Published on March 21, 2008
Lindahl Lecture 1: The Economics of Cities: Lindahl Lecture 1: The Economics of Cities Edward L. Glaeser Harvard University Cities Result from Three Forces: Cities Result from Three Forces Agglomeration Economies and Social Interactions These are the magic of urban areas Urban Technologies: Bricks and Mortar, Trains and Cars Government Policies Both local and national The Plan of these Lectures: The Plan of these Lectures In lecture 1, I will focus on agglomeration economies– what makes cities productive and attractive. In lecture 2, I will focus on social interactions and other effects of proximity In lecture 3, I will address both the urban technologies and the role of government Plan of this Lecture: Plan of this Lecture Overview of Urban Economics Measuring Agglomeration based on location patterns Lessons from Urban Growth Urban Labor Markets Learning, Information and Cities– some theory The Heart of Urban Economics: Spatial Equilibrium: The Heart of Urban Economics: Spatial Equilibrium Workers must be indifferent across space: U(Wages, Amenities, Prices)=U Higher wages must be offset be either lower amenities or higher prices. Firms must be as well: Profits(Wages, Prices, Productivity)=0 Higher wages must be offset by either higher productivity or higher prices. There is also a housing supply equilibrium that will be addressed in lecture 3. An Easy Example: An Easy Example Assume wages are fixed at w– and that commuting costs equal t*distance from the city– then spatial equilibrium implies that rents must decline by t*distance from city. Rents or housing values will be higher in areas with higher amenities or better schools. Housing Prices and Temperature 1990: Housing Prices and Temperature 1990 Why locate together? : Why locate together? Cities can come in principle for two reasons First, a desire to be next to some exogenous attribute, like a mine or a port Second, a desire to be next to the other inhabitants of the city Why Cities?: Why Cities? As such, cities are defined as the absence of physical space between people and firms They always occur in an attempt to eliminate transportation costs for goods, people and ideas The empirical questions revolve around which are these are more important Moving Goods, People and Ideas: Moving Goods, People and Ideas Cities are originally about moving goods Every large city in the U.S. before 1880 is on a river and most are where the river meets the sea. Local Feedback where producers move to be close to consumers (Krugman). Moving People: Moving People Modern big cities specailize in business services– these require fact to face contact. Cities allow works to switch employers and industires, which provides insurance and better search. Proximity to other people isn’t just productive, it’s also fun (city as marriage market). Moving Ideas: Moving Ideas Ideas, like everything else, move better over short distances (face-to-face) Jaffe, Trajtenberg and Henderson show that patent citations are geographically localized. Idea-intensive industries (finance, the arts) remain core parts of urban growth. Urban edge in idea production makes cities important (Athens, Florence). The Impact of Proximity: The Impact of Proximity While city location is a choice, it is also interesting because it shapes outcomes Firms may be more productive in dense areas (Ciccone and Hall, 1996) Workers may learn more quickly in dense areas It may be easier to steal in dense areas Our beliefs are formed by our neighbors Measuring Agglomeration (Ellison, Glaeser, JPE, 1997): Measuring Agglomeration (Ellison, Glaeser, JPE, 1997) How should we measure the amount of agglomeration or people or industry? I think measures should generally be model-driven, i.e. reflect a parameter in some sort of a model. Assume profits have the following form: Where we assume: Where we assume Individual shocks follow a weibull distribution The spillover effect takes on a value of 1 with probability gs The mean and variance of profits are: Together these assumptions give us that: : Together these assumptions give us that: Properties of the Index: Properties of the Index Easy to compute with available data Easy benchmark with no spillover/natural advantage version Comparable across industries with different sizes of firms Comparable across different levels of aggregation Not good at dealing with issues of actual location Facts on agglomeration: Facts on agglomeration Median estimate of gamma is .026; mean is .051. A few industries are extremely concentrated– fur