Upload
others
View
9
Download
0
Embed Size (px)
Citation preview
City Decline in an Urbanizing World- Preliminary -
Luis E. Quintero and Paula Restrepo
October 23, 2019
Urbanization studies and policy focuses on managing urban growth for the devel-
oping world. However, a large portion of the developing world present urban decline.
The first contributions of the paper will be to simply present this fact with a large
scale dataset that spans more than 2 decades for the Europe and Central Asia region.
The economic effect of population decline is quantified through the use of city level
Night Lights data. The second contributions if to specify a model that explains deliv-
ers predictions and test them. The first prediction, that market access has a negative
impact on city performance in a context of population decline is is different from the
usual positive effect found in the literature, which focuses on places with growing pop-
ulation. Second, the model suggests that productivity will increase after population
declines, especially in large cities. The first one is supported by the data, while the
second is not, although some endogeneity concerns remain. The tests are performed
for 5,548 cities between 1989 and 2013. We take particular notice of factors that are
of particular importance to this region, such as the existence of monotowns, in which
many countries have faced challenges transitioning from a centrally planned economy
to a market-based economy following the collapse of the Soviet Union.
Keywords: Migration, Urban Decline, Urban infrastructure, Eastern Europe, Gravity
Models.
1
1 Introduction
Policy and academic research on city dynamics in the developing world overwhelmingly
focuses on the growth of cities in both area and population, as well as in the increasing
urbanization of countries Geography (2009); Glaeser and Xiong (2017); Chauvin et al. (2017).
Desmet and Henderson (2015) warns against using without caution what we have learned
from developed countries to predict what will happen in developing countries, which further
motivates our inquiry into this particular world region. City growth and overall urbanization
are tightly linked to overall development (Henderson, 2014).
The objective and contributions of this paper are threefold. First, it presents evidence on
the striking phenomenon of population decline in Eastern Europe and Central Asia (ECA)
using a novel dataset that spans more than 3 decades for all cities in the countries of the
region. This evidence is aimed at motivating the stylized fact that a large portion of the
developing world has entered a demographic transition before attaining the typical develop-
ment levels associated with it. Second, it present a simple model (in progress) that provides
predictions of the role of market access and other determinants of population redistribution
in a context of population decline. The model is based on Brezis and Krugman (1997) model
of life cycle of cities. In our case, we do not model a change in technology, but instead a shock
to the city population. This model provides two main predictions: first, that in a context
when a group of cities face a negative population shock, population is redistributed to places
with larger markets; as a consequence, market access acts more as a push than a pull factor,
increasing the population movement out of those cities with large market access. Second,
that the state of technology and infrastructure is durable. As a consequence, in the short
run, when population declines cities increase their productivity. This effect is even stronger
in cities with more knowledge and infrastructure, which we proxy with local market size.
Third, it tests these predictions. The first one is confirmed. Our estimation finds that mar-
ket potential has a negative effect on population growth on cities that are already declining,
which supports the model’s idea that inhabitants located in a shrinking city j perceive larger
markets located closer to them as incentives to migrate out of j and not as potential markets
to sell products that would make staying in j more attractive. This is a robust result to
different specifications of market potential, inclusion of instrumental variables to control for
market potential endogeneity, and different treatments of the error structure in estimation.
This new empirical results contrasts with the existing literature (see for example Henderson
and Wang (2007)) and provides justification for allowing heterogeneity in the reaction of
urban systems to positive and negative shocks. The second result, whether productivity is
increased after population decreases, gives us mixed results. The estimation use novel data
2
to measure aggregate proximity to other markets, or market access (including using driving
distances instead of geodesic distances between cities to incorporate costs of poor transport
infrastructure or hazardous topography), as well as size of local market using night lights
(NLs) to measure economic activity.
2 Population Shock in the ECA Region
Most countries in Eastern Europe and Central Asia (ECA)1region experienced a negative
population shock at the end of the 1980s. The total population of the country either declined
or its growth rate was significantly slowed down. Figure 1c shows population growth rates
for the world. Since 1990, ECA suffered a strong decline in its population growth rates,
reaching absolute population decline in 1998. This region presented growth rates that are
lower than those of richer industrialized economies which have gone through a demographic
transition2, while having economic development levels of other developing region with much
larger growth rates (marker size denotes GDP per capita of the region).
Being an already highly urbanized region, the ECA countries have seen their urban
system suffer the effects of this negative population shock. 1c shows that many of these
countries have experienced a decline in more than 80% of their cities. The average number
of cities with absolute population loss in ECA countries whose populations have decelerated
are 77% between 1990 and 2000, and 72% between 2001 and 2012, which is dramatic.3 This
decline is dramatic. In contrast, for instance, Glaeser and Gyourko (2005) study population
decline in US metropolitan areas. They find only 6.72% of the counties considered losing
population, with an average loss of 9%.
These demographic changes are closely related to widespread structural changes that took
place during the transition from centrally planned to market oriented economies, especially
the loosening of migration restrictions which increased migration outflows, and the loss of
child bearing benefits that incentivized an accelerated decline in fertility.4As discussed, these
1ECA refers to the low and medium income countries as classified by the World Bank (high-incomeeconomies, those with a GNI per capita of $12,476 or more are excluded). The countries in this region are: Al-bania, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Georgia, Kazakhstan, Kosovo, Kyr-gyz Republic, Macedonia, Moldova, Montenegro, Romania, Russian Federation, Serbia, Tajikistan, Turkey,Turkmenistan, Ukraine, and Uzbekistan. Our analysis leaves out Armenia, Azerbaijan, Bosnia and Herze-govina, Kosovo, Macedonia, Montenegro, Turkmenistan due to data availability constraints.
2Myrskyla et al. (2009) review the conclusive body of evidence of a negative correlation between fertilityand industrial development, a theory that posits that at high levels of economic development societiesundergo a transition led by lower fertility rates that results in lower population growth rates. ECA seemsto have gone through such transition without the development levels usually associated with it.
3City population comes from our main dataset, collected from censuses and projections for 5,548 citiesin 17 countries. See detail in section 5.
4Figure ?? shows net migration and fertility rates for the region, and their decline, which approximately
3
national trends have permeated the urban system. National fertility and migration rates,
alone, explain 52 percent of ECA’s cities population growth variance (table 6).
Mobility. Strict mobility restrictions kept the urban systems in an equilibrium that was
far from that implied by individual decisiones in a market systems. From Thunen and
Hall (1966) and Fujita and Krugman (1995), to the more recent quantitative spatial mod-
els surveyed in Redding and Rossi-Hansberg (2017), spatial equilibrium, the methodological
workhorse that leads most of our modern understanding of cities, assumes wages, prices, pop-
ulation, and the housing stock are endogenous and jointly determined. This determination
is driven by agent’s making location decisions, both within and across cities, following arbi-
trage opportunities that appear as a response to shock, until prices adjust and remove any
incentives for moving. During the centrally planned regimes, these price adjustments were
not possible, and resources were constantly in a state of suboptimality. Not only was move-
ment of workers strictly controlled both across and within countries, trade across locations
and the local housing rents were exogenously determined by the central government too. As
a consequence, for example, both population and economic activity was more widespread in
space than observed in market based economies during the same decades (Markevich and
Mikhailova, 2013). Henderson and Wang (2007) found that more than a decade after tran-
sition, countries which urbanized under planned economies had significant lower levels of
urban concentration (measured by Spatial Gini Coefficients) when compared to the rest of
the world.
Starting on 1922, a general ban on emigration was established (Light, 2012), which was
further strengthened after WWII. Immigration was similarly blocked, even removing family
reunification provisions. This implied that in the 1970s, foreign nationals only accounted
for 0.05 percent of all workers in the USSR (Levcik, 1977). Internal migration restrictions
presented an even starker contrast with market economies. Residence was regulated centrally
by the emission of propiska, the residence permit, and receiving one for another location was
not a right. Furthermore, all major cities were subject to special limitations on the number
of propiskas, as well as to restrictions on the categories of persons who were eligible to receive
a propiska. By the late Soviet period, the cities that required a propiska included all capitals
of Soviet national republics, cities with a population in excess of 500,000, and several smaller
towns considered attractive for migration, as well as medium-size cities that were declared
closed-cities for national security purposes (as they were home to sensitive military, industrial
or scientific facilities). Besides residence, mere internal movement to major areas required an
internal passport5 People in the rest of the country did not receive passports, which resulted
coincides with regime transition.5requirement of internal passports was initially limited to 25 major cities and a 100-kilometer strip along
4
11.
21.
41.
61.
82
Popu
latio
n re
lativ
e to
196
0 (A
lban
ia)
.91
1.1
1.2
1.3
1.4
1.5
1.6
Popu
latio
n re
lativ
e to
196
0
-40 -20 0 20 40Year relative to break
Bulgaria Serbia Ukraine Russian BelarusRomania Poland Georgia Moldova Albania
(a) Declining population after trend discontinu-ity.
11.
52
2.5
3Po
pula
tion
rela
tive
to 1
960
-40 -20 0 20 40Year relative to break
Kazakhstan Kyrgyz
(b) Slowing population after trend discontinu-ity.
-0.5
0
0.5
1
1.5
2
2.5
3
3.5
1985 1990 1995 2000 2005 2010 2015
Annu
al P
opul
atio
n G
row
th
East Asia & Pacific Western EuropeECA Middle East & North AfricaSub-Saharan Africa North AmericaLatin America & Caribbean
(c) Population Growth by regions.
0 .2 .4 .6 .8 1Percentage of shrinking cities (by population)
UkraineSerbiaRussia
RomaniaPoland
MoldovaKyrgyz
KazakhstanGeorgiaBulgariaBelarusAlbania
1989-2000 2000-2011
(d) Share of shrinking cities.
Figure 1: Population Shock in ECA. (a,b) Population evolution around the time of the largesttrend discontinuity detected in a 55 year (1960-2015) window by country. (c) PopulationGrowth and GDP 1985-2015. The size of the markers is determined by GDP in constant2010 US dollars. Population from UN World Population Prospects. GDP from World Banknational accounts data, and OECD National Accounts data. (d) Percentage of all citieslosing population detected as urban areas in 1996 by NLs analysis.
5
in the majority of the population not being able to move, or even visit, the most developed
parts of the country. The whole system imposed restrictions on people’s movements that
would strengthen agglomeration economies. The removal of these restrictions at the time
of transition into a market economy naturally created strong adjustment in the location of
people and firms. It was like suddenly opening floodgates and allowing movement to a new
equilibrium.
