Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
Lecture 7: Industrial Concentration and Systems ofCities
WWS 582a
Esteban Rossi-Hansberg
Princeton University
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 1 / 40
Duranton and Overman (2005)
Need a way to measure industrial concentration
What seems like industrial concentration might be misleading due to thesmall number of plants
I The law of large numbers does not apply
Ellison and Glaeser (1997) dartboard approachI E.g. 75% of the employees in the U.S. vacuum cleaner industry work in one offour main plants
I Even if these plants locate separately, four locations must account for at least75% of the employment in this industry without it being localized in anymeaningful way
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 2 / 40
Some Examples1082 REVIEW OF ECONOMIC STUDIES
(a) Basic Pharmaceuticals(SIC2441)
(b) Pharmaceutical Preparations(SIC2442)
(c) Other Agricultural and ForestryMachinery (SIC2932)
(d) Machinery for Textile, Apparel andLeather Production (SIC2954)
FIGURE 1
Maps of four illustrative industries
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 3 / 40
Characteristics of a Measure of Localization
Any test of localization should rely on a measure which:1 is comparable across industries;2 controls for the overall agglomeration of manufacturing;3 controls for industrial concentration;4 is unbiased with respect to scale and aggregation
Their measure considers the distribution of distances between pairs ofestablishments in an industry and compares it with that of hypotheticalindustries with the same number of establishments which are randomlydistributed conditional on the distribution of aggregate manufacturing
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 4 / 40
The Measure of Localization
Data selection: Sometimes cutting out the lower tail might be useful becauseestablishments do something else
Let dij denote the distance between establishments i and j then theK−density at distance d is given by
K (d) =1
n (n− 1) hn−1∑i=1
n
∑j=1+1
f(d − dijh
)where f is the Gaussian kernel, n is the number of establishments, and h isthe bandwidth.
Need to use a Kernel comes form the error in measuring "effective distance"
Note that the number and size of plants is taken as given
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 5 / 40
The Examples1084 REVIEW OF ECONOMIC STUDIES
(a) Basic Pharmaceuticals(SIC2441)
(b) Pharmaceutical Preparations(SIC2442)
(c) Other Agricultural and ForestryMachinery (SIC2932)
(d) Machinery for Textile, Apparel andLeather Production (SIC2954)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
FIGURE 2
K -density, local confidence intervals and global confidence bands for four illustrative industries
the entire industry population. If, instead of a census, we had a random sample of firms from eachindustrywe would need to worry about the statistical variation due to the estimation of the actualK -density. Applications of the techniques developed below to samples of firms from particularindustries could allow for this statistical variation to be taken into account but the exhaustivenature of our data means that we are able to ignore it in what follows (seeEfron and Tibshirani,1993, andQuah,1997, for further discussion of these issues as well asDavison and Hinkley,1997, for a discussion more focused on point patterns).
Thesecond difference stems from the fact that the spatial nature of our data implies strongdependence between the bilateral distances that are used to calculate the density. This strongdependence arises because the observations of interest are actually the points that generatethese bilateral distances. Even if the underlying points are independently located, the bilateraldistances between these points will not be independent.6 This has implications for the samplingtheoryof our estimator,K A(d). In situations where the observations are independent (or only
6. See below for more on this issue.
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 6 / 40
Counterfactuals
Still need to control for the overall tendency of manufacturing to agglomerate
Draw locations for plants in feasible "sites" and calculate the counterfactualdensity fixing the number of plants
I Sampling without replacement
Run many simulations and take the top 5% and lowest 5% in the generatedsample
The result is a confidence interval for what would happen if the industry wasrandomly located
When for industry A, KA(d) > KA(d), this industry is said to exhibitlocalization at distance d (at a 5% confidence level)
Symmetrically, when KA(d) < KA(d), this industry is said to exhibitdispersion at distance d (at a 5% confidence level)
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 7 / 40
The Index
Define an index of localization as
γA (d) ≡ max(KA(d)−KA(d), 0
)To reject the hypothesis of randomness at distance d because of localization(dispersion), we only need γA (d) > 0
Similarly the dispersion index is calculated as
ψA (d) ≡ max(KA(d)− KA(d), 0
)Similarly global measures are calculated when we use a bound determined bythe top 5% or realizations at any given distance
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 8 / 40
The Examples1084 REVIEW OF ECONOMIC STUDIES
(a) Basic Pharmaceuticals(SIC2441)
(b) Pharmaceutical Preparations(SIC2442)
(c) Other Agricultural and ForestryMachinery (SIC2932)
(d) Machinery for Textile, Apparel andLeather Production (SIC2954)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
0·005
0·004
0·003
0·002
0·001
0·0000 20 40 60 80 100 120 140 160 180
Distance (km)
FIGURE 2
K -density, local confidence intervals and global confidence bands for four illustrative industries
the entire industry population. If, instead of a census, we had a random sample of firms from eachindustrywe would need to worry about the statistical variation due to the estimation of the actualK -density. Applications of the techniques developed below to samples of firms from particularindustries could allow for this statistical variation to be taken into account but the exhaustivenature of our data means that we are able to ignore it in what follows (seeEfron and Tibshirani,1993, andQuah,1997, for further discussion of these issues as well asDavison and Hinkley,1997, for a discussion more focused on point patterns).
