NYC, 7 November 2014 Empirics of housing, land use, and location choices Gilles Duranton Wharton Housing • At the heart of extremely large allocations • Closely related to other large allocations • Several traditions: – Housing as an asset – The macroeconomic impact of housing – Social / public policy – Location choice • Advantages of focusing on housing as a location problem – The demand and supply of housing is about locations – Housing as part of a broader set of urban issues – Connected to transport and commercial real estate 2 Roadmap • A few illustrations • Theory reminder: housing and accessibility • The empirics of gradients • Sorting within cities • Suburbanisation • Job decentralisation • Housing and the regulation of land use Mostly based on our recent Handbook chapter with Diego Puga 3 Land use in Paris Multi-family residential Single-family residential Commercial Transport Open space 4 Land use in Paris 100% Share of land by use Open space Transport 75% 50% Built−up 25% 0% −30 −20 −10 0 10 20 Distance to Notre Dame (km.) 5 30 Land use in Paris 100% Share of built−up land by use Commercial 75% 50% Single−family residential 25% Multi−family residential 0% −30 −20 −10 0 10 20 Distance to Notre Dame (km.) 6 30 Land prices in Berlin Source: Ahlfeldt, Redding, Sturm, and Wolf (2014) 7 ds before World War I, declined sharply during the war and e interwar period. In the second half of the 20th century, per year on average.Housing prices over time: France 2012. Source: Knoll, Schularick, and Steger (2014) 8 Figure 6: France, 1870–2012. rowth of a little more than 1 percent. Note that this is lower than average annual GD capita growth of about 1.8 percent for the sample average. That is to say, house pric e risen significantly over the past 140 years relative to the consumer prices but have lagg Housing prices over time: World me growth in most countries. We will return to this point later. Source: Knoll, Schularick, and Steger (2014) Figure 15: Mean and median real house prices, 14 countries. 9 A quick refresher on accessibility models • Accessibility is a key determinant of the demand for housing – Accessibility has value – It implies other location characteristics that residents value • The most famous accessibility model is the monocentric model originally developed by Alonso (1964), Mills (1967), and Muth (1969) • With amm, accessibility is modelled in a very simple but useful way: – ‘Reasonable’ first-order description of cities – Models key disadvantages of bigger cities (more expensive housing, longer commutes) – Structures our examination of the empirical literature 10 The accessibility - housing price tradeoff • Provided location preferences are not fully idiosyncratic, locations with better accessibility will be more desirable • As a result, floorspace in these locations will fetch a higher price • There will be tradeoff between household travel costs and housing prices: as accessibility declines housing price should decline to offset the increase in travel costs exactly • General models of accessibility are hard because the location choice of everyone depends on that of everybody else, including employers • The monocentric model simplifies the problem by imposing an exogenous location of jobs (at the centre) and by reducing the travel problem to commutes 11 Main predictions Five gradients: 1. Housing prices decrease with distance to the cbd 2. Because of lower prices, housing consumption by residents increase with distance to the cbd 3. Profit maximisation by competitive constant-returns builders leads land prices to decrease with distance to the cbd 4. This also leads the capital/land ratio to decrease with distance to the cbd 5. In turn, the previous gradients imply that population density decreases with distance to the cbd 12 The ‘forgotten’ predictions The monocentric model also offers a number of quantitative predictions 1. The Alonso-Muth condition for housing prices 2. An equivalent Alonso-Muth condition for land prices that adjusts for density 3. Housing development should be proportional to the ratio of land price gradients to housing price gradients 4. Differential land rent at the cbd should be proportional to city population 5. Proportionality between transport costs and differential land rent aggregates 13 Main challenges to the monocentric model 1. Household heterogeneity 2. Housing is durable 3. The location of jobs is endogenous 4. Land and housing markets are subjects to frictions and regulated 14 The empirics of gradients: population density • Large literature starting with Clark (1951) • Rings vs area approaches • Generally negative density gradients except in some heavily constrained cities • Gradients become flat in far suburbs • Richer patterns at a micro-scale (subcenters, etc) 15 The empirics of gradients: housing prices, land prices, development intensity and housing consumption • Some work on housing price • Often has a hard time to evidence gradients • Work by Cheshire and Sheppard (E 1995) and Ahlfeldt (JRS 2011) is most advanced • Some work on land prices • Again supportive of negative gradients • Close to no work on the intensity of development (except McMillen bc 2006) • Even less on housing consumption 16 Limited knowledge: data availability • Historically a binding constraint • Usually a small number of cities (often one) • Highly selected cities • A lot of within country variation in gradients mostly unexplored • Some work in ‘time series’ but limited (McMillen JUE 1996) 17 Limited knowledge: methods • How to define centres? • How to define subcentre? (McMillen JUE 2001) • Residential density is not area density (or is it?) • Unit housing prices are hard to obtain and hedonics are potentially problematic – Standard specifications can be derived from the amm model using an approximation that is true only locally – The model expects house characteristics to be determined jointly with distance to the cbd • Endless debates about functional forms for the spatial decay 18 A misconstrued debate • Joint test of the (assumed) monocentric assumption and the Alonso-Muth tradeoff • But no city is truly monocentric • The interesting questions regard the Alonso-Muth tradeoff • Some work that considers accessibility but – It imposes arbitrary accessibility indices – Or model-based indices that are never validated by looking at actual travel behaviour 19 More work is obviously needed • Take theory more seriously – Assess the Alonso-Muth condition without imposing monocentricty – Take accessibility seriously • Assess the quantitative predictions of amm models (Combes, Duranton and Gobillon, 2013) • Develop tools to assess how monocentric cities are (?) • More ‘quantitative’ work as in Rappaport (2014)? • More structural work like Ahfeldt, Redding, Sturm, and Wolf (2014)? 20 Household heterogeneity • The monocentric city model can easily be extended to have multiple income groups • If commuting is paid in numeraire, the rich will live farther from the cbd provided housing is a normal good • With commuting paid in time, the rich will live farther from the cbd provided the income elasticity of demand for land is greater than the income elasticity of the value of commuting time • Richer forms of heterogeneity have not been explored much (including parcel heterogeneity) 21 Patterns of location choices: sorting and segregation Some facts for the us • Historically (LeRoy and Sonstelie JUE 1983, Lee and Lin, 2014): – Rich in the center – Decentralisation of the rich with the streetcar and the car – Recentralisation since the 1970s – Less decentralisation in centres with strong amenities • Today (Brueckner and Rosenthal REStat 2009 and others): – Generally positive income gradients – Bell-shaped beyond some distance – Discontinuities at the boundaries of the main municipality – Considerable income overlap – Some rich us centres are exceptions – Opposite patterns in Europe 22 Can the accessibility / land price tradeoff explain these patterns? • Need to estimate the income elasticity of the demand for land (/housing)... • ... and the income elasticity of commuting costs • Wheaton (AER 1977) – Use results from the San Francisco bart surveys – Find both elasticities to be about 0.25 23 • Glaeser, Kahn, and Rappaport (JUE 2008) – Estimate demand with log parcel size = c + a log income + e – ols income elasticity of the demand for land about 0.1 – Imputing land consumption for apartments raises it to 0.3 – Land price is missing from the regression: downward bias – Instrument income with education raises the elasticity to 0.5 – Parcel size may not capture true land consumption. Using log area per person also raises the elasticity to 0.5 with ols – Income elasticity of commuting costs argued to be 1 – But travel is not only time. A simple back of the envelop suggests about 0.5 for poorer households – GKR conclude about the importance of transit Since the two elasticities may be close, the am tradeoff may be a weak determinant of weak patterns (?) 24 Alternative explanations • Amenities (Brueckner, Thisse, and Zenou EER 1999) – Contrast Paris with Detroit – But no real empirics (and real empirics are particularly hard on this) – Although appealing as an explanation, amenities may generate many patterns depending on shape of the demand for land, etc 25 • Housing durability (Brueckner and Rosenthal, REStat 2009) – Controlling for the age of the housing stock explains away the income gradient away in us cities – The age of the housing stock is endogenous – Instrumenting using 20-year lags confirm the findings • ‘Flight from blight’ – Crime (Cullen and Levitt REStat 1999) – Racial preferences (Boustan QJE 2010) – Changes in the school system (Baum-Snow and Lutz AER 2011) • Fiscal motives / Tiebout sorting (Inman AER 1995, de Bartolome and Ross JUE 2003) Not much evidence except for recent structural work (Bayer and McMillan 2012) 26 Patterns of residential land development The sprawl ‘confusion’: • Low density • Scattered development • Causes (car-based travel) • Consequences 27 Facts about residential decentralisation In us metropolitan areas • Central city share of residents (Boustan and Shertzer, 2012) – 1940: 56% – 2000: 32% • Flattening of population gradients • Constant scatteredness of development: 1976-1992 (Burchfield, Overman, Puga, and Turner, 2006) • Similar evidence for a sample of world cities (Angel, Parent, and Civco, 2012) 28 A fall in commuting costs to explain residential decentralisation? • Early evidence is ambiguous (e.g., Brueckner and Fansler, 1983) • Main issues: choice of measure of commuting costs and endogeneity of these measures • Baum-Snow (QJE 2007) – Counts radial rays of us Interstate Highways – Instruments them with a corresponding measure from the 1947 highway plan – Finds strong evidence that highways led to residential decentralisation: -9% in central city population for each extra ray 29 • Puzzle: the monocentric model predicts decentralisation but not a central city decline (cheaper travel should lead to growth for central cities, albeit less than suburbs) • ‘Blight factors" might be at play • Rising income may also have played a big role (Margo JUE 1992) 30 What explains new development fragmentation? Burchfield, Overman, Puga, and Turner (2006) • Look at the share of open space within 1 km of new developments over 1992-2006 in us metropolitan areas • Find evidence about: – Car-based cities – Uncertainty (as predicted by models of uncertain land developments) – Low past growth – Temperate climate – More decentralised production sectors (manufacturing) – Hills (but not mountains) – Aquifers – Municipal fragmentation 31 Consequences of sprawl • More driving? (Bento, Cropper, Mobarak, and Vinha, REStat 2005 and a large planning literature) – But the effects are modest – Not well identified • Reduced social interactions? – Again, poorly identified – The better evidence (Glaeser and Gottlieb US 2006, or Brueckner and Largey JUE 2008) says the opposite • Obesity? – Still poorly identified in most cases – Eid, Overman, Puga, and Turner (JUE 2008) using panel data find no effect 32 Facts about employment decentralisation In us metropolitan areas • Only 24% of jobs within 5 km of the cbd in 1996 • Share of central city jobs went from 61% in 1960 to 34% in 2000 • Employment decentralisation has been considerable but less than residential decentralisation • Employment decentralisation started later but is highly correlated with residential decentralisation • Skilled jobs have decentralised less, manufacturing jobs have decentralised more • Considerable heterogeneity across cities 33 Diffuse decentralisation or movement towards subcentres? • Some evidence about subcentres (McMillen and Smith JUE 2003) • No measure of "diffuseness" of decentralisation? • Glaeser and Kahn (BWPUA 2001) claim diffuse for the us • Garcia-Lopez, Hemet and Viladecans (2015) provide evidence of subcentre growth and emergence for Greater Paris 34 Wage gradients According to models of endogenous business locations (eg Ogawa and Fujita, 1980): • Firms tradeoff land costs vs. agglomeration • But also get closer to their workers when they decentralise • And may thus pay them less → wage gradient • This gradient is expected to be weak relative to the housing price gradient 35 Evidence about wage gradients • Workers with greater commutes have higher wages • But this is a prediction of the monocentric model as residents with higher wages live further from the cbd • When firms have market power firms will pay workers with longer commutes more even in absence of gradients (Manning, LE 2003) • Also: workers commute much more than the travel minimising solution suggests (‘wasteful commuting’) • Also: commute times have remained fairly stable (‘commuting time paradox’) 36 Another implication of job decentralisation: spatial mismatch • First proposed by Kain (1968) to explain low minority employment rates in us central cities • Fact: Access to jobs strongly declined for minority residents in central cities • A variety of mechanisms possibly at play: – Search is more difficult – Wages net of commuting costs may get below a reservation level – More discrimination in white suburbs 37 • Typical regression 1: neighbourhood labour market participation on access to jobs and controls • Typical regression 2: neighbourhood employment on race • Identification issue: neighbourhood choice is not exogenous • Raphael (JUE 1998) focuses on young workers more likely to live with their parents • Policy solutions – Move people where the jobs are? (mto results are not encouraging) – Move jobs where workers are? (Literature on placed-based policies raises a lot of doubts) – Subsidise travel? 