Upload
new-economic-school
View
600
Download
3
Embed Size (px)
DESCRIPTION
14 июля / Суздаль Началась первая летняя школа трёхлетнего проекта РЭШ "Роль географии в экономике: теория и эмпирика".
Citation preview
1
Empirics of Economic Geography, part 1
Suzdal, 2010
2
Stylized facts of spatial development patterns1. Inequality (!)
3
Earth at night
4
Rappaport, Jordan & Sachs, Jeffrey D, 2003. " The United States as a Coastal Nation," Journal of Economic Growth, Springer, vol. 8(1), pages 5-46, March.
5
Stylized facts of spatial development patterns1. Inequality (!)
Physical geography plays an important part Role of natural endowment?
Yes, in agricultural societies Yes, for primary industries and if transport costs are
high
2. Persistence Physical geography plays a role in initial
concentration of economic activity Lock-in effects work long after the initial factors
become unimportant
6
7
Stylized facts of spatial development patterns1. Inequality (!)
Physical geography plays an important part Role of natural endowment?
Yes, in agricultural societies Yes, for primary industries and if transport costs are high
2. Persistence Physical geography plays a role in initial concentration of
economic activity Lock-in effects work long after the initial factors become
unimportant (agglomeration externalities)
8
Agglomeration externalities Questions for the empirical research:
Concentration increases productivity (?) Is this true? What do the data say? How to measure?
How do firms choose location? Do firms prefer to cluster?
What factors determine concentration? Why concentrate in a particular location? Location-specific Industry-specific
9
The determinants of spatial concentrationKim, S., 1995 “Expansion of Markets and the Geographic distribution of
economic activities”, Quarterly Journal of Economics 110:881-908
Let Is,t – measure of concentration (t = time, s = industry)
Estimate
Is,t = Xs,t β + γs + δt + εs,t
choose a set of explanatory variables Xs,t (size of plant, share of raw materials)
Why are industries concentrated? Because it is profitable (increasing returns)?
Test: are industries with stronger increasing returns more concentrated spatially? Answer: yes.
10
The determinants of spatial concentration, cont. Omitted variables problem
Omitted variable can drive both concentration and plant size
What is missing? Trade costs Intermediate goods Demand factors (elasticity of substitution) Production factors and production technologies History ….
Omitted variables lead to bias in β estimates Solution (partial): panel data with individual effects
Omitted variables that vary with time and industry are still a problem
11
The determinants of spatial concentration, cont. How do we measure concentration?
(S. Kim had Gini coefficients) Ellison G., and E.L. Glaeser 1997, “Geographic
Concentration in U.S. Manufacturing Industries: a Dartboard Approach”, Journal of Political Economy 105: 889-927 Index has to be neutral to the # of firms in the industry
But the topic of measuring concentration needs a whole lecture Choice of industrial disaggregation level Choice of spatial unit, etc … for the next session
12
The determinants of local productivity Consider a production function:
yj = Aj (sjlj)μ kj1-μ
Aj – technology level, sj – labor productivity What do A,s depend on? Parameterize, estimate
production function. Problems? The usual: 1)endogeniety, 2) omitted variables
If we have wage data Use firm’s FOC
13
Firm’s maximization problem πj = ∑b pjbyjb – wjlj – rjkj max
pjb – price in region b, yjb – quantity exported into region b
let pj = ∑b pjb(yjb/yj)
πj = pjyj – wjlj – rjkj
From FOC: wj = μpjAjsj
μ(kj/lj)1-μ and rj = (1-μ)pjAjsjμ(kj/lj)-μ
or wj = μ(1-μ)(1-μ)/μ sj (pjAj /rj1-μ)1/μ
or (wage equation):
ln wj = const + ln sj + 1/μ ln pj + 1/μ ln Aj –(μ-1)/μ ln rj
14
Estimating wage equations:ln wj = const + ln sj + 1/μ ln pj + 1/μ ln Aj –(μ-1)/μ ln rj
Parameterize A and/or s Test whether they increase with density
example: ln wrs = α + β ln denr + εrs
Problems? Omitted variables, again Endogeniety
density is endogenous self-sorting Solutions: IV, panel data
15
Combes, Duranton & Gobillon 2008. "Spatial wage disparities: Sorting matters!," Journal of Urban Economics, vol. 63(2), pages 723-742. Large panel data on individual workers
worker fixed effect (= ability) lifetime learning (proxied by age) industry effect time effect
Compare with the results obtained with average regional wages
worker sorting explains 50% of agglomeration economies estimate individual data: double the density productivity ↑ by 2% aggregate data: …. by 5%
Sorting matters, reduces estimates. But agglomeration economies remain >0
16
How do firms choose location? Logit framework for empirics of location
choice: Carlton (1983) used McFadden’s logit model 1. Several alternative locations, firm chooses one
2. Ur - location-specific factors, common
3. εr - firm-specific shock for each location, individual
4. Locate in r iff πr* > πs* for all s.
5. If F(εr) = exp(exp(-εr)),
then P{choose location r} = exp(Ur)/∑s exp(Us) 6. Linearize, estimate
17
What are the factors that attract firms?
…exogenous Low costs Higher productivity Being closer to consumers, to save on trade
costs Market potential
18
Market potential: history of thought Old idea (Harris, 1954)
MPr = ∑s Ys/drs
Evidence from growth literature: proximity to the developed countries increases GDP
NEG models theory-based explicit expression for RMP
RMP = ∑s φrsμsYsPsσ-1
(since π = cr1-σ/σ * (∑s φrsμsYsPs
σ-1) – Fr = cr1-σ/σ *M – Fr )
19
Head & Mayer, 2004 (φ =?)