Thesecond difference stems from the fact that the spatial nature of our data implies strongdependence between the bilateral distances that are used to calculate the density. This strongdependence arises because the observations of interest are actually the points that generatethese bilateral distances. Even if the underlying points are independently located, the bilateraldistances between these points will not be independent.6 This has implications for the samplingtheoryof our estimator,K A(d). In situations where the observations are independent (or only
6. See below for more on this issue.
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 9 / 40
Findings Across Industries
1090 REVIEW OF ECONOMIC STUDIES
TABLE 1
Localizationat three thresholds for four-digit industries
Percentage of four-digit industries localized at:
5 km 5 km only 5 and 30 km only 5 and 150 km only 5, 30 and 150 km39·3 6·4 22·6 0·9 9·4
30 km 30 km only 30 and 150 km only38·9 6·0 0·9
150 km 150 km only17·1 6·0
(a) Global localization
100
90
80
70
60
50
40
30
20
10
00 20 40 60 80 100 120 140 160 180
Distance (km)
(b) Global dispersion
100
90
80
70
60
50
40
30
20
10
00 20 40 60 80 100 120 140 160 180
Distance (km)
(c) Local localization
100
90
80
70
60
50
40
30
20
10
00 20 40 60 80 100 120 140 160 180
Distance (km)
(d) Local dispersion
100
90
80
70
60
50
40
30
20
10
00 20 40 60 80 100 120 140 160 180
Distance (km)
FIGURE 3
Number of four-digit industries with local/global localization and dispersion
across distances, but not across the two figures.13 It is immediately apparent that the extent oflocalizationis much greater at small distances than large distances. As before, dispersion doesnot show any marked pattern. The important conclusion we draw here is that localization tendsto take place mostly at fairly small scales.
13. This is because for an industry that exhibits localization the density is unbounded from above whereas thedensityof an industry that exhibits dispersion is bounded from below by zero.
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 10 / 40
Taking Intensity into AccountDURANTON & OVERMAN TESTING FOR LOCALIZATION 1091
(a) Global localization
0·240·220·200·180·160·140·120·100·080·060·040·020·00
0 20 40 60 80 100 120 140 160 180Distance (km)
(b) Global dispersion
0·240·220·200·180·160·140·120·100·080·060·040·020·00
0 20 40 60 80 100 120 140 160 180Distance (km)
FIGURE 4
Index of global localization and dispersion by distance
0·50
0·45
0·40
0·35
0·30
0·25
0·20
0·15
0·10
0·05
0·000 20 40 60 80 100120140160180200220240
FIGURE 5
Distribution of global localization and dispersion by four-digit industries
Differences between industries
We now turn to the examination of differences between industries. We start by constructinga measure of the extent to which different industries deviate from randomness. Proceedingas before, for each industryA we can define the following cross-distance indices:0A ≡∑180
d=00A(d), and9A ≡∑180
d=09A(d). Respectively, these measures are the sum for eachindustry of the index of global localization and dispersion across all levels of distance. Toillustrate the variations in industry outcomes, we rank industries by decreasing order of theseindices and plot them inFigure 5. The upper line is the measure of localization, the lower that ofdispersion.As is immediately clear, there are a few industries that show very high localization ordispersion, but the majority of industries do not see such extreme outcomes. This highly skeweddistribution of localization confirms previous findings (Ellison and Glaeser(1997),Maurel andSedillot (1999),Devereuxetal. (2004)).
