[PPT] - Brixen Workshop and Summer School on International Trade and Finance PowerPoint Presentation

SLIDE 1

Brixen Workshop and Summer School on International Trade and Finance Part 4: Quasi-Experimental Geography Daniel Sturm London School of Economics

1

SLIDE 2

Brixen Workshop 2012

2 -

Daniel Sturm

1 Introduction

This part of the workshop starts from the old question what

determines the spatial distribution of economic activity.

Economic activity is highly unevenly distributed across countries

and also across regions within countries.

A key question is what causes these differences in the level of

economic activity.

We will focus mainly on the distribution of economic activity

across regions or cities within countries.

SLIDE 3

Brixen Workshop 2012

3 -

Daniel Sturm

There are at least two broad explanations for the unequal

distribution of empirical activity across space within countries: – Fundamentals: differences in the fundamental productivity of locations. – Agglomeration Forces: Proximity to other economic agents increases the attractiveness of a location.

These two mechanisms are obviously not exclusive and can both
perate at the same time.
The key empirical question is to what extent observed patterns of

activity are explained by these two mechanism.

SLIDE 4

Brixen Workshop 2012

4 -

Daniel Sturm

Whether fundamentals or agglomeration forces are responsible for

the pattern of economic activity has important implications for the persistence of spatial equilibria.

Suppose for example that only agglomeration forces are operating.
In this case the location of economic activity is rather arbitrary - a

location is attractive because other workers are locating there.

This is a bit like choosing a nightclub: club A is “cool” because all

the “cool” people go to club A rather than club B.

If instead only fundamentals are operating then the distribution of

activity is determined by the distribution of fundamentals.

SLIDE 5

Brixen Workshop 2012

5 -

Daniel Sturm

If agglomeration forces rather than fundamentals dominate the

spatial distribution of activity this also has key policy implications.

In this case regional policy could try to move the distribution of

activity between different equilibria.

With a temporary subsidy regions could try to attract a “critical

mass” of economic activity.

Once this critical mass has been established, the location remains

attractive even when the subsidy has ended.

SLIDE 6

Brixen Workshop 2012

6 -

Daniel Sturm

1.1 Overview

The theoretical roots of economic geography
Fundamentals versus agglomeration forces
Empirical evidence: Bombing
Empirical evidence: German division
Empirical evidence: Portage
Outlook for future research

SLIDE 7

Brixen Workshop 2012

7 -

Daniel Sturm

2 The Origins of Economic Geography

Krugman (1979) started the new trade theory.
This paper already contains almost all the ideas necessary for the

new economic geography model of Krugman (1991).

We will briefly review the key building blocks of Krugman (1979)

before looking at what economic geography has added to this model.

SLIDE 8

Brixen Workshop 2012

8 -

Daniel Sturm

2.1 Basic Assumptions of Krugman (1979)

There are two countries.
Labor is the only factor of production. Endowments L and L∗.
Many firms, which produce differentiated products i ∈ [1, n] .
Preferences are:

U =

n

∑

i=1

v(ci) (1) with v′ > 0 and v′′ < 0. These preferences incorporate “love of variety”.

SLIDE 9

Brixen Workshop 2012

9 -

Daniel Sturm

Technology is:

li = α + βxi (2) with α, β > 0.

Note that this production function has increasing returns to scale.
Firms are monopolistically competitive, i.e. they ignore the effects
f their decisions on others and set prices like a monopolist.
For now assume that there is no trade between the two countries.

SLIDE 10

Brixen Workshop 2012

10 -

Daniel Sturm

2.2 Equilibrium

In autarky the equilibrium in each country is a simply trade-off:

– Love of variety means that ceteris paribus consumers want as much variety as possible. – However, increasing returns mean that each new variety requires the fixed labor requirement α.

The number of varieties that a country produces in equilibrium

depends entirely on its labor endowment.

Key result: the larger country will produce more variety in

equilibrium and offer higher real wages.

SLIDE 11

Brixen Workshop 2012

11 -

Daniel Sturm

2.3 Allowing Migration

Krugman points out at the end of his 1979 paper that if migration

is allowed people will want to move from small to large countries.

We can also interpret “countries” as regions or cities.
This incentive for people to agglomerate in the large country or

region continues to exist even if we allow trade in goods.

The migration incentive only disappears if the “world is flat” and

there are zero trade costs.

You have already learned earlier this week that empirically the

earth is anything but flat.

SLIDE 12

Brixen Workshop 2012

12 -

Daniel Sturm

3 Fundamentals versus Path Dependence

The simple Krugman (1979) model with migration can capture the

basic idea of multiple equilibria and path dependence.

Consider a situation in which initially the two regions have the

same population.

Now suppose some workers move from region one to region two.
This shift increases real wages in region two and induces further

migration into region two.

This cumulative causation will draw the entire population of region
ne into region two.

SLIDE 13

Brixen Workshop 2012

13 -

Daniel Sturm

If instead workers had moved from region two to region one, then

region one would have ended up with all workers.

The model therefore predicts that there are multiple equilibria.
Which region becomes the large region depends on potentially

small historical accidents.

SLIDE 14

Brixen Workshop 2012

14 -

Daniel Sturm

3.1 The Role of Fundamentals

In the simple model considered above the two regions are equally

“good” locations for workers ex-ante.

Apart from the number of workers locating in a region there are no
ther factors that determine the attractiveness of a location.
An alternative reason why workers locate in a region are differences

in fundamentals that determine the productivity of regions.

A very basic way of capturing this idea in our framework are

differences in the marginal labor requirement β between regions.

SLIDE 15

Brixen Workshop 2012

15 -

Daniel Sturm

Suppose for example that β is lower in region one because of a

favorable climate or access to navigable waterways.

If such a difference in productivity is large enough it will eliminate

the possibility of path dependence.

The more productive region will attract all workers even if a

temporary shock has displaced many or all workers to the less productive region.

SLIDE 16

Brixen Workshop 2012

16 -

Daniel Sturm

3.2 Jargon: First versus Second Nature

The forces of cumulative causation are in the literature often

referred to as “second nature” advantages of a location.

In contrast differences in the fundamental attractiveness of

locations are referred to as “first nature” advantages of a location.

SLIDE 17

Brixen Workshop 2012

17 -

Daniel Sturm

3.3 Jargon: Multiple Equilibria

Macroeconomists typically reserve the term “multiple equilibria” to

models where expectations determine which equilibrium is selected.

Cases in which historical accidents determine which of several

potential equilibria is selected are called “multiple steady-states”

r “path-dependence”.
Other fields are less religious about this and refer to any situation

where there are several potential equilibria as “multiple equilibria”.

