The Price of Residential Land in Large U.S. Cities

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Price of Residential Land in Large U.S. Cities Morris A. Davis and Michael G. Palumbo 2006-25 NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Price of Residential Land in Large U.S. Cities Morris A. Davis Michael G. Palumbo University of Wisconsin Federal Reserve Board ∗ May 2006 Abstract Combining data from several sources, we build a database of home values, the cost of housing structures, and residential land values for 46 large U.S. metropolitan areas from 1984 to 2004. Our analysis of these new data reveal that since the mid-1980s residential land values have appreciatedoveramuchwiderrangeofcitiesthaniscommonlybelieved. And,since1998,almost alllargeU.S.citieshaveseensignificantincreasesinrealresidentiallandprices. Averagingacross the cities in our sample, by year-end 2004, the value of residential land accounted for about 50 percent of the total market value of housing, up from 32 percent in 1984. An implication of our results is that the future course of home prices — their average rate of appreciation and their volatility — is likely to be determined even more by the course of land prices than used to be the case. ∗ Correspondence to: mdavis@bus.wisc.edu and michael.g.palumbo@frb.gov. This research was conducted while DaviswasattheFederalReserveBoard. WeappreciatecommentsandsuggestionsofferedbyGlennFollette,Jonathan Heathcote, Robert Martin, Patrick McCabe, Raven Saks, and Tom Tallarini. as well as research assistance provided by Elizabeth Ball, Teran Martin, and Tonia Cary. The views expressed are our own and do not necessarily reflect the views of the Board of Governors or the staff of the Federal Reserve System.

1 Introduction In this paper, we extend the methods proposed by Davis and Heathcote (2004) to decompose home values in 46 of the largest U.S. metropolitan areas into the value of housing structures and and the market value of residential land. Thus, this paper introduces new data for studying secular trends and cyclical dynamics of home prices over the period from 1984 through 2004, and emphasizes some new facts about residential land values over the past twenty years that should help in thinking about the future course of home prices around the country. For us, learning about home prices means studying land prices. This is because housing structures can be easily produced and, thus, should be supplied elastically to the market. So, the replacement cost of any existing housing structure should be tightly linked to the costs of building a similar structure – the costs of building materials and wages in the construction industry. By contrast, the land, location, and amenities associated with an existing home (“land”, for short) cannot necessarily be easily reproduced. Land’s relatively inelastic supply means that its market value should largely be determined by demand-side factors, such as household incomes, interest rates, or even speculative activity. Moreover, substantial differences in residential land values across metropolitan areas means that land values can, at times, paint a somewhat different picture of pricing dynamics than home values would seem to imply. Following Davis and Heathcote (2005), our measurement and analytical framework centers hp on the idea that the percentage change in home prices in city j during period t (denoted g ) jt equals the weighted average of the percentage change in construction costs (gcc) and the change in jt residential land prices (g lp ):1 jt g j h t p = ω j l t−1 g j lp t +(1−ω j l t−1 )g j c t c. (1) Here, the weight ωl equals the share of home value accounted for by the market value of jt−1 residential land at the beginning of the period. We take observations on the percentage change in home prices in major MSAs from Freddie Mac’s Conventional Home Price Index (CMHPI), and we obtain construction costs at the city level from data published by R.S. Means Company. To infer the percentage change in land prices, we compute the weights in equation (1) at a benchmark date with estimates based on data on home values and housing characteristics from the Metropolitan American Housing Surveys that are available for 46 large U.S. metropolitan areas (“cities”, for short). Given these estimates at 1If ph denotes an index for home prices in city j for period t, ghp = ph jt −1 . jt jt ph jt−1 1

the benchmark date, we apply a dynamic equation that is compatible with (1) to derive a full time-series for land’s share of home values (back to 1984) based on observed changes in home prices and construction costs in each metro area. Using these new data, we show that across the wide range of cities in our sample — along the coasts and across the nation’s interior — land prices have significantly outpaced construction costs since 1984, driving up land’s share of home value (ωl ) by an average of almost 20 jt percentage points over this period. Although we estimate that residential land accounted for less than a quarter of home value in quite a number of large U.S. metro areas twenty years ago, these days that is true only in Oklahoma City. Another striking result is just how widespread the strength of land prices has been in the current housing boom — taken to have begun at the end of 1998. We show that in 43 of the 46 large metropolitan areas in our sample, a rapid pace of land price appreciation has pushed up land’s share of home value markedly in just the past six years. To be sure, since 1998 land has appreciated at the fastest pace in cities along the East and West coasts, where new residential land was arguably in shortest supply. In these areas — where in 1998 land already accounted for a large share of home value — home prices and land prices tell quite a similar story. But, because in 1998 land was not so expensive in places like Houston, Kansas City, Milwaukee, Minneapolis, Pittsburgh, St. Louis, and Tampa, our data on land prices show the significant imprint that has been left by the recent housing boom — an imprint which is understated to an important extent in data on home prices. We also emphasize that even though residential land has appreciated significantly, on net, over the past past twenty years, for most large metro areas the path has been more of a roller coaster ride than a steady upward march. Indeed, we show that 39 of the 46 cities in our sample have experienced a clear peak in the real residential land price index, and in many of these cities it has taken 10 years or more for land prices to fully recover from their previous troughs. Our point estimates for residential land values and their price indexes are derived using several formulas, different sources of data, and a few assumptions about unobserved quantities. However, our main results are rather robust, as they come from interpreting the sizes of changes in Freddie Mac’s CMHPI relative to changes in construction costs, measured by data from R.S. Means. Consider, for example, the cases of Minneapolis-St. Paul and San Francisco. Few would disagree that residential land is relatively inexpensive in the former city; our methods estimate land’s share of home value in 1984 to have been 0.12 in Minneapolis-St. Paul and 0.75 in San 2

Francisco. Now, suppose that construction costs were flat in real terms in both cities so that equation (1) reduces to g j h t p = ω j l t−1 g j lp t for j = {SF,MSP}. If land prices had increased at about the same rate in both cities lp lp (g = g ), then we would have expected real real home prices to have risen about j=SF,t j=MSP,t 6-1/4 times faster in San Francisco than in Minneapolis-St. Paul after 1984 (ωl /ωl = 0.75/0.12 = 6.25). But according to the CMHPI, from 1984 through j=SF,t−1 j=MSP,t−1 2004, home prices in San Francisco rose only about 3 times as fast as in Minneapolis-St. Paul. Thus, we infer that land values must have risen faster in Minneapolis-St. Paul than in San Francisco. All told, we estimate that by 2004, land’s share of home value jumped by 34 percentage points in Minneapolis-St. Paul — to 46 percent — whereas in San Francisco, the share increased by 13 percentage points to 88 percent. To be sure, Minneapolis-St. Paul is an extreme case from our sample, but it may help to clarify that our main results stem from recognizing that in places where land is relatively inexpensive and when land prices are stable, one would expect home prices to move closely with construction costs. And, if one observes home prices outpacing construction costs in places where land has been relatively inexpensive, land must be appreciating at a fairly rapid clip. The data we bring to bear on these issues is similar to that of Glaeser et al. (2005) and others, but our specific estimation methods differ somewhat from theirs, as do our our points of emphasis and conclusions. For example, Gyourko and Saiz (2004) compare construction costs and home prices around the country, but focus on the distribution of land values within metropolitan areas. A similar emphasis on differences across neighborhoods within MSAs led Glaeser et al. (2004, p. 2) to state that “In the sprawling cities of the American heartland, land remains cheap . . .” In contrast, we focus on the average value of residential land across an MSA, recognizing that our estimation methods implicitly put relatively greater weight on homes in the more expensive neighborhoods in each MSA. And, although land may be cheap for a sizable fraction of the homes in each large city, we report that, on average, land commands a significant share of home value in most of them. Indeed, as shown in table 6 (toward the end of the paper), by year-end 2004, land accounted for just under half of home value in the median metropolitan area in our sample (Denver). And, even among the bottom quartile of cities in our sample, land’s 3

share of home value averaged 29 percent in 2004, up from just 8 percent in 1984. To be sure, land has remained much less expensive across the “heartland” than in cities along the coasts, but over the past two decades — particularly in the past six years — it has become much more expensive just about everywhere. We should note that in this paper we do not take a firm stand on just why land values have soared everywhere, or on whether current or historical valuations look about “right” around the country. Glaeser et al. (2005) and Quigley and Raphael (2005) have argued that zoning restrictions have played a large role in land-price appreciation, at least in some major metro areas. Zoning restrictions would hold down the elasticity of supply of residential land, and thus might explain the surge in the value of land associated with existing homes. Additionally, Campbell et al. (2006) have argued that real interest rates, which have trended down over the past two decades and have been near historic lows in recent years, have also played an important role in stimulating the demand for housing. According with the logic outlined above, we would expect the effects of low interest rates to lead to a particularly rapid appreciation of residential land across the country, although variation in zoning regulations could imply different rates of appreciation in different cities. Further, Davidoff (2005) has argued that the price of land capitalizes the net present value of income opportunities in each metro area, and recent changes to land prices may reflect the extent to which these opportunities have changed. At this point though, we have left to future research an assessment of the quantitative significance of differential zoning regulations, interest rates, or other factors on land valuations across the country. That said, we emphasize the implication of our data and analysis that, with residential land having appreciated so significantly over the past twenty years around the country, the future course of land prices is expected to play an even more prominent role in governing home prices — in terms of average appreciation rates and volatility — in the next two decades. The next section of the paper is a detailed description of our source data and methods for estimating land’s share of home value and generating a constant-quality price index for residential land across large U.S. metropolitan areas. Section 3 reports evidence on the average pace of appreciation and variability of land prices across our sample of metropolitan areas since 1984, with a particular emphasis on the patterns seen in the current housing boom. The final section of the paper discusses the implications of our main empirical results. 4

