Building cities Vernon Henderson, Tanner Regan and Tony Venables January 24, 2016
Motivation Buildings and land are typically about 60% of private wealth in nations. Growing cities require new housing on the extensive margin of development With city population and income growth: demand for space increases. That shifts up rent gradients, where land nearest city centre most valuable. Rising rents imply needed densification, reconstruction & increased building heights to conserve on land African cities growing very fast: primate cities about 4% a year and housing prices rising Nairobi house prices up by 350% since 2000 (CPI by about 250%)
Motivation However for densification and reconstruction, capital in buildings is durable. Therefore: Extensive densification of population requires reconstruction into taller or broader (less green space) buildings Construction decisions are based on expectations. Formal sector development requires effective private property rights To secure financing & insurance Need to mitigate risk of law suits (if develop illegally ), expropriation, and the like. 62% of urban population in African cities are in slums (UN Habitat 2012; State of the Worlds Cities ): Generally poor property rights?
Issues about building patterns Motivation Involve long term commitment of so much of a nation s wealth: (1) How efficient is the allocation of construction between formal and informal (slum) sectors? (2) How are these allocations affected by planning and property right regimes? (3) Are cities built high and dense enough? Role of property rights How are decisions affected by expectations? Role of financing? (4) How are they affected by public investment (infrastructure) decisions
Motivation We know little about this process Little work in economics which studies the dynamic and spatial evolution of cities built with durable capital No work with formal and informal sectors and issues in property rights Little [no?] empirical work which tracks the redevelopment and infilling of the built-environment for entire cities
Model context Growing monocentric city, where access to centre is at a premium Given growth in city productivity relative to the rest of the economy, house rents are rising at a fixed growth rate and city population is growing Rising underlying land values
Model: sunk costs / durability with perfect foresight Formal and informal sectors: Choose height to build for each piece of land upon (re)construction. Formal sector. Buildings are putty-clay, with fixed height for long intervals of time. After first formal sector development, at some point need reconstruction to accommodate rising rents and demanded densification Informal sector. Buildings are putty (legos). Cheap to build one storey but marginal costs of height increases are much higher than in formal sector Cost of conversion: agricultural or informal sector land to first formal sector development. Cost to obtain property rights. Can vary over time and location. Also cost of permanent infrastructure/utilities. Nairobi example
Predictions of model Formal sector on intensive margin of development Develop informal into formal Informal develops at extensive margin, at city edge Redevelopment in formal nearer the city centre In example used, trigger price of space (same across time and space) upon which develop or redevelop achieve that price later further out Heights decline with distance from centre and are lower in informal sector An informal sector property at the same distance from the centre with a higher conversion cost will be developed later into formal sector usage, than a property with a lower conversion cost. Higher trigger price and built to different height--- and that differential persists in the next stage Key issues in development: Higher conversion costs cause a city to develop later and more slowly Expectations: if don t anticipate price rises will under-build
Same first time (D) conversion costs Segment of land with high conversion cost Formal (2) Building height Formal (1) Building height Formal (2) Formal (1) Informal Informal time Distance from CBD time Distance from CBD Distance constant Time constant Under assumptions: Trigger price: which is same across time and space to move to new height Hit that at later times for lots further out.
Model: sunk costs / durability with perfect foresight As rents grow, at each x want heights to rise. But with non-malleable capital only do that at intervals. At time 1 and x =5 probably h=2 is a height that is too high for initial rents; While, when redevelop (time < 50), h=2 is too low for rents. Of course that is why redevelopment occurs. Expectations about price rises affect heights. Building height 8+ x 3 (100) Perfect foresight Building height 5+ x 4 (100) Under-forecast (by 20% ) annual rise of rents x 2 (100) x 3 (100) x 2 (100) x 1 (100) x 1 (100) x 0 (100) x 0 (100) x 1 (1) x 0 (1) Distance from CBD x 0 (1) Distance from CBD Development at 1, 50 and 100 years, simulated for specific parameters
Empirical work Focused on Nairobi Data Built Environment data is both individual buildings and a grid with over 23,000 potential cells, which are 150m by 150m and populated with data on land cover and building counts for each of the years 2004 and 2013/15. Two main sources: Two cross-section of slums and building outlines, circa 2004 and 2015, based on aerial photo 2015 has heights (LIDAR) Some portions of the city with outline and height for 2009 and 2011 SPOT satellite data (built cover and roads) [2004 and 2013: 1.5 2.5 meter resolution]: roads Rent and height data Geo- referenced household level data from the 2012 `Kenya: State of the Cities' survey by the National Opinion Research Center (NORC) and the World Bank. Houses rents and characteristics for a sample that is stratified between informal and formal areas Use property listings data set scraped from property24.co.ke. We focus on the vacant plot listings with information on price and plot area, which represent about 80% of all vacant land listings, but in the formal sector.