goods (.6), costume jewelry (.3) Many are not– cane sugar refining, A few really change when we correct for plant size (vaccuum cleaners) Does Natural Advantage Explain Agglomeration (AER, 1999): Does Natural Advantage Explain Agglomeration (AER, 1999) To extend the JPE paper, we try to control for local characteristics What do the variables mean?: What do the variables mean? The delta is industry specific and allows different industries to respond differently to costs The beta is the coefficient to estimate that is “cost” specific (i.e. electricity, labor, etc.) The y variables are state cost specific (i.e. price of electricity in kansas) The z variables are industry cost specific (i.e. how much does that industry use that input) The Empirical Strategy: The Empirical Strategy Regress state/industry shares on characteristics and then ask how much is explained Characteristics include energy prices, labor costs, proximity to the coast, proximity to consumers, etc. Some are quantities; some are prices. Overall Results: Overall Results Controlling for all of these variables reduces the mean gamma across industries from .051 to .048. When we allow 2 and 3 digit industry dummies (we are using 4 digit industries) concentration falls to .045 and .041 Since many industries aren’t very concentrated, this explains some portion of those industries, but little of the highly concentrated industries. The Dynamics of Industrial Concentration (REStat, 2002): The Dynamics of Industrial Concentration (REStat, 2002) How permanent are concentrations of industries– Krugman (1991), e.g. Empirical Results: Empirical Results Use the Census Longitudinal Research Database with plant level data for all manufacturing Estimates of beta are around -.06 for 5 year patterns Mean reversion would cause concentration to decline by 12 percent every 5 years, But this is made up for by the concentration of new firms An Extension: New Births, Closures, Etc. : An Extension: New Births, Closures, Etc. We can extend the methodology to look at what sort of changes create mean reversion Closures are more likely in places with initial concentration New Openings are more likely in places with less initial concentration (equally of affiliated and unaffiliated plants) Co-Agglomeration (NBER Working Paper, 1997, Dumais + E/G): Co-Agglomeration (NBER Working Paper, 1997, Dumais + E/G) To add to our knowledge of the sources of agglomeration, we look at which industries colocate near one another. Changes specification: regress growth in employment on presence of other industries in initial period. Levels specification; regress employment on presence of other industries (BUT THERE IS A REFLECTION PROBLEM) All measures of colocation are normalized to have standard deviation of 1 Suppliers and Consumers: Suppliers and Consumers Use Input-Output matrices to calculate the extent that an industry buys to or sells from other industries. Use that matrix to calculate the extent that a state or MSA is supplier or customer In levels, .06 for customers, .01 for suppliers. In State changes, .04 and .03 In MSA changes, .02 and 0 Labor Supply: Labor Supply Use occupation data to figure out who uses the same type of workers Calculate a similarity index across and ask which places have industries that use similar workers In state levels, the coefficient is .41 In MSA changes, the coefficient is.43 In State Changes, the coefficient is .18 Idea Flows: Idea Flows Option 1: Use the Scherer input output matrix for patent flows Option 2: Use patents of co-ownership, excluding those firms with supply/demand relationship In levels, the coefficients are .04 and .03 State change coefficients are -.01 and .06 MSA changes coefficients are 0 and .08 Urban Growth Underpinnings: Urban Growth Underpinnings This implies that: This implies that City size will be a function of consumer amenities, fixed factors of production, reservation utility levels and so forth. Wages will also rise with productivity– this is being offset by lower amenity levels. These equations are then first differenced to provide estimating equations To close the model: To close the model Assume changes in A and changes in C are functions of initial characteristics and then regress changes in population (or employment) and income on initial conditions. Higher employment means either more productivity or better amenities Higher wages means either more productivity or worse amenities. Testing the New Growth Theory (GKSS, JPE, 1992): Testing the New Growth Theory (GKSS, JPE, 1992) Under what