2 Data and Methods In this section, we describe exactly how we merge different sources of data to compute quarterly time-series estimates, for 46 major MSAs in the United States, of (a) the average value of land as a fraction of average home value and (b) the growth rates of residential land prices (constant-quality). For each MSA, the estimation process occurs in 3 steps that are each discussed in detail below.2 The complete set of data we create and use (except for the R.S. Means data) are available upon request. For reference, the data labels original to each source (CMHPI, R.S Means, and BEA) are listed in table 1. 2.1 Merging house price, construction cost, and household data In the first step, we merge MSA-level data from three different sources. We use the MSA-specific Conventional Mortgage House Price Index (“CMHPI”), produced by Freddie Mac, for information on changes of prices of existing homes; MSA-indexes for the growth and level of construction costs as published by R.S. Means (2004); and an estimate of the number of households in each MSA that we create from data from the Bureau of Economic Analysis (“BEA”) and the Census Bureau. Changes in home prices. The CMHPI is a repeat-transactions price index for existing homes published quarterly.3 Changes in this price index provide an estimate of growth in house prices holding quality roughly fixed between two consecutive periods. Appendix A presents evidence that the published level of the MSA-specific CMHPI contains significant measurement error, and describes a state-space model that we use to filter the quarterly CMHPI for each city.4 Changes in construction costs. The book “Square Foot Costs,” published by R.S. Means, contains time-series price indexes for residential construction costs for most major cities in every state, with annual observations beginning in 1982.5 Since the published index values refer to “January” of each year, we shift the series slightly and relabel the published index value for January of year y as the index value for the fourth quarter of year y−1. We generate a quarterly index for each city by assuming constant quarterly growth rates between years. The indexes can 2We use the words “city” and “MSA” interchangeably, although all our data are for MSAs. 3The raw CMHPI data are available for download at http://www.freddiemac.com/finance/cmhpi/. 4AsshownbyKingandRebelo(1993),thestate-spacemodelencompassesthewell-knownHodrick-Prescottfilter. ThisfilteringdoesnotmateriallyalteranyofourresultsbecausethenoiseintheCMHPIisnotsopronouncedafter 1984. 5As reported in table 1, R.S. Means does not publish construction cost indexes for Oakland and San Jose. For these two MSAs, we use the construction cost index for San Francisco. 5

be combined with time-series information (also from R.S. Means) on residential construction costs nationwide to construct dollar costs-per-square-foot for building single-family homes in each city since 1982. As described below, we merge these square foot construction costs with data from the Metropolitan American Housing Surveys to estimate the value of residential structures in each city. Changes in the number of households. We cannot find time-series on the number of households in each MSA, which we use to proxy the pace of construction of new homes (described later). Instead, we create an estimate by dividing annual data on the population in each MSA, published by the BEA, by annual data on aggregate U.S. household size from the Census Bureau. We convert the data to a quarterly basis by assuming that the annual data refers to the second quarter of each year and by assuming constant quarterly growth rates between years. Our data for household size in the aggregate U.S. comes from Table HH-4, “Households by Size: 1960 to Present,” of the Current Population Survey (CPS) Reports.6 The BEA population-by-MSA data are available in the Regional Economic Accounts, Local Area Annual Estimates, Table CA1-3, “Personal income and population summary estimates,” for Metropolitan Statistical Areas.7 The BEA publish population estimates for all CSAs (“Consolidated Statistical Areas”), MSAs, Metropolitan Divisions, and Micropolitan Statistical Areas. For almost all our cities, we use the MSA estimates; for Los Angeles and Anaheim, Dallas and Fort Worth, and San Francisco and Oakland we use the Metropolitan Division data; and for the New York MSA, we add together the New York-White Plains-Wayne and Nassau-Suffolk Metropolitan Divisions. Of course, the assumption that household size is the same across MSAs is most likely incorrect. However, for our calculations on changes in land prices to be accurate, we only require that the percent change in the number of households is correct, not the actual number of households. 2.2 Creating Benchmark Structures Shares In the next step of the process, we combine micro data for a few key variables from the Metropolitan American Housing Survey, denoted throughout as AHS-M, with data on construction costs from R.S. Means to estimate a benchmark structures share of house value for each city. The specific MSAs surveyed and dates of the survey that are included in our study are 6These reports are available at http://www.census.gov/population/www/socdemo/hh-fam.html. 7These data are available at http://www.bea.doc.gov/bea/regional/reis/. 6

listed in the rightmost column of table 1. For each MSA we use data from the most recent AHS-M, with the exception of New York, Los Angeles, Chicago, Philadelphia, and Detroit. For these cities, we use data from the 1989, 1991, or 1993 AHS-M. For these cities, a specific AHS-M is not collected after 1993, rather the cities are oversampled in the national AHS. We do not use the national AHS because the top-code value for home values has been fixed at $350,000 for some time, and it is therefore quite difficult to reliably calculate average home values in these cities. For example, in the 2003 national AHS more than 40 percent of the observations of owner-occupied single-family detached units in the Los Angeles MSA are top-coded.8 We use the following set of variables from each AHS-M:9 • tenure and nunit2. tenure characterizes the owned/rented/vacant status of the unit. nunit2 specifies whether the structure is single-family detached or attached or in a multiple-unit building. Our sample includes only owner-occupied single-family detached dwellings. • built, cellar, garage, floors, and unitsf. built records the year the structure was built, cellar whether the unit has a partial or full basement, garage indicates whether the unit has an attached or detached garage, floors the number of floors of the structure, and unitsf is the finished square footage of the structure. We use these variables, along with data from R.S. Means, to compute the new building cost of the structure according to the procedure described later in this section. • value. value denotes the self-reported market value of the housing unit. • weight. weight specifies the sampling weight of the unit reported in the AHS-M. We discard from our sample any housing unit that is missing data for any of these 9 variables. In some cases, built brackets the year in which the house was built, in which case the midpoint of the bracket is chosen. Also, cellar had to be recoded slightly: We specify that a housing unit has a basement if it has a basement under all or part of the building, but not a concrete slab, crawl space, or “something else” under the building. Finally, unitsf and value are top-coded at or around the 97th percentile for each city in each AHS-M. We do not adjust the 8The top-coded percentages for this set of homes in the New York, Chicago, Philadelphia, and Detroit MSAs in the 2003 national AHS are 38, 16, 14, and 6 percent, respectively. 9A full description of each of these variables can be found in the AHS codebook. The current codebook can be downloaded from http://www.huduser.org/Datasets/ahs/AHS Codebook.pdf. 7

square-footage of the unit for top-coding but we multiply the top-code of value by 1.5, an adjustment we believe is approximately correct based on results in Davis and Heathcote (2004). The raw unweighted number of observations that meet all of our criteria are listed in table 2. The median number of observations for each AHS-M sample is just under 1,800, with a minimum sample of about 800 (single-family owner-occupied) for the New York metro area and a maximum of more than 2,500 for Salt Lake City. For each AHS-M and each housing unit in our sample, we calculate an estimate of the cost of rebuilding the structure if it were brand new as of the AHS-M date. To do this, we start by estimating a regression equation to approximate the cost per square foot of rebuilding any given housing unit in 2003:Q4 for a single-family home in an average U.S. city (denoted by R.S. Means as the “National 30-city average”). The estimated equation, which primarily is meant to account for the nonlinear relationship between building costs per square foot and the size of a residential structure, takes the form: Predicted cost per square foot = $77.8625+$11.675∗cellar−$4.50∗I(floors ≥ 2) (2) +$0.027∗d∗(1900−unitsf)−$0.008∗(1−d)∗(unitsf −1900). I(.) is an indicator function; it is equal to 1 if the expression in parentheses is true, 0 otherwise. The dummy variable d is set to 1 if the reported square footage of the unit is less than 1900 square feet, 0 otherwise. The predicted cost-per-square-foot equation (2) captures the facts that, as suggested by the R.S. Means data, a basement increases the building cost per square foot by about 15 percent, multiple-story structures cost less per square foot to build than single-story structures, and the average cost per square foot declines with the total square-footage of the unit, with a kink in the rate of decline at 1900 square feet. To summarize, the estimated coefficients in equation (2) provide a parsimonious way to pool the residential construction costs published by R.S. Means for many different sizes of single-family housing structures with different attributes. In our particular application, the cost per square foot in (2) roughly applies to an “average” structure, with three-quarters brick and one-quarter wood veneer, and a basement that is half-finished.10 10R.S. Means estimates cost-per-square-foot for 11 possible values for total square-feet of living area, for each of four housing units of different quality (“Economy”, “Average”, “Custom”, and “Luxury”), for six different styles of structure for each quality (“1 story”, “1-1/2 Story”, “2 Story”, “2-1/2 Story”, “Bi-Level”, and “Tri-Level”), and for multiple exterior wall and basement options. See the R.S. Means book for details. 8

To convert the cost per square foot to a total cost for an average U.S. city at year-end 2003, we multiply the cost-per-square foot from (2) by the reported square footage of the unit and then add $10,000 if the unit has a garage (cost taken from R.S. Means). Finally, to convert the total cost from an average U.S. city in 2003:4 to the appropriate MSA at the date of the AHS-M survey, we multiply this total cost by R.S. Means Index for the AHS-M MSA, date of AHS-M survey . (3) R.S. Means Index for the National 30-city average, 2003:4 An example might help clarify how these calculations work. Suppose we wish to calculate the cost of a new single-family home to be built new in the Washington DC MSA in 1998:2, and suppose the home is 2,500 square-foot, with two stories, a garage, and a basement. According to (2), the nationwide cost-per-square foot in 2003:4 would be $77.8625+$11.675−$4.50−$0.008∗600 = $80.24. (4) And, the total nationwide construction cost in 2003:4 (including the garage) would be $80.24∗2,500+$10,000 = $210,594. (5) Converting the nationwide construction cost in 2003:4 to the cost for the Washington DC MSA in 1998:2 requires applying the DC area’s 1998:2 factor, 110.07 $210,594∗ = $174,286, (6) 133.0 where 110.07 is the (estimated) R.S. Means index value for Washington, DC in 1998:2 and 133.0 is the R.S. Means Index value for the National 30-city average in 2003:4. Once we have calculated the cost of building the structure brand new, we depreciate the structure to better estimate its true replacement cost (or market value of the structure). The way to think about depreciation in this context is that it measures the expense required to bring an existing aged structure up to “like-new” standards. This includes expenditures on physical repairs, such as fixing a roof, as well as expenditures on functional improvements, such as improving the insulation. In our calculations, the depreciation on a structure is simply a function of its age. Let n refer to the new building cost of the structure associated with household i in period t and s i,t i,t refer to the replacement cost of the structure after accounting for depreciation. We calculate 1 agei,t s = n ∗ , (7) i,t i,t (cid:18)1+δ(cid:19) 9

