 7

7 8 1 , 6 0 3

0 . 6 5

0 . 8 0

2 5 0 , 6 1 1

3 2 0 , 8 7 3

0 . 6 4

0 . 6 4

1 0 2 , 2 4 3

9 2 , 2

0 1

0 . 5 0

0 . 4 0

2 0 0 9 1 8 4

, 4 4 1

1 2 9 , 0 7 1

0 . 6 8

0 . 6 8

4 6 9 , 6 9 2

5 2 7 , 9 4 0

0 . 6 8

0 . 7 5

—

2 7 7 , 6 8 0

—

0 . 6 1

9 7 , 7

7 8

7 0 , 9

5 9

0 . 5 0

0 . 3 4

N o t e :

T h i s t a b l e s h o w s t h e e v o l u t i o n o f t o t a l l a n d v a l u e a n d t h e s h a r e o f p r o p e r t y v a l u e s a c c o u n t e d f o r b y l a n d , b

y P r i m a r y M e t r o p o l i t a n S t a t i s t i c a l A r e a . F o r e a c h m e t r o p o l i t a n a r e a , e s t i m a t e s a r e s h o w n

f o r b o t h

t h e h e d o n i c m e t h o d f o r d e r i v i n g l a n d v a l u e s a n d t h e r e p l a c e m e n t c o s t m e t h o d .


16/30


FIGURE 3Evolution of Total Land Value (left) and Structural Value (right) for Four Metropolitan Areas


17/30


from assessor data. In principle, the assessordata describe the universe of housing trans-actions, whereas the CMHPI is limited totransactions where: (1) the transaction was a

repeated sale, and (2) the buyer took out aconventional mortgage that was purchased orinsured by Freddie Mac or Fannie May.

Figure 4 27 compares the evolution of av-erage property values in the two datasets. Theassessor data suggest slightly smaller in-creases in property values during the boom.One possible explanation is that less expen-sive transactions were more likely to be as-sociated with unconventional mortgages. An-other explanation is that new houses built overthis period tended to be located near the urbanfringe, where land values (and property val-ues) were lower. In any case, the differencesbetween the two measures of average propertyvalue in Figure 4 are dwarfed by the differ-ences in estimated land values in Figure 3.Thus, the differences between the hedonic andreplacement cost estimates for land value ap-pear to be tied to methodology, not the un-derlying data.

Data differences aside, the main economicimplication of our comparison between thehedonic and replacement cost estimates isthat, during the boom, the market value of structures may have exceeded their replace-ment costs. To further investigate this possi-bility, we examine the spatial variation in theevolution of land values within each metro-politan area.

Within-Metropolitan Heterogeneity in theEvolution of Land Values (1998–2009)

Figure 5 28 illustrates the spatial heteroge-neity in land values across counties in thegreater San Francisco and Boston metropoli-tan areas. 29 The left-most maps display theland value of the average residential property

27 Figures are produced using our assessor data and datafrom Freddie Mac’s CMHPI as documented by Davis andPalumbo (2008).

28 These gures were produced using our hedonic ap-proach to estimating land value. Black represents greatestabsolute change, positive or negative.

29

Charlotte and Miami have only one or two countieswith available assessor data, so their maps are less interest-ing.

sold in 1998, the change in average land valueduring the boom (1998–2006), and the changein average land value during the bust (2006–2009). In the San Francisco metropolitan area

the counties with the highest land values in1998 are San Francisco and San Mateo, fol-lowed by Marin and Santa Clara. These samecounties experienced the largest increases inland value during the boom and the smallestdecreases during the bust. Looking at the left-most maps in Figure 5 for the Boston metro-politan area reveals a similar pattern.

The right-most maps in Figure 5 focus onthe ratio of land value to total property value.The three maps display the average land sharein 1998 and the subsequent changes duringthe boom and bust periods. Focusing rst onthe greater San Francisco metropolitan area,we see that areas with higher land shares in1998 (e.g., San Francisco and San Mateo)tended to see drops in land share during theboom and increases during the bust. Again, asimilar pattern emerges in the Boston metro-politan area.