We will follow this more relaxed definition of multiple equilibria.

SLIDE 18

Brixen Workshop 2012

18 -

Daniel Sturm

3.4 Policy Implications

Whether location patterns are determined by fundamentals (first

nature) or agglomeration forces (second nature) has important policy implications.

Governments spend considerable amounts of money trying to shift

economic activity from one location to another.

Implicitly, a key rational for such policies is that agglomeration

forces and path dependence are important.

SLIDE 19

Brixen Workshop 2012

19 -

Daniel Sturm

The hope is that temporary subsidies can create a “critical mass”
f economic activity in a location that makes the location

self-sustaining.

If instead location patterns are mainly determined by fundamentals

temporary interventions are futile.

SLIDE 20

Brixen Workshop 2012

20 -

Daniel Sturm

3.5 The Key Empirical Question

To what extent is the observed distribution of economic activity

determined but fundamentals or agglomeration forces and path dependence?

This question has generated a lot of discussion over the last

decades.

For some proponents of economic geography the answer to this

question is almost self-evident (see the quote on the next slide):

SLIDE 21

Brixen Workshop 2012

21 -

Daniel Sturm

“[T]he dramatic spatial unevenness of the real economy - the disparities between densely populated manufacturing belts and thinly populated farm belts, between congested cities and desolate rural areas; the spectacular concentration of particular industries in Silicon Valleys and Hollywoods - is surely the result not of inherent differences among locations but of some set of cumulative processes, necessarily involving some form of increasing returns, whereby geographic concentration can be self-reinforcing.” (Fujita et al., 1999, p. 2)

SLIDE 22

Brixen Workshop 2012

22 -

Daniel Sturm

3.6 Empirical Approaches

More systematic approaches that try to distinguish between

fundamentals and agglomeration forces have taken two main forms: – Evidence on the persistence of spatial equilibria – Exploiting the impact of temporary shocks

A number of papers show that the spatial distribution of

population and economic activity is surprisingly persistent.

SLIDE 23

Brixen Workshop 2012

23 -

Daniel Sturm

Davis and Weinstein (2002) report, for example, that for Japan the

correlation between population densities in 1600 and 1998 is a staggering 0.76 (raw correlation) and 0.83 (rank correlation).

While such evidence is striking it could in principle be due to:

– Fundamentals are key and fundamentals do not change much

ver time.

– There are strong agglomeration forces that lock in location patterns once established.

SLIDE 24

Brixen Workshop 2012

24 -

Daniel Sturm

The most promising empirical approach to distinguish between

fundamentals and agglomeration forces are temporary shocks.

An ideal experiment in our modeling framework would be as

follows: – Initially all workers are located in region one. – A temporary shock dislocates all (or most) workers to region two. – If workers return to region one after the shock has disappeared, this would be strong evidence for fundamentals being key. – If instead workers remain in region two even after the shock has disappeared, this would be strong evidence for path dependence.

SLIDE 25

Brixen Workshop 2012

25 -

Daniel Sturm

4 Empirical Testing: Bombing

Davis and Weinstein (2002) were the first to exploit a large scale

natural experiment to distinguish between agglomeration forces and fundamentals.

In particular they exploit the Allied bombing of Japanese cities

during the Second World War as a large but temporary shock.

Destruction due to bombing was not only large but also very

asymmetric across Japanese cities.

SLIDE 26

Brixen Workshop 2012

26 -

Daniel Sturm

The basic idea behind their approach is to examine whether the

partial destruction of cities during the war had permanent effects

n city size.
If this was the case, this would be powerful evidence for multiple

equilibria and path-dependence.

SLIDE 27

Brixen Workshop 2012

27 -

Daniel Sturm

4.1 Historical Background

The bombing of Japan during the Second World War devastated

the targeted 66 cities, destroying approximately half their housing stock.

Approximately three hundred thousand Japanese were killed.
Particularly famous are obviously Hiroshima and Nagasaki due to

the nuclear bombs dropped on them.

SLIDE 28

Brixen Workshop 2012

28 -

Daniel Sturm

However, the majority of cities in the sample suffer essentially no

destruction (and this includes several large cities).

Despite its scale, the bombing was clearly a temporary shock.
While it killed many people, destroyed houses and other

infrastructure it did not change the fundamental attractiveness of locations.

A possible exception is the nuclear radiation in Hiroshima and

Nagasaki, which we will return to.

SLIDE 29

Brixen Workshop 2012

29 -

Daniel Sturm

4.2 Data

Data covers 303 Japanese cities with population in excess of

30,000 in 1925.

The population data is available every 5 years.
The only exception is the 1945 census which took place in 1947.
One measure of the intensity of the shocks are the dead or missing

city residents.

SLIDE 30

Brixen Workshop 2012

30 -

Daniel Sturm

4.3 Empirical Approach

Davis and Weinstein (2002) motivate their key regression with a

simple statistical model.

Suppose that the log of the share of city i’s population in total

population at time t is sit and is equal to: sit = Ωi + εit (3) where Ωi is the initial size and εit are city specific shocks.

The persistence of these shocks is modeled as:

εit+1 = ρεit + νit+1 (4)

SLIDE 31

Brixen Workshop 2012

31 -

Daniel Sturm

If ρ ∈ [0, 1] equals one city growth would be a random walk with

temporary shocks having permanent effects.

If ρ is instead equal to zero temporary shocks only have temporary

effects on the distribution of population across cities.

Taking first differences of (3) results in:

sit+1 − sit = εit+1 − εit (5)

Now (repeatedly) substituting (4) in (5) results in the final

estimating equation: sit+1 − sit = (ρ − 1)νit + [νit+1 + ρ(ρ − 1)εit−1] (6)

SLIDE 32

Brixen Workshop 2012

32 -

Daniel Sturm

If ρ equals one (6) simplifies to

sit+1 − sit = νit+1 (7) and city population follow a random walk.

Equation (6) is implemented empirically by regressing:

pgrowth65−47,i = β0 + β1pgrowth47−40,i + ui (8) where population growth between 1947 and 1940 is the measure of the innovation (i.e. war time shock).

Note that β0 will capture aggregate population growth across all

cities and hence changes in shares are equal to growth rates.

SLIDE 33

Brixen Workshop 2012

33 -

Daniel Sturm

4.4 Instrumental Variable Strategy

Note that the measure of innovation in (8) vit will depend on past

shocks to city growth (εit−1).

To overcome this problem Davis and Weinstein (2002) instrument

for vit with war-time destruction (deaths and housing stock lost).

The identifying assumption is that war related destruction was

determined by military factors, which are uncorrelated with pre-war growth shocks.