conditions are new ideas created? Marshall/Arrow/Romer– high concentration, big firms Jacobs– diversity, lots of little firms Michael Porter– high concentration– little firms The Empirical Test: The Empirical Test Using city-industry (i.e. steel in Pittsburgh) growth between 1956 and 1987, GKSS look at what predicts growth: Concentration of the industry (share in city relative to share in U.S.) Initial Employment in the City-Industry Competition (or firm size relative to national average) Diversity in other industries Results: Results Average firm size (competition) always predicts more growth– what does it mean? There is substantial mean reversion Relative size is sometimes good/sometimes bad (no clear pattern) Diversity is good in our paper (not obvious how robust) Subsequent Work: Climate and the Consumer City: Subsequent Work: Climate and the Consumer City In 1900, cities had to locate in places where firms had a productive advantage. In 2000, cities increasingly locate in places with attractive amenities. The move to warm, dry places. The continued resilience of a few big consumer cities (NYC, Chicago, Boston, San Francisco). Climate is the most reliable predictor of city growth: Climate is the most reliable predictor of city growth Best thought of as a regional effect: Best thought of as a regional effect Other correlations between pop. growth and consumer amenities:: Other correlations between pop. growth and consumer amenities: 35 percent correlation with temperature and 12 percent with dryness 24 percent correlation with proximity to ocean (Rappaport + Sachs) 14 percent correlation with theaters In France, 45 percent correlation with restaurants and 33 percent with hotel rooms In UK, 31 percent correlation with tourist nights Other facts : Other facts Real wages used to decline with city size, now they rise (to be discussed later) Amenities (high housing prices relative to wages) strongly predict later population growth Housing price growth in central cities has boomed Reverse Commuting has increased Urban Growth is Very Persistent: Urban Growth is Very Persistent The Rise of the Skilled City (JME, 1995, BWPUA, 2004): The Rise of the Skilled City (JME, 1995, BWPUA, 2004) One fact that is regularly observed is the more skilled cities grow more quickly (Cityscape, 1994) Simon and Nardinelli show this going back to 1880. Are skilled cities more innovative? Is the productive value of being around skilled workers rising? What does the rise of the skilled city mean?: What does the rise of the skilled city mean? Or, perhaps are skilled cities become more attractive places to live? Test using wage changes, housing price changes and income changes The skill premium (i.e. the extra wages associated with being around skilled people) are rising quickly Housing prices rise almost enough to keep real wages constant Skills and City Growth: Skills and City Growth Also predicts growing income: Also predicts growing income Interpretation: Interpretation The natural interpretation of this is that skills are working through labor demand, not labor supply. But it is true that the skills effect still works within metro areas (which are common labor markets) One startling fact is that skills matter for older, colder places, not newer warmer cites: the reinvention hypothesis Declining Regions: Declining Regions Growing Region (the West): Growing Region (the West) The Reinvention Hypothesis: The Reinvention Hypothesis An alternative interpretation– skills matter in times of shock (Schultz, Welch). Skilled cities excel because they permit innovation. As such, the key to reinvention is to keep skilled people from leaving. Boston’s Growth is one of Reinvention: Boston’s Growth is one of Reinvention In 1630, Winthrop comes to Boston for consumption, not production reasons. City on the Hill-- a religious community. All other colonies are about production. Original export industry is some fishing and selling goods to new immigrants. American’s first city: American’s first city Boston is founded in 1630 with 150 settlers. Location is determined by the Charles river and clean water. Population rises to 7,000 in 1690. Population is 17,000 is 1740 when the city is overtaken by Philadelphia. The 1640 Crisis and Its Resolution: The 1640 Crisis and Its Resolution In the early 1640s, the flow of immigrants subsides. English revolution Bostonians respond by reinvention, not exit. Respond by selling basic foodstuffs and wood, but now to other colonies. The Colonial Model for Boston: The Colonial Model for Boston New