where age is the age of the structure of housing unit i, in years, at date t and δ is the annual i,t rate of depreciation. We set δ = 0.015 and discuss the implications of this choice in Appendix B. Our final step with the AHS-M data, we calculate a benchmark MSA-wide average structures share for the period corresponding to the AHS-M survey date, denoted ωs, as in period t, as t weight ∗s i,t i,t ωs = Pi . (8) t weight ∗value i,t i,t Pi Where weight and value refer to the AHS-M variables associated with housing unit i in i,t i,t period t and the summation in the numerator and denominator is over all households in our included sample for that particular MSA. A nice property of this estimate of ωs is that it does not t require that s and value are exactly accurate for every i.11 Rather, our estimate is consistent i,t i,t even in the presence of additive measurement error in s and value as long as the expected i,t i,t value of the measurement error is 0. That is, if in expectation, homeowners accurately report the value of their house, and, in expectation or on average, our estimates of replacement cost within an MSA are correct, then our estimate of the structures weight in the MSA is not biased. 2.3 Extrapolating Benchmark Structures Shares In the final step of our procedure, we extrapolate our benchmark structures share for each MSA to uncover a quarterly time-series of structures shares. First, remember that we consider the total value of housing in any MSA in any period t, denoted as phh , as the sum of the replacement cost t t of structures in that MSA, pss , and the market value of the land in that MSA, pll , that is, t t t t phh = pss +pll . (9) t t t t t t Next, we use the observation that the total nominal value of structures in an MSA at period t+1, ps s , is equal to the total nominal value in period t, pss , revalued for changes to construction t+1 t+1 t t costs, plus nominal net new structures — that is, new structures less depreciation. We write this identity as ps ps s = pss t+1 +ps ∆s (10) t+1 t+1 t t (cid:18) ps (cid:19) t+1 t+1 t where ps /ps accounts for revaluation due to changes in construction costs and ps ∆s t+1 t t+1 t+1 (cid:0) (cid:1) denotes nominal value of net new structures. 11ThisisunlikeGyourkoandSaiz(2004),whographsomethinglikethedistributionofvaluei,t/si,twithinanMSA. 10

Now, assume that the nominal value of net new structures in an MSA is equal to some proportion, call it θ , of the nominal value of net new housing value in that MSA, denoted t ph ∆h , t+1 t+1 ps ∆s = θ ph ∆h . (11) t+1 t+1 t t+1 t+1 This is the same assumption used by Davis and Heathcote (2004).12 Inserting (11) into (10) produces ps ps s = pss t+1 +θ ph ∆h . (12) t+1 t+1 t t (cid:18) ps (cid:19) t t+1 t+1 t To finish, divide both sides of (12) by the nominal value of housing at t+1, ph h : t+1 t+1 ps ps t+1 s t+1 = ps t s t (cid:16) t p + s t 1 (cid:17) h t +θ ∆h t+1 . (13) ph t+1 h t+1 ph t h t ph t+1 (cid:18)h t+1 (cid:19) t h t+1 (cid:18) ph (cid:19) t Note that we used the identity ph h = phh ph t+1 ht+1 when dividing pss ps t+1 by t+1 t+1 t t (cid:18) ph t (cid:19)(cid:16) ht (cid:17) t t (cid:16) ps t (cid:17) ph h . t+1 t+1 Define the total structures share of aggregate house value in an MSA in period t as ωs = ps t st. By definition — see equation (9) — this equals (1−ωl) from equation (1). t ph t ht t Substituting this into (13) yields ps t+1 ωs = ωs (cid:16) ps t (cid:17) h t +θ ∆h t+1 . (14) t+1 t ph t+1 (cid:18)h t+1 (cid:19) t h t+1 (cid:18) ph (cid:19) t Equation (14) gives us a formula we can use to update our structures share in an MSA from its benchmark value that we calculated in the last section using the AHS-M micro data; that is, given a structures share in period t, ωs, the growth rate of construction costs ps /ps , the t t+1 t (cid:0) (cid:1) growth rate of home prices ph /ph , a value for θ (the structures intensity of nominal net new t+1 t t (cid:16) (cid:17) housing), and a proxy for the growth rate of the real housing stock, structures and land, (h /h ), we can calculate a new structures share, ωs . t+1 t t+1 Notice the implications of equation (14). First, in the absence of growth in the housing stock, i.e. h = h , the structures share in t+1 simply equals the structures share in t, adjusted t+1 t for growth in construction costs relative to house prices. In growing cities, that is, h > h , the t+1 t growth of construction costs relative to existing home prices matters less in determining the 12DavisandHeathcoteassumethat,onaverage,thenominalvalueofstructuresaccountsforroughly87.5percent ofthenominalvalueofnewhousing. Thisestimate,whichwasobtainedduringconversationswithstaffattheCensus Bureau, is based on an unpublished Census study from 1999. 11

structures share next period. Instead, the share of new homes accounted for by structures plays a role, since new homes account for a nonzero fraction of the total stock next period. To implement this equation for each MSA, we start by benchmarking the structures share at the appropriate date to our estimate of the structures share derived from AHS-M data that was detailed earlier. To update this benchmark share — that is, to produce a continuous quarterly time series of the structures share from 1984:4 through 2004:4 according to equation (14)– we use MSA-level construction cost indexes from R.S. Means and the corrected CMHPI for ps /ps t+1 t (cid:0) (cid:1) and ph /ph , respectively.13 t+1 t (cid:16) (cid:17) To complete the updating, we make two more assumptions. First, we assume that (h /h ) t+1 t is proportional to growth in the number of households in an MSA. This assumption is consistent with the Davis and Heathcote (2004) data on the real stock of housing and data from the Census Bureau on the number of households in the U.S.14 Second, we assume that the fraction of new home value accounted for by the structure is exp(3.243∗ωs) θ = t . (15) t 1+exp(3.243∗ωs) t This specification of θ allows developers to vary the land-intensity of new homes with the average t land-intensity in the MSA. Since ωs is, by definition, never less than 0 or more than 1, θ is t t bounded from below by 0.50 and from above by 0.96, and the function is concave between those values (see figure 1). We chose the scale parameter 3.243 such that when the average structures share in an MSA is 0.60, the structures share of new housing is 0.875, values that are consistent with the assumptions in Davis and Heathcote (2005) and roughly consistent with the Census Bureau’s data on construction value put in place.15 For a few midwestern cities early in the sample period, our algorithm implies near-zero point estimates for land’s average share of home value; we set land’s share to 0.05 in these few cases. Finally, we use a transformation of equation (1) to compute percent changes for a 13We should note that it is unclear that changes to the R.S. Means construction cost indexes fully incorporate changes in builders’ margins. If not, fluctuations in builders margins will be attributed to the value of land. We’re confident that our results are not importantly affected by this consideration. 14According to the Davis and Heathcote data, the aggregate real stock of housing grew at 1.45 percent per year between 1975 and 2003; for comparison, the aggregate number of households grew 1.51 percent per year over the same time period. 15Our results would be qualitatively similar if we simply set θt =0.875 for all MSAs. 12

Figure 1 Assumed Relationship between Structures Share of Home Value for Existing Homes (ωs) t and for Newly Built Homes (θ ) t exp(3.243∗ωs) θ = t t 1+exp(3.243∗ωs) t 0.95 0.85 0.75 0.65 0.55 0.45 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 0.85 0.90 0.95 Structures Share, Existing Homes semoH weN ,erahS serutcurtS Note. This figure plots the function used in this study to map the share of home values accounted for by residential structures for the stock of existing homes to the share for newly built homes. (constant-quality) index of residential land prices: 1 g j lp t = ωl [g j h t p −(1−ω j l t−1 )g j c t c]. (16) jt−1 lp g is the value-weighted average growth rate of residential land containing the existing stock of jt homes in MSA j between periods t−1 and t. As long as the growth rate of construction costs gcc jt hp lp and home prices g are derived from constant-quality price indexes, then g is, by construction, jt jt lp a constant-quality growth rate. Note that g is not a “dollars-per-acre” concept, nor is it jt lp necessarily related to growth in the price of farmland on the outskirts of an MSA. g simply jt tracks the growth rate of the price of the combined set of attributes of existing homes that make 13

these homes more expensive than the replacement cost of their structures, including premiums for location and other local amenities. 3 Results The algorithm of the previous section results in a database of quarterly observations on the components of home values from 1984 through 2004 for 46 large U.S. metropolitan areas. More specifically, we have estimated average values for the stock of single-family, owner-occupied homes and their structure and land components, and we have constructed price indexes for residential land, as well.16 In this section, we describe the basic trends uncovered by these new data, focusing on 5 broad geographic regions — cities in the Midwest, Southeast, Southwest, and along the East and West coasts.17 The data show some variation across cities within these regions, but the regional variation is predominant. We describe changes in the components of home value from 1984 through 1998, then focus on the housing boom that has affected most of the country since 1998. After documenting the trends in land values since 1984, we show that most cities across the country have experienced a significant land pricing-cycle since the mid-1980s, in which the real price of residential land reached a significant peak followed by a long period of recovery. In large cities in the southwest, the peak occurred around 1985 — essentially following a boom in energy production in that region; in other cities, the peaks were around 1990. Only for a handful of large midwest cities have real residential land prices exhibited a fairly steady upward march over the past two decades. 3.1 Components of home value in 1984 Table A shows that in the mid-1980s homes were, on average, much less expensive in large U.S. cities in the Midwest and the Southeast than along the East and West coasts. The regions differed little in terms of their average replacement cost of residential structures, but there were large regional differences in the value of residential land. In 2004 dollars, the average residential 16These data are available upon request. In the text, we include tables reporting data at the regional level; tables 3 through 6 at the end of the paper include data for all 46 cities in our sample, and they show how the cities are distributed among the 5 broad geographic regions. 17In aggregating to the regional and full-sample level, we report simple averages across cities — not weighting by populationorhomevalue. Thedistributionofhomevaluesissufficientlyskewedthatweightedaverageswouldclosely resemble the patterns shown for the cities located just along the East and West coasts. 14