As an additional robustness test the 1925 to 1940 growth rate is

included as an additional explanatory variable.

SLIDE 34

Brixen Workshop 2012

34 -

Daniel Sturm

4.5 Results

Before estimating (8) we will have a look at a simple scatter plot
f city growth rates.
The following slide contains the key estimation results.

SLIDE 35

Brixen Workshop 2012

35 -

Daniel Sturm

THE AMERICAN ECONOMIC REVIEW

(5)

Sit+1 = Sit +

12it?I*

If p E [0, 1), then city share is stationary and any shock will dissipate over time. In other words, these two hypotheses can be distin- guished by identifying the parameter p. One approach to investigating the magnitude

f p is to search

for a unit root. It is well known that unit root tests usually have little power to separate p < 1 from p = 1. This is due to the fact that in traditional unit root tests the inno- vations are not observable and so identify p with very large standard errors. A major advantage

f our data

set is that we can easily identify the innovations due to bombing. In particular, since by hypothesis the innovation, vit, is uncorre- lated with the error term (in square brackets), then if we can identify the innovation, we can

btain an unbiased

estimate

f p.

An obvious method

f looking at the innova-

tion is to use the growth rate from 1940 to 1947. However, this measure of the innovation may contain not only information about the bombing but also past growth rates. This is a measure- ment error problem that could bias our estimates in either direction depending

n p. In order

to solve this, we instrument the growth rate from 1940-1947 with buildings destroyed per capita and deaths per capita.20 We can obtain a feel for the data by consid- ering the impact of bombing on city growth rates. As we argued earlier, if city growth rates follow a random walk, then all shocks to cities should be permanent. In this case, one should expect to see no relationship between historical shocks and future growth rates. Moreover, if

ne believes that there is positive serial corre-

lation in the data, then one should expect to see a positive correlation between past and future growth rates. By contrast, if one believes that location-specific factors are crucial in under- standing the distribution

f population,

then

ne

should expect to see a negative relationship between a historical shock and the subsequent growth rate. In Figure 1 we present a plot of

20 The actual

estimating equation is Si60

Si47 = (
1)1V47

+

[vi60

+ P(P

1)~i34]

Our measure

f the

innovation is the growth rate between 1940 and 1947 or

Si47

Si40

= ^t47

+ [Pi34

8i40].

This

is clearly

correlated with the error term in the estimating equation, hence we instrument. 1.0-

r-

I'*

0.- cr

2

0o ? 0o(~

(^& Q^&(

Oo

0.5

0.5 Growth Rate 1940-1947

FIGURE 1. EFFECTS OF BOMBING ON CITIES WITH

MORE THAN 30,000 INHABITANTS Note: The figure presents data for cities with positive casu- alty rates only.

population growth between 1947 and 1960 with that between 1940 and 1947. The sizes of the circles represent the population of the city in

1925. The figure reveals a very clear negative

relationship between the two growth rates. This indicates that cities that suffered the largest population declines due to bombing tended to have the fastest postwar growth rates, while cities whose populations boomed conversely had much lower growth rates thereafter. In Table 2, we present a regression showing the power of our instruments. Deaths per capita and destruction per capita explain about 41 per- cent of the variance in population growth of cities between 1940 and 1947. Interestingly, although both have the expected signs, destruc- tion seems to have had a more pronounced effect on the populations

f cities. Presumably,

this is because, with a few notable exceptions, the number of people killed was only a few percent

f the city's population.

We now turn to test whether the temporary shocks give rise to permanent

effects. In order

to estimate equation (4), we regress the growth rate of cities between 1947 and 1960 on the growth rate between 1940 and 1947 using deaths and destruction per capita as instruments for the wartime growth rates. The coefficient

n

growth between 1940 and 1947 corresponds to (p - 1). In addition, we include government subsidies to cities to control for policies de- signed to rebuild cities. If one believes that cities follow a random

1280 DECEMBER 2002

SLIDE 36

Brixen Workshop 2012

36 -

Daniel Sturm

DAVIS AND WEINSTEIN: GEOGRAPHY OF ECONOMIC ACTIVITY

TABLE 2-INSTRUMENTAL VARIABLES EQUATION

(DEPENDENT VARIABLE = RATE OF GROWTH IN CITY

POPULATION BETWEEN 1940 AND 1947)

Independent variable Coefficient Constant 0.213 (0.006) Deaths per capita

0.665

(0.506) Buildings destroyed per capita

2.335

(0.184)

R2:

0.409 Number

f observations:

303 Note: Standard errors are in parentheses.

walk or that catastrophes can permanently alter the size of cities, then one should expect the coefficient on the 1940-1947 growth rate, (p - 1), should be zero. If one believes that the temporary shocks have only temporary effects, then the coefficient on 1940-1947 growth should be negative. A coefficient of -1 indi- cates that all of the shock was dissipated by 1960. The results are presented in Table 3. The coefficient on 1940-1947 growth is -1.048, indicating that at this interval p is approximately

zero. This means that the typical city com-

pletely recovered its former relative size within 15 years following the end of World War II. Given the magnitude

f the destruction,

this is quite surprising. Apparently, U.S. bombing of Japanese cities had no impact on the typical city's size in 1960. This strongly rejects the hypothesis that growth in city-size share is a random walk. As expected, reconstruction subsidies seem to have had a positive and statistically signifi- cant impact on rebuilding

cities. However, the

economic impact is quite modest. Our estimates suggest that a one standard deviation increase in reconstruction expenses would increase the size

f a city in 1960 by 2.2 percent.

Reconstruction expenses probably had a small effect because both the sums spent and the variance in the sums were small. In the cities that suffered the heaviest destruction-Tokyo, Hiroshima, Na- gasaki, and Osaka-our estimates indicate that government reconstruction expenses accounted for less than one percentage point of their cu- mulative growth between 1947 and 1960. Given

TABLE 3-TWO-STAGE LEAST-SQUARES ESTIMATES OF IMPACT OF BOMBING ON CITIES (INSTRUMENTS: DEATHS PER CAPITA AND BUILDINGS DESTROYED PER CAPITA)

Dependent Dependent variable = variable= growth rate growth rate

f population
f population

between between 1947 and 1947 and 1960 1965 Independent variable (i) (ii) (iii) Growth rate of population

1.048
0.759
1.027

between 1940 and 1947 (0.097) (0.094) (0.163) Government reconstruction 1.024 0.628 0.392 expenses (0.387) (0.298) (0.514) Growth rate of population 0.444 0.617 between 1925 and 1940 (0.054) (0.092) R2: 0.279 0.566 0.386 Number

f observations:

303 303 303

Note: Standard errors are in parentheses.

that cumulative growth in these cities over that period was between 55 and 96 percent, we conclude that reconstruction expenses had rela- tively small impacts. As we noted earlier, a major reason for this was that reconstruction policies, like most Japanese regional policies, disproportionately sent money to rural areas. As a result, the four biggest per capita recipients

f

assistance were small northern cities that were never targeted by U.S. bombers. One potential problem with these results is that it is possible that the United States inad- vertently targeted cities based on underlying growth rates. While this was not an explicit strategy, we cannot rule it out. If the United States bombed rapidly growing cities more heavily, then we may be biasing our results downwards. We therefore repeated

ur

exercise adding the growth rate between 1925 and 1940 to our list of independent

variables. This im-

proves the fit, but does not qualitatively change the results, although the coefficient on 1940- 1947 growth falls to -0.76. This implies that bombed cities had recovered

ver three-fourths
f their lost growth by 1960. One reasonable

question to ask, then, is whether these cities ever returned to their prewar trajectories

r not.

If one believes in the location-specific model, then one should expect that the coefficient on

VOL. 92 NO. 5 1281

SLIDE 37

Brixen Workshop 2012

37 -

Daniel Sturm

4.6 Hiroshima and Nagasaki

One possible objection to the results so far is that the largest part
f war-time population changes is due to refugees rather than

deaths.

Existing social networks may facilitate the return of refugees to

their home cities.

As a results war-time bombing may not have been sufficient to
vercome the force of social networks.
The following (famous) graph looks at the case of Hiroshima and

Nagasaki where 20.8 and 8.5 percent of the population are estimated to have died.

SLIDE 38

Brixen Workshop 2012

38 -

Daniel Sturm

THE AMERICAN ECONOMIC REVIEW

c 0o

a.

1925

1930 1935 1940 1947 1950 1955 1960 1965 1970 1975 Year FIGURE 2. POPULATION GROWTH

wartime growth should asymptotically ap- proach unity as the end period increases. In the last column

f Table

3 we repeat the regression,

nly now extending the endpoint to 1965 in-

stead of 1960. The estimated coefficient now reaches -1.027. That is, after controlling for prewar growth trends, by 1965 cities have en- tirely reversed the damage due to the war. Again, the impact of reconstruction subsidies also lessens as we move into the future. To- gether, these results suggest that the effect of the temporary shocks vanishes completely in less than 20 years. One possible objection to our interpretation is that in most cases, the population changes cor- responded much more to refugees than deaths. Of the 144 cities with positive casualties, the average number

f deaths

per capita was only 1 percent. Most of the population movement that we observe in our data is due to the fact that the vast destruction

f buildings forced people to

live elsewhere. However, forcing them to move

ut of their

cities for a number

f years

may not have sufficed to overcome the social networks and other draws of their home cities. Hence it may seem uncertain whether they are moving back to take advantage

f particular

character- istics of these locations or simply moving back to the only real home they have known. However, there are two cases in which this argument cannot be made: Hiroshima and Na- gasaki. In those cities, the number

f deaths

was such that if these cities recovered their popula- tions, it could not be because residents who temporarily moved out of the city returned in subsequent

years. We have already noted that
ur data underestimates

casualties in these cit-

ies. Even so, our data suggest that the nuclear

bombs immediately killed 8.5 percent of Na- gasaki's population and 20.8 percent of Hiro- shima's population. Moreover given that many Japanese were worried about radiation poison- ing and actively discriminated against atomic bomb victims, it is unlikely that residents felt an unusually strong attachment to these cities or that

ther

Japanese felt a strong desire to move

there. Another reason why these cities are in-

teresting to consider is that they were not par- ticularly large or famous cities in Japan. Their 1940 populations made them the 8th and 12th largest cities in Japan. Both cities were close to

ther

cities of comparable size so that it would have been relatively easy for other cities to absorb the populations of these devastated cities. In Figure 2 we plot the population

f these

two cities. What is striking in the graph is that even in these two cities there is a clear indica- tion that they returned to their prewar growth trends. This process seems to have taken a little longer in Hiroshima than in other cities, but this is not surprising given the level of destruction.

1282 DECEMBER 2002

SLIDE 39

Brixen Workshop 2012

39 -

Daniel Sturm

4.7 Implications

The results of Davis and Weinstein (2002) show a stunning

persistence in city size even after horrific war-time devastation.

This has important implications for attempts to use regional policy

to shift economic activity between different spatial equilibria.

This is summarized in the following quote from Davis and

Weinstein (2002):

SLIDE 40

Brixen Workshop 2012

40 -

Daniel Sturm

“An important practical question, then, is whether such spatial catastrophes are theoretical curiosa or a central tendency in the

data. Our results provide an unambiguous answer: Even nuclear

bombs have little effect on relative city sizes over the course of a couple of decades. The theoretical possibility of spatial catastrophes due to temporary shocks is not a central tendency borne out in the data.” (Davis and Weinstein (2002), p. 1284)

SLIDE 41

Brixen Workshop 2012

41 -

Daniel Sturm

4.8 Extensions

The findings of Davis and Weinstein (2002) have been extended in

a number of directions.

Davis and Weinstein (2008) use data on the industry structure of

Japanese cities.

They show that not only total population levels but also the

industrial composition of cities recovers quickly after the war.

Brackman et. al. (2004) replicate the analysis of Davis and

Weinstein (2002) on West German data.

They find that also in this context cities recover quickly from the

war-time shock (but their results are not as clear cut).

SLIDE 42

Brixen Workshop 2012

42 -

Daniel Sturm

Another application of this idea is Miguel and Roland (2011), who

look at the case of Vietnam. “The United States Air Force dropped in Indochina, from 1964 to August 15, 1973, a total of 6,162,000 tons of bombs and

ther ordnance. U.S. Navy and Marine Corps aircraft expended

another 1,500,000 tons in Southeast Asia. This tonnage far exceeded that expended in World War II and in the Korean

War. The U.S. Air Force consumed 2,150,000 tons of munitions

in World War II 1,613,000 tons in the European Theater and 537,000 tons in the Pacific Theater and 454,000 tons in the Korean War.” (Clotfelder (1995) as citied in Miguel and Roland (2011))

SLIDE 43

Brixen Workshop 2012

43 -

Daniel Sturm

4.9 Conclusion

The evidence presented in the literature following Davis and

Weinstein (2002) is extremely powerful.

Spatial patterns seem to be highly persistent and resilient to

shocks of horrific magnitudes.

This should give pause to policy-makers who attempt to use

temporary policy interventions to shift the economy between different equilibria.