38 Causes of job decentralisation • Residential decentralisation? • Transportation (Baum-Snow, 2014) • Crime, housing stock decay, racial preferences, changes in the school system • Changes in communication technology allowing firms to separate different production units 39 Some continuity despite considerable decentralisation • Subcentre continuity in Los Angeles (Redfearn RSUE 2009) • Transit continuity in Los Angeles (Brooks and Lutz unp 2013) • Urban continuity (Bleakley and Lin QJE 2012, Davis and Weinstein JRS 2008) • But major changes within cities when the stock of housing disappears (Siodla 2014, Hornbeck and Keniston 2014) 40 Land use regulations • Perhaps the single most important determinant of the supply of new housing • Regulations limit both the type and intensity of development • A challenging empirical topic • But not really a challenge for the monocentric model 41 Measuring land use regulations First, some measurement issues need to be discussed • We can measure land cover • Land use is much harder • There are many forms of land use regulations, particularly in the us • Should we focus on land use regulations or on the building code? 42 development. Gyourko and Saiz (2006) document a large degree of heterogeneity in structure production costs across local markets, which is correlated with a number of supply shifters including the extent of construction worker unionization, the level of local wages, local topography as reflected in the presence of high hills and mountains, and the local regulatory environment as measured by an index of Internet chatter on construction regulations. .8 1 1.2 Index, 1980 = 1 1.4 1.6 1.8 2 2.2 2.4 Real Construction Costs and House Prices Over Time 1980 1985 1990 1995 2000 2005 2010 2015 year Real House Price Real Construction Cost Source: Gyourko and Molloy (2014). us Construction costs are the cost of an Note. Construction the costCompany of an Economy Quality from R.S. Price Economy Quality home fromcosts R.S.are Means deflated byhome the Consumer Means Company deflated by the Consumer Price Index. House prices are the Index. House prices are the repeat-sales index published by CoreLogic deflated repeat-sales index published by CoreLogic deflated by the price index for excluding housing personal services. consumption expenditures excluding housing services. The deflator was calculated by the authors from data published by the Bureau of Economic Analysis. 43 Land price as the difference. How to measure land use regulations Three main approaches 1. Use simple economics: • Glaeser, Gyourko and Saks (JLawEc 2005) measure the difference between the price and marginal cost of an extra floor in Manhattan condos and call this gap the regulatory tax • Finding: Manhattan condo c. 2000 were at least 50% more expensive than in absence of regulation • GGS also estimate the regulatory tax for single family homes for a cross-section of cities ; it goes from zero in Birmingham, Cincinnati, and Houston to nearly 20% in Boston, over 30% in Los Angeles, and upwards of 50% in the San Francisco Bay Area 44 2. ‘Deep but narrow’ • Glaeser, Shuetz, and Ward (2006) and Glaeser and Ward (JUE 2009) use extremely detailed data for municipalities in the Boston area including prohibitions of irregularly-shaped lots, extensive wetlands restrictions, septic system regulations, etc • They also collect detailed data on undeveloped land... • ... and on building activity over nearly 100 years • They draw a strong connection between lack of development and minimum lot size 45 3. ‘Shallow but wide’: • Gyourko, Saiz, and Summers (US 2008) collected survey information for 2,611 us communities • They estimate indices of the stringency of the local land use environment (wrluri index) • Strong correlations across the component indices • Saiz (QJE 2010) develops measures of land availability for major us urban areas, excluding wetlands, water bodies, slopes above 15 degree • He finds a strong correlation with the wrluri index • That allows him to estimate housing supply elasticities that have been widely used in subsequent work 46 Motives for land use regulations: non-conforming uses • Stull (AER 1974) • Provides a strong rationale for separating users (zoning) • A similar case can be made in case of positive externality across similar users (e.g., Ogawa and Fujita JRS 1980) • Preferences for open space can also justify open space protection (Turner, JUE 2005) 47 • Makes a good case for open space protection • But we do not know how much space should remain undeveloped nor whether one big park is better than two small ones (and land prices may offer a poor guide for action) • Outside of the separation of dirty manufacturing, it is hard to justify full Euclidian zoning • Externalities from non-conforming uses also make it hard to justify restrictions that provide a ceiling on development for a given type of users (e.g., minimum lot size) 48 Motives for land use regulations: congestion • Perhaps large congestion externalities in cities • More general issue behind: how much land should be allocated to roads and parking • Once a subject of interest (Solow Vickrey JET 1971) • Land use regulation can be used to implement the first-best (Pines and Sadka JUE 1985) • But informationally very demanding for a benevolent