Besides the mobility of population, firm location was severely restricted. One major
consequence of this control was the existence of cities, often called monotowns, that had
weak sectoral diversification (Kuzmenko and Soldak, 2010). Mono-towns can be found in all
of the ex-Soviet countries. In Ukraine alone, for example, 111 of the 456 cities in Ukraine
are considered monotowns even today. In Russia, there were 319 monotowns in 2015, cor-
responding to 30 percent of the cities and 12 percent of the total urban population.6 Just
as with labor mobility, transition to market economies allowed firms to relocate without a
strong central coordination. Many exited the market when confronted with international
competition and significantly smaller national markets after the fragmentation of previously
united territories. At the city level, mono-functional cities were found to be particularly
vulnerable due to their narrow economic specialization, such as has been studied for Detroit
(Owens et al., 2017; Glaeser and Ponzetto, 2007).
Fertility. Structural changes in fertility are the other component of the negative popula-
tion shock. The global fertility rate has fallen sharply going from an expected 2.7 children
by woman before 1989 to 1.6, well below replacement levels, in 2005, with some countries
going significantly below.
Under centrally planned economies, fertility rates in Eastern Europe were distorted by
constant interference to keep them high. The significant reduction of longer paid maternity
leaves, the end of guaranteed retention of workplace seniority during maternity leave, the
end of birth incentives policies7, and the overall increase of uncertainty (Billari and Filipov,
2004) and access to better contraception, increased the marriage age and lowered fertility
rates after the end of centrally planned economies in the region.
the USSR’s western border, but was later expanded to include towns, district centers, places with agriculturalequipment, areas within a 100-kilometer radius of large cities, frontier zones, building sites, and state farms,al locations known as restricted areas (Matthews, 1993).
6The definition of monotown varies by country. In Ukraine, monotowns are defined as cities where themajority of the economically active population works in enterprises focused on one of two economic sectorsthat also support a substantial part of the city’s budget. In Russia, the definitions requires satisfying tworequirements: the share of the enterprises of the same industry exceeds 25 percent of the local employment,or its production share is at least 50 percent of local total
7Birth incentives were allowances per children, which increased for the second and third child. In Bulgaria,for instance, besides this incentive, a year of unpaid vacation was guaranteed until children were 3 and itcontributed to the mother’s pension
6
0
0.5
1
1.5
2
2.5
3
3.5
-10
-8
-6
-4
-2
0
2
4
6
1960 1970 1980 1990 2000 2010
Ferti
lity
Rate
Net M
igra
tion
per 1
000
Net Migration per 1000 Fertility Rates
Figure 2: Aggregate Fertility Rates and Net migration rates for the Eastern Europe and Cen-tral Asia (ECA) region. Dashed line is the average period of the structural break estimated.Source: UN World Population Prospects 2017 revision.
Structural break. We want to analyze the impact that negative population shocks have
on urban systems. We more rigorously test for whether there was a population structural
break as suggested by the institutional changes discussed above. We follow Gendron-Carrier
et al. (2017) analysis of trend break8 in the value of population growth by estimating a series
of regressions of population growth on a step function, allowing the timing of the step to
change the study period.9 Let i = 1, ..., I index countries cities and t index years between
1959 and 2015. Define the following families of indicator variables,
Dit(j, j′) =
1 t ∈ j, ..., j′
0 otherwise(1)
Dit(t ≷ j′) =
1 t ≷ j′
0 otherwise(2)
Equation (2) describes a step function beginning j and ending j′ years from the considered
break candidate. We run the following regressions
PopitPopi1960
= α0,j + α1,jt+ α2,jtDit(j, k) + α3,jtDit(j, k) + εit,j (3)
for all j ∈ {−s ∗ T, ..., 0, ....s ∗ T} for each country i. T is the total number of periods.
8who in turn use the method in Andrews (1993), and Hansen (2000), and the statistics correction inAndrews (2003)
9Gendron-Carrier et al. (2017) know precisely the break they want to test, the time of opening of a subwaystation, and define variables around this candidate break. In contrast, in our analysis we test all possiblecandidates in a fixed width window that covers 75% of our available periods
7
We set k = T+sT2
, which effectively removes an upper tail of consideration in the break
trend. For our main analysis we set s = 0.75% following Gendron-Carrier et al. (2017). The
corresponding k is = 49.
Figures 4-7 show the Wald statistics corresponding to α3,j.10. The countries in figures
4 and 5 show a clear change in their population, with initial positive growth and then
absolute decline. The Wald statistics, which correspond to the tren factor, show this pattern.
We choose the higher value of the statistic, in absolute value, to determine the strongest
structural break. We label the years that are associated with a strong turn to market
economies, and that are linked to the changes in migration and fertility policies mentioned
above. The values of the Wald statistics are well above critical values derived from Andrews
(2003).11
Similarly, countries in figure 6 show a break in population dynamics, with initial positive
growth, followed by a still positive but slower change. The same conclusions are reached
from the statistical analysis for the countries in the first group. Finally, countries in figure
7 do not show a negative population shock, which is the type of population of interest. This
figure shows 2 former Soviet countries that do not show a negative population structural
change, as indicated by the Wald statistics low value.12
Additionally, having detected the break following the procedure above for countries in
the first two groups (figures 4 to 6 ), we also perform Chow (Chow, 1960) tests for the
equivalency of both trend and level equivalencies for the sample before and after the break
(within the {−s∗T : s∗T} range ). These Chow tests reject equality of the coefficient values
across subsamples, in favor of a significant structural change in the evolution of population
growth. Results are available upon request.
This section showed that there was negative population shock for 12 out of the 14 coun-
tries analyzed in the ECA region. The timing of the shock coincides with the strong structural
reforms that happened at the end of the 1980s. The main reforms that could be associated
with the loss of population were those that loosened restriction on mobility, and those that
10Recall break tests are performed on population change; population levels are shown, together with theWald statistics for α3,j from the regressions of the form shown in equation 3 for illustration
11Andrews (2003) provides (asymptotic) critical values for the Wald statistics values we have just generated,a sup-Wald that test the hypothesis of parameter constancy for the subsample, for α3,j = 0 for all j. For ourcase, where the break affects only one parameter and when trimming 25% from the boundaries of the sample,the 5% critical value is 7.87, less than the largest (absolute) value we observe for the Wald statistic. FollowingGendron-Carrier et al. (2017), we regard the test with cuation since our estimation framework differs fromthe one for which the value is derived in small ways. Still, the difference is so large we would expect thehypothesis test result to be valid. We fous on detecting whether there was one important structural breakthat affects population trends before our analysis. Thus we do not repeat the procedure in subsamples todetect multiple breaks as in Bai and Perron (1998).
12It shows another country in the region, Turkey, as well as Germany and UK for comparison, none withsignificant negative population shocks.
8
reduced the incentives for having children and resulted in an early demographic transition not
typical of regions in developing stages. Even after absolute losses of population, suboptimal
locations of people and firms inherited from centrally planned economies also incentivized
strong reallocation of the remaining resources and population. We are interested in assessing
the impact of a strong population decline in the whole urban system. This dramatic loss of
population that affected whole countries and their cities is a convenient exogenous shock to
test the reaction of the urban system in terms of redistribution of the remaining population
and economic activity.
An important concern remains with respect to the exogenenity of this shock. Although
the economic reforms were indeed unexpected, there were simultaneous events, including wars
and slow transition to a market system. To assuage these concerns, our results, the impact of
the negative population shock is additionally calculated 11 years after the transition, where
population decline is still happening but when the region’s economic systems have stabilized
and any violent conflict had already long ended. Figures 8 and 9 shows economic indices
constructed to measure, to the extent that this is possible, integration to a market economy
system. The indices indicate that, at the time of our analysis, the economic systems of the
countries are at least as market oriented as those of other developing regions.
3 Theoretical Model
Brezis and Krugman (1997) introduce a model aimed at explaining the life cycle of cities
as a consequence of techonological innovation. Incorporating a spatial dimension to the
Schumpeterian concept of creative destruction (Schumpeter, 1975), who considered it the
essential fact of capitalism. In their model, the appearance of a new technology changes
the relative productivity of two cities considered. Productivity depends on the accumulated
production over time (learning and accumulated knowledge). As a consequence, a larger city
might keep the older technology because it has a higher productivity given the accumulated
knowledge. The small city might adopt the new one because it has implies a higher marginal
growth at low accumulated production levels.
We adopt this model but do not simulate the introduction of a new technology but instead
explore what happens when there is shock to the population, as has been experience in ECA
cities. The decrease in the available labor supply changes the dynamics of productivity and
non tradeable prices. As a consequence, when we consider two cities, there is movement of
the remaining labor between the two cities, until a new spatial equilibrium is reached. We
reproduce some details here for completeness and then include two propositions about the
effects of a negative shock to population.
9
Initially we describe the economy in one city, and then include multiple cities to analyze
movements of factors as a reaction to shocks. The model considers an economy with a given
labor force L. This supply is divided into the two sectors, one technically progressive, i.e.
manufacturing, and another technically stagnant. The latter uses with a constant-returns
technology. The goods in this sector are non tradeable across cities. Brezis and Krugman
(1997) denote it as food. It can easily be considered as housing as well. Manufacturing
workers are denoted with number from 0 to m − 1. Housing workers are denoted with
numbers m to L.
One unit of housing is produced by one unit of land and one worker through Yh =
min(Lh, T ). The utility function of the housing worker is Cobb-Douglas U(Qmj, Qhj) =
QβmjQ
1−βhj where j denotes the housing worker j. Production of housing occurs outside of
the central business district (CBD). Commuting costs reduce the value of housing located
farther away from the CBD. This is modeled as a reduction of the produced good from the
perspective of the housing producer a la Samuelson’s iceberg costs. Dj is the remaining
value of a unit of housing produced to take into account the commuting costs. Assuming
a decay rate of δ we get Dj = e−δ(j−m). This gives us the following budget constraint:
pmQmj +phQhj = phDj−Rj ≡ Ej, with p is price of manufactures and housing (with m and
h subindices respectively). Rj is the land rented to produce the unit of housing and phDj is
the revenue received for the unit located m+ j away from the CBD. E)j is thus the income
of housing workers, including the imputed rent of their own housing.