SLIDE 44

Brixen Workshop 2012

44 -

Daniel Sturm

While the use of bombing as a temporary shock is ingenious, it

also has a number of potential problems: – Even though bombing kills people and destroys housing and social networks it does not destroy legal titles and operating permits. – As a result rebuilding cities may simply be easier than starting a new settlement from scratch. – Similarly, even after a nuclear blast there is going to be remaining useful infrastructure, which may make it cheaper to rebuild in the old location.

SLIDE 45

Brixen Workshop 2012

45 -

Daniel Sturm

5 Empirical Testing: German division

Redding, Sturm and Wolf (2011) use an alternative natural

experiment to bombing to investigate the impact of temporary shocks: German division and reunification.

Key idea: Does economic activity which was relocated in response

to division returns to its pre-war pattern after reunification.

They focus on a particular industry which is both likely to be prone

to multiple steady-states and for which a wealth of data is available: air transportation.

SLIDE 46

Brixen Workshop 2012

46 -

Daniel Sturm

5.1 Road Map

Sketch of the theoretical model
Data and empirical strategy
Basic finding
Further evidence
Conclusion

SLIDE 47

Brixen Workshop 2012

47 -

Daniel Sturm

5.2 Theoretical Model

There are three cities and demand for air travel between any two

cities is a decreasing function of the price (derived from micro-foundations in the paper).

There is a monopoly airline which has to choose whether to

connect all three cities directly or to use one city as a “hub”.

The airline’s marginal costs depend on distance flown and it has to

pay a fixed cost F > 0 for any bilateral connection.

Finally, creating a hub involves sunk costs H > 0.

SLIDE 48

Brixen Workshop 2012

48 -

Daniel Sturm

5.3 Basic Implications of the Model

Operating a hub based in city i will generate higher per-period

profits if the saving in fixed costs F exceeds the loss in variable profits due to the longer distance flown on indirect connections: ωi = F −

πD

kj − πI kj

Without loss of generality index cities so that ω1 ≥ ω2 ≥ ω3.
Denote the corresponding present discounted values by

Ω1 ≥ Ω2 ≥ Ω3.

SLIDE 49

Brixen Workshop 2012

49 -

Daniel Sturm

There are multiple steady-state hub locations if for several cities i:

Ωi > H and Ωj − Ωi < H

∀j = i

In contrast, the equilibrium hub location is unique if for one city i:

Ωi > H, and Ωi − Ωj > H

∀j = i

Therefore, the existence of multiple steady-state locations depends
n the size of the sunk cost relative to variation in economic

fundamentals.

SLIDE 50

Brixen Workshop 2012

50 -

Daniel Sturm

5.4 Temporary Shocks

Suppose parameters are such that city one and two are potential

equilibrium locations for the hub and the hub is located in city one.

A shock S such as division will shift the location of the hub

between multiple equilibria if: – It is large enough to relocate the hub: Ω2 − (Ω1 − S) > H – It is ultimately reversed to an extent that both locations are again potential equilibria: |Ω2 − (Ω1 − S′)| < H

SLIDE 51

Brixen Workshop 2012

51 -

Daniel Sturm

5.5 Data

The basic dataset consists of the number of departing passengers

at the major German airports from 1927 - 1938 and 1950 - 2002.

Additional datasets are:

– Departing passengers at the largest European airports in 1937 and 2002. – Economic characteristics of the locations of major German airports. – Bilateral passenger flows between the major German airports and all worldwide destinations in 2002.

SLIDE 52

Brixen Workshop 2012

52 -

Daniel Sturm

Figure 1.—The Location of the German Airports in our Sample

SLIDE 53

Brixen Workshop 2012

53 -

Daniel Sturm

Figure 2.—Airport Passenger Shares

Departing Passengers at the Ten Main German Airports

SLIDE 54

Brixen Workshop 2012

54 -

Daniel Sturm

5.6 Further Evidence

The evidence presented so far is suggestive that there has been a

shift between multiple steady-state locations.

An alternative explanation is that economic fundamentals have

changed so much between the pre-war and reunification periods that Berlin is no longer a potential equilibrium hub location.

There are several additional pieces of evidence to rule out this

alternative explanation.

SLIDE 55

Brixen Workshop 2012

55 -

Daniel Sturm

5.6.1 Comparison with other Countries

As the table on the next slide shows, the experience of Germany is

highly unusual.

In all other countries the largest airport before the war is also the

largest airport today (and is in the largest city).

SLIDE 56

Brixen Workshop 2012

56 -

Daniel Sturm

822 THE REVIEW OF ECONOMICS AND STATISTICS

Table 3.—The Largest Airports of European Countries, 1937 and 2002 (1) (2) (3) (4) Market Share of Market Share of Rank of Largest Largest Airport, Largest Airport, Largest Airport, Airport 1937 in 1937 1937 2002 2002 Austria Vienna 94.1 76.5 1 Belgium Brussels 65.6 89.9 1 Denmark Copenhagen 96.2 91.7 1 Finland Helsinki 80.3 73.7 1 France Paris 70.2 61.4 1 Germany Berlin 30.8 35.0 4 Greece Athens 43.9 34.7 1 Ireland Dublin 100.0 78.1 1 Italy Rome 35.7 34.5 1 Netherlands Amsterdam 62.3 96.4 1 Norway Oslo 75.6 45.8 1 Portugal Lisbon 100.0 46.3 1 Spain Madrid 43.5 26.8 1 Sweden Stockholm 56.9 61.9 1 Switzerland Zurich 55.7 62.0 1 United Kingdom London 52.7 65.6 1

The countries are the EU 15 countries without Luxembourg (which had no airport prior to World War II and had only one airport in 2002) and Norway and Switzerland. The prewar data for Austria refer to the year

1938. The prewar data for Spain are the average over 1931 to 1933. As in the case of Berlin, we aggregate airports when cities have more than one airport. See the data appendix for detailed references to the sources.

SLIDE 57

Brixen Workshop 2012

57 -

Daniel Sturm

5.6.2 The Role of Market Access

The model suggests that bilateral departures depend on:

– Remoteness from other locations (market access) – Local economic fundamentals (in particular population) – An airport’s role as a hub.

The contribution of market access can be estimated with the help
f a gravity model.

SLIDE 58

Brixen Workshop 2012

58 -

Daniel Sturm

In particular one can estimate:

ln

Aij
= si + mj + δ ln
distij
+ uij
The estimated coefficients can be used to decompose total

passenger departures as follows:

Ai = ∑

j

Aij =

Si MAi where

MAi = ∑

j

dist

δ ij

Mj

SLIDE 59

Brixen Workshop 2012

59 -

Daniel Sturm

Figure 3.—The Role of Market Access

SLIDE 60

Brixen Workshop 2012

60 -

Daniel Sturm

5.6.3 What Explains the Size of Frankfurt?

Frankfurt’s size is almost entirely due to its status as a transit hub.
Furthermore economic fundamentals affect the volume of traffic in

the way suggested by the model.