Overall, the spatial heterogeneity in theevolution of land values within the Bostonand San Francisco areas seems somewhatcounterintuitive. The counties that experi-enced the least volatility in land values duringthe boom-bust cycle are the same countiesthat we would expect to have the most inelas-tic supply of housing. This pattern is the op-posite of what the prior literature has observedabout the variation in land values betweenmetropolitan areas (e.g., Davis and Palumbo2008). Figure 2 provides an example of thestylized fact that housing prices (and land val-ues) tend to be more volatile in metropolitan

areas where the supply of land is relativelyinelastic. 30 Why would the volatility of landvalues be so different within a metropolitanarea?

To further investigate the relationship be-tween land value and housing supply, Table 4summarizes trends in land values and permitsissued for the construction of new housingunits in the San Francisco Bay Area. Column1 reports the baseline number of owner-oc-

30

This is despite the fact that many have been concernedabout gentrication in urban areas, which is likely to dampenthe volatility.


18/30


FIGURE 4Comparing Our Assessor Data to Freddie Mac’s Conventional Mortgage Home Price Index

cupied housing units by county from the 2000Census, and Column 2 reports the number of new permits for construction of single-familyresidential housing units. The counties areranked by Column 3, which reports the ratioof Column 2 to Column 1. The ratios are

smallest for San Francisco and its adjacentcoastal counties (Marin and San Mateo). Thesame is true if we look at the ratio of all newpermits to all housing units in Column 4. Thisratio is much higher for San Francisco be-cause it includes permits to build apartmentunits. It also includes all housing units in thedenominator, regardless of occupancy status.In the absence of county-level estimates forthe supply elasticity of housing, Columns 3–4 provide a crude proxy for the responsiveness

of housing supply during the boom.Comparing the ratios in Columns 3–4 withthe values of land and property in Columns

5–12 highlights ve interesting trends. 31 First,at the start of the boom period, property val-ues and land values were higher in countieswhere the supply of housing was less respon-sive. This is true whether we look at the me-dian self-reported property values in Column

5, the mean of actual transaction prices in Col-umn 6, or our estimates for mean land valuesin Column 7. Second, land tends to representa smaller share of total property value in coun-ties where the housing supply is more respon-sive (comparing Columns 6 and 7). Third,while the counties with the least responsivehousing supply experienced the largest nom-inal increases in land values during the boom(Column 8), these increases were relativelysmall in percentage terms (Column 9). Fourth,

31 All ve trends are also present in the Boston area. Forbrevity, we provide a table with results in the Appendix.


19/30


FIGURE 5Within Metropolitan Land Value Heterogeneity (left) and Heterogeneity in the Evolution of Land Shares

(right) , San Francisco Bay Area and Boston Area(gure continued on following page)


20/30


FIGURE 5Within Metropolitan Land Value Heterogeneity (left) and Heterogeneity in the Evolution of Land Shares

(right) , San Francisco Bay Area and Boston Area (continued)

during the bust, the counties with the least re-sponsive housing supply experienced thesmallest decreases in land values in both nom-inal and percentage terms (Columns 10–11).

Finally, and perhaps most strikingly, the coun-ties with the least responsive housing supplyhad large net gains in land value between

1998 and 2009, whereas the fastest growingcounties (Contra Costa, Napa, and Solano)lost most of the land value that had accumu-lated during the boom. Overall, these trends

support our initial hypothesis that the BayArea counties with the most volatile propertyvalues and land values during the boom-bust


21/30


T A B L E 4

H o u s i n g U n i t s , P

r o p e r t y V a l u e s , a

n d L a n d V a l u e s i n t h e S a n F r a n c i s c o B a y A r e a , 1 9 9 8 – 2

0 0 9

1

2

3

4

5

6

7

8

9

1 0

1 1

1 2

C o u n t y , S a n

F r a n c i s c o

M e t r o p o l i t a n

A r e a

O w n e r -

O c c u p i e d

U n i t s : 2 0 0 0

( t h o u s a n d s )

S F R

P e r m i t s :

2 0 0 0 – 2

0 0 7


N e w S F R

P e r m i t s

2 0 0 0 –

2 0 0 7 /

B a s e l i n e

O

w n e r -

O c c u p i e d

U n i t s : 2 0 0 0

A l l N e w

P e r m i t s

2 0 0 0 –

2 0 0 7 /

B a s e l i n e

O w n e r -

O c c u p i e d

U n i t s : 2 0 0 0

M e d i a n

P e r c e i v e d

H o u s i n g

V a l u e :