lot in 1984 was worth just $14,000 in the Midwest, $135,000 along the West Coast, and $62,000 across our entire sample of large cities.18 As of year-end 1984, on average, residential land accounted for just 11 percent of home value in cities in the Midwest, 55 percent of value in cities along the West Coast, and 32 percent of value across our full sample. Table A Components of Home Value in 1984 by Geographic Region Memo: Home Structure Land Land’s share Region value value value of value ($1000s) ($1000s) ($1000s) (percent) a. Midwest 120 106 14 11 b. Southeast 129 94 36 27 c. Southwest 158 100 58 35 d. East Coast 172 105 67 38 e. West Coast 226 91 135 55 f. Full sample 162 100 62 32 2004 dollars. Unweighted averages across sample-cities in each region. Components may not sum to totals due to rounding. 3.2 Changes in home value, 1984 through 1998 The following table B documents the cumulative changes in the components of home value between 1984 and 1998 in the 5 geographic regions. In real terms — that is, relative to the core PCE price index — homes became considerably more valuable in 4 of the 5 regions — the exception being in cities in the Southwest.19 In real terms, average home values in the Southwest and the share of home value accounted for by the market value of residential land was lower in 1998 than in 1984. By contrast, the two other regions of the country that had relatively low home values in 1984 — the Midwest and the Southeast — experienced significant increases, on net, over the next 15 years, and the lion’s share of those increases can be traced to very fast appreciation of residential land. Indeed, as reported in the table, land’s share of home values rose by 16 percentage points and 9 percentage points, respectively, in Midwest and Southeast cities from 1984 through 1998. Appreciating land values also pushed up home values, in real terms, in cities 18We convert to 2004 dollars using the BEA’s chain-weighted price index for personal consumption expenditures excluding food and energy items. 19Inthatregion,veryhighenergypricesfromthelate1970shadprovidedasubstantialboosttoeconomicactivity, and, based on the Freddie Mac data, evidently, resulted in an exceptional pace of home price appreciation ending in the mid-1980s. 15

along the East and West coasts, but the average increases in land’s share of home value — 3 and 6 percentage points, respectively, over this period — were not as large as in the Midwest and Southeast. Looking across all the large cities in our sample, the real value of average residential lots increased 50 percent from 1984 through 1998, and land’s share of home value increased 8 percentage points, from 32 percent to 40 percent. Table B Change in Components of Home Value by Geographic Region — 1984 through 1998 Cumulative change in: Change in Home Structure Land land’s share Region value value value of value (pct) (pct) (pct) (pctg pts) a. Midwest 26 2 208 16 b. Southeast 14 0 53 9 c. Southwest -9 -4 -17 -4 d. East Coast 24 8 49 3 e. West Coast 39 20 51 6 f. Full sample 22 5 48 8 In real terms; unweighted averages across sample-cities in each region. 3.3 Changes in home value, 1999 through 2004 Table C indicates how widespread across the country the recent housing boom has been. All 5 regions have seen substantial real increases in average home values since 1998 — about 25 percent (cumulatively) in large cities in the Midwest, Southeast, and Southwest, and around 80 percent along the East and West coasts. In addition, although construction costs around the country have generally outpaced consumer price inflation — leading to increases in the real value of residential structures on the order of 10 to 18 percent since 1998 — the more important story has been a widespread rapid appreciation of residential land. We estimate increases in the market value of residential lots around 50 percent in the Southeast and Southwest, 75 percent in the Midwest, and around 125 percent along the East and West coasts. Thus, land’s share of home value has risen considerably in each of the 5 regions of the country, up 7 to 10 percentage points in the South and Midwest and 13 or 18 percentage points along the coasts. Indeed, among the 46 large cities in our sample, only Charlotte and Salt Lake City show lower land shares of home value in 2004 than in 1998, and Memphis’s share only edged up by 1 percentage point. Since 1998, the largest increases in land’s share of home value were registered in 16

Providence, RI (26 percentage points), New York City (23), Minneapolis/St. Paul (21), St. Louis (18), and Washington, DC (18). In St. Louis, land’s 30 percent share of home value was still well below our sample-average (51 percent), but was appreciably greater than the 12 percent share recorded just six years earlier. Since 1998, home values in St. Louis rose 34 percent in real terms — well below the sample-average pace — but the relatively low value of residential lots in 1998 led this to translate into more than a 200 percent cumulative increase in the real value of residential land — right up there with the other fastest increases in our sample (Sacramento and San Bernardino, CA, and Providence, RI). Table C Change in Components of Home Value by Geographic Region — 1999 through 2004 Cumulative change in: Change in Home Structure Land land’s share Region value value value of value (pct) (pct) (pct) (pctg pts) a. Midwest 28 9 75 10 b. Southeast 26 15 45 7 c. Southwest 24 10 52 8 d. East Coast 77 14 115 18 e. West Coast 81 18 145 13 f. Full sample 56 13 105 11 In real terms; unweighted averages across sample-cities in each region. 3.4 Components of home value in 2004 As can be seen in table D, by year-end 2004, single-family owner-occupied homes remained much more expensive in cities along the East and West coasts ($376,000 and $568,000, respectively) than in the other regions of the country, where the average was near $185,000. Our estimates of the value of residential structures for homes along the coasts were not much greater than those for the other 3 regions, so that nearly all of the difference in home values reflected differences in the value of their land components. The average lot was worth about $75,000 in cities in the Midwest, Southeast, and Southwest, but was valued at $245,000 on the East Coast, and $440,000 in West Coast cities. At year-end 2004, we estimate that land’s share of home value had risen to 75 percent along the West Coast, 65 percent on the East Coast, compared with about 40 percent in the other 3 regions and 51 percent across the entire sample of 46 cities. 17

Still, despite the wider differences in home values across the country in 2004, we find that 4 of the 5 regions saw substantial increases in land values and land shares since 1984. Midwest cities saw their share rise to 36 percent from just 11 percent twenty years earlier — the largest percentage-point increase of the 5 regions — and these cities saw the largest cumulative increase in average land values as well, averaging more than a four-fold increase over the twenty-year period. On net, the slowest average rates of increase in home and land values were found for cities in the Southwest, and, by our estimates, there were several cities in that region for which average home values in 2004 remained below their 1984 levels (in real terms) — Dallas, Fort Worth, Houston, Houston, Oklahoma City, and San Antonio. Overall, though, we estimate that, on average, real land values rose 26 percent since 1984 in our Southwest cities, and land’s share of home value edged up 4 percentage points, on net, to 38 percent at year-end 2004.20 Table D Components of Home Value in 2004 by Geographic Region Memo: Home Structure Land Land’s share Land’s share Region value value value of value in 1984 ($1000s) ($1000s) ($1000s) (percent) (percent) a. Midwest 192 119 73 36 11 b. Southeast 187 108 79 42 27 c. Southwest 179 106 73 38 35 d. East Coast 376 131 245 64 38 e. West Coast 568 128 440 74 55 f. Full sample 307 120 187 51 32 Unweighted averages across sample-cities in each region. 3.5 Changes in the distribution of land’s share of home value, 1984 through 2004 The widespread net increase in land’s share of home value across the U.S. since the mid-1980s is evident in figure 2, which shows the cumulative distribution function of land’s share of home value across the 46 large cities in our sample as of year-end 1984, 1998, and 2004. As was consistent with the relatively fast appreciation of real land values in the Midwest and Southeast from 1984 through 1998, figure 2 shows a relatively large rightward shift in the distribution of land share for 20These numbers were boosted by real increases in residential land values in New Orleans, Denver, and Salt Lake City,whichweincludedintheSouthwestgroupingbasedontheirsimilartime-seriespathsforlandandhomevalues (discussed below). 18

cities in the lower two-thirds of the distribution. By contrast, for cities with the largest land shares, the line segment in 1998 lies just about on top of the 1984-segment, indicating that cities shuffled their order at the top of the distribution in that period; but, overall, there was not a material net increase in land’s share in the most expensive cities. Between 1998 and 2004, the entire distribution function for land’s share of home value shifted noticeably to the right, with somewhat larger increases generally occurring among cities in the top half of the distribution. At year-end 2004, the average share of home value we attribute to residential land ranged from a low of about 25 percent in Oklahoma City to nearly 90 percent in San Francisco. The range from lowest to highest is about the same as in 1984, as land’s share of home value was less than 5 percent in a handful of cities in the middle of the country — running from Buffalo down to Pittsburgh and over to St. Louis, for example — and reached about 75 percent in San Francisco and Anaheim. 3.6 Changes in the distribution of residential land values, 1984 through 2004 Figure 3 shows how far the distribution of average real land values shifted between 1984 and 1998, and then again over the past six years. Consistent with the patterns evident in figure 2, real land values in cities in the lower half of the distribution can be seen to have shifted by proportionately more from 1984 to 1998 (note the log scale for the x-axis). Although the entire distribution shifted further to the right between 1998 and 2004, in recent years the disproportionate increases in real land values occurred in cities in the top half of the distribution. 3.7 Volatility of real land prices since 1984 The previous subsections have emphasized net changes in the components of home value, in real terms, over a rather long period of time — 1984 through 1998 — and in the current housing boom — 1999 through 2004. In the course of that discussion, we mentioned that real land and home values in large cities in the Southwest have taken quite a roller coaster ride, and it was not until the early 2000s that many of those cities saw their average real home values return to levels last registered in the mid-1980s! This subsection emphasizes that the majority of large cities in other regions of the country has also experienced significant and prolonged decline in real land prices — generally in the latter 1980s or early 1990s, when national indexes of existing home prices fell in real terms. 19

Figure 2 Cumulative Distribution of Land’s Share of Home Value across Metropolitan Areas in Selected Years .Fraction of U.S. metro areas 1.00 1.00 0.75 0.75 1984 1998 2004 0.50 0.50 0.25 0.25 0.00 0.00 0.00 0.20 0.40 0.60 0.80 1.00 Land’s share o.f home value Note. This figure plots cumulative distribution functions of land’s share of home value across our sample of 46 large metropolitan areas in 1984, 1998, and 2004. Real land prices in the Southwest after 1985. Figure 4 plots indexes of real land prices across 9 cities in the southwestern U.S. that experienced a peak near early-1985. The indexes are normalized so that their value in 1985:Q1 is 100, and separate indexes are shown for the median city in each quarter after 1985:Q1 (the black line) and for the cities representing the 20th and 80th percentiles of the distribution (the blue and red lines).21 According to figure 4, the median city in this group — Houston — saw its land price index fall 50 percent, cumulatively, in real terms, over the five years ended in 1989. Although real land prices in Houston began rising gradually in 1990, our estimates imply that the index did not fully return to its early-1985 level 21The 9 cities are: Dallas, Denver, Fort Worth, Houston, New Orleans, Oklahoma City, Phoenix, Salt Lake City, and San Antonio. For Phoenix, the index is set to 100 in 1986:Q1 because 1985 saw a decent-sized increase in real landpricesthere. Notethatthe20thand80thpercentilesarecomputedforeachquarter,andthereisalittleshuffling among cities in the distribution over the time period shown. 20