SLIDE 61

Brixen Workshop 2012

61 -

Daniel Sturm

Figure 4.—Transit and Local Passenger Departures, 2002

SLIDE 62

Brixen Workshop 2012

62 -

Daniel Sturm

Figure 5.—Local Departures and Local GDP

are those who traveled less than 50 kilometers to an airport. Local GDP is calculated from the population of all municipalities within 50 kilometers of an airport and the GDP per capita

SLIDE 63

Brixen Workshop 2012

63 -

Daniel Sturm

5.6.4 Simulating the Impact of Moving the Hub

The final piece of evidence is a simulation of the impact of

relocating the hub on volumes of four types of passengers: – International Transits – Domestic Transits – Ground Transits – Local Departures

Table 5 summarizes the impact of relocating the hub.

SLIDE 64

Brixen Workshop 2012

64 -

Daniel Sturm

HISTORY AND INDUSTRY LOCATION: EVIDENCE FROM GERMAN AIRPORTS 829

Table 5.—Estimated Impact of Relocating the Air Hub from Frankfurt on Total Passenger Departures across the 15 German airports (1) (2) (3) (4) Estimated Change in Estimated Change in Estimated Change in Estimated Percentage Alternative Location of Air Transit Ground Transit Total Passenger Change in Total the Air Hub Passengers Passengers Departures Passenger Departures Berlin −407,498 −1,862,056 −2,232,380 −3.38 Dusseldorf 148,590 − 18,331 125,759 0.19 Hamburg −332,672 −1,644,620 −1,852,323 −2.80 Munich 566,039 − 865,146 − 422,204 −0.64

The table reports the estimated change in passenger departures across the 15 German airports as a result of the hypothetical relocation of the air hub from Frankfurt to each of the alternative locations. All air transit passengers who currently change planes at Frankfurt are assumed to instead fly via the alternative airport, and the coefficient on distance from column 4 of table 4 is used to infer the change in the number of air transit passengers as a result of the change in distance traveled caused by the relocation of the hub. The logarithm of ground transit departures is regressed on the logarithm of the distance-weighted sum of GDP in all German counties, and the estimated coefficient is used to infer how the number of ground departures currently observed in Frankfurt would change if it instead had the distance-weighted GDP of the alternative location of the

hub. See the main text for further discussion. Total bilateral departures for the 15 German airports in 2002 were 66,134,047. Total bilateral departures from Frankfurt airport in 2002 were 23,782,604.

SLIDE 65

Brixen Workshop 2012

65 -

Daniel Sturm

At 10 Euro per passenger and a 3 percent discount rate the NPV
f a change in the total passenger volume by 2.5 million would be

0.86 billion Euro.

The construction of new terminal facilities at Berlin which are

designed to be a third of the size of Frankfurt were estimated to cost just over 2 billion Euro (current costs >4.5 billion Euro).

A third runway at Heathrow is currently costed at 10 billion

Pounds.

SLIDE 66

Brixen Workshop 2012

66 -

Daniel Sturm

5.7 Why Frankfurt?

Several historical accidents propelled Frankfurt ahead of alternative

locations for the hub in the post-war period: – Frankfurt airport was captured by U.S. troops in March 1945. – From 1948 Frankfurt became the European terminal for the U.S. Military Air Transport Service (MATS). – During the Berlin airlift of 1948-9, Frankfurt was the

perational center for the U.S. contribution to the airlift.

SLIDE 67

Brixen Workshop 2012

67 -

Daniel Sturm

5.8 Conclusion

Redding, Sturm and Wolf (2011) show that German division even

though ultimately temporary seems to have permanently moved the location of Germany’s hub airport.

They present a number of pieces of evidence that this shift is due

to a shift between equilibria rather than a change in fundamentals.

The rise of Frankfurt was a long process and it is not clear that

Frankfurt would have survived as the new hub if reunification had happened in the early 1960s.

This suggests that for temporary shocks to permanently change

location decisions they need to be fairly long-lasting.

SLIDE 68

Brixen Workshop 2012

68 -

Daniel Sturm

While the paper provides suggestive evidence there are least two
pen questions:

– To what extent are economic activities other than air hubs subject to multiple equilibria? – What policy instruments could generate the same impact as German division?

SLIDE 69

Brixen Workshop 2012

69 -

Daniel Sturm

6 Empirical Testing: Portage

In a paper that has just been published in the QJE Bleakley and

Lin use a somewhat different empirical approach to the question of fundamentals versus agglomeration and path dependence.

They do not look for another kind of temporary shock.
Instead they investigate the impact of a natural feature that was

valued historically but has become obsolete a long time ago.

In particular they look at so called “portage” sites.

SLIDE 70

Brixen Workshop 2012

70 -

Daniel Sturm

Portage refers to the carrying of boats or their cargo over land to

avoid obstacles such as rapids or water falls.

During the early settlement of the US most goods traveled by boat
n rivers and portage sites were natural places for exchange and

commerce.

However, the importance of portage sites vanished with the

construction of canals and the railroads.

Proximity to a portage site has been economically irrelevant for at

least 100 years.

This is illustrated in the following graph.

SLIDE 71

Brixen Workshop 2012

71 -

Daniel Sturm

SLIDE 72

Brixen Workshop 2012

72 -

Daniel Sturm

6.1 Data

The paper uses three main datasets to measure the contemporary

density of economic activity in different parts of the US: – County level population – Census tract population data from the 2000 census – Nighttime light intensity data for 2003

SLIDE 73

Brixen Workshop 2012

73 -

Daniel Sturm

6.2 Where Were Portage Sites?

The paper considers three areas in which portage sites were

located in the US.

The most impressive and important of these is the so-called “fall

line”.

The following maps show the location of the fall line and economic

activity around the fall line.

SLIDE 74

Brixen Workshop 2012

74 -

Daniel Sturm

FIGURE A.1 The Density Near Fall-Line/River Intersections

SLIDE 75

Brixen Workshop 2012

75 -

Daniel Sturm

FIGURE II Fall-Line Cities from Alabama to North Carolina

SLIDE 76

Brixen Workshop 2012

76 -

Daniel Sturm

FIGURE IV Fall-Line Cities from North Carolina to New Jersey

SLIDE 77

Brixen Workshop 2012

77 -

Daniel Sturm

SLIDE 78

Brixen Workshop 2012

78 -

Daniel Sturm

6.3 The Continuing Prominence of Portage Sites

The maps on the previous slides show a striking pattern.
The places at which the fall line intersects major rivers are also

today clearly visible agglomerations.