In spatial equilibrium we require that all housing workers have equal income. This
determines rents. The income of the worker producing the most distance housing L is equal
to any other j, so EL/(pβmp
1−βh D1−β
L ) = Ej/pβmp
1−βh D1−β
j . To simplify notation, it is assumed
that the most distant housing worker L’s rent is zero. Substituting this into the budget
constraint and using the equalization of income condition we find the utility of each housing
worker in terms of the iceberg loss of value of the farthest unit
Uh =γpfDL
pβm(DLph)1−β= γDβ
Lpβhp−βm (4)
with γ = ββ(1− β)1−β. The rent is then determined for unit j,
Rj = phDj[1− (DL/Dj)β] (5)
Housing workers and landowners consume housing too. The income generated at a loca-
tion j is divided by housing workers and landowners, each spending (1− β) of their income
in housing according to the Cobb-Douglas preferences. β units of housing are available to
10
be sold to others. The net supply of housing to manufacturers Sh is obtained by integrating
β times the supply, which is the amount of value that remains of each unit delivered to
manufacturing workers Dj
Sh = β
∫ L
m
Djdj =β
δ(1− eδ(L−m) (6)
The manufacturing sector has a production function Ymi = a in which each manufac-
turing worker i produces a certain amount a that depends on the accumulated city level
production. This captures learning by doing and other type of sectoral knowledge spillovers.
a is determined by a = a(K(t)), where
K(t) =
∫ t
0
∫ m
0
Ymi(τ)didτ (7)
and
a = ΓevK
evK + µ(8)
K is a stock, so a is take as given at each point in time.
The budget constraint of a worker is pmQmi + phQhi = pma ≡ Ei and his utility after
maximization is
UM = γE
pβmp1−βh
= γap1−βm pβ−1
h (9)
Consumption of housing by manufacturing workers is then Qhi = (1−β)Ei
ph= (1−β)apm
ph
3.1 Short run equilibrium
Given a level of labor force L and productivity a, we can determine the allocation of land and
labor into the two sectors, and determine rents. We assume the manufactures to be tradable
with no additional costs, and so define manufactures as the numeraire (pm = 1). Equilibrium
requires that no further movement of labor occurs between sectors, which equalizes real
income of workers in both sectors
DβL = a/ph (10)
and that the market for housing clear, which implies
Dh = (1− β)ma
ph=β
δ(1− e−δ(L−m)) (11)
where Dh is the demand for housing, and the supply comes from equation 7.
Intersection of equations 10 and 11 determine the equilibrium allocation of labor and
11
relative price of housing.
There are two important parameters that affect this equilibrium in the following ways:
Proposition 1 The welfare of all individuals in the city rises with a. An increase of 1
percent in a increases utility in β.
This comes from the fact that a enters equations 10 and 11 as a/ph, and thus an increase in
a is matched by an increase in the price of housing, with no reallocation of labor 13 . The
result follows from this and the utility equation 9
Proposition 2 An increase in L reduces the income of the typical worker.
This comes from equation 9, since the utility of a worker is negatively related to ph, and an
increase in L raises ph, we get the previous result. The intuition is that in a larger city, some
housing units must be farther away, implying a lower utility from housing (or higher prices)
and thus lower utility.
Proposition 3 From the previous propositions, it follows that the utility in a city is an
increasing function of productivity a and decreasing function on L
3.2 Multiple cities and a population shock
If we introduce two cities in the model, spatial equilibrium that no incentives exist to induce
movement of labor between the two cities. Now assume that there is a negative population
shock that affects both cities equally, so that Lkτ = Lk for cities k = 1, 2, where L is the
population after the shock, and 0 < τ < 1, and L1 > L2. The initial effect on the equilibrium
is provided by the following:
Proposition 4 The decline in L increases the income of the typical worker. The decline in
population is larger in the larger city 1. The increase in income is larger too in city a.
The first result follows the same logic as proposition 2, but with the inverse direction in the
changes. Again, the intuition is that in a city that becomes smaller, some housing units that
were farther away exit the market, implying a higher utility from housing (or lower prices)
and thus higher utility.
Proposition 5 The differences in incomes induces movement of labor from the small city
2 to the large city 1.
13m is unaffected by an increase in a is an artifact of Cobb-Douglas preferences.
12
This result comes from the combination of two effects. First, since productivity depends on
the historical accumulation of production and knowledge, and since the current production
has negligible effect in a, the differences in productivity remain between the small and large
cities. The gap between of the new lower housing prices determined by the intersection
of equations 10 and 11 in each city and the productivity becomes greater after the loss of
population. There are thus incentives for labor to reallocate. How fast this movement occurs
depends on whether we assume mobility costs or not, but also on how fast the accumulated
productions decrease. This depends on the parameters values chosen 14.
4 Data
Previous discussion suggests that population reorganization, as well as pure population de-
cline, will occur in cities as a response to negative shocks, such as a sectoral shock, a change
in fertility rates or a change in the domestic market access of cities (coming for example
from a division of a country in two). In order to test these predictions, we use data on cities
in the ECA region, which has experienced a strong negative population shock. We use data
on economic activity, population and market access to test the predictions. We consider the
construction of datasets with variables that are comparable across 15 countries in the region
to be an important contribution and suggest it is a useful tool for related research. The core
empirical analysis in this report is based on a city database - hereafter referred to as the
Cities in ECA database - composed of 5,549 cities and covering 15 countries of the Eastern
Europe and Central Asia region as defined by the World Bank Group. ECA refers to the low
and medium income countries as classified by the World Bank (high-income economies, those
with a GNI per capita of $12,476 or more are excluded). The countries included are: Alba-
nia, Armenia, Azerbaijan, Belarus, Bosnia and Herzegovina, Bulgaria, Georgia, Kazakhstan,
Kosovo, Kyrgyz Republic, Macedonia, Moldova, Montenegro, Romania, Russian Federation,
Serbia, Tajikistan, Turkey, Turkmenistan, Ukraine, and Uzbekistan. The city database also
contains information on cities in the United Kingdom (UK) and Germany, which are used
as a benchmark from Western Europe. The city database contains information for each city
along three dimensions: demographic, spatial and economic.
14See for example figure 5 in Brezis and Krugman (1997) to see what the movement between the twocities would look like under a simulation. Their simulation considers the introduction of a new technology.However, the allocative labor effects between are similar to the negative population shock we discuss
13
4.1 Population
Demographic data was obtained from official sources and verified by in-country experts. An
effort was made to have city definitions be comparable. Population data collected for cities
fall on or around three years of analysis: 1989, 1999 and 2010 (or the latest year available)
—hereinafter referred to as year 1, year 2 and year 3, respectively. Population data is
collected for all municipalities as small as 1,000. Despite having a shorter time frame than
similar datasets, its larger size scope allows for conclusions to be applicable to all tails of the
urban system population distribution15. Population data is based on administrative units.
Data includes all administrative units with urban denomination in each country. Additional
descriptive statistics are available in table 5. Reports by country for all of the available data
is available in an online appendix.
To control for population general patterns, we also use data on natural population growth
(the difference between the birth rate and the death rate), fertility rates, and net migration
at the country level from Census reports and other statistical publications from national
statistical offices.
4.2 Economic Activity
Given the lack of available disaggregated and comparable economic data at the city level,
Nighttime Lights (NLs) provide a unique dataset that can be used as a base for assessing
economic trends. We use Nighttime Lights (NLs) as a proxy of economic activity. This NL
information is measured from satellites, aggregated to roughly 250,000 quarter degree (lon-
gitude/latitude) grid squares. The NLS analysis presented in this study is based on a Global
Night Time Lights Urban Extents and Growth Patterns Product developed by the World
Bank team. This product uses Nighttime Lights (NLs), produced by the Defense Meteoro-
logical Satellite Program (DMSP) - Optical Line Scanner (OLS) database and maintained by
NOAA. We use the radiance calibrated NLs data, which addresses saturation issues (Roberts
et al., 2015). NLs offer a versatile and global dataset at a relatively fine spatial resolution.
There is evidence of a strong positive correlation between NLs growth and real GDP
growth at the country level, and more recently at the regional/subnational level (Henderson
et al. (2011)) 16. Henderson et al. (2011) discuss the benefits of using this data and present
evidence of its validity. There are several advantages of using this data. First, NLs data
15As a contrast, Henderson and Wang (2007) build a data set on all metro areas over 100,000 from 1960to 2000; the UN Statistics Division has a dataset since 1950, for cities with more than 300,000 inhabitants.
16This relationship can be heterogenous depending on the sectoral concentration (Henderson et al., 2016)or growth and decline dynamics (Quintero et al., 2017) past patterns of growth over time. We do notincorporate that heterogeneity here.
14
provide a globally consistent data set that is comparable, across countries. Second, unlike
other global economic datasets, it is sampled uniformly (Henderson et al., 2012), and its mea-
surement error is not related to development levels. Measurement errors are thus expected
to affect any area with a uniform probability, whereas official statistical data measurement
error affects disproportionately smaller cities in countries with poorer institutions and ability
to collect this data, which could introduce a harmful bias. Third, NLs provide information
about economic activity at levels of geographical disaggregation for which economic data is
generally absent, which is the case of cities in ECA. We perform tests similar to those in
Henderson et al. (2012) using subnational Gross Regional Domestic Product (GRDP) and
the corresponding aggregate NLs data, and find robust positive correlations that at least
partially support the use of NLs at other aggregation levels (see table 7). This test assesses
the relationship between NLs (radiance calibrated) and regional GDP for 16 of the 17 ECA
countries analyzed (Moldova does not produce subnational GDP data). For all 16 coun-
tries we found a positive and significant relationship between NLs and GDP levels. We also
tested– by aggregating the full set of countries, the relationship between GDP growth and
NLS growth at the subnational level and found it to be positive and statistically significant
with a coefficient of 0.182 for NUTS 2 regression (0.081 for NUTS3) and R-square of 0.788
(0.826).
5 Data
5.1 Market Access
The new economic geography has used measures of market access as an important determi-
nant of trade. The market access measure is related to gravitational models of migration
pioneered by Ravenstein (1889). This concept has been extended to explain migration (Lewer
and Van den Berg, 2008) and city growth rates. Au and Henderson (2006) develop a model
for city growth, in which they include not just the usual external economies of scale and
local diversification, but also the new economic geography’s local market and market access
effects. The main role of market access here is the idea that increasing exposure to nearby
markets benefits the city economy through higher demand and lower transportation costs for
city exports. Larger distance implies higher costs and higher product sale prices, which trade
theory usually models with the iceberg costs setup. However, as discussed, higher market
potential can also negatively affect city growth if its main effect is to serve as a pull factor
for local population to out-migrate. We want to test which of these predictions are true in
a context of population decline. Head and Mayer (2004); Overman et al. (2001) provide a
15
comprehensive review of the literature that develops the relationship between trade across
space and market access.