2 0 0 0


M e a n S a l e

P r i c e : 1 9 9 8


M e a n L a n d

V a l u e :

1 9 9 8


M e a n L a n d

V a l u e

C h a n g e :

1 9 9 8 t o

2 0 0 6


M e a n L a n d

V a l u e

C h a n g e :

1 9 9 8 t o

2 0 0 6 ( % )

M e a n L a n d

V a l u e

C h a n g e :

2 0 0 6 t o

2 0 0 9


M e a n L a n d

V a l u e

C h a n g e :

2 0 0 6 t o

2 0 0 9 ( %

) M e a n L a n d

V a l u e

C h a n g e :

1 9 9 8 t o

2 0 0 9


S a n

F r a n c i s c o

7 9 . 5

5

0 . 5 8

0 . 0 1

0 . 0 5

3 9 6

4 0 8

3 5 8

3 2 2

9 0

9 7

− 1 4

4 1 9

S a n M a t e o

1 3 5 . 6 1

5 . 6 7

0 . 0 4

0 . 0 4

4 6 9

4 3 0

3 2 8

3 3 2

1 0 1

− 4 2

6

2 8 9

M a r i n

5 5 . 1

2

3 . 2 7

0 . 0 6

0 . 0 4

5 1 5

4 9 8

2 6 4

3 2 4

1 2 3

− 1 0 2

1 7

2 2 2

S a n t a C l a r a

2 9 1 . 7 7

1 8 . 0

8

0 . 0 6

0 . 0 8

4 4 6

4 4 6

2 8 6

2 7 7

9 7

− 8 2

1 5

1 9 5

A l a m e d a

2 5 1 . 1 7

1 6 . 2

9

0 . 0 6

0 . 0 6

3 0 3

3 1 6

2 0 0

2 7 9

1 3 9

− 1 4 1

2 9

1 3 8

S a n t a C r u z

4 3 . 4

3

4 . 1 7

0 . 1 0

0 . 0 6

3 7 8

2 7 3

2 2 5

2 3 8

1 0 6

− 2 0 0

4 3

3 8

S o n o m a

9 1 . 6

1

1 1 . 7

9

0 . 1 3

0 . 1 0

2 7 3

2 5 0

1 4 8

2 5 5

1 7 3

− 1 5 6

3 9

9 9

C o n t r a

C o s t a

2 1 0 . 3 4

3 4 . 4

1

0 . 1 6

0 . 1 2

2 6 8

3 0 3

1 7 2

2 6 9

1 5 7

− 2 2 6

5 1

4 3

N a p a

2 3 . 4

9

4 . 3 0

0 . 1 8

0 . 1 2

2 5 1

2 3 1

1 3 7

2 9 1

2 1 2

− 2 0 9

4 9

8 2

S o l a n o

7 5 . 9

7

1 3 . 9

4

0 . 1 8

0 . 1 3

1 7 8

1 8 8

1 0 1

2 0 1

1 9 9

− 1 8 0

6 0

2 1

N o t e :

C o l u m n 1 i s b a s e d o n t h e 2 0 0 0 C e n s u s . C o l u m n 2 i s b a s e d o n a n n u a l c o u n t s o f p e r m i t s f o r s i n g l e - f a m i l y r e s i d e n t i a l ( S F R ) c o n s t r u c t i o n r e p o r t e d b y t h e S O C D S b u i l d i n g p e r m i t s d a t a b a s e p r o v i d e d

b y w w w . h

u d u s e r . o

r g . C

o l u m n 3 i s t h e r a t i o o f C o l u m n 2 t o C o l u m n 1 . C o l u m n 4 d i v i d e s t h e t o t a l n u m b e r o f p e r m i t s f o r n e w h o u s i n g u n i t s ( a l l t y p e s ) b e t w e e n 2 0 0 0 a n d 2 0 0 7 b y t h e y e a r 2 0 0 0 s t o c k o f

h o u s i n g

u n i t s ( a l l t y p e s ) . T

h e n u m e r a t o r i s r e p o r t e d b y t h e C a l i f o r n i a S t a t i s t i c a l A b s t r a c t . T

h e d e n o m i n a t o r i s r e p o r t e d b y t h e 2 0 0 0 C e n s u s . C o l u m n 5 i s t h e m e d i a n s e l f - r e p o r t e d h o u s i n g v a l u e b y c o u n t y ,

f r o m t h e 2 0 0 0 C e n s u s . C o l u m n 6 i s t h e a v e r a g e t r a n s a c t i o n p r i c e f r o m o u r c o u n t y a s s e s s o r d a t a

. C o l u m n s 7 – 1 2 a r e b a s e d o n o u r e s t i m a t e s f o r l a n d v a l u e o f t h e a v e r a g e S F R p r o p e r t y i n e a c h c o u n t y .