planner • Many other instruments seem more appropriate (tolls, parking pricing, etc) • No empirical knowledge, in part because we do not know about land use in cities 49 Motives for land use regulations: Tiebout • Tiebout equilibrium requires head taxes • But not generally feasible • Many countries use a property tax • But it induces the poor to chase the rich • Fiscal zoning can be used to re-establish a Tiebout equilibrium • Exclusionary zoning may not be only fiscal (e.g., racism) • Review of the evidence by Ihlanfeldt (2004): some support for the fiscal motive but the evidence for other forms of exclusionary zoning is mixed today • In any case, separating sorting by income (fiscal) and sorting by race (purely exclusionary) is difficult 50 Motives for land use regulations: Political economy • Several variants: 1. Under limited mobility or with idiosyncracy in preferences for specific locations, ‘monopoly’ local landowners may want to restrict the supply of housing (Fischel 1985) 2. Although some developments may increase local land values they are risky and local residents may be risk-averse (Fischel 2001, Ortalo-Magne and Prat AEJ 2014) 3. Conflict between owners of developed vs. underdeveloped land (Hilber and Robert-Nicoud JUE 2013) • They all point to the role of local homeowners • This rings true in the us, less clear in other countries 51 • Regressing the stringency of land use regulation on homeownership raises an obvious simultaneity issue • Hilber and Robert-Nicoud use the fraction of childless married couples as instrument for the share of homeowners • Rationale: These couples are more likely to have stable high disposable incomes making them eligible for mortgages • Exclusion restriction: they should not otherwise affect regulations • Overall, they find small effects • Glaeser and Ward (JUE 2009) instrument current with lagged homeownership (1940 or 1970) • This avoids reverse-causation but maybe not omitted variables • They also finds small effects 52 Do restrictive regulations raise land prices? Theoretically ambiguous • If the demand for locations is perfectly elastic it cannot • If demand is imperfectly elastic, restricting quantities will raise prices 53 Large empirical literature on restrictiveness and land prices: • Mostly in cross-section • Survey by Quigley and Rosenthal (2005): more restrictive regulations lead to higher prices and the effects are often large • For instance Mayer and Somerville (JUE 2000): A one month increase in approval time for a new development reduces building permits by 10 percent. • Glaeser and Ward (JUE 2009) study Massachusetts townships over 1980-2002. The measured negative effects of wetlands bylaws and subdivision rules on construction are larger in panel than in cross-section (restrictions are implemented in areas with more-than-average construction activity) 54 Non-price and non-quantity effects of regulations: • Bertaud and Brueckner (JUE 2005) show that building height restrictions lead to more sprawl and longer commutes • Geshkov and DeSalvo (JRS 2012): minimum lot sizes and height restrictions contribute to sprawl, while maximum lot sizes, limits on construction, minimum density requirements and impact fees reduce sprawl 55 Welfare effects of regulations: Glaeser, Gyourko and Saks (2005) • They estimate crowding cost for housing in NY by looking at price differentials across floors • They also estimate a broad congestion externality by looking at rents as a function of city population and income in a broad cross-section of cities • The costs of these two externalities are small relative to the regulatory tax • They also estimate a positive fiscal contribution of new residents • They conclude at negative welfare effects 56 Turner, Haughwout, and van der Klaauw (E 2014) • Distinguish three effects of land use regulation 1. Own-lot effect: the direct cost of regulatory constraints on how land is used 2. External effect: value of regulation for owners of nearby land (assumed positive following an externality argument) 3. Supply effect: regulation reduces the supply of developable land. • Data: – Cost-Star land price data – wrluri data to measure regulations – 2006 National Land Cover Database to measure development 57 • Estimation – They use border discontinuities between municipalities (with different regulations) – Identifying assumption: unobserved traits of land are uncorrelated with regulations across borders – Control for many covariates and use only straight borders • Main result: a one standard deviation increase in the index of regulation reduces land prices by 36 percent ; welfare would be improved by relaxing land use regulations • See also Cheshire and Sheppard (E 2002) 58 Conclusion The empirical literature on land use is underdeveloped. Time is ripe for a new effort: • Big improvements in data availability • But this is not enough: – Better connection to theory is needed – Identification concerns must be dealt with • Land use is much broader than just ‘land’ 59