Figure 3 Cumulative Distribution of Residential Land Values across Metropolitan Areas in Selected Years F. raction of U.S. metro areas 1.00 1.00 0.75 0.75 1984 1998 2004 0.50 0.50 0.25 0.25 0.00 0.00 10 50 200 350 500 800 Land v.alues (Thousands of 2004 dollars per home; log scale) Note. This figure plots the cumulative distribution functions of average real residential land values across our sample of 46 large metropolitan areas in 1984, 1998, and 2004. until 1999 — 15 years later! Denver’s experience is reflected in the red line: There, real land prices fell, cumulatively, by 60 percent from 1985 through 1991; however, the recovery in that city was much sharper, and by the mid-1990s Denver’s index of real land prices had returned to its 1985-level. By 1999 (the last period shown in figure 4), the index of real land prices was two-and-a-half times as high as it had been 15 years earlier. By contrast, San Antonio — whose experience is reflected in the blue line — saw a remarkably large drop in real land prices, and by 1999 the level of the index in that city had recovered only about halfway. Indeed, we estimate that after a fairly rapid period of appreciation from 1999 through 2004, the index of real land prices in San Antonio finally returned to its 1985-level. Peaks in real land prices in cities elsewhere across the U.S. Moving beyond the 9 Southwest cities in which real land prices peaked around 1985, 30 of the remaining 37 cities in our sample 21

Figure 4 Real Residential Land Prices in Southwest Metropolitan Areas after 1985 C . ities with a peak near 1985 Index = 100 in 1985:Q1 250 250 Quarterly data 200 200 150 150 80th percentile 100 100 Median 50 50 20th percentile 0 0 0 5 10 15 Years aft.er 1985 Note. This figure plots the 20th, 50th, and 80th percentiles of the distribution of real land prices in 7 Southwestcitiesoverthefifteenyearsfollowingthepeakexperiencedaroundearly1985. For6ofthe7cities inthisgroup,theindexofrealresidentiallandpricesisnormalizedto100in1985:Q1; forPhoenix,theindex is set to 100 in 1986:Q2. experienced a peak sometime after 1986 — figure 5 uses a “butterfly chart” to summarize those episodes. To generate figure 5, we identified for each of these 30 cities the quarter in which their real land price index reached a “local” peak, normalized the level of the price index in the peak-quarter to 100, and then computed the relative level of the index in all quarters around the peak. The black line is the median normalized index among the 30 cities, and the blue and red lines, respectively, denote the 20th and 80th percentiles across the distribution of cities at each quarter surrounding their respective peaks. The left-hand portion of the graph represents the behavior of real residential land prices three years before the peak-quarter, and the right-hand portion shows prices in the three years following the peak. Thus, considering the path of the “median” line, figure 5 reveals that 15 of the 30 large U.S. 22

Figure 5 Real Residential Land Prices around Previous Peaks C . ities with a peak after 1986 Index = 100 at peak 110 110 Quarterly data 100 80thpercentile 100 Median 90 90 80 80 20thpercentile 70 70 60 60 50 50 40 40 30 30 -3 -2 -1 0 1 2 3 Years aro.und peak Note. Thisfigureplotsthe20th,50th,and80thpercentilesofthedistributionofreallandpricesfor30cities that experienced a peak in between 1987 and 1992. The figure shows the paths for real land prices from three years before a peak to three years after the peak. For each of the 30 cities in this group, an index of real land prices is normalized to 100 in the peak-quarter. cities in this broad group have experienced, at some point since 1986, a cumulative, net three-year decline in real land prices of 16 percent or more. This broad set of cities includes Boston (a 24 percent three-year decline through 1991:Q4), Kansas City (30 percent, 1990:Q3), Los Angeles (19 percent, 1992:Q4), New York (28 percent, 1991:Q2), Sacramento (24 percent, 1993:Q4), San Diego (15 percent, 1993:Q1), San Francisco (18 percent, 1992:Q4), St. Louis (26 percent, 1990:Q3) and Washington DC (12 percent, 1992:Q4). Figure 5 does not show the full recovery period for this group of cities, but for the median city (Tampa) it took a full ten years for the real land price index to return to the level at its previous peak. In a number of large cities — including Los Angeles, Philadelphia, Providence, RI, and Sacramento — real land prices did not reach their 1990 peaks until 2001 or 2002, well into the current housing boom. Considering the portion above the median in figure 5, 15 cities in our sample experienced a 23

relatively mild cycle for land prices around 1990 — their cumulative real decline was generally less than 10 percent and the level of their real land price index had returned to its peak level by the mid-1990s. Indeed, by the time the current housing boom was getting underway toward the end of 1998, their real land price index was considerably above the level at the time of the previous peak. This group includes Charlotte, Detroit, Memphis, Miami, and Minneapolis-St. Paul. With the exception of Charlotte and Memphis — where land prices have languished in real terms since 1998 — this group of cities continued to see a rapid expansion of residential land values through 2004. We note that for most of these cities that experienced a peak in real residential land prices around 1990, the subsequent real depreciation involved a stagnation of land prices in nominal terms that was eroded over time by an increase in core consumer prices. That is, the price index for personal consumption expenditures excluding food and energy items in the National Income and Product Accounts (NIPA) — which is the index we use to convert nominal values and price indexes into real terms — rose about 15 percent over three-year periods from 1989 through 1994. This is about the same order of magnitude as our estimate of real peak-to-trough declines in residential land prices for most of these cities, so our data do not suggest widespread, outright nominal declines in land prices. Still, the minority of cities in this group that are estimated to have experienced real land-price declines around 20 percent are also estimated to have seen their nominal land-price indexes fall in the peak-to-trough period. Midwest cities that have not experienced a previous peak in land prices. According to our estimates, 7 large cities in the midwest have seen a more smooth upward march in real land prices and average land values since 1984, rather than the roller coaster experience of the majority. This group, which includes Chicago, Cincinnati, Indianapolis, and Milwaukee, registered increases in home prices that outpaced construction costs and general price inflation year after year since 1984. In general, for cities in this group, land accounted for a small portion of home value in 1984 — about 10 percent. By 1998, however, land’s share of home value had risen to 30 percent, and, by 2004, the share in these cities had nearly reached 40 percent, not too far below the average across all cities in our sample. 4 Discussion This paper has introduced methods we developed to build a new database for measuring the evolution of residential land values across large U.S. metropolitan areas since the mid-1980s. We 24

have not yet used the data to estimate models capable of explaining just which economic factors have caused changes in land prices in different areas at different times, but we have documented, for the first time, some key facts that a model would need to explain. In particular, we have shown that, over the past twenty years, residential land has become relatively more expensive in just about every large metro area in the U.S. — not only in places along the east and west coasts of the country, as some have suspected — though the pace of appreciation has, of course, varied considerably from region to region. Moreover, we have demonstrated that the current housing boom, which began around the end of 1998, has left its imprint in the form of a rapid appreciation of residential land values just about everywhere. In addition, we have shown that, at some point since 1984, the majority of large U.S. cities have experienced one pronounced price-cycle in which residential land lost value for an extended period of time, usually following several years of particularly rapid appreciation. In real terms, land prices have generally taken several years to go from peak to trough, and the subsequent recovery from these price-declines has generally occurred at a more gradual pace. To us, the most important implication of our findings is that, looking forward, cycles in land prices will shape the contour of home values to a greater extent than they have in the past — because in just about every large U.S. metro area land’s share of home value is now much higher than it used to be. More specifically, land’s greater share of home value could mean faster home-price appreciation, on average, and possibly larger swings in home prices. To gauge the possible magnitudes, we consider how current land values would translate into future home-price appreciation in cities along the East and West coasts should land prices and construction costs repeat their average performance (in real terms) in recent history. From 1984 through 1998 (ignoring the current boom), these two regions experienced average annual real increases in land prices of 4.2 percent and 4.7 percent, respectively; over the same period, their real construction costs fell by an average of 0.3 percent and 0.8 percent, respectively. In 1984 and 2004, land accounted for 38 percent and 64 percent of home value, on average, in large cities along the East Coast; in cities along the West Coast, land’s share was 55 percent in 1984 and 74 percent in 2004. In table E, we plug these values into equation (1) to compute, for each region, the percentage increase in home prices resulting from a repeat-experience of land prices and construction costs from 1984 through 1998. Our calculations imply that simply by taking into account the more expensive land values currently in place we would expect real home prices to accelerate by more than 1 percentage point per year in cities along both coasts. So, even if land 25

prices were to increase from now on at the average pace seen before the current boom, home prices might rise more quickly, on average, than they did before. Table E Effect of Higher Land Share on Prospective Home-Price Appreciation Using Land’s Share in 1984: East Coast: 1.4% = 0.38∗(4.2%)+(1−0.38)∗(−.3%) West Coast: 2.2% = 0.55∗(4.7%)+(1−0.55)∗(−.8%) Using Land’s Share in 2004: East Coast: 2.6% = 0.64∗(4.2%)+(1−0.64)∗(−.3%) West Coast: 3.3% = 0.74∗(4.7%)+(1−0.74)∗(−.8%) Acceleration in Home Prices from Higher Land Shares: East Coast: 1.2 ppt = 2.6%−1.4% West Coast: 1.1 ppt = 3.3%−2.2% The consequences for future home-price volatility could be just as significant because we would expect cycles in home prices to continue to be driven by cycles in real land prices. Again, in our framework, variance of home prices depends on the variances of land prices and construction costs, and the greater current share of home value accounted for by residential land has significantly pushed up the weight on land-price volatility.22 Of course, it is possible that some of the factors driving up residential land prices so significantly over the past twenty years could also work to decrease their volatility, which would offset the simple “accounting effect” of land’s greater share of home value. We see this to be an important avenue for future research. 22There is a positive covariance over time between real land prices and construction costs that also affects the variance of home prices. 26