22/30


T A B L E 5

C o r r e l a t i o n b e t w e e n I n c r e a s e i n L a n d S h a r e ( 1 9 9 8 – 2 0 0 6 ) a n d B a s e l i n e L a n d S h a r e ( 1 9 9 8 )

1

2

3

4

5

6

7

8

9

1 0

D e p . V

a r i a b l e =

( L a n d S h a r e i n

2 0 0 6 –

L a n d S h a r e i n 1 9 9 8 )

S a n F r a n c i s c o

B o s t o n

M i a m i

C h a r l o t t e

L o s A n g e l e s

D e t r o i t

O a k l a n d

P i t t s b u r g h

S a n J o s e

T a m p a

1 9 9 8 L a n d S h a r e

− 0 . 5 7 4 * * *

( − 0 . 0 6 2 )

− 0 . 4 4 8 * * *

( − 0 . 0 4 1 )

− 0 . 5 4 0 * * *

( − 0 . 0 6 0 )

− 0 . 5

2 0 * * *

( − 0 . 0 8 1 )

− 0 . 7 3 2 * * *

( − 0 . 0 2 6 )

− 0 . 6 4 6 * * *

( − 0 . 0 5 1 )

− 0 . 7 1 0 * * *

( − 0 . 0 3 8 )

− 0 . 3 9 3 * * *

( − 0 . 0 8 6 )

− 0 . 7 5 4 * * *

( − 0 . 0 6 8 )

− 0 . 6 3 9 * * *

( − 0 . 0 9 5 )

C o n s t a n t

0 . 3 5 9 * * *

( − 0 . 0 4 8 )

0 . 3 0 6 * * *

( − 0 . 0 2 4 )

0 . 3 8 3 * * *

( − 0 . 0 4 4 )

0 . 2 3 5 * * *

( − 0 . 0 4 5 )

0 . 5 2 8 * * *

( − 0 . 0 2 0 )

0 . 3 7 5 * * *

( − 0 . 0 3 7 )

0 . 4 8 3 * * *

( − 0 . 0 2 9 )

0 . 2 3 7 * * *

( − 0 . 0 5 7 )

0 . 5 1 9 * * *

( − 0 . 0 5 2 )

0 . 3 8 6 * * *

( − 0 . 0 5 8 )

N u m b e r o f C e n s u s t r a c t s

1 0 3

3 5 9

1 4 1

7 0

6 5 0

1 8 0

2 4 7

4 8

1 4 2

1 4 3

N u m b e r o f h o u s i n g t r a n s a c t i o n s

4 7 , 6 8

7

1 8 1 , 6 1 7

1 3 2 , 4 9 8

8 2 , 2 9 1

3 1 2 , 0 5 9

8 8 , 5

7 0

1 7 2 , 7 3 5

2 4 , 5

9 2

7 5 , 8

6 3

1 0 0 , 5 2 6

R - s q u a r e d

0 . 4 5 8

0 . 2 5 5

0 . 3 6 5

0 . 3

7 5

0 . 5 5 4

0 . 4 7 9

0 . 5 9 1

0 . 3 1 2

0 . 4 6 7

0 . 2 4 2

N o t e :

S t a n d a r d e r r o r s i n p a r e n t h e s e s .

* * * p

<

0 . 0 1

.

cycle also had the most elastic housing sup-plies.

Finally, we report an intriguing pattern inour estimates for the ratio of land value to

total property value. We further disaggregateour results to the level of a Census tract andregress the change in the land

Date post:	07-Aug-2018
Category:	Documents
Upload:	john-johnson
View:	213 times
Download:	0 times

LE - The Value of Residential Land and Structures During the Great Housing Boom and Bust

Documents