To give some idea of the reality underlyingFigure 5,Table 2lists the 10 most localizedindustriesand the 10 most dispersed. Interestingly, more than a century afterMarshall(1890),Cutlery (SIC2861) is still amongst the most localized industries. Six textile or textile-related
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 11 / 40
Particular Industries
1092 REVIEW OF ECONOMIC STUDIES
TABLE 2
Mostlocalized and most dispersed four-digit industries
SIC92 Industry 0 or9
Most localized
2214 Publishing of sound recordings 0·4701711 Preparation and spinning of cotton-type fibres 0·4112231 Reproduction of sound recordings 0·4031760 Manufacture of knitted and crocheted fabrics 0·3211713 Preparation and spinning of worsted-type fibres 0·3192861 Manufacture of cutlery 0·3141771 Manufacture of knitted and crocheted hosiery 0·2901810 Manufacture of leather clothes 0·2031822 Manufacture of other outerwear 0·1812211 Publishing of books 0·178
Most dispersed
1520 Processing and preserving of fish and fish products 0·2003511 Building and repairing of ships 0·1131581 Manufacture of bread, fresh pastry goods and cakes 0·0942010 Saw milling and planing of wood, impregnation of wood 0·0822932 Other agricultural and forestry machinery 0·0671551 Operation of dairies and cheese making 0·0641752 Manufacture of cordage, rope, twine and netting 0·0623615 Manufacture of mattresses 0·0501571 Manufacture of prepared feeds for farm animals 0·0492030 Manufacture of builders’ carpentry and joinery 0·047
industries are also in the same list together with three media-based industries. These highlylocalizedindustries are fairly exceptional. In contrast, the mean industry (after ranking industriesby their degree of localization) is barely more localized than if randomly distributed. It is mostlyfood-related industries together with industries with high transport costs or high dependence onnatural resources that show dispersion.
Our main focus in this paper is on the proportion of manufacturing sectors that are localized.However, it is interesting to notice that a number of industries that appear inTable 2are fairlysmall in terms of overall employment. This raises the question as to whether the percentage ofmanufacturing workers employed in localized industries is above or below the percentage ofsectors that are localized. Weighting sectors by their share in manufacturing employment, wefind that 67% of U.K. manufacturing employers work in sectors that are localized. This showsthat localized sectors tend to have a larger share of manufacturing employment. Offsetting this,however, is the fact that the employment share weighted mean of the index of globalization,0A, is 30% lower than the unweighted mean of the index. That is, larger sectors tend to be lessstrongly localized.
Finally, it is also interesting to notice that for many (two-digit) branches, related industrieswithin the same branch tend to follow similar patterns.Table 3 breaks down localizationof industries by branches. For instance nearly all Food and Drink industries (SIC15) orWood, Petroleum, and Mineral industries (SIC20, 23 and 26) are not localized. By contrast,most Textile, Publishing, Instrument and Appliances industries (SIC17–19, 22 and 30–33) arelocalized. The two main exceptions are Chemicals (SIC24) and Machinery (SIC29). In thesetwo branches, however, the more detailed patterns are telling. Chemical industries such asFertilisers (SIC2415) vertically linked to dispersed industries are also dispersed whereas thoselike Basic Pharmaceuticals (SIC2441) or Preparation of Recorded Media (SIC2465) verticallylinked to localized industries are themselves very localized. The same holds for machinery: Other
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 12 / 40
Particular IndustriesDURANTON & OVERMAN TESTING FOR LOCALIZATION 1093
TABLE 3
Localizationby two-digit branch
Two-digit branch Number of No. global No. globalfour-digit localization localizationindustries ≤60 km >60 km
15 Food products and beverages 30 1 016 Tobacco products 1 1 017 Textiles 20 16 918 Wearing apparel, dressing, etc. 6 6 319 Tanning and dressing of leather, footwear 3 3 320 Wood and products of wood, etc. 6 0 021 Pulp, paper and paper products 7 2 122 Publishing, printing and recorded media 13 13 823 Coke, refined petroleum products 3 0 024 Chemical and chemical products 20 8 825 Rubber and plastic products 7 1 326 Other non-metallic mineral products 24 4 227 Basic metals 17 11 1028 Fabricated metal products 16 9 1229 Other machinery and equipment 20 6 930 Office machinery and computers 2 2 231 Electrical machinery 7 2 532 Radio, televisions and other appliances 3 3 333 Instruments 5 3 434 Motor vehicles, trailers, etc. 3 1 335 Other transport equipment 8 2 236 Furniture and other products 13 4 5
Aggregate 234 98 92
Agricultural and Forestry Machinery (SIC2932) is very dispersed like most agriculture-relatedindustries,whereas Machinery for Textile, Apparel and Leather Production (SIC2954) is verylocalized like most textile industries.