20
21
22
23
24
25
Head and Mayer, 2004: conclusions Market potential attracts firms, but…
Harris’ measure performs better than Krugman’s Nonstructural agglomeration controls are
significant why do firms cluster? DSK: because of downstream
linkages (customers). But this is not the whole story. Are there important upstream and horizontal linkages? Or omitted variable bias?
Wages are +, unemployment -, subsidies – endogeniety
26
Stability of spatial development patterns Why are agglomerations so persistent?
Location fundamentals are good Agglomeration externalities are strong Which are more important for regional
development? Regional policy is useless if location fundamentals
prevail in long run
27
(Davis, Weinstein (2002), “Bones, Bombs, and Breakpoints…” “Bones” part – regional archeological data Lessons:
persistence in density levels and rank (Could this be explained by the physical geography?)
same level of concentration before XXth century, increasing after
closure to trade decreases concentration
Variation in regional population density is great for all historical periods and persistent through history. Does this look like random growth? Zipf’s law holds (Could be explained by the features of
physical geography)
28
29
30
31
32
(Davis, Weinstein (2002), “Bones, Bombs, and Breakpoints…”, cont.
“Bombs and Breakpoints” – WWII bombing data
Estimate ρ (How? Use bombing data for identification)
Lessons: mean-reversion: ρ = 0 by 1965 (even if controlled for
reconstruction efforts!) No breakpoints (no multiple equilibria)?
Si = i +it where it+1 = it + vit+1
33
34
35
(Davis, Weinstein (2002), “Bones, Bombs, and Breakpoints…”, cont. Lessons, cont:
mean-reversion: ρ = -1 by 1965 (even if controlled for reconstruction efforts!)
Tails of the distribution are “nice”. Where are the breakpoints?
What is a role of increasing returns and location fundamentals? Historically: first geography, then IR At present: unknown. War shocks destroy infrastructure, but
do not destroy either location fundamentals or agglomeration externalities.
Other countries? Miguel, Roland (2005): in Vietnam spatial structure
unaffected by bombing in long run
36
37
Davis, Weinstein (2008) “A Search For Multiple Equilibria In Urban Industrial Structure”, JRSc The war shock to the industrial production is
even larger
38
39
Cities were specialized
40
But there is still mean-reversion in aggregate manufacturing shares…
41
And even in the shares of each industry separately
Single spatial eq’m?
42
43
More natural experiments: Germany
Germany bombing (Garretsen, Schramm, Brakman (2003)): mean-reversion in West Germany no mean-reversion in East Germany ρGer < ρJap
Is it due to longer adjustment period, or is Germany converging to a different equilibrium?
Bosker, Brakman, Garretsen,
Schramm (2005):
2 equilibria,
geography matters
44
More natural experiments: Germany, cont. German division and reunification
Redding, Sturm, Wolf, (2007) “History and Industry Location: Evidence from German Airports” Major air hub has shifted from Berlin to Frankfurt after division … and did not shift back to Berlin after unification …while if not for the division, Berlin would be a hub
45
46
Redding, Sturm, Wolf, (2007) “History and Industry Location: Evidence from German Airports” Is this a change in economic fundamentals or
multiple equilibria? Measure the “fundamentals” and compare Frankfurt and
Berlin From international data on cities and air traffic: Berlin today
has all the economic advantages to be a hub Projected losses of passenger traffic from relocating the hub
are small relative to sunk costs
47
multiple possible hub locations
48
German division and reunification: role of distance and market access Growth literature: development and market access
correlate. Causality? Redding, Sturm (2008) “The Costs of Remoteness: Evidence
from German Division and Reunification”, AER German division = a shock to market access for the
cities close to the border What are the consequences for development?
Strategy: write a NEG model, estimate some parameters, calibrate the model to German cities, compare estimations with calibrations find the parameters significant impact of division negligible impact of reunification
49
50
Setup questions
Division and reunification = shocks to market access (market potential) for the cities on the border
were the shocks unexpected? Does it lead to the decline of the affected cities? (treatment
group) yes, it does
Are there alternative explanations for the decline? pre-existing dynamics and changes in industry structure differences in wartime destruction, refugees economic integration with the western neighbors fear of armed conflict need to rule out
51
Theoretical framework Helpman (1998) “The size of regions”
fixed number of cities = C, each endowed with nontradeables = HC (amenity)
Consumption: μ = share of tradeables (differentiated as per D-S, σ>1), 1-μ = share of amenities
city-to-city transport cost matrix: Tij
labor: inelastic supply=1, free mobility get the (long-run!) size of the city as a fnc of:
consumer price index firm market access index amenity stock
52
Calibrations: μ=2/3, σ=4, T = (distance)φ, φ=1/3 after division T→∞ for the cross-border terms
53
The effect is stronger for small cities (HME!)
54
Estimations
Errors: clustering by city (why?) Why need β?
55
56
57
58
Results:
small cities suffer more, indeed 75 km ≈ border cutoff Regional policy act (1965) = aid to the cities
on the immediate border (<25 km) all the effects are underestimated!
+ robustness checks
59
Quantitative analysis grid search over parameters (μ,σ,φ) in single-
equilibrium range (σ(1-μ)>1)
60
Parameter of interest = distance coefficient (1-σ)φ ≈ 1.6 > than in international trade gravity estimates and < than in interregional trade (= 1.76) Land vs sea transport?
61
Structural changes (robustness check)
62
Destruction, refugees and migration
63
Western integration
64
Reunification
65
Reunification:
The change in market access is much smaller than from division
convergence between East & West recovery of the border cities