References [1] Campbell, S., Davis, M., Gallin, J. and R. Martin (2006), “What Moves Housing Markets,” mimeo. Available at: http://www.morris.marginalq.com/rentprice-final.pdf. [2] Davis, M., and J. Heathcote (2004), “The Price and Quantity of Residential Land in the United States,” Finance and Economics Discussion Series 2004-37, Federal Reserve Board. More recent version at: http://www.morris.marginalq.com/2005-10-Davis-Heathcote-Land.paper.pdf. [3] Davidoff, T. (2005), “A House Price is not a Home Price: Land, Structures, and the Macroeconomy,” mimeo. [4] Davis, M., and J. Heathcote (2005), “Housing and the Business Cycle,” International Economic Review 46(3), pp. 751-784. [5] Glaeser, E., Gyourko, J. and R. Saks (2005), “Why Have Housing Prices Gone Up?” NBER Working Paper no. 11129. [6] Gyourko, J., and A. Saiz (2004), “Is there a Supply Side to Urban Revival?” mimeo. [7] King, R., and S. Rebelo (1993), “Low Frequency Filtering and Real Business Cycles,” Journal of Economic Dynamics and Control 17, pp. 207-233. [8] Quigley, J. and S. Raphael, 2005, “Regulation and the High Cost of Housing in California,” American Economic Review, forthcoming. 27

Appendix A Measurement Error, CMHPI As noted in the text, the MSA-level CMHPI seems to be measured with significant error, and the measurement error is responsible for much of the observed volatility in the house price indexes. For example, as shown by Davis and Heathcote (2004) for the national CMHPI, measurement errors in the level of the index can explain the high degree of negative autocorrelation in percentage changes computed from the series. To see this, suppose that the log of the observed housing price index, log ph , is equal to the log of the true price index, log ph∗ , plus some i.i.d. t t (cid:16) (cid:17) (cid:16) (cid:17) measurement error e ∼ N 0,σ2 , that is t e (cid:0) (cid:1) log ph = log ph∗ +e . (17) t t t (cid:16) (cid:17) (cid:16) (cid:17) The first difference of (17) is ∆log ph = ∆log ph∗ +∆e . (18) t t t (cid:16) (cid:17) (cid:16) (cid:17) The left-hand side of (18) is the observed growth rate of the price index and, depending on the properties of ∆log ph∗ , the observed growth rate could be negatively autocorrelated since ∆e t t (cid:16) (cid:17) will be negatively autocorrelated. To purge the national OFHEO price index of measurement error, Davis and Heathcote (2004) detrend the OFHEO for overall consumer price inflation,23 assume that the true real growth rate of the OFHEO is a random walk (∆log ph∗ = u with u an i.i.d. draw from t t t (cid:16) (cid:17) N 0,σ2 ), and finally assume that u is uncorrelated with e for all t and s. Given this u t s (cid:0) (cid:1) framework, Davis and Heathcote estimate the variance of e and u using the Kalman Filter. t t They uncover the real sequence of ph∗ using the Kalman Smoother, and convert the real sequence t to a nominal sequence by adding back consumer price inflation. Since we measure land prices residually — that is, we use data on house prices and construction costs to infer land prices — any measurement error in the metro-area CMHPI price indexes will feed through to our land price series. We feel it is important to generate, as best as possible, a measurement-error free version of the CMHPI, and, thus, we apply the Davis and Heathcote procedure to data for each MSA. In estimation, we allow for a break in the variance of 23DavisandHeathcotedetrendbythepriceindexforpersonalconsumptionexpendituresexcludingfoodandenergy in the National Income and Product Accounts (“NIPA”) as published by the BEA, line 23 of NIPA table 2.3.4. 28

the measurement error in each series anywhere between 1980:1 and 1992:4 for each MSA, and choose the break date to maximize the estimated log-likelihood of the sample. Given our estimates of σ2 and σ2 (before and after the break date), we construct a measurement-error free u e series of the real level and growth rate of the CMHPI using the Kalman Smoother. We call these estimates our “corrected” real estimates, and convert the corrected real estimates to nominals by factoring in the NIPA price index for personal consumption expenditures excluding food and energy (the “core PCE price index”).24 Three results stand out from our work. First, estimated break-dates for the measurement error process typically occur in the mid 1980s: The median break-date is 1984:2, which is why the analysis in our paper begins in 1984. Second, for almost all MSAs covered by the CMHPI, the variance of the measurement error is much larger in the earlier part of the sample than in the latter part. The median ratio of the standard deviation of the measurement error in the early part of the sample to the standard deviation in the later part is 6.3. Third, much of the variance in the observed growth rate in the MSA-specific CMHPI data seems to reflect measurement error. For example, we find that in the period from 1992:1 through 2004:4, the median ratio of the standard deviation of the error-corrected growth rates of home prices to the standard deviation of the published growth rate is 0.51. B Checking δ Obviously, our estimates of the land and structures share of homes in each MSA is sensitive to our assumptions, but some sensitivity analyses we have run suggest that our choice of δ is potentially important. Specifically, if we were to have chosen a lower (higher) value for δ, our estimated replacement cost of structures would account for a higher (lower) fraction of house value. Taking the new building cost data from R.S. Means as accurate, we can justify our value of δ in three ways. First, δ = 0.015 is almost exactly the value used by the BEA when it constructs its estimate of aggregate stock of residential structures (Davis and Heathcote 2005). Second, a lower estimate of δ will imply that the value of land, on average, was negative in many Midwestern cities in the mid 1980s. Third, we can approximate the aggregate land share using our MSA-level data and a back-of-the-envelope formula; a value of δ = 0.015 gives a back-of-the-envelope estimate that is 24Ourestimatesofσ2 andσ2foreveryMSA-levelCMHPIindexpublishedbyFreddieMac(includingmanyoutside u e the 46-city sample used in this paper) are available upon request. 29

quite close to the more carefully constructed estimate produced by Davis and Heathcote (2004). To make these calculations, we first estimate the number of households in single-family owner-occupied housing units in each MSA by multiplying the percentage of households in each MSA that live in single-family owner-occupied housing (derived from AHS-M weights) by our estimate of the total number of households. This estimate is shown in the first column of table A.1. Next, we multiply the number of households living in single-family owner-occupied housing units by the average value of these houses (and the land associated with these houses) to derive the total value of housing and land for these homes in each MSA. The third column in table A.1 lists our estimate of the aggregate value of housing, by MSA, for the stock of single-family owner-occupied homes in 2000:2. The fourth column shows the fraction of U.S. total home value (for single-family owner-occupied homes) that is accounted for by each MSA; for each MSA, it is calculated as the value in the third column divided by $9,037.3 billion.25 In 2000, our sample of about 24 million households (44 percent of the total number of households in single-family owner-occupied housing units) accounts for $5,066 billion in house value, about 56 percent of U.S. total value. The average price of the housing units in our sample of MSAs is $212,000. According to the 2000 Decennial Census of Housing, the average home value in the U.S. for the single-family owner-occupied stock was $167,000 in 2000:2, implying the average price of homes that are not included in our sample was $109,000, about the average value of homes in Buffalo in 2000. Our aggregate land share will be 0.56∗w¯l +0.44∗X, where w¯l is the value-weighted average t t share of house value attributable to land in our sample of MSAs and X is the fraction of house value attributable to land for the 44 percent of aggregate house value for which we do not know land’s share. We find that w¯l is about 50.2% in 2000:2. For the aggregate land share to be equal t to 40 percent in 2000 (about the estimate we get using the Davis and Heathcote method for the single-family owner-occupied stock), X must be 27 percent, approximately the same as land’s share in Houston 2000:2. 25According to calculations using the 2000 Decennial census, the market value of single-family owner-occupied homes inthe entire U.S. was$9,037.3 billion. For reference, usingdata fromthe 1990Census, weestimate the value of single-family owner-occupied homes in the U.S. to have been $5,508 billion; in 1990:2, the estimated value of housing across our sample of MSAs is $3,119 billion. 30