The market access for city i aggregates such market exposure to all other individual
markets (all other cities in the country), assigning weights to markets proportional to their
size and inversely to their distance. Using a notation similar to Henderson and Wang (2007),
we use the following measure of market access 17 to obtain a measure of market potential,
MP, for each city:
MAi(t) =
Njt−1∑k∈j|j 6=i
nk(t)
dik(12)
where n is a measure of income, dik is the distance between city i and j in country j,
Njt is the total number of cities in county j at time t. This measure is a a gravity index.
It separates the external and local market access in the market potential measure, and so
excludes the interaction of city j with itself (which we capture with measurements of local
market in our estimation). This specification is a simplification of market access that assumes
that prices are equal across cities, that market size measurement is a proxy for income, and
that transportation costs between j and k are proportional to distance.
In particular, for our main specification, we use an exponent of -1 for distance and assume
a proportional relationship between costs and distance. Hanson (2005) use an exponential
form to connect distance and costs. Yet, the vast empirical literature estimating gravity
equations suggests that dik should be a power function of distance of the form as the log of
trade flows is unanimously found to decrease linearly with the log of distance (with slope
near -1) (Head and Mayer, 2004). Additionally, following earlier gravity model literature
we also test using a squared distance in the denominator of equation 12. We introduce this
robustness test to capture differences in the results, which suggest that further estimating
or calibrating the exponent is important.
To compute distance, we first use geodetic distance using equations introduced by Vin-
centy (1975), which assume that the Earth is an oblate spheroid, and hence are more accu-
rate than methods that assume it is spherical. Thomas and Featherstone (2005) prove this
method maintains submillimeter accuracy between all locations near the sea level18
17The specification we use is a differenced and linearized version of Au and Henderson (2006) under anassumption of perfect population mobility. Specifically we take Au and Henderson (2006) for real incomeper worker in a city, equate it to the national real income in national labor markets (under perfect mobility),takes logs, difference, and then linearize.
18Most real-life data points will not be located near sea level. At a reference altitude of 3.2 km (LakeTiticaca) an error is expected in proportion to the increase in radius of the earth or 0.05%. In the citiesin our dataset, this does not seem to be an issue. Only Tajikistan and Kyrgyztan have small towns withaltitudes higher than 3.2 km with populations of less than 5,000 (hence, very low weight in any market access
16
NL populationgeodetic distance MP (NL, d) MP (pop, d)geodetic distance squared MP (NL, d2) MP (pop, d2)driving distance MP (NL, dd) MP (pop, dd)driving distance squared MP (NL, dd2) MP (pop, dd2)
Table 1: Using the different market size and distance combinations we get 8 different mea-sures.
Alternatively, we use driving distance to measure distance equation 12 using driving
optimal distances calculated from the Open Source Routing Machine (OSRM) and Open-
StreetMap19. We expect driving distance to be a more accurate measure of transport costs
because they incorporate costs of poor transport infrastructure or hazardous topography,
which could be of particular importance for urban systems in developing countries. Despite
the additional work this requires, we find no qualitative difference in the results derive with
market access calculated with geodetic or driving distances.
Market size, nk(t), is measured in two ways, using either population or NLs as measures
of market size (to proxy total city income) as described in sections 4.1 and 4.2. Using
either measurement changes the number of observations and the level of observations in our
regressions. Population is measured by each administrative unit (generally municipalities).
The captured NL footprint, in contrast, cannot be separated between municipalities whose
NL emissions touch in space, forming agglomerations, as described in the following section.
For these agglomerations we only have one measurement of NLs so we aggregate them into
one when using market access with NLs.
5.2 Urban form
To classify areas into urban and rural, which is equivalent to delineate urban footprints, we
use NLs. We follow the methodology in Roberts et al. (2015). The algorithm to determine
urban extents consists of random selection of satellite images in each country, followed by
a determination of whether they show built infrastructure that corresponds to urban areas
using color matching. After multiple random draws, a minimum NLs intensity level threshold
for urban areas is determined. This threshold is allowed to vary by country to incorporate
differences in electricity infrastructure.
Agglomerations are groupings of cities who work as a single functional entity, sharing
calculation). Only Tajikistan has high average altitudes.19Subsamples of distance calculations derived from Google maps were tested and no significant changes
were found between these two methods of distance calculation. Google maps driving distances, which undergomore strict validation, could not be used for the whole sample because of query restrictions in its API
17
labor and housing markets. As a consequence, we expect any type of agglomeration benefits
and spillover to be shared as well. NLs urban thresholds are used to identify agglomerations,
which determined as groups of municipalities whose footprints intersects with one another.
The status of agglomeration or single city will of course over time and we trace that evolution
with historical NLs data. Figure 3 shows how, for example, Kiev grew to absorb cities that
were separated from its core between 1996 and 2010). We control for the derived urban
structure (i.e. single city, agglomeration center, and periphery) in our estimation.
We identify a total of 352 agglomerations composed of a total of 2,358 cities in the
17 countries studied. Each agglomeration is composed, on average, of 6.7 cities. Among
agglomerations we differentiate between the centers of agglomerations, which correspond to
the largest city in the agglomeration, and the cities surrounding the centers of agglomeration
to which we refer as surrounding agglomeration. Single cities are cities who, contrary to
agglomerations, have an urban footprint contained within a single administrative area. There
are a total of 2,243 single cities in the Cities in ECA database.
We also create secondary cities indicators to include as controls. Secondary cities are
cities that fall in the top 20 percent of cities in a given country in terms of population size.
The capital city is not considered in this category and is controlled for separately. This
definition allows to compare countries whose urban systems are very different.
5.3 Industrial structure
Additionally, the role of industrial structure and central planning is explored for the subsam-
ple of Russian cities, using a dataset on data that identifies 224 mono towns in Russia. As
discussed in ??, mono-towns are Russian cities whose economies were dominated by a single
industry. This was a common feature during Soviet times, where towns were selected to
dedicate most of its resources to supporting a single industry. The list used in city-database
is an official list of the mono-towns in Russia taken from Kuzmenko and Soldak (2010). This
classification is obtained by applying specific criteria: the share of the largest or several
enterprises of the same industry (or operating in the same market) exceeds 25 percent of the
number of employed in the city/town, or the share of one industry is more than 50 percent
of the total production in the city/town. As of 2015, there are 3193 mono-towns in Russia,
which corresponds to 30 percent of the cities in Russia and 12 percent of the total urban
population in 2010.
18
5.4 Location fundamentals
Following Henderson and Wang (2007) we include six location fundamentals as controls
to avoid missing variable bias, : (i) distance to border, (ii) distance to coast, (iii) forest
coverage, (iv) annual precipitation, (v) average temperature in January and (vi) land usabil-
ity. Distance to border and coast is measured by geodetic distance. Forest coverage comes
from the Food insecurity, poverty, and environment global GIS database (FGGD), by FAO
Food Insecurity and (FGGD). Each pixel cover 5 arc-minutes and the variable includes the
percentage of the area belonging to forest. Average minimum January temp and average
annual precipitation is modeled from historical data published on the World Bank’s Climate
Change Knowledge Portal (originally derived from the Climatic Research Unit Time Series)
(Jones and I., 2013). We use health of vegetation as a proxy for land usability. This is
obtained from the MODIS (Moderate Resolution Imaging Spectroradiometer ) Normalized
Difference Vegetation Index (NDVI). We use the maximum 2015 NDVI value (basically the
healthiest reading that occurred during 2015, by cell). The MODIS NDVI is produced by
the US Geological Survey (USGS) (USGS). NDVI is calculated from the visible and near-
infrared light reflected by vegetation. Healthy vegetation absorbs most of the visible light
and reflects near-infrared light. Unhealthy vegetation reflects more visible light and less
near-infrared light. In terms of the index, a larger number means healthier vegetation.
6 Estimation
Push and pull factors are captured in a measurement of a local market size and an outside
market potential. Our model provides a useful estimating equation:
∆pi = β1ni + β2MAi(nk, dik, k ∈ j|j 6= i) + controlsi,c (13)
∆pi is the population change (p3/p2) between periods 2 and 3 20 , our main variable of interest.
ni is the initial local market size. MAi is the market access defined in 12. The notation
here makes it explicit that MA for city i depends on both the market size and distance
measurement used. Different combinations are used as described in table 12. Controls are
location fundamentals, migration, fertility, and natural population growth rates discussed
in the previous sections. Migration, fertility and natural population growth rates vary by
country c. All variables are introduced in logarithms.
We are particularly interested in the effect of market access in a context of population
20Recall periods 1, 2 and 3 are determined in each country by the years in the second column of table 5
19
decline, as discussed in our motivation and theory. To capture heterogeneity in the role of
market access we modify equation 13 to allow the effect of the outside market effect (β2) to
vary depending on the observed population dynamics of the previous period. In estimation
this is incorporated through an interaction term with an indicator function I{∆p > 0},which refers to the lagged change in population.
∆pi = β1ni + β2MAi + β3I{∆p > 0} ∗MAi + controlsi,c (14)
6.1 Instrumentation
We are interested in the effect of market access in the loss of population. Market access can be
endogenous in the sense that there are omitted variables that could affect both market access
and population dynamics. For instance, natural features as a body of water can provide
advantages that affect population growth, and at the same time affect the probability of
more towns locating closer together in nearby areas, increasing market access. Indeed, this
endogeneity is suggested empirically when performing Wu-Hausman tests of endogeneity.
We use instrumental variables that affect population change only through their effect on
market access. We choose the distance to nearest cost, distance to nearest international
border; a measure of city centrality calculated as the distance of each city to the centroid
of the country (the most central location in the polygon that makes the country’s map);
and a measurement of ranking of the city’s market size in the country urban system. In 2
stage IV estimation, these instruments perform well in a first stage, and pass Sargan’s test of
overidentifying restriction tests in all specifications. We are currently working on including
lagged variables as instruments.
6.2 Preliminary results
The table 4 shows the main results. Model (2) shows the usual result of a positive impact
market access on population growth. However, as suggested in the theory discussion, we
expected the effect to be different in contexts of population decline. When interacted with
a dummy that identifies country lagged (periods 1 to 2) population loss, market access
has a negative impact on population growth (positive impact on population decline) even
when instrumenting to control for endogeneity (model (4)), introducing robust variance
estimates, and controlling for local population country wide trends through measures of
natural population growth and fertility, local environmental amenities, and institutional
structures that might benefit capitals and secondary cities. Further heterogeneity is allowed
20
in models (6),(8), and (10). Indeed, the impact of market access is negative for cities in
regions that are declining (6), in cities that have previously declined (8), and overall in
combinations of surrounding population decline (10). This is the main empirical result that
contrasts with other studies that study market access in contexts of population growth.