England exported to other colonies 73 percent to the Southern Colonies and Caribbean (1770) 13 percent to England Goods were basic commodities 35 percent is fish (to West Indies 1770) 32 percent livestock 21 percent woodstock Basic Model : Basic Model Land in Virginia and Haiti is worth more growing tobacco and sugar The North has little it can export to Europe, so its land is worth less and it grows commodies. North is poorer than South in the 1700s. The 19th Century Reinvention: The 19th Century Reinvention But after 1790, Boston begins to grow again. Growth from 18,000 in 1790 to 90,000 in 1840 Kept pace with national population growth. It is maritime, not manufacturing. 10,000 in maritime trades 5,000 in manufacturing (less than Lowell) Boston as a Share of the U.S.: Boston as a Share of the U.S. What Happened?: What Happened? Boston’s port is still inferior to NYC. Between 1821 and 1841, Boston’s share of trade drops from 21 percent to 10 percent. But Bostonians increasingly own and man the ships. Boston’s share of registered tonnage rises from 45 to 58 percent between 1811 and 1851. “Yankees captures New York Port around 1820 and dominated its activity until the Civil War” (Albion, 1931). What Happened, Continued: What Happened, Continued Boston’s comparative advantage was in human capital– both at the high end (merchants) and in sailers. Over the 1790-1840 period, technology and politics increased globalization of trade. China trade and South Africa Whaling far from New England Clipper Ships The human capital became more important than the port location. Live by the Clipper Ship, Die By …: Live by the Clipper Ship, Die By … In the 1840s, steam ships start becoming more important than sail. Boston’s human capital becomes far less valuable. Boston’s loses it maritime dominance, never to regain it. But Reinvention Once Again: But Reinvention Once Again Over the 1840-1920 period, Boston would continue to boom. Manufacturing replaced maritime. Improvements in engine technology helped the city in two ways Freed Manufacturing form river power Created Rail Networks And then there’s the Irish: And then there’s the Irish Boston starts becoming Irish in the 1840s. The Potato famine coincides with last era of Boston maritime dominance. As a result, its cheaper for the Irish to go from Liverpool to Boston than to NYC– this will not be true for later migrants. The Twentieth Century: The Twentieth Century Manufacturing left cities Car cities replaced higher density areas People fled cold places The rich fled redistributive cities. In the 1970s, Boston was in bad shape: In the 1970s, Boston was in bad shape Population had been declining for decades The economy was in shambles Housing cost less than new construction in most of the area. But since 1980, the city has surged: But since 1980, the city has surged Population has grown modestly The economy has grown robustly Housing prices have soared. Economic Growth since 1980: Economic Growth since 1980 In 1980, per capita income is the Boston Metro Area was $7547 which meant it ranked 61st in the nation. In 1994, personal income was 26,093 tenth in the nation. In 1996, average annual pay was 34,383, sixth in the nation. Middlesex County Employment: Middlesex County Employment Professional, Scientific and Technical Services– 110,000 jobs or 13 percent Educational Services– 64,000 jobs or 7 percent Administrative and Support Services– 64,00 jobs or 7 percent Computer and Electronic Manufacturing– 58,000 jobs or 7 percent/ Urban Wages (JOLE, 2001): Urban Wages (JOLE, 2001) Wages are higher in big cities than in small towns This is a nominal wage difference, not a real wage difference There is no labor supply puzzle, but there is a labor demand puzzle. b Nominal Wages and City Size(Slope=.073, R-Squared=.3): Nominal Wages and City Size (Slope=.073, R-Squared=.3) Real Wages and City Size 1970: Real Wages and City Size 1970 Real Wages and City Size Today: Real Wages and City Size Today Is it Selection? : Is it Selection? Wage Premium for metropolitan area residence .2-.35 depending on source What about the real wage facts today Controlling for standard omitted factors (education, industry, occupation) makes little difference Controlling for AFQT in the NLSY makes no difference More on selection: More on selection Parent’s location when used as an instrument predicts higher wages today. But individual fixed effects regressions do generally eliminate much of the city effect .28 to .05 in PSID .24 to .1 in NLSY What’s going on here? The Learning