Table 1 List of Data Sources and Data Labels CMHPI R.S.Means BEAPopulation AHS-M AHS-Mdate ORANGECOUNTYCAPMSA Anaheim SantaAna-Anaheim-Irvine,CAMetropolitanDivision Anaheim-SantaAna,CAPMSA** 2002 ATLANTAGAMSA Atlanta Atlanta-SandySprings-Marietta,GA(MSA) Atlanta,GAMSA 1996 BALTIMOREMDPMSA Baltimore Baltimore-Towson,MD(MSA) Baltimore,MDMSA 1998 BIRMINGHAMALMSA Birmingham Birmingham-Hoover,AL(MSA) Birmingham,ALMSA 1998 BOSTONMA-NHPMSA Boston Boston-Cambridge-Quincy,MA-NH(MSA) Boston,MA-NHCMSA 1998 BUFFALO-NIAGARAFALLSNYMSA Buffalo Buffalo-NiagaraFalls,NY(MSA) Buffalo,NYCMSA** 2002 CHARLOTTE-GASTONIA-ROCKHILLNC-SC Charlotte Charlotte-Gastonia-Concord,NC-SC(MSA) Charlotte,NC-SCMSA 2002 CHICAGOILPMSA Chicago Chicago-Naperville-Joliet,IL-IN-WI(MSA) Chicago,ILPMSA 1991*** CINCINNATIOH-KY-INPMSA Cincinnati Cincinnati-Middletown,OH-KY-IN(MSA) Cincinnati,OH-KY-INPMSA** 1998 CLEVELAND-LORAIN-ELYRIAOHPMSA Cleveland Cleveland-Elyria-Mentor,OH(MSA) Cleveland,OH-KY-INPMSA** 1996 COLUMBUSOHMSA Columbus Columbus,OH(MSA) Columbus,OHMSA 2002 DALLASTXPMSA Dallas Dallas-Plano-Irving,TXMetropolitanDivision Dallas,TXPMSA 2002 DENVERCOPMSA Denver Denver-Aurora,CO(MSA) Denver,COMSA 1995 DETROITMIPMSA Detroit Detroit-Warren-Livonia,MI(MSA) Detroit,MIPMSA 1993*** FORTWORTH-ARLINGTONTXPMSA FortWorth FortWorth-Arlington,TXMetropolitanDivision Ft. Worth-Arlington,TXPMSA 2002 HARTFORDCTPMSA Hartford Hartford-WestHartford-EastHartford,CT(MSA) Hartford,CTMSA 1996 HOUSTONTXPMSA Houston Houston-SugarLand-Baytown,TX(MSA) Houston,TXPMSA 1998 INDIANAPOLISINMSA Indianapolis Indianapolis,IN(MSA) Indianapolis,INMSA** 1996 KANSASCITYMO-KSMSA KansasCity KansasCity,MO-KS(MSA) KansasCity,MO-KSMSA 2002 LOSANGELES-LONGBEACHCAPMSA LosAngeles LosAngeles-LongBeach-Glendale,CAMetropolitanDivision LosAngeles-LongBeach,CAPMSA** 1989*** MEMPHISTN-AR-MSMSA Memphis Memphis,TN-MS-AR(MSA) Memphis,TN-AR-MSMSA 1996 MIAMIFLPMSA Miami Miami-FortLauderdale-MiamiBeach,FL(MSA) Miami-Ft. Lauderdale,FLCMSA 2002 MILWAUKEE-WAUKESHAWIPMSA Milwaukee Milwaukee-Waukesha-WestAllis,WI(MSA) Milwaukee,WIPMSA 2002 MINNEAPOLIS-ST.PAULMN-WIMSA Minneapolis Minneapolis-St. Paul-Bloomington,MN-WI(MSA) Minneapolis-St. Paul,MN-WIMSA 1998 NEWORLEANSLAMSA NewOrleans NewOrleans-Metairie-Kenner,LA(MSA) NewOrleans,LAMSA 1995 NEWYORKNYPMSA NewYork NewYork-Nassau-Suffolk-Orangeˆ NewYork-Nassau-Suffolk-Orange,NYPMSA 1991*** NORFOLK-VIRGINIABEACH-NEWPORTNEWS Norfolk VirginiaBeach-Norfolk-NewportNews,VA-NC(MSA) Norfolk-VirginiaBeach-NewportNews,VAMSA 1998 OAKLANDCAPMSA SanFrancisco Oakland-Fremont-Hayward,CAMetropolitanDivision Oakland,CAˆ 1998 OKLAHOMACITYOKMSA OklahomaCity OklahomaCity,OK(MSA) OklahomaCity,OKMSA 1996 PHILADELPHIAPA-NJPMSA Philadelphia Philadelphia-Camden-Wilmington,PA-NJ-DE-MD(MSA) Philadelphia,PA-NJPMSA** 1989*** PHOENIX-MESAAZMSA Phoenix Phoenix-Mesa-Scottsdale,AZ(MSA) Phoenix,AZMSA** 2002 PITTSBURGHPAPMSA Pittsburgh Pittsburgh,PA(MSA) Pittsburgh,PAMSA 1995 PORTLAND-VANCOUVEROR-WAPMSA Portland Portland-Vancouver-Beaverton,OR-WA(MSA) Portland,OR-WAPMSA 2002 PROVIDENCE-FALLSRIVER-WARWICKRI-MA Providence Providence-NewBedford-FallRiver,RI-MA(MSA) Providence-Pawtucket-Warwick,RI-MAPMSA 1998 ROCHESTERNYMSA Rochester Rochester,NY(MSA) Rochester,NYMSA 1998 SACRAMENTOCAPMSA Sacramento Sacramento-Arden-Arcade-Roseville,CA(MSA) Sacramento,CAPMSA 1996 SALTLAKECITY-OGDENUTMSA SaltLakeCity SaltLakeCity,UT(MSA) SaltLakeCity,UTMSA 1998 SANANTONIOTXMSA SanAntonio SanAntonio,TX(MSA) SanAntonio,TXMSA 1995 RIVERSIDE-SANBERNARDINOCAPMSA Riverside Riverside-SanBernardino-Ontario,CA(MSA) Riverside-SanBernardino-Ontario,CAPMSA** 2002 SANDIEGOCAMSA SanDiego SanDiego-Carlsbad-SanMarcos,CA(MSA) SanDiego,CAMSA** 2002 SANFRANCISCOCAPMSA SanFrancisco SanFrancisco-SanMateo-RedwoodCity,CAMetropolitanDivision SanFrancisco,CA 1998 SANJOSECAPMSA SanFrancisco SanJose-Sunnyvale-SantaClara,CA(MSA) SanJose,CAPMSA 1998 SEATTLE-BELLEVUE-EVERETTWAPMSA Seattle Seattle-Tacoma-Bellevue,WA(MSA) Seattle-Everett,WAPMSA 1996 ST.LOUISMO-ILMSA St. Louis St. Louis,MO-IL(MSA) St. Louis,MO-ILMSA 1996 TAMPA-ST.PETERSBURG-CLEARWATERFL Tampa Tampa-St. Petersburg-Clearwater,FL(MSA) Tampa-St. Petersburg,FLMSA 1998 WASHINGTONDC-MD-VA-WVPMSA Washington Washington-Arlington-Alexandria,DC-VA-MD-WV(MSA) Washington,DC-MD-VAMSA 1998 **FromAHSdocumentation: ”Sameareasincebeginning. Allotherareaschangeboundariesovertime.” ***MostrecentAHS-Misnotusedduetotop-codingissues. Seetextfordetails. SˆumofNewYork-WhitePlains-Wayne,NY-NJMetropolitanDivisionandNassau-Suffolk,NYMetropolitanDivision. 31

Table 2 Observations Used to Benchmark Structures Share MSA AHS-M year Number of Observations Anaheim 2002 1,582 Atlanta 1996 1,990 Baltimore 1998 1,316 Birmingham 1998 2,264 Boston 1998 1,071 Buffalo 2002 1,391 Charlotte 2002 2,289 Chicago 1991 1,313 Cincinnati 1998 1,470 Cleveland 1996 1,474 Columbus 2002 2,029 Dallas 2002 2,082 Denver 1995 2,181 Detroit 1993 1,986 Fort Worth 2002 1,924 Hartford 1996 2,065 Houston 1998 1,650 Indianapolis 1996 2,242 Kansas City 2002 2,233 Los Angeles 1989 1,190 Memphis 1996 1,924 Miami 2002 1,378 Milwaukee 2002 1,637 Minneapolis/St. Paul 1998 2,237 New Orleans 1995 1,424 New York 1991 791 Norfolk 1998 1,629 Oakland 1998 1,715 Oklahoma City 1996 2,032 Philadelphia 1989 1,049 Phoenix 2002 1,975 Pittsburgh 1995 1,894 Portland 2002 2,321 Providence 1998 1,232 Rochester 1998 1,897 Sacramento 1996 1,760 Salt Lake City 1998 2,513 San Antonio 1995 1,797 San Bernardino 2002 2,262 San Diego 2002 1,573 San Francisco 1998 1,132 San Jose 1998 1,684 Seattle 1996 2,077 St. Louis 1996 1,868 Tampa 1998 1,768 Washington, DC 1998 1,336 32

Table 3 Components of Home Value by Region of the U.S.: Cumulative Changes from 1984 through 2004 Homevalue Landvalue Structurevalue —–percent(cumulative)—– Fullsample 89.8% 203.8% 18.9% Byregion: Midwest 60.2% 437.3% 11.9% Southeast 44.8% 121.2% 15.6% Southwest 13.0% 26.2% 5.3% EastCoast 119.3% 266.3% 22.1% WestCoast 151.2% 225.0% 41.2% Citieswithinregions: Midwest Buffalo 50.7% 765.5% 13.1% Chicago 106.1% 422.9% 24.3% Cincinnati 56.3% 662.5% 3.2% Cleveland 60.3% 1208.6% -0.2% Columbus 56.3% 244.2% 11.5% Detroit 102.1% 1214.6% 43.5% Indianapolis 38.0% 605.6% 6.1% KansasCity 35.5% 161.3% 11.7% Milwaukee 79.7% 568.0% 9.7% Minneapolis/St. Paul 83.6% 591.6% 13.3% Pittsburgh 50.6% 682.1% 17.3% Rochester 11.4% 52.6% 0.8% St. Louis 48.2% 788.9% 9.3% Southeast Atlanta 45.6% 104.6% 25.4% Birmingham 46.0% 339.2% 6.5% Charlotte 43.6% 45.1% 41.6% Memphis 25.7% 181.5% 1.2% Tampa 61.5% 199.6% 12.1% Southwest Dallas -6.8% -26.5% 21.0% Denver 54.6% 169.5% 11.9% FortWorth -12.1% -33.2% 5.0% Houston -1.0% 12.7% -6.2% NewOrleans 20.3% 95.9% -10.0% OklahomaCity -14.8% -28.9% -9.3% Phoenix 37.4% 26.6% 54.0% SaltLakeCity 44.7% 537.6% 1.8% SanAntonio -10.4% -18.4% -7.2% EastCoast Baltimore 100.3% 220.5% 19.2% Boston 142.6% 266.6% 18.3% Hartford 73.4% 229.7% 11.2% Miami 102.7% 145.9% 41.2% NewYorkCity 170.7% 466.4% 30.1% Norfolk 77.2% 124.8% 11.6% Philadelphia 109.7% 454.2% 19.7% Providence 166.6% 805.1% 14.4% WashingtonDC 120.2% 217.3% 35.0% WestCoast Anaheim 147.1% 166.0% 87.0% LosAngeles 143.7% 215.7% 32.1% Oakland 158.4% 232.5% 44.1% Portland 117.0% 436.7% 19.3% Sacramento 135.8% 299.9% 36.9% SanBernardino 110.8% 161.3% 58.1% SanDiego 177.5% 241.6% 53.8% SanFrancisco 179.4% 230.3% 27.4% SanJose 162.7% 217.3% 45.6% Seattle 125.5% 343.9% 23.7% 33