5. ESTABLISHMENT SIZE AND LOCALIZATION
Four main conclusions emerge so far: (i) 52% of industries are localized, (ii) localization mostlytakes place at small scales, (iii) deviations from randomness are very skewed across industriesand (iv) industries that belong to the same branch tend to have similar localization patterns. Thesefindings may be driven by particular types of establishments or particular sectoral definitions. Togain insights about the size of localized establishments and the scope of localization, we replicateour analysis with alternative samples of plants and alternative sectoral definitions. This sectiondeals with size issues; questions relating to scope are examined in the next section.
Note that the issue of size may be particularly crucial as firm-size distributions arevery skewed in most industries. In our population of plants, 36% of establishments employtwo persons or less and represent only a very small fraction (2·4%) of total manufacturingemployment. The issue of firm size is also important from a policy perspective. Policiesencouraging dispersion are not likely to be very successful if it is only small establishments thatcan be dispersed, whereas clustering policies might be more difficult to implement if it is onlylarge establishments that cluster. Finally, the type of establishments, big or small, that cluster ordisperse is potentially very informative about the relevance of particular theories.
at Princeton University on A
pril 4, 2012http://restud.oxfordjournals.org/
Dow
nloaded from
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 13 / 40
Desmet and Rossi-Hansberg (2013)
Why do people live in particular cities? Agents canI be more productive (productivity advantages, high price of tradeables)I enjoy the city (amenities, geography)I frictions may be low (urban costs, taxes, infrastructure, other market frictions)
We use a simple urban theory to calculate these components for the U.S.economy
I "Wedge" analysis as in Chari, et al. (2007)I With and without externalities in productivity and amenities
Wide set of counterfactual exercises that explain the relative importance ofthese characteristics for welfare
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 14 / 40
Findings
We find that eliminating differences in any of these characteristics leads toI Small changes in welfareI Large population reallocations
Externalities have an overall small effect but lead to important city selection
The effect of productivity and amenity shocks is substantially reduced by theurban structure
Provide a simple methodology to compare urban systems across countriesI Illustrate using the cases of the U.S. and China
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 15 / 40
The Model
Standard model of a system of cities with:I Elastic labor supply so that labor taxes create distortionsI Nt identical agents choose where to live and workI Cities have idiosyncratic productivities and amenitiesI Mono-centric cities that require commuting infrastructures that citygovernments provide by levying labor taxes
I City governments can be more or less effi cient in the provision of the publicinfrastructure. We refer to this variation as a city’s excessive frictions.
Later add externalities in productivity and amenities
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 16 / 40
Technology
Goods are produced in I mono-centric circular cities with sizes NitCities have a local level of productivity Ait . Production in a city i in period tis given by
Yit = AitKθitH
1−θit
The standard first order conditions of this problem are
wit = (1− θ)YitHit
= (1− θ)yithit
rt = θYitKit
= θyitkit
Capital is freely mobile across locations so there is a national interest rate rtWe can then write down the “effi ciency wedge”which is identical to the levelof productivity, Ait , as
Ait =Yit
K θitH
1−θit
=yit
kθith1−θit
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 17 / 40
Preferences
Agents order consumption and hour sequences according to
∞
∑t=0
βt [log cit + ψ log (1− hit ) + γi ]
where γi denotes the amenities associates with city i
The problem of an agent with capital k0 is therefore
max{it ,cit ,hit ,kit}∞
t=0
∞
∑t=0
βt [log cit + ψ log (1− hit ) + γi ]
subject to
cit + xit = rtkit + withit (1− τit )− Rit − Titkit+1 = (1− δ) kit + xit ,
In steady state kit+1 = kit and xit = δkit . Furthermore, we assume kit issuch that rt = δ (capital is at the Golden Rule level)
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 18 / 40
The Labor Wedge
The simplified budget constraint of the agent becomes
cit = withit (1− τit )− Rit − Tit .