Regarding the controls, we see expected correlations. Cities in formerly communist coun-
tries suffer from a greater decline even after controlling for general population trends and
local environmental amenities. This can be explained by the many distortions in location
and industry structures that were imposed in cities in these countries (see section ??). This
is further studied in table 3 for only Russia. The mono town indicator, which captures
concentration of a single industry, is correlated with population decline as expected. Indus-
trial concentration would have made it more difficult for cities to adapt to different shocks,
including population transition and introduction to a market economy. However, endogene-
ity concerns are not addressed for this variable here. Another variable included to capture
the effect of cities established by central planning is a dummy equal to 1 if the city was
founded during the USSR (foundation date between 1922 and 1991). The correlation is not
significant. Although the share of these cities is not negligible (17%), we believe we might be
missing many cities that were not officially founded during these times but grew significantly
by centralized decree.
Secondary and capital cities decline less. This could reflect institutions that favor these
cities in the country (we are already controlling for local size of the market, which should
incorporate local agglomeration effects). Surprisingly, cities with larger local markets are
losing more population, although this result is much less robust to the different specifica-
tions. Another interesting result, is the role of agglomerations. regressions that use NL as
a measurement of market size are run at the agglomeration not the municipality level, as
NLs footprints cannot be identified separately. The aggregate population growth, in the case
of agglomerations, is the dependent variable used. The agglomerations are numerous but
have actually been declining (35% of the cities in period 2012 in the sample belong to an
agglomeration vs 42% in period 1989). Agglomeration economies would suggest a positive
effect coming from the integration of cities. However, it could be the case that frictions and
coordination issues from separate local administrative bodies could be negatively affecting
the performance of these cities and pushing people out Henderson and Venables (2009).
Owens et al. (2017) investigate this coordination problems for Detroit metropolitan area.
Estimation performed with population based market access (tables 10 and 11) allows us
to separate market access between municipalities that belong to the same agglomeration.
These suggest a process of suburbanization, where the core of the agglomeration is losing
population with respect to single cities, but the periphery of the agglomerations are growing
21
more than both the core and other single cities.
Appendices C.2 and C.1 show additional results. Estimation performed with population
based market access is weaker in both the sense of showing lower fit and less statistical in the
local market and market access variables. However, it is important to notice this estimation
does not satisfy our theoretical predictions at the country level. Even though the result that
higher market access has a negative impact for cities that are located in regions that are
losing population and for cities that were themselves declining in previous periods is robust
to this specification, the result does not translate to the declining context defined at the
country level. We believe the results that are using NLs are a more reliable measure of the
market size (in the sense of effective demand) in any case.
As for the tables that show results using geodetic distances, there are virtually no qualita-
tive differences. Although the idea of using driving distances to better capture transportation
costs in developing countries which might have insufficient transport infrastructure remains
attractive, our current results suggest that this additional step makes no difference in the
context of ECA. Countries with different development levels might warrant this.
The heterogeneous effect is stronger in countries that were formerly communist (table
??), which could be connected to the common history of locating population and cities
through central decisions, which in turn could be affecting how optimal these locations were
in the first period. This is interesting, but we are currently working on developing a more
clear understanding of the phenomenon.
Some locational fundamentals matter. Being close to the coast and having higher av-
erage temperatures (in January) is positive correlated with city population growth. These
variables can have multiple meanings in regards to their relationship with the existence,
persistence and growth of cities. Coastal cities likely have a higher access to international
markets (international market potential); while the coast can also be considered a natural
amenity. Average temperature can also be considered a natural amenity, with people having
preferences from milder winters.
22
Table 2: All pooled OLS - IV (NL-Driving Distance) - Population Growth
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)OLS IV OLS IV OLS IV OLS IV OLS IV
Market
local mkt 0.017∗ -0.000 0.016∗ 0.014∗ 0.016∗ 0.014 0.013∗ -0.013∗ 0.013∗ 0.005∗
(10.10) (-0.03) (9.35) (6.34) (9.84) (0.69) (8.25) (-2.07) (7.82) (2.20)
mkt access -0.021∗ 0.031∗ -0.013∗ -0.006 -0.017∗ 0.094 -0.010∗ 0.071∗ -0.005∗ 0.025∗
(-13.71) (2.05) (-7.38) (-1.20) (-9.72) (1.53) (-5.84) (4.02) (-2.39) (4.46)
mkt access×I{∆countryp < 0} -0.010∗ -0.017∗ -0.007∗ -0.016∗
(-7.88) (-10.85) (-5.71) (-9.94)
mkt access×I{∆regionp < 0} -0.007∗ -0.242∗ -0.001 -0.006∗
(-6.49) (-2.64) (-0.94) (-4.10)
mkt access×I{∆p < 0} -0.016∗ -0.039∗ -0.014∗ -0.030∗
(-14.64) (-16.25) (-12.52) (-20.13)
Former communist -0.123∗ -0.036 -0.146∗ -0.159∗ -0.122∗ -0.166 -0.130∗ -0.033 -0.145∗ -0.160∗
(-14.03) (-1.35) (-15.94) (-14.71) (-14.00) (-1.47) (-15.32) (-1.13) (-16.36) (-14.24)
Population Fundamentals
nat. pop ∆ 0.002∗ 0.003∗ 0.002∗ 0.001∗ 0.002∗ -0.005 0.002∗ 0.002∗ 0.001∗ 0.000(6.68) (6.25) (4.68) (3.24) (6.20) (-1.42) (5.48) (3.24) (4.02) (0.36)
net migration -0.001 -0.026∗ -0.006∗ -0.010∗ -0.003∗ -0.030 -0.004∗ -0.039∗ -0.007∗ -0.018∗
(-0.97) (-3.55) (-4.07) (-3.66) (-2.16) (-1.09) (-3.27) (-4.64) (-5.36) (-6.50)
Urban Structure
Agglomeration -0.032∗ -0.033∗ -0.025∗ -0.020∗ -0.031∗ 0.010 -0.030∗ -0.030∗ -0.026∗ -0.017∗
(-4.15) (-3.64) (-3.35) (-2.66) (-4.05) (0.26) (-4.14) (-2.99) (-3.53) (-2.18)
Capital city 0.002 0.138∗ 0.011 0.042 0.005 -0.010 0.018 0.196∗ 0.024 0.080(0.07) (2.24) (0.38) (0.95) (0.17) (-0.04) (0.62) (2.87) (0.81) (1.73)
Location Fundamentals
pct forest 0.000 -0.000 0.000 0.000 0.000 0.002 0.000∗ -0.000 0.000∗ 0.000∗
(0.89) (-1.81) (1.24) (1.28) (1.12) (1.46) (2.27) (-0.30) (2.46) (3.43)
precipitation 0.000 0.000∗ 0.000 -0.000 0.000 -0.001∗ 0.000 0.000∗ -0.000 -0.000∗
(1.50) (3.20) (0.35) (-0.52) (0.47) (-2.18) (0.56) (2.10) (-0.36) (-2.69)
temperature 0.001 -0.003∗ 0.001∗ 0.002∗ 0.000 -0.003 -0.000 -0.006∗ 0.000 -0.001(1.31) (-2.33) (2.47) (2.87) (0.94) (-0.62) (-0.99) (-4.63) (0.01) (-1.91)
land usability -0.000 -0.001∗ -0.000 -0.000 -0.000 0.003 -0.000 -0.001∗ -0.000 -0.000(-0.59) (-3.34) (-0.45) (-0.47) (-0.15) (1.41) (-0.69) (-3.75) (-0.52) (-1.02)
Constant 0.064∗ 0.114∗ 0.091∗ 0.113∗ 0.063∗ -0.018 0.093∗ 0.197∗ 0.109∗ 0.171∗
(2.45) (3.29) (3.46) (4.19) (2.42) (-0.13) (3.68) (4.99) (4.31) (6.08)Observations 2501 2491 2501 2491 2501 2491 2501 2491 2501 2491
R2 0.262 . 0.280 0.265 0.275 . 0.321 . 0.330 0.212
Adjusted R2 0.259 . 0.277 0.261 0.271 . 0.317 . 0.326 0.208
t statistics in parentheses∗ p < 0.05
23
Table 3: Russia OLS - IV (NL-Driving Distance)
(1) (2) (3) (4) (5) (6) (7) (8)OLS IV OLS IV OLS IV OLS IV
Market
local mkt 0.016∗ 0.016∗ 0.014∗ 0.012 0.008∗ 0.007∗ 0.008∗ 0.006∗
(7.19) (6.82) (6.41) (1.12) (4.00) (3.26) (3.90) (2.98)
mkt access -0.005 -0.000 0.010∗ 0.028 0.022∗ 0.029∗ 0.026∗ 0.037∗
(-1.40) (-0.03) (2.85) (0.38) (6.79) (3.88) (7.88) (4.71)
mkt access×I{∆regionp < 0} -0.014∗ -0.019 -0.006∗ -0.007∗
(-9.15) (-0.38) (-4.49) (-4.82)
mkt access×I{∆p < 0} -0.027∗ -0.029∗ -0.024∗ -0.026∗
(-17.48) (-18.06) (-14.95) (-15.65)
Industrial Structure
Monotown -0.037∗ -0.037∗ -0.024∗ -0.019 -0.017∗ -0.015∗ -0.013 -0.011(-4.73) (-4.78) (-3.15) (-0.41) (-2.54) (-2.33) (-1.93) (-1.63)
founded 1922-1991 0.012 0.013 -0.001 -0.004 0.005 0.006 -0.000 0.000(1.29) (1.39) (-0.17) (-0.08) (0.68) (0.76) (-0.03) (0.04)
Urban Structure
Agglomeration -0.018 -0.017 -0.010 -0.004 -0.012 -0.010 -0.009 -0.006(-1.45) (-1.37) (-0.84) (-0.12) (-1.14) (-0.98) (-0.85) (-0.57)