Table 4 Components of Home Value by Region of the U.S.: Cumulative Changes from 1984 through 1998 Homevalue Landvalue Structurevalue —–percent(cumulative)—– Fullsample 21.7% 48.2% 5.3% Byregion: Midwest 25.6% 207.7% 2.3% Southeast 14.8% 53.0% 0.2% Southwest -9.0% -17.2% -4.3% EastCoast 23.9% 49.4% 7.6% WestCoast 38.8% 51.4% 19.9% Citieswithinregions: Midwest Buffalo 29.1% 422.5% 8.4% Chicago 44.6% 180.1% 9.6% Cincinnati 29.4% 440.7% -6.6% Cleveland 36.9% 884.3% -7.7% Columbus 31.4% 165.6% -0.6% Detroit 64.7% 717.9% 30.3% Indianapolis 23.0% 474.9% -2.5% KansasCity 5.8% 46.1% -1.8% Milwaukee 32.7% 250.3% 1.5% Minneapolis/St. Paul 15.0% 137.8% -2.0% Pittsburgh 22.6% 209.1% 12.8% Rochester -1.5% 5.0% -3.2% St. Louis 10.3% 166.9% 2.1% Southeast Atlanta 13.9% 29.0% 8.8% Birmingham 22.2% 216.9% -4.0% Charlotte 26.5% 30.6% 21.3% Memphis 12.3% 143.6% -8.4% Tampa -1.3% 23.2% -10.1% Southwest Dallas -22.5% -43.8% 7.6% Denver 12.0% 58.5% -5.2% FortWorth -24.3% -50.0% -3.4% Houston -21.1% -45.6% -11.8% NewOrleans -6.5% 16.4% -15.7% OklahomaCity -28.1% -66.8% -13.1% Phoenix -1.8% -22.3% 29.7% SaltLakeCity 37.9% 542.1% -6.0% SanAntonio -25.0% -62.7% -10.0% EastCoast Baltimore 22.1% 46.7% 5.4% Boston 33.9% 60.3% 7.3% Hartford 14.7% 59.9% -3.3% Miami 12.3% 10.4% 15.0% NewYorkCity 40.7% 93.8% 15.5% Norfolk 2.6% 2.2% 2.8% Philadelphia 30.9% 117.1% 8.4% Providence 36.2% 175.8% 2.9% WashingtonDC 19.8% 26.5% 13.9% WestCoast Anaheim 20.3% 11.4% 48.4% LosAngeles 23.7% 32.1% 10.8% Oakland 33.9% 42.4% 20.7% Portland 72.6% 291.4% 5.7% Sacramento 15.6% 21.2% 12.2% SanBernardino 2.7% -21.5% 28.0% SanDiego 25.6% 24.1% 28.3% SanFrancisco 61.1% 74.4% 21.4% SanJose 64.0% 81.7% 26.0% Seattle 64.9% 183.9% 9.4% 34

Table 5 Components of Home Value by Region of the U.S.: Cumulative Changes from 1999 through 2004 Homevalue Landvalue Structurevalue —–percent(cumulative)—– Fullsample 56.0% 105.0% 12.8% Byregion: Midwest 27.6% 74.6% 9.4% Southeast 26.1% 44.6% 15.4% Southwest 24.2% 52.3% 10.0% EastCoast 77.0% 145.2% 13.5% WestCoast 81.0% 114.7% 17.7% Citieswithinregions: Midwest Buffalo 16.8% 65.7% 4.4% Chicago 42.6% 86.7% 13.4% Cincinnati 20.8% 41.0% 10.5% Cleveland 17.1% 33.0% 8.2% Columbus 18.9% 29.6% 12.1% Detroit 22.7% 60.7% 10.1% Indianapolis 12.2% 22.8% 8.8% KansasCity 28.1% 78.9% 13.7% Milwaukee 35.5% 90.7% 8.1% Minneapolis/St. Paul 59.6% 190.8% 15.6% Pittsburgh 22.8% 153.0% 4.0% Rochester 13.1% 45.2% 4.1% St. Louis 34.4% 233.1% 7.0% Southeast Atlanta 27.8% 58.7% 15.3% Birmingham 19.5% 38.6% 11.0% Charlotte 13.5% 11.2% 16.7% Memphis 12.0% 15.6% 10.5% Tampa 63.7% 143.1% 24.6% Southwest Dallas 20.2% 30.8% 12.4% Denver 38.0% 70.1% 18.1% FortWorth 16.1% 33.6% 8.7% Houston 25.4% 107.2% 6.3% NewOrleans 28.7% 68.3% 6.7% OklahomaCity 18.5% 113.9% 4.4% Phoenix 39.9% 62.9% 18.7% SaltLakeCity 4.9% -0.7% 8.3% SanAntonio 19.4% 118.8% 3.1% EastCoast Baltimore 64.1% 118.5% 13.1% Boston 81.2% 128.6% 10.2% Hartford 51.2% 106.2% 15.1% Miami 80.5% 122.8% 22.8% NewYorkCity 92.4% 192.3% 12.7% Norfolk 72.8% 119.9% 8.6% Philadelphia 60.2% 155.3% 10.5% Providence 95.8% 228.2% 11.2% WashingtonDC 83.8% 150.9% 18.5% WestCoast Anaheim 105.4% 138.8% 26.0% LosAngeles 96.9% 139.0% 19.2% Oakland 93.1% 133.5% 19.4% Portland 25.7% 37.1% 12.9% Sacramento 104.0% 230.0% 22.1% SanBernardino 105.2% 232.8% 23.5% SanDiego 121.0% 175.2% 19.9% SanFrancisco 73.5% 89.4% 5.0% SanJose 60.2% 74.6% 15.6% Seattle 36.8% 56.3% 13.1% 35

Table 6 Land’s Share of Home Value by Region of the U.S., 1984 to 2004 1984 1998 2004 —–share—– Fullsample 0.320 0.397 0.509 Byregion: Midwest 0.107 0.265 0.362 Southeast 0.267 0.359 0.415 Southwest 0.346 0.308 0.384 EastCoast 0.376 0.461 0.644 WestCoast 0.550 0.608 0.738 Citieswithinregions: Midwest Buffalo 0.050 0.202 0.287 Chicago 0.205 0.398 0.521 Cincinnati 0.081 0.337 0.393 Cleveland 0.050 0.360 0.408 Columbus 0.193 0.389 0.424 Detroit 0.050 0.248 0.325 Indianapolis 0.053 0.249 0.273 KansasCity 0.159 0.220 0.307 Milwaukee 0.125 0.331 0.466 Minneapolis/St. Paul 0.121 0.251 0.458 Pittsburgh 0.050 0.126 0.260 Rochester 0.205 0.219 0.281 St. Louis 0.050 0.121 0.300 Southeast Atlanta 0.255 0.288 0.358 Birmingham 0.119 0.308 0.357 Charlotte 0.559 0.577 0.565 Memphis 0.136 0.295 0.305 Tampa 0.264 0.329 0.489 Southwest Dallas 0.586 0.425 0.462 Denver 0.271 0.383 0.472 FortWorth 0.448 0.296 0.341 Houston 0.274 0.189 0.312 NewOrleans 0.286 0.357 0.466 OklahomaCity 0.279 0.129 0.233 Phoenix 0.606 0.479 0.558 SaltLakeCity 0.080 0.373 0.353 SanAntonio 0.284 0.141 0.258 EastCoast Baltimore 0.403 0.484 0.645 Boston 0.501 0.600 0.757 Hartford 0.285 0.397 0.541 Miami 0.587 0.578 0.713 NewYorkCity 0.322 0.444 0.674 Norfolk 0.419 0.418 0.593 Philadelphia 0.207 0.343 0.547 Providence 0.193 0.390 0.654 WashingtonDC 0.467 0.493 0.674 WestCoast Anaheim 0.760 0.704 0.819 LosAngeles 0.608 0.649 0.787 Oakland 0.607 0.646 0.781 Portland 0.234 0.531 0.579 Sacramento 0.376 0.394 0.638 SanBernardino 0.511 0.390 0.633 SanDiego 0.658 0.651 0.811 SanFrancisco 0.749 0.811 0.885 SanJose 0.682 0.756 0.824 Seattle 0.318 0.548 0.626 36

Table A.1 Total Home Value in 2000, by MSA, and as a fraction of U.S. Total26 Number of Households Total Home Value Percentage of MSA (millions) Average Home Value ($billions) U.S. Total Los Angeles 1.39 $292,978 $408.04 4.52% New York 1.34 $298,608 $401.25 4.44% Chicago 1.50 $213,725 $320.48 3.55% Boston 0.75 $329,916 $247.40 2.74% Washington, DC 0.77 $259,465 $200.12 2.21% San Jose 0.31 $596,146 $184.97 2.05% Anaheim 0.47 $373,424 $175.38 1.94% Oakland 0.44 $398,061 $173.47 1.92% Detroit 1.02 $150,546 $153.81 1.70% San Francisco 0.23 $662,740 $151.86 1.68% Atlanta 0.86 $174,456 $150.58 1.67% Philadelphia 0.81 $182,811 $148.54 1.64% Seattle 0.55 $264,350 $146.46 1.62% San Diego 0.45 $315,524 $141.50 1.57% Miami 0.71 $198,389 $139.97 1.55% Minneapolis/St. Paul 0.69 $181,982 $125.76 1.39% Phoenix 0.64 $191,011 $121.75 1.35% San Bernardino 0.65 $183,335 $118.87 1.32% Houston 0.87 $133,004 $115.45 1.28% Dallas 0.69 $166,073 $114.84 1.27% Denver 0.42 $211,938 $88.87 0.98% Baltimore 0.42 $201,935 $84.48 0.93% Portland 0.40 $207,863 $82.39 0.91% St. Louis 0.58 $136,723 $79.76 0.88% Cleveland 0.45 $167,222 $75.42 0.83% Cincinnati 0.42 $166,848 $70.01 0.77% Sacramento 0.34 $201,412 $67.77 0.75% Pittsburgh 0.53 $110,121 $57.95 0.64% Kansas City 0.41 $139,434 $57.52 0.64% Columbus 0.33 $165,840 $54.37 0.60% Providence 0.30 $180,214 $53.93 0.60% Tampa 0.41 $129,033 $52.28 0.58% Milwaukee 0.29 $164,459 $47.99 0.53% Charlotte 0.28 $171,937 $47.85 0.53% Fort Worth 0.37 $122,065 $45.28 0.50% Norfolk 0.28 $148,796 $42.40 0.47% Hartford 0.22 $186,811 $41.50 0.46% Indianapolis 0.33 $125,267 $41.17 0.46% Salt Lake City 0.23 $175,263 $39.49 0.44% Memphis 0.25 $126,773 $31.76 0.35% Birmingham 0.23 $134,957 $30.73 0.34% San Antonio 0.32 $93,108 $29.96 0.33% New Orleans 0.24 $125,306 $29.50 0.33% Rochester 0.22 $121,127 $27.13 0.30% Buffalo 0.23 $111,554 $26.20 0.29% Oklahoma City 0.22 $89,243 $19.38 0.21% 26All variables (including households) refer to single-family owner-occupied units. 37