Nairobi 2004 Nairobi 2012/15 Kibera
Table 1. House rent and land price hedonics (1) (2) (3) (4) Ln Rent per Ln Rent per Ln Price per Ln Price per sqm sqm sqm sqm Distance to Centre -0.124*** -0.0920*** -0.152*** -0.175*** (0.0245) (0.0255) (0.0151) (0.0246) Slum=1-1.860*** -1.175*** (0.302) (0.337) Slum=1 X Distance to Centre 0.186*** 0.133*** (0.0397) (0.0393) Lot size -0.0276*** -0.0180* (0.00829) (0.0100) Coordinates estimated=1-0.320** -0.324** (0.124) (0.122) (Suggested) Residential=1-0.444 (0.309) (Suggested) Residential=1 X Distance to Centre 0.00626 (0.0309) Constant 6.546*** 3.789*** 10.74*** 11.75*** (0.217) (0.782) (0.103) (0.372) Controls No Yes No No Listing Month No No Yes Yes Observations 764 691 561 334 R-squared 0.154 0.350 0.570 0.661 Standard errors in parentheses* p<0.10, ** p<0.05, *** p<0.01 Controls Include: Fraction of neighbourhood that is not built, tenancy agreement is written formal, number of bath rooms, piped water to dwelling, piped water within 50m, toilet facility in house, connected to electricity, wall is made from iron sheet or tin, wall is made from brick stone or block, roof is made from iron sheet, floor is made from earth or clay, access road is paved or tarmacked, access road is gravel, dwelling or room in multi-story, piped water to dwelling, one flood in recent rainy season, 2-3 floods in recent rainy season, 3+ floods in recent rainy season, and number of floors
Cross-sectional (in dynamic context) layout of city: driven by rents NORC: predicted rent price differences from centre per sq meter of floor space. 2012. ln pˆ ( x) aˆ BZ ˆ ˆ x In formal sector Controls for many house and neighbourhood characteristics Error bands on prediction Sales price of vacant land per square meter. Controls for (use, roads (?) and) distance. Much steeper land price gradient Have NORC house rents 2012 Have market asking sale prices on vacant land 2015 Land prices: much steeper gradient Price at centre 13x price at edge
Cross-sectional layout of city: individual building sizes 2015
Residential heights NORC data: Residential Height: Slum vs non-slum
Cross-sectional layout of city: grid cover 2015, Heterogeneity: smooth with bigger cell size! Includes roads 43% 28% At edge is 30% of centre Issue: What is a Building? Consider row houses, commercial complexes with tower and non-tower portions, U-shaped dwelling complex with central garden and the like Separate or not Visual counting from satellite images by humans or machines On the ground by planners 3x as many buildings at edge as at centre
Dynamics To track reconstruction, demolition and in-fill: Need to overlay 2004 and 2015 images: in progress Can talk today about change by grid square in Building and road coverage Building counts
Changes in cover Paved roads 2 lanes 10 m; 4 lanes 20 (conservative) Likely greater width in city centre especially if one adds in sidewalks Increased demand for road space near centre for commuting. shopping and delivery purposes: Solow & Vickrey, Riley: 1970 s
Growth rate of counts and average roof size % coverage inn 2015 - % coverage in 2006 Growth Counts 2006-2015 Change Cover 2006-2015 Growth Size 2006-2015 km to Center 0.157*** 1.462*** -0.0410*** (0.00878) (0.0756) (0.00288) Ln Count 2006-0.152*** (0.0204) Ln Count 2006 X km to Center -0.0190*** (0.00287) Cover 2006-0.0872*** (0.0218) Cover 2006 X km to Center -0.00959** (0.00326) Ln Avg Roof Size 2006-0.403*** (0.00966) Slum=1 0.124 1.675-0.487*** (0.188) (1.536) (0.129) Slum=1 X km to Center 0.0190-0.151 0.0427*** (0.0280) (0.231) (0.00787) Slum=1 X Ln Count 2006 0.0706 (0.0501) Slum=1 X Ln Count 2006 X km to Center -0.0180* (0.00745) Slum=1 X Cover 2006 0.202*** (0.0523) Slum=1 X Cover 2006 X km to Center -0.0184* (0.00821) Slum=1 X Ln Avg Roof Size 2006 0.0273 (0.0239) Constant 0.580*** 1.931*** 2.063*** (0.0593) (0.497) (0.0565) Observations 5809 6096 5809 Standard errors in parentheses* p<0.05, ** p<0.01, *** p<0.001
Heterogeneity by initial counts: mean and mean plus 2 s.d. s Drop in count implies reconstruction?
What about slum costs? Look at overall pattern of development of slums Lack of development into formal sector near centre: Kibera as poster-child High costs of redevelopment Rent differentials at same location on standard house : slum & nonslum: Reflects differential land values: Evidence of misallocation?
Misallocation of land Land not in highest and best use. Inability to redevelop old slums.
Government more costly to redevelop? Role of political connections
Kibera 1000 acres given by British to Nubians in 1912, as a settlement for their military support, but not titled Nubians occupied some of area Rest over time illegally allocated by local chiefs and bureaucrats Run by slum lords. Majority owned by politicians Supposedly 60-80% rate of return (although bribe costs unknown) Can t redevelop. Slumlords and Nubians have no title. Political opposition: Redevelopment would be loss of rent profits for participating politicians. Need to buy them out Back of the envelope calculation base on rent differentials and land prices in formal sector suggests lost land value is for the 1000 acres in USA dollars is $1.86 billion
Summary City development involves Informal sector (low cost & low height) at the fringes Conversion to formal sector Redevelopment of formal sector (at increasing building heights) as city grows Issues Conversion costs Role of history and heterogeneity within city Role of expectations