Capital city 0.035 0.041 0.030 0.044 0.044 0.051 0.040 0.051(0.41) (0.48) (0.37) (0.53) (0.62) (0.72) (0.58) (0.73)
Location Fundamentals
pct forest -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000(-0.74) (-0.90) (-0.36) (-0.66) (-0.59) (-0.80) (-0.40) (-0.76)
precipitation -0.000 -0.000 0.000 0.000 -0.000 -0.000 -0.000 0.000(-0.41) (-0.14) (0.39) (0.33) (-0.45) (-0.09) (-0.03) (0.54)
temperature 0.003∗ 0.002∗ 0.002∗ -0.000 0.001∗ 0.001 0.001 -0.000(3.77) (2.19) (2.25) (-0.01) (2.05) (0.71) (1.42) (-0.11)
land usability -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000 -0.000 -0.000 -0.000(-2.92) (-2.96) (-2.58) (-2.14) (-1.60) (-1.63) (-1.54) (-1.71)
Constant -0.007 -0.027 -0.062 -0.136 -0.037 -0.062 -0.059 -0.101∗
(-0.17) (-0.53) (-1.73) (-0.47) (-1.19) (-1.45) (-1.91) (-2.29)Observations 694 694 694 694 694 694 694 694
R2 0.179 0.177 0.268 0.241 0.433 0.428 0.449 0.440
Adjusted R2 0.167 0.165 0.257 0.229 0.424 0.419 0.439 0.430
t statistics in parentheses∗ p < 0.05
24
Table 4: All pooled OLS - IV (NL-Driving Distance) - Productivity
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)OLS IV OLS IV OLS IV OLS IV OLS IV
Population growth
Growth btn yr 1 and yr2 0.035 0.018 0.017 0.017 0.015 0.118 0.017 0.018 0.016 0.017(0.86) (0.89) (0.89) (0.89) (0.77) (1.72) (0.90) (0.91) (0.84) (0.89)
Market
local mkt 0.007 0.004 0.001 0.003 -0.001 0.045 -0.002 0.004 -0.002 0.002(1.36) (0.78) (0.12) (0.61) (-0.11) (1.83) (-0.34) (0.64) (-0.34) (0.46)
mkt access -0.006 -0.019 -0.001 -0.013 0.001 -0.274∗ 0.004 -0.015 0.009 -0.009(-1.22) (-1.49) (-0.22) (-1.62) (0.20) (-2.05) (0.66) (-1.10) (1.55) (-0.91)
mkt access×I{∆countryp < 0} 0.077∗ 0.006 0.080∗ 0.013(3.77) (0.15) (3.93) (0.34)
mkt access×I{∆regionp < 0} -0.007 0.320 -0.004 -0.000(-1.72) (1.95) (-1.02) (-0.10)
mkt access×I{∆p < 0} -0.009∗ -0.006 -0.009∗ -0.008(-2.34) (-1.26) (-1.99) (-1.50)
Former communist -0.360∗ -0.187∗ -0.023 -0.164 -0.126∗ -1.582∗ -0.116∗ -0.166∗ 0.035 -0.125(-5.30) (-3.57) (-0.42) (-1.87) (-2.75) (-2.19) (-2.56) (-2.86) (0.60) (-1.34)
Population Fundamentals
nat. pop ∆ -0.001 0.010∗ 0.015∗ 0.011∗ 0.010∗ 0.016∗ 0.010∗ 0.010∗ 0.015∗ 0.011∗
(-0.50) (9.07) (8.44) (3.71) (8.72) (3.98) (8.75) (9.00) (8.47) (3.91)
net migration -0.006 0.012∗ 0.023∗ 0.011 0.005 0.065∗ 0.005 0.012∗ 0.023∗ 0.013(-1.61) (2.09) (4.03) (1.35) (1.40) (2.15) (1.43) (2.09) (3.90) (1.59)
Urban Structure
Agglomeration 0.012 -0.017 -0.012 -0.015 -0.011 -0.085 -0.012 -0.018 -0.011 -0.016(0.40) (-0.75) (-0.53) (-0.68) (-0.47) (-1.33) (-0.51) (-0.77) (-0.48) (-0.71)
Capital city -0.175 -0.349∗ -0.204∗ -0.336∗ -0.226∗ -0.568∗ -0.220∗ -0.348∗ -0.190∗ -0.333∗
(-1.65) (-3.19) (-2.66) (-3.10) (-2.94) (-2.03) (-2.86) (-3.17) (-2.47) (-3.08)
Location Fundamentals
pct forest 0.000 0.000 0.000 0.000 0.000 -0.002 0.000 0.000 0.000 0.000(0.33) (0.74) (0.35) (0.60) (0.45) (-1.24) (0.41) (0.86) (0.53) (0.79)
precipitation -0.000 -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000∗
(-1.95) (-4.90) (-3.12) (-3.77) (-4.71) (-2.83) (-4.50) (-4.72) (-2.80) (-3.53)
temperature -0.005∗ -0.000 -0.001 -0.000 -0.001 0.005 -0.001 -0.001 -0.001 -0.001(-3.57) (-0.23) (-0.79) (-0.30) (-0.44) (1.26) (-0.66) (-0.42) (-1.14) (-0.57)
land usability -0.001 -0.000 -0.000 -0.000 -0.000 -0.001 -0.000 -0.000 -0.000 -0.000(-1.95) (-0.79) (-1.50) (-1.00) (-1.35) (-0.96) (-1.39) (-0.75) (-1.45) (-0.92)
Constant 0.115 0.080 -0.127 0.056 0.032 1.581∗ 0.027 0.059 -0.189 0.011(1.17) (0.96) (-1.33) (0.43) (0.38) (2.00) (0.32) (0.68) (-1.93) (0.08)
Observations 958 1481 1486 1481 1486 1481 1486 1481 1486 1481R2 0.061 0.151 0.164 0.155 0.158 . 0.159 0.152 0.169 0.158Adjusted R2 0.049 0.144 0.157 0.148 0.150 . 0.152 0.145 0.160 0.150
t statistics in parentheses∗ p < 0.05
25
7 Conclusions
In this paper, we present evidence on the striking phenomenon of population decline in
Eastern Europe and Central Asia (ECA) using city level data for 3 decades. This analysis
encompasses cities of all sizes in the city size distribution in each country. The evidence shows
that ECA has entered a demographic transition before attaining the typical development
levels associated with it. Then, we extend the model in Brezis and Krugman (1997) to provide
predictions for the consequences of a shock that reduces population. We test empirically the
two main predictions: that population will be redistributed to places with larger markets,
and that emigration will be higher in cities with large market access. Second, that the
state of technology and infrastructure is durable, and that in the short run, population
decline will be accompanied by productivity growth. We confirm the first. Our estimation
finds that market potential has a negative effect on population growth on cities that are
already declining. This is a robust result to different specifications of market potential,
inclusion of instrumental variables to control for market potential endogeneity, and different
treatments of the error structure in estimation. This new empirical results contrasts with the
existing literature (see for example Henderson and Wang (2007)) and provides justification
for allowing heterogeneity in the reaction of urban systems to positive and negative shocks.
The second result, whether productivity is increased after population decreases, gives us
mixed results.
References
Andrews, Donald WK, “Tests for parameter instability and structural change with un-
known change point,” Econometrica: Journal of the Econometric Society, 1993, pp. 821–
856.
, “Tests for parameter instability and structural change with unknown change point: A
corrigendum,” Econometrica, 2003, pp. 395–397.
Au, Chun-Chung and J Vernon Henderson, “Are Chinese cities too small?,” The
Review of Economic Studies, 2006, 73 (3), 549–576.
Bai, Jushan and Pierre Perron, “Estimating and testing linear models with multiple
structural changes,” Econometrica, 1998, pp. 47–78.
26
Billari, Francesco C and Dimitur Filipov, Education and the transition to motherhood:
A comparative analysis of Western Europe, Vienna Institute of Demography, Austrian
Academy of Sciences, 2004.
Brezis, Elise S and Paul R Krugman, “Technology and the life cycle of cities,” Journal
of Economic Growth, 1997, 2 (4), 369–383.
Chauvin, Juan Pablo, Edward Glaeser, Yueran Ma, and Kristina Tobio, “What
is different about urbanization in rich and poor countries? Cities in Brazil, China, India
and the United States,” Journal of Urban Economics, 2017, 98, 17–49.
Chow, Gregory C, “Tests of equality between sets of coefficients in two linear regressions,”
Econometrica: Journal of the Econometric Society, 1960, pp. 591–605.
Desmet, Klaus and J Vernon Henderson, “The geography of development within coun-
tries,” Handbook of Regional and Urban Economics, 2015, 5, 1457–1517.
Fujita, Masahisa and Paul Krugman, “When is the economy monocentric?: von Thunen
and Chamberlin unified,” Regional science and urban Economics, 1995, 25 (4), 505–528.
Gendron-Carrier, Nicolas, Marco Gonzalez-Navarro, Stefano Polloni, and
Matthew A Turner, “Subways and urban air pollution,” 2017.
Geography, Reshaping Economic, “World development report,” The World Bank,
Washington DC, 2009.
Glaeser, Edward L and Giacomo AM Ponzetto, “Did the death of distance hurt
Detroit and help New York?,” Technical Report, National Bureau of Economic Research
2007.
and Joseph Gyourko, “Urban decline and durable housing,” Journal of political econ-
omy, 2005, 113 (2), 345–375.
and Wentao Xiong, “Urban Productivity in the Developing World,” Technical Report,
National Bureau of Economic Research 2017.
Hansen, Bruce E, “Testing for structural change in conditional models,” Journal of Econo-
metrics, 2000, 97 (1), 93–115.
Hanson, Gordon H, “Market potential, increasing returns and geographic concentration,”
Journal of international economics, 2005, 67 (1), 1–24.
27
Head, Keith and Thierry Mayer, “The empirics of agglomeration and trade,” Handbook
of regional and urban economics, 2004, 4, 2609–2669.
Henderson, J Vernon, “Urbanization and the Geography of Development,” 2014.
and Anthony J Venables, “The dynamics of city formation,” Review of Economic
Dynamics, 2009, 12 (2), 233–254.
and Hyoung Gun Wang, “Urbanization and city growth: The role of institutions,”
Regional Science and Urban Economics, 2007, 37 (3), 283–313.
, Tim L Squires, Adam Storeygard, and David N Weil, “The Global Spatial
Distribution of Economic Activity: Nature, History, and the Role of Trade,” Technical
Report, National Bureau of Economic Research 2016.
Henderson, V, A. Storeygard, and D. Weil, “A Bright Idea for Measuring Economic
Growth,” American Economic Review: Papers and Proceedings, 2011, 101 (3), 194–199.
, , and , “Measuring Economic Growth from Outer Spaces,” American Economic
Review, 2012, 102 (2), 994–1028.