Bleakley and Lin (2012) also provide statistical evidence that these

agglomerations are not random. – The table on the next slide shows that portage sites have a statistically significantly higher density today. – The following table shows that portages sites that served larger upstream areas are larger today.

SLIDE 79

Brixen Workshop 2012

79 -

Daniel Sturm

TABLE I PROXIMITY TO HISTORICAL PORTAGE SITE AND CONTEMPORARY POPULATION DENSITY (1) (2) (3) (4) (5) (6) (7) (8) Basic Other spatial controls Additional fixed factors Other samples Specifications: State fixed effects Distance from various features Climate variables Aquifer Share Mean elevation Atlantic Rivers only Within 100mi

f the fall line

Explanatory variables: Panel A: Census Tracts, 2000, N = 21452 Dummy for proximity to portage site 1.113 1.009 1.118 1.041 0.979 1.077 0.838 1.039 (0.340)∗∗∗ (0.321)∗∗∗ (0.243)∗∗∗ (0.316)∗∗∗ (0.330)∗∗∗ (0.316)∗∗∗ (0.401)∗∗ (0.319)∗∗∗ Distance to portage site, natural logs −0.617 −0.653 −0.721 −0.460 −0.562 −0.577 −0.572 −0.764 (0.134)∗∗∗ (0.128)∗∗∗ (0.118)∗∗∗ (0.121)∗∗∗ (0.123)∗∗∗ (0.118)∗∗∗ (0.177)∗∗∗ (0.142)∗∗∗ Panel B: Nighttime Lights, 1996–97, N = 65000 Dummy for proximity to portage site 0.504 0.445 0.490 0.500 0.506 0.522 0.495 0.391 (0.144)∗∗∗ (0.127)∗∗∗ (0.161)∗∗∗ (0.144)∗∗∗ (0.147)∗∗∗ (0.155)∗∗∗ (0.151)∗∗∗ (0.100)∗∗∗ Distance to portage site, natural logs −0.188 −0.159 −0.151 −0.186 −0.196 −0.138 −0.130 −0.212 (0.065)∗∗∗ (0.065)∗∗ (0.090) (0.061)∗∗∗ (0.065)∗∗∗ (0.059)∗∗ (0.101) (0.060)∗∗∗

SLIDE 80

Brixen Workshop 2012

80 -

Daniel Sturm

TABLE II UPSTREAM WATERSHED AND CONTEMPORARY POPULATION DENSITY (1) (2) (3) (4) (5) Basic Other spatial controls Water power Specifications: State fixed effects Distance from various features Explanatory variables: Panel A: Census Tracts, 2000, N = 21452 Portage site times upstream watershed 0.467 0.467 0.500 0.496 0.452 (0.175)∗∗ (0.164)∗∗∗ (0.114)∗∗∗ (0.173)∗∗∗ (0.177)∗∗ Binary indicator for portage site 1.096 1.000 1.111 1.099 1.056 (0.348)∗∗∗ (0.326)∗∗∗ (0.219)∗∗∗ (0.350)∗∗∗ (0.364)∗∗∗ Portage site times horsepower/100k −1.812 (1.235) Portage site times I(horsepower > 2000) 0.110 (0.311)

SLIDE 81

Brixen Workshop 2012

81 -

Daniel Sturm

6.4 Why Did Portage Sites Survive?

There are two broad explanations why portage sites have survived

nearly hundred years after portage has become irrelevant.

One view is that the density of economic activity is determined by

fundamentals.

If this is the case, the decline in the importance of portage should

eliminate portage sites in the long-run.

The reason that this has not happened yet could be large sunk

investments undertaken when portage was still important that depreciate slowly over time.

SLIDE 82

Brixen Workshop 2012

82 -

Daniel Sturm

An alternative view is that there are multiple spatial equilibria.
Once established agglomerations of economic activity are held

together by the usual (second nature) agglomeration forces.

The historical portage advantage has in this view only helped to

select an equilibrium location for an agglomeration from many possible equilibria.

SLIDE 83

Brixen Workshop 2012

83 -

Daniel Sturm

The paper presents evidence that is consistent with multiple

equilibria rather than a slow decline to the long-run equilibrium: – Portage sites actually grow relative to other sites after portage becomes irrelevant. – Portage sites do not seem to have an excess supply of any

bvious factor of production (the historical sunk investment).

SLIDE 84

Brixen Workshop 2012

84 -

Daniel Sturm FIGURE V Portage and Population Density, 1790–2000

SLIDE 85

Brixen Workshop 2012

85 -

Daniel Sturm

TABLE III PROXIMITY TO HISTORICAL PORTAGE SITE AND HISTORICAL FACTORS

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Baseline Railroad network length, 1850 Distance to RR hub, 1850 Literate white men, 1850 Literacy rate white men, 1850 College teachers per capita, 1850

Manuf. /

agric., 1880 Non-agr. share, 1880 Industrial diversity (1-digit), 1880 Industrial diversity (3-digit), 1880 Water power in use 1885, dummy Explanatory variables: Panel A. Portage and historical factors Dummy for proximity to portage site 1.451 −0.656 0.557 0.013 0.240 0.065 0.073 0.143 0.927 0.164 (0.304)∗∗∗ (0.254)∗∗ (0.222)∗∗ (0.014) (0.179) (0.024)∗∗∗ (0.025)∗∗∗ (0.078)∗ (0.339)∗∗∗ (0.053)∗∗∗ Panel B. Portage and historical factors, conditioned on historical density Dummy for proximity to portage site 1.023 −0.451 0.021 −0.003 0.213 0.022 0.019 0.033 −0.091 0.169 (0.297)∗∗∗ (0.270) (0.035) (0.014) (0.162) (0.019) (0.019) (0.074) (0.262) (0.054)∗∗∗ Panel C. Portage and contemporary density, conditioned on historical factors Dummy for proximity to portage site 0.912 0.774 0.751 0.729 0.940 0.883 0.833 0.784 0.847 0.691 0.872 (0.236)∗∗∗ (0.236)∗∗∗ (0.258)∗∗∗ (0.187)∗∗∗ (0.237)∗∗∗ (0.229)∗∗∗ (0.227)∗∗∗ (0.222)∗∗∗ (0.251)∗∗∗ (0.221)∗∗∗ (0.233)∗∗∗ Historical factor 0.118 −0.098 0.439 0.666 1.349 1.989 2.390 0.838 0.310 0.331 (0.024)∗∗∗ (0.022)∗∗∗ (0.069)∗∗∗ (0.389)∗ (0.164)∗∗∗ (0.165)∗∗∗ (0.315)∗∗∗ (0.055)∗∗∗ (0.015)∗∗∗ (0.152)∗∗