The first order conditions of this problem are given by
1cit= λit ,
andψ
11− ht
= wit (1− τit ) λit ,
So the labor wedge τ is given by
(1− τit ) =ψ
(1− θ)
cit1− hit
hityit
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 19 / 40
Commuting Costs and Land RentsCities are mono-centric, all production happens at the center, and people livein surrounding areas characterized by their distance to the center, d
Each agent lives on one unit of land and commutes from his home to work.Commuting is costly in terms of goods, T (d) = κd
We normalize the price of agricultural land to zero. Since land rents arecontinuous in equilibrium, R (d) = 0.
Since all agents in a city are identical,
Rit (d) + T (d) = T (dit ) = κdit
Hence
Rit (d) + T (d) = κ
(Nitπ
) 12
all d
Average land rents are equal to
ARit =2κ
3
(Nitπ
) 12
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 20 / 40
Government Budget Constraint and Frictions
The government levies a labor tax, τit , to pay for the transportationinfrastructure
I It requires κgit workers per mile commuted to build and maintain urbaninfrastructure. So
G (hitwit ,TCit ) = githitwitκTCit = githitwitκ23
π−12N
32it
I Hence, git is inversely related to the effi ciency of the government in providingurban infrastructure
I The government budget constraint is then given by
τithitNitwit = githitwitκ23
π−12N
32it
which implies that the “labor wedge” can be written as
τit = gitκ23
(Nitπ
) 12
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 21 / 40
Characterization of Equilibrium
Labor market equilibrium satisfies ∑Ii=1 Nit = Nt and all agents receive thesame utility level u
So given (Ait ,γit , git ) we can calculate Nit all i
In equilibriumI More productive cities are largerI Cities with larger amenities are largerI Larger cities have more frictions, but this tradeoff depends on how effi cientlocal governments are in providing urban infrastructure
F “Excess frictions” make cities smaller
We explore these derivatives using data on U.S. cities and paying attention tothe general equilibrium effects
I The empirical results are consistent with the theory
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 22 / 40
Testing the model with US data
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 23 / 40
Identifying City Characteristics
Need to calculate the triplet (Ait ,γit , git ) from the data
Obtain "effi ciency wedge" from
Ait =yit
kθith1−θit
I We can do this with or without capital data
Calculate "labor wedge" from
(1− τit ) =ψ
(1− θ)
cit1− hit
hityit
Then obtain ln git from
ln τit = α+12lnNit + ln git
Use model to obtain γit so as to match size distribution of cities with u = 10
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 24 / 40
Data
Data for all MSA in the U.S. between 2005-2008I Cities with population greater than 50 000, consistently measured after 2003
Consumption: No readily available data on consumption at MSA levelI Use retail earnings and adjust using national averagesI For housing consumptions use gross rents
Capital: use U.S. sectoral capital stocks and allocate it to MSAs according totheir shares in sectoral earnings
Hours worked: use Current Population Survey but eliminate MSAs with lessthan 50 observations
Housing rental prices: use American Community Survey
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 25 / 40
Parameters
Let ψ = 1.4841 and θ = 0.3358 as in McGrattan and Prescott (2009).
Let r = δ = 0.02 (assumptions on capital)
Useln τit = α+
12lnNit + ε5it
then we can identify κ from the estimate of α as the model implies. Weestimate κ = 0.0017
I We use κ = 0.002 but do robustness checks with other values of κ
If we eliminate all characteristics, welfare would increase by 3.26% and allcities would have 1 million 68 thousand people
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 26 / 40
The Effect of Kappa
10 11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Effect of given all Shocks
= 0.0005, Utility = 10.49 = 0.001, Utility = 10.31Actual: = 0.002, Utility = 10 = 0.004, Utility = 9.39 = 0.006, Utility = 8.79
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 27 / 40
Counterfactuals Without One Shock
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Model Utility = 10
Counterfactuals Without One Shock, = 0.002 , = 0 , = 0
ActualModeled
10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.1217, Reallocation = 0.367
ActualAvg. Efficiency
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0191, Reallocation = 0.19913
ActualAvg. Amenities
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0886, Reallocation = 0.4399
ActualAvg. Exc. Frictions
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 28 / 40
Counterfactuals With Only One Shock
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Model Utility = 10
Counterfactuals with Only One Shock, = 0.002 , = 0 , = 0
ActualModeled
-5 0 5 10 15 20-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0605, Reallocation = 0.44004
ActualEfficiency Only
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.291, Reallocation = 0.63365
ActualAmenities Only
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0499, Reallocation = 0.14564
ActualExc. Frictions Only
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 29 / 40
Reallocation