Insecurity, Poverty Food and Environment Global GIS Database (FGGD), “Land
use patterns and land cover,” Technical Report, Food and Agriculture Organization of the
United Nations (FAO) 2007.
Jones, P. and Harris I., “Version 3.21 of High Resolution Gridded Data of Month-by-
month Variation of Climate,” Technical Report, The World Bank/U.o.E. Anglia/NCAS
British Atmospheric Data Centre January 2013.
Kuzmenko, LM and MO Soldak, “Monofunctional Cities: Problems, Support Provision
and Development,” Economic Bulletin Donbasu, 2010.
Levcik, Friedrich, “Migration and Employment of Foreign Workers in Comecon Countries
and Their Problems,” Eastern European Economics, 1977, 16 (1), 3–33.
Lewer, Joshua J and Hendrik Van den Berg, “A gravity model of immigration,”
Economics letters, 2008, 99 (1), 164–167.
Light, Matthew A, “What does it mean to control migration? Soviet mobility policies in
comparative perspective,” Law & Social Inquiry, 2012, 37 (2), 395–429.
Markevich, Andrei and Tatiana Mikhailova, “Economic geography of Russia,” 2013.
28
Matthews, Mervyn, The passport society: Controlling movement in Russia and the USSR,
Westview Pr, 1993.
Myrskyla, Mikko, Hans-Peter Kohler, and Francesco C Billari, “Advances in de-
velopment reverse fertility declines,” Nature, 2009, 460 (7256), 741.
Overman, Henry G, Stephen J Redding, and Anthony J Venables, “The economic
geography of trade production and income: a survey of empirics,” 2001.
Owens, Raymond E, Esteban Rossi-Hansberg, and Pierre-Daniel G Sarte, “Re-
thinking Detroit,” 2017.
Quintero, Luis, Paula Restrepo, Mark Roberts, and Benjamin Stewart, “Shedding
light on measuring economic growth from outer-space: learning from the BRICS,” 2017.
Ravenstein, Ernest George, “The laws of migration,” Journal of the royal statistical
society, 1889, 52 (2), 241–305.
Redding, Stephen J and Esteban Rossi-Hansberg, “Quantitative spatial economics,”
Annual Review of Economics, 2017, 9, 21–58.
Roberts, M., B. Stewart, M. Prakash, and K. McWilliams, Global Night Time Lights
Urban Extents and Growth Patterns Product. Alpha Version World Bank 2015.
Schumpeter, Joseph, “A. 1942. Capitalism,” Socialism and Democracy. New York. Harper
and Row, 1975.
The Heritage Foundation, “Index of Economic Freedom,” Technical Report 2018.
The World Bank, “Ease of Doing Business Index,” Technical Report 2018.
Thomas, CM and WE Featherstone, “Validation of Vincenty’s formulas for the geodesic
using a new fourth-order extension of Kivioja’s formula,” Journal of Surveying engineering,
2005, 131 (1), 20–26.
(USGS), US Geological Survey, “MODIS Normalized Difference Vegetation Index
(NDVI).”
Vincenty, Thaddeus, “Direct and inverse solutions of geodesics on the ellipsoid with ap-
plication of nested equations,” Survey review, 1975, 23 (176), 88–93.
29
Figure 3: Growth of Kiev into an agglomeration
von Thunen, Johann Heinrich and Peter Geoffrey Hall, Von Thunen’s isolated state
: an English edition of Der isolierte Staat, 1st ed ed., Oxford ; London : Pergamon Press,
1966. Abridged and translated from the 2nd German ed.
A Population Shock
30
End of Communism
Turks forced displacement .9.9
51
1.05
1.1
1.15
Tren
d C
oeffi
cien
t Wal
d St
at
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Bulgaria
End of Communism
Balkan Wars 11.
051.
11.
151.
2Tr
end
Coe
ffici
ent W
ald
Stat
-20
-15
-10
-50
Popu
latio
n re
lativ
e to
196
0
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Serbia
End of Communism
11.
051.
11.
151.
21.
25Tr
end
Coe
ffici
ent W
ald
Stat
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Ukraine
End of Communism 11.
051.
11.
151.
21.
25Tr
end
Coe
ffici
ent W
ald
Stat
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Russia
End of Communism 11.
051.
11.
151.
21.
25Tr
end
Coe
ffici
ent W
ald
Stat
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Belarus
End of Communism 11.
051.
11.
151.
21.
25Tr
end
Coe
ffici
ent W
ald
Stat
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Romania
Figure 4: Population Structural Break. Declining Population Pattern.
31
End of CommunismPro market reforms 1
1.1
1.2
1.3
Tren
d C
oeffi
cien
t Wal
d St
at
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Poland
End of CommunismExpulsion of Abkhazia 11.
11.
21.
31.
4Tr
end
Coe
ffici
ent W
ald
Stat
-50
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Georgia
End of Communism
First election
11.
11.
21.
31.
41.
5Tr
end
Coe
ffici
ent W
ald
Stat
-50
-40
-30
-20
-10
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Moldova
End of Communism
11.
21.
41.
61.
82
Tren
d C
oeffi
cien
t Wal
d St
at
-150
-100
-50
0Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Albania
Figure 5: Population Structural Break. Declining Population Pattern (continued).
End of Communism
11.
21.
41.
61.
8Tr
end
Coe
ffici
ent W
ald
Stat
-15
-10
-50
Popu
latio
n re
lativ
e to
196
0
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Kazakhstan
End of Communism
11.
52
2.5
3Tr
end
Coe
ffici
ent W
ald
Stat
-10
-50
5Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Kyrgyz
Figure 6: Population Structural Break. Slowing Population Pattern.
32
11.
52
2.5
3Tr
end
Coe
ffici
ent W
ald
Stat
-20
24
6Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Turkey
End of Communism
11.
52
2.5
33.
5Tr
end
Coe
ffici
ent W
ald
Stat
-6-4
-20
2Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Uzbekistan
End of Communism
12
34
Tren
d C
oeffi
cien
t Wal
d St
at
-20
24
Popu
latio
n re
lativ
e to
196
0
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Tajikistan
End of Communism
11.
051.
11.
15Tr
end
Coe
ffici
ent W
ald
Stat
-20
24
6Po
pula
tion
rela
tive
to 1
960
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
Germany
11.
051.
11.
151.
21.
25Tr
end
Coe
ffici
ent W
ald
Stat
-50
510
Popu
latio
n re
lativ
e to
196
0
1960
1965
1970
1975
1980
1985
1990
1995
2000
2005
2010
2015
Year
Trend Coefficient Wald Stat Population Relative to 1960
UK
Figure 7: Population Structural Break. Growing Population Pattern.
33
B Transition to a Market Economies
4050
6070
80Ec
onom
ic Fr
eedo
m In
dex
1995 1999 2003 2007 2011 2015 2019
East Asia & Pacific ECAHigh income: OECD Latin America & CaribbeanMiddle East & North Africa South AsiaSub-Saharan Africa
Figure 8: Economics Freedom Index. The Economic Freedom Index measures how econom-ically free societies are, where freedom is understood as no government obstruction to thefree movement of labor, capital, and goods (The Heritage Foundation, 2018).
3040
5060
7080
Ease
of D
oing
Bus
ines
s In
dex
2004 2006 2008 2010 2012 2014 2016 2018
East Asia & Pacific ECAHigh income: OECD Latin America & CaribbeanMiddle East & North Africa South AsiaSub-Saharan Africa
Figure 9: Ease of Doing Business Index. The Ease of Doing Business Index measures howfair and friendly economies are to medium and small private firms (The World Bank, 2018).
C Additional Tables
34
Country Period Population annual change %population % of cities shrinkingTotal Urban shrinking cities all >30k >100k
Albania 1989-2001 -0.2 1.08 14.12 27.42 10 0Albania 2001-2011 -0.55 1.65 47.25 82.26 60 0
1989-2001 -0.16 0.47 - - - -Belarus
2001-2014 -0.39 0.22 26.87 70.8 43.33 21.431989-2001 -0.87 -0.47 - - - -
Bulgaria2001-2013 -0.81 -0.42 67.66 94.7 91.11 55.561989-2002 -0.67 -1.06 96.45 94.44 87.5 100
Georgia2002-2014 0.2 0.35 6.92 31.48 0 01989-2001 0.4 0.4 - - - -
Germany2001-2014 -0.11 0.06 45.19 61.14 55.5 38.961989-1999 -0.62 -0.68 59.8 69.86 67.92 68.18
Kazakhstan1999-2015 0.85 0.54 5.8 21.92 13.21 01989-1999 1.25 0.52 25.76 75.61 68 75
Kyrgyz Rep.1999-2013 1.21 1.2 12.49 42.86 33.33 01989-2000 -0.04 -0.17 74.97 55.77 80 100
Moldova2000-2015 -0.16 -0.31 40.88 81.13 80 01989-2003 0.06 0.13 - - - -
Poland2003-2011 -0.04 -0.21 64.06 52.94 68.21 82.051992-2002 -0.51 -0.71 95.52 93.57 95.45 100
Romania2002-2011 -0.93 -0.73 90.41 90.86 92.54 901989-2000 -0.01 0.004 50.15 65.19 54.51 50.92
Russia2000-2010 -0.27 -0.23 42.15 73.61 63.04 48.171991-2002 -0.09 0.43 50.9 46.37 55 60
Serbia2002-2011 -0.36 -0.03 50.94 71.91 51.28 11.111989-2000 1.75 0.02 - - - -
Tajikistan2000-2014 2.05 2.03 2.38 5.26 7.69 01989-2000 1.61 2.73 - - - -
Turkey2000-2012 1.31 2.19 7.77 59.23 12.77 4.171991-2001 0.29 0.36 21.33 28.24 28.62 29.41
UK2001-2011 0.65 0.98 8.32 14.95 12.14 6.251989-2001 -0.43 -0.34 83.29 80 79.41 73.33
Ukraine2001-2013 -0.59 -0.35 75.48 82.06 81.02 75.561990-2000 1.87 1.15 11.88 10.17 9.84 22.22
Uzbekistan2000-2014 1.56 1.33 5.85 11.86 8.2 11.11
Table 5: Descriptive Statistics of the population dataset. Source: National Statistics Offices.
35
City Population Growth 2000-2010National Natural Increase Rate 0.395***
(60.20)National Net Migration Rate 1.331***
(68.87)Cons -3.751
(61.97)R2 0.52N 5,476** p<0.05; *** p<0.01
Table 6: Migration and fertility role in explaining variation in individual city population.Cities in Europe and Central Asia: Regional Database and TransMonEE Database.