Notes. This table displays estimates of equation 1, with the below noted modifications. In Panels A and B, the outcome variables are historical factor densities, as noted in the

column headings. The main explanatory variable is a dummy for proximity to a historical portage. Panel B also controls for historical population density. In Panel C, the outcome variable is 2000 population density, measured in natural logarithms, and the explanatory variables are portage proximity and the historical factor density noted in the column

heading. Each panel/column presents estimates from a separate regression. The sample consists of all U.S. counties, in each historical year, that are within the watersheds of rivers

that cross the fall line. The estimator used is OLS, with standard errors clustered on the 53 watersheds. The basic specification includes a polynomial in latitude and longitude, a set

f fixed effects by the watershed of each river that crosses the fall line, and dummies for proximity to the fall line and to a river. Reporting of additional coefficients is suppressed.

Data sources and additional variable and sample definitions are found in the text and appendixes.

SLIDE 86

Brixen Workshop 2012

86 -

Daniel Sturm

TABLE IV PROXIMITY TO HISTORICAL PORTAGE SITE AND CONTEMPORARY FACTORS

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Housing units, 1990 Median rents, 1990 Median values, 1990 Interstates, 2000 Major roads, 2000 Rail, 2000 Travel time to work, 1990 Crime, 1995 Born in state, 1990 Water use, 1995 Federal expend., 1997 Gov’t. empl., 1997 Explanatory variables: Panel A. Portage and contemporary factors Dummy for proximity to portage site 0.910 0.110 0.108 0.602 0.187 0.858 −0.554 1.224 0.832 0.549 1.063 1.001 (0.243)∗∗∗ (0.040)∗∗∗ (0.053)∗∗ (0.228)∗∗ (0.071)∗∗ (0.177)∗∗∗ (0.492) (0.318)∗∗∗ (0.186)∗∗∗ (0.197)∗∗∗ (0.343)∗∗∗ (0.283)∗∗∗ Panel B. Portage and contemporary factors, conditioned on contemporary density Dummy for proximity to portage site 0.005 0.014 −0.001 0.159 −0.064 0.182 −0.447 −0.007 −0.025 −0.153 0.032 0.114 (0.015) (0.020) (0.038) (0.108) (0.054) (0.110) (0.513) (0.058) (0.046) (0.145) (0.091) (0.077)

Notes. This table displays estimates of equation (1), with exceptions noted here. In Panels A and B, the outcome variables are current factor densities (natural log of the ratio
f quantity per square mile), as noted in the column headings. (The exceptions are house rent and value, which are in logs but not normalized by area, and travel times, which are

in minutes.) The coefficient reported is for proximity to historical portage sites. Panel B also controls for current population density. Each cell presents estimates from a separate

regression. The sample consists of all US counties, from the indicated year, that are within the watersheds of rivers that cross the fall line. The estimator used is OLS, with standard

errors clustered on the 53 watersheds. The specification includes a polynomial in latitude and longitude, a set of fixed effects by the watershed of each river that crosses the fall line, and dummies for proximity to the fall line and to a river. Reporting of additional coefficients is suppressed. Data sources and additional variable and sample definitions are found in the text and appendixes.

SLIDE 87

Brixen Workshop 2012

87 -

Daniel Sturm

6.5 Conclusion

Bleakley and Lin (2012) show that a long irrelevant natural

advantage – portage – is clearly reflected in the density of contemporary economic activity.

They argue that portage has simply helped to select one of many

possible equilibria for an agglomeration.

Historical accidents therefore have profound long-run consequences

in a world of path-dependence.

SLIDE 88

Brixen Workshop 2012

88 -

Daniel Sturm

The results nevertheless also leave a number of open questions:

– What is the critical level of economic activity that a place needs to achieve to become self-sustaining? – There are plenty of abandoned gold-rush cities in the US that died after the local gold-mines closed. – Furthermore, the paper has little to say about the potential role

r effectiveness of policy in shaping spatial equilibria.

SLIDE 89

Brixen Workshop 2012

89 -

Daniel Sturm

7 Outlook for future research

While research in this area has produced a number of interesting

insights over the last years, there are still plenty of open questions.

Let me highlight three areas in which our understanding is still

particularly incomplete: – How general are multiple spatial equilibria? Which industries are particularly prone to multiple equilibria? – Can economic policy play a meaningful role in shaping the spatial distribution of economic policy?

SLIDE 90

Brixen Workshop 2012

90 -

Daniel Sturm

– Finally what are the welfare consequences of spatial equilibria? If the natural advantage of portage sites has evaporated, should cities ideally be located elsewhere?

SLIDE 91

Brixen Workshop 2012

91 -

Daniel Sturm

8 Bibliography Bleakley, Hoyt and Jeffrey Lin (2012) “Portage and Path Dependence,” Quarterly Journal of Economics, 127(2), 587-644. Brakman, Steven, Harry Garretsen, and Marc Schramm (2004) “Bombing of German Cities during WWII and its Impact on City Growth,” Journal of Economic Geography, 4(2), 201218. Davis, Donald and David Weinstein (2002) “Bones, Bombs, and Break Points: The Geography of Economic Activity,” American Economic Review, 92(5),1269-1289.

SLIDE 92

Brixen Workshop 2012

92 -

Daniel Sturm

Davis, Donald and David Weinstein (2008) “A Search for Multiple Equilibria in Urban Industrial Structure,” Journal of Regional Science 48(1), 29-65. Fujita, Masahisa, Paul Krugman, and Anthony J. Venables (1999) The Spatial Economy: Cities, Regions and International Trade, Cambridge: MIT Press. Krugman, Paul (1979) “Increasing Returns, Monopolistic Competition, and International Trade,” Journal of International Economics, 9(4), 469-479. Krugman, Paul (1991) “Increasing Returns and Economic Geography,” Journal of Political Economy, 99(3), 483-499.

SLIDE 93

Brixen Workshop 2012

93 -

Daniel Sturm

Miguel, Ted and Gerard Roland (2011) “The Long-run Impact of Bombing Vietnam,” Journal of Development Economics, 96(1), 1-15. Redding, Steve, Daniel Sturm and Nikolaus Wolf (2011) “History and Industry Location: Evidence from German Airports,” Review of Economics and Statistics, 93(3), 814-831.