Hypothesis: The Learning Hypothesis If cities increase human capital only slowly, then this can explain the individual fixed effect results without selection Urban dummy is small for young workers (under 10 percent) But rises more than 15 percent over time Also true in fixed effect regressions– a 15 percent increase over time Analysis of Movers: Analysis of Movers Ashenfelter dip before leaving or moving to cities 7 percent gain or so within a few years Increasing wage gains over time The NLSY results show somewhat quicker wage growth People who leave cities don’t face wage losses. Learning in Cities (Journal of Urban Economics, 1999): Learning in Cities (Journal of Urban Economics, 1999) To understand the previous section, a brief model with two skill levels (the paper does a more general distribution). Your probability of becoming skilled involves (1) meeting a skilled person in your industry and (2) imitating that person (with prob. C) If the share of skilled people in an area is “s,” then the probability of becoming skilled from each interaction is cs/I. More on learning: More on learning The key assumption is that the number of meetings is a function of city size or density, or D(N) where N is population. The probability of becoming skilled in a period equals 1-(1-cs/I)D(N) If city rent+transports=aN/2, and unskilled wages=w and the gain from being skilled is V then Closing the learning model: Closing the learning model Spatial equilibrium requires (1-(1-cs/I)D(N) )V-aN/2=(1-cs/I)D(N’) V-aN’/2 The gains from extra learning are offset by higher rents. If there are just two locations– one with no learning and the other, a city then Comparative Statics: Comparative Statics City size rises with returns to learning, discount factor and falls with A. The skill level of the city will itself also be a function of the learning parameters. With multiple skill levels, the skill distribution is uniform. Information Technology and the Future of Cities (Journal of Urban Economics, 1998): Information Technology and the Future of Cities (Journal of Urban Economics, 1998) So cities exist in part to speed information flows Doesn’t that mean that information technology will kill cities? Not so fast– the key question is whether face-to-face interactions and electronic interactions are complements or substitutes A Simple Model: A Simple Model Step 1– learn reservation value (denoted j with cumulative distribution R(j) and choose whether or not to collaborate Step 2– learn match quality a, which means match returns are af(i) where “i” is intensity Step 3– produce intensity using elecronic media (phones) or face-to-face Phones vs. Face-to-Face: Phones vs. Face-to-Face Two technologies differ in their fixed costs and in their power iP=BPT and if=Bf(T-T), where Bf>Bp Phones don’t have fixed costs, but they are worse at creating intimacy. Use phones whenever desired “i” is low. Solving the model: Solving the model There are two cutoff values for “a” The lower values determines a level of a at which is makes sense to end the relationship A higher values above which people use face to face interactions Better electronic technologies are increases in BP, which impacts several margins Improvements in Technology: Improvements in Technology First it decreases the cutoff of “a” at which you interact at all. Second, it increases the cutoff at which you switch from phones to face-to-face. Third, it lowers the cutoff for the initial participation decision. What does this mean?: What does this mean? First, improvements in technology may actually increase the amount of face-to-face contact, by increasing the number of people who work together. Second, if cities are a technology for lowering the fixed costs of face-to-face, then demand for cities will rise if improvements in technology raise face-to-face contact. Third, the key condition for this to hold is that people in cities use phone technology more. What does the data say?: What does the data say? Fact # 1: Phones and cities go together across countries, and over time. Fact # 2: Business travel has risen over the past 20 years (face to face) Fact # 3: Co-authorship and other forms of interaction are rising steadily. Fact # 4: High tech industries are particularly likely to urbanized More on the data: More on the data Fact # 5: Silicon Valley is clustered Fact # 6: People in cities often use electronic forms of interaction more, not less. Overall– there is no compelling case that cities and technology are complements, but none that they are substitutes either.