Calculate reallocation following Davis and Haltiwanger (1992) by adding thenumber of new workers in expanding cities
I Same effi ciency: 37% reallocation and welfare gains of 1.2%Example: New York would lose 77% of its population
I Same amenities: 20% reallocation and welfare gains of 0.2%Example: San Diego would lose 42% of its population
I Same excessive frictions: 44% reallocation and welfare gains of 0.8%Example: Trenton would gain 326% of its population
So very large reallocations, but small welfare gainsI Reallocation in the U.S. economy amounts to around 2.1% over 5 years
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 30 / 40
Geographic Distribution Without Differences in Amenities: Without Differences in Efficiency:
Without Differences in Excessive Frictions:
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 31 / 40
Changing the Level of Excessive Frictions
11 11.5 12 12.5 13 13.5 14 14.5 15-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Excessive Frictions Counterfactuals: US
All Excessive Frictions at 90th Percentile, Utility = 9.2023All Excessive Frictions at 50th Percentile, Utility = 9.7505All Excessive Frictions at 10th Percentile, Utility = 10.1663
So overall cost of excessive frictions is significant in levels: Role for policy
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 32 / 40
Adding Production Externalities
LetAit = AitN
ωit
where Ait is an exogenous characteristic and ω governs the elasticity ofproductivity with respect to size
Fairly robust estimate of ω in the literature, so use ω = 0.02
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 33 / 40
Counterfactuals with Production Externalities
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Model Utility = 10
Counterfactuals Without One Shock, = 0.002 , = 0.02 , = 0
ActualModeled
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.1094, Reallocation = 0.37565
ActualAvg. Efficiency
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0189, Reallocation = 0.21897
ActualAvg. Amenities
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0963, Reallocation = 0.47717
ActualAvg. Exc. Frictions
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 34 / 40
Adding Externalities in Amenities
Letγit = γitN
ζit
where γit is an exogenous characteristic and ζ governs the elasticity ofproductivity with respect to size
I As in the case of production externalities we let ζ = 0.02
Reallocation and welfare changes very similar
Less dispersion of city characteristics tends to decrease utility in the presenceof externalities
City selection effect is stronger
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 35 / 40
Counterfactuals Without One Shock and Both Externalities
11 12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Model Utility = 10
Counterfactuals Without One Shock, = 0.002 , = 0.02 , = 0.02
ActualModeled
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0784, Reallocation = 0.40752
ActualAvg. Efficiency
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 10.0324, Reallocation = 0.30766
ActualAvg. Amenities
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 9.9585, Reallocation = 0.49123
ActualAvg. Exc. Frictions
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 36 / 40
With Externalities but Only Average Characteristics
6 8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Actual, Utility = 10No Shocks, = = 0.02, Utility = 9.991No Shocks, = = 0.04, Utility = 9.877No Shocks, = = 0.06, Utility = 9.703
For ω = 0.02, 131 cities with only 613 agents and 61 with 3320745
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 37 / 40
Comparing with China
12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Model Utility = 10
China: Counterfactuals Without One Shock, = 0.001 , = 0 , = 0
ActualModeled
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 14.6992, Reallocation = 0.64395
ActualAvg. Efficiency
8 10 12 14 16 18-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 11.2977, Reallocation = 0.5001
ActualAvg. Amenities
12 13 14 15 16 17-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Counterfactual Utility = 9.8496, Reallocation = 0.070892
ActualAvg. Exc. Frictions
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 38 / 40
Comparing with China
12.5 13 13.5 14 14.5 15 15.5 16 16.5-6
-5
-4
-3
-2
-1
0
ln(population)
ln(p
rob
> p
opul
atio
n)
Excessive Frictions Counterfactuals: China
All Excessive Frictions at 90th Percentile, Utility = 9.2718All Excessive Frictions at 50th Percentile, Utility = 9.8461All Excessive Frictions at 10th Percentile, Utility = 10.2043
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 39 / 40
Conclusions
System of cities in U.S. such that large changes in city characteristics (orpolicy) have small effect on welfare but large effect on reallocation
With externalities, city selection becomes an important part of reallocation
Implies that the losses from lack of mobility are likely smallI Small mobility costs would yield negative welfare effects
More generally: Paper provides a simple GE methodology to compare urbansystems
I Identify main characteristics of citiesI Understand the effect of shocks and policyI Assess magnitude of welfare gains at stake
F Small in the US, but much larger in China
ERH (Princeton University ) Lecture 7: Ind. Conc. and Systems of Cities 40 / 40