36
Country L(NLS) Constant Observations R2
Albania 1.24** -0.48 12 0.80-0.37 -4.01
Belarus 1.25** -6.43 6 0.84-0.28 -3.59
Bulgaria 1.17*** -6.04*** 140 0.72-0.06 -0.65
Georgia 0.88* -1.19 7 0.6-0.36 -3.22
Germany 0.72*** 0.95*** 1,980 0.41-0.02 -0.2
Kazakhstan* 0.50** 21.20*** 28 0.13-0.24 -2.83
Kyrgyz Republic 0.92*** 0.21 7 0.66-0.15 -1.58
Poland 0.61*** 0.87*** 325 0.94-0.03 -0.31
Romania 1.07*** -4.92*** 210 0.67-0.07 -0.74
Russia 0.33*** 6.82*** 456 0.98-0.03 -0.38
Serbia* 1.26*** -1.87 25 0.83-0.22 -2.24
Tajikistan* 0.92*** 13.17*** 8 0.99-0.02 -0.16
Turkey 1.40*** 0.21 52 0.74-0.14 -1.7
UK 0.56*** 2.56*** 840 0.28-0.04 -0.44
Ukraine* 0.85*** -0.69 135 0.5-0.08 -0.87
Uzbekistan 1.01*** 1.94 39 0.95-0.13 -1.44
Table 7: Correlation of NLS and regional GDP: using Nighttime lights (NLS) as a proxy foreconomic activity.
37
C.1 Robustness: NL based market access and geodetic distance
38
Table 8: All pooled OLS - IV (NL-Geodetic Distance)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)OLS IV OLS IV OLS IV OLS IV est9 est10
Market
local mkt 0.013∗ -0.001 0.012∗ 0.006∗ 0.012∗ -0.001 0.007∗ -0.005∗ 0.007∗ 0.001(7.84) (-0.40) (7.59) (2.58) (7.38) (-0.13) (4.77) (-2.05) (4.57) (0.47)
mkt access -0.014∗ 0.070∗ -0.010∗ 0.026∗ -0.004 0.111∗ -0.002 0.066∗ 0.004 0.036∗
(-4.31) (5.67) (-3.19) (2.68) (-1.23) (3.61) (-0.63) (6.31) (1.23) (4.20)
mkt access×I{∆countryp < 0} -0.004∗ -0.005∗ -0.003∗ -0.004∗
(-6.13) (-5.55) (-5.27) (-4.93)
mkt access×I{∆regionp < 0} -0.007∗ -0.084∗ -0.002∗ -0.004∗
(-12.19) (-2.61) (-4.63) (-6.62)
mkt access×I{∆p < 0} -0.015∗ -0.016∗ -0.014∗ -0.015∗
(-26.95) (-24.71) (-23.91) (-25.40)
Former communist -0.008 0.053∗ -0.024∗ -0.000 -0.006 0.016 -0.018∗ 0.029∗ -0.028∗ -0.010(-0.91) (4.33) (-2.67) (-0.02) (-0.71) (0.49) (-2.34) (2.88) (-3.63) (-0.82)
Population Fundamentals
nat. pop ∆ 0.010∗ 0.012∗ 0.008∗ 0.009∗ 0.009∗ -0.005 0.007∗ 0.008∗ 0.006∗ 0.006∗
(26.56) (17.29) (20.00) (15.26) (23.12) (-0.78) (20.81) (13.81) (15.11) (10.62)
net migration 0.010∗ -0.004 0.007∗ 0.000 0.006∗ -0.043∗ 0.005∗ -0.007∗ 0.001 -0.005∗
(8.73) (-1.53) (5.34) (0.06) (4.87) (-2.69) (4.79) (-2.74) (1.07) (-2.15)
Urban Structure
Agglomeration -0.030∗ -0.018∗ -0.025∗ -0.019∗ -0.026∗ 0.017 -0.026∗ -0.017∗ -0.022∗ -0.016∗
(-4.01) (-2.21) (-3.44) (-2.56) (-3.61) (0.69) (-4.03) (-2.32) (-3.39) (-2.34)
Capital city 0.026 0.131∗ 0.028 0.081 0.016 -0.159 0.017 0.080∗ 0.016 0.037(0.86) (2.79) (0.94) (1.90) (0.55) (-1.28) (0.65) (2.26) (0.61) (1.17)
Location Fundamentals
pct forest -0.000 -0.000∗ -0.000 -0.000 -0.000 0.000 0.000 -0.000 0.000 0.000(-0.61) (-2.93) (-0.61) (-1.83) (-0.43) (0.47) (0.93) (-1.08) (0.90) (0.09)
precipitation 0.000 0.000∗ -0.000 0.000 -0.000 -0.000 -0.000 0.000∗ -0.000 -0.000(0.60) (3.36) (-0.70) (0.89) (-0.16) (-1.43) (-0.49) (2.50) (-1.82) (-0.26)
temperature 0.001∗ -0.000 0.002∗ 0.002∗ 0.001 -0.001 -0.001 -0.002∗ 0.000 -0.000(2.00) (-0.27) (4.01) (3.06) (1.68) (-0.78) (-1.32) (-3.11) (0.66) (-0.14)
land usability -0.000∗ -0.001∗ -0.000 -0.001∗ -0.000 0.001 -0.000 -0.001∗ -0.000 -0.000∗
(-2.21) (-5.22) (-1.66) (-2.95) (-1.66) (0.76) (-1.57) (-4.54) (-0.92) (-2.24)
Constant 0.081∗ -0.476∗ 0.087∗ -0.147∗ 0.026 -0.603∗ 0.088∗ -0.351∗ 0.073∗ -0.128∗
(2.42) (-5.56) (2.63) (-2.31) (0.80) (-3.21) (2.99) (-4.89) (2.47) (-2.27)Observations 2506 2496 2506 2496 2506 2496 2506 2496 2506 2496R2 0.395 0.230 0.404 0.370 0.429 . 0.531 0.426 0.541 0.517Adjusted R2 0.392 0.226 0.401 0.367 0.426 . 0.529 0.424 0.538 0.514
t statistics in parentheses∗ p < 0.05
39
Table 9: Russia OLS - IV (NL-Geodetic Distance)
(1) (2) (3) (4) (5) (6) (7) (8)OLS IV OLS IV OLS IV OLS IV
Market
local mkt 0.017∗ 0.015∗ 0.014∗ 0.014∗ 0.008∗ 0.006∗ 0.007∗ 0.006∗
(7.39) (6.63) (6.57) (3.44) (3.85) (3.23) (3.76) (3.07)
mkt access -0.020∗ 0.008 -0.003 0.017 0.004 0.020∗ 0.009 0.027∗
(-2.85) (0.53) (-0.41) (0.65) (0.72) (1.96) (1.60) (2.63)
mkt access×I{∆regionp < 0} -0.007∗ -0.003 -0.003∗ -0.003∗
(-9.40) (-0.41) (-4.44) (-6.01)
mkt access×I{∆p < 0} -0.013∗ -0.014∗ -0.012∗ -0.013∗
(-18.53) (-15.72) (-15.86) (-14.96)
Industrial Structure
Monotown -0.035∗ -0.038∗ -0.022∗ -0.031 -0.014∗ -0.014∗ -0.010 -0.010(-4.48) (-4.81) (-2.83) (-1.63) (-2.08) (-2.11) (-1.47) (-1.43)
founded 1922-1991 0.007 0.012 -0.006 0.006 0.002 0.004 -0.003 -0.001(0.81) (0.87) (-0.65) (0.27) (0.28) (0.40) (-0.41) (-0.12)
Urban Structure
Agglomeration -0.018 -0.016 -0.007 -0.011 -0.013 -0.012 -0.008 -0.007(-1.39) (-1.58) (-0.60) (-0.68) (-1.21) (-1.32) (-0.82) (-0.81)
Capital city 0.041 0.044∗ 0.012 0.030 0.015 0.016 0.005 0.005(0.48) (3.37) (0.15) (0.89) (0.22) (1.38) (0.07) (0.42)
Location Fundamentals
pct forest -0.000 -0.000 0.000 -0.000 -0.000 -0.000 0.000 -0.000(-0.53) (-1.10) (0.06) (-0.73) (-0.20) (-0.68) (0.06) (-0.38)
precipitation -0.000 -0.000 -0.000 0.000 -0.000 -0.000 -0.000 0.000(-1.01) (-0.15) (-0.12) (0.20) (-1.01) (-0.37) (-0.56) (0.11)
temperature 0.003∗ 0.002 0.002∗ 0.002 0.002∗ 0.001 0.001∗ 0.001(4.49) (1.79) (3.39) (1.17) (2.67) (1.00) (2.29) (0.64)
land usability -0.000∗ -0.000∗ -0.000∗ -0.000∗ -0.000 -0.000 -0.000 -0.000(-2.43) (-2.58) (-2.15) (-2.40) (-1.14) (-1.43) (-1.11) (-1.48)
Constant 0.152∗ -0.093 0.009 -0.172 0.026 -0.117 -0.025 -0.185(2.15) (-0.69) (0.14) (-0.77) (0.44) (-1.25) (-0.42) (-1.95)
Observations 697 697 697 697 697 697 697 697R2 0.181 0.162 0.274 0.228 0.454 0.448 0.469 0.462Adjusted R2 0.169 0.150 0.263 0.216 0.445 0.439 0.460 0.453
t statistics in parentheses∗ p < 0.05
40
C.2 Robustness: Population base market access and driving dis-
tance
41
Table 10: All pooled OLS - IV (Population-Driving Distance)
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)OLS IV OLS IV OLS IV OLS IV est9 est10
Market
local mkt 0.018∗ 0.012∗ 0.017∗ 0.011∗ 0.017∗ 0.012∗ 0.005 0.001 0.004 -0.002(4.67) (3.06) (4.49) (2.68) (4.66) (2.77) (1.49) (0.25) (1.07) (-0.11)
mkt access 0.005 0.029∗ 0.003 0.035∗ 0.018∗ 0.029∗ 0.037∗ 0.053∗ 0.037∗ 0.064∗
(0.91) (3.00) (0.66) (3.87) (3.44) (2.98) (8.00) (5.97) (7.78) (5.35)
mkt access×I{∆countryp < 0} 0.003∗ 0.003 0.008∗ 0.007∗
(1.99) (1.91) (5.67) (4.16)
mkt access×I{∆regionp < 0} -0.016∗ 0.003 -0.004∗ -0.005∗
(-11.79) (0.24) (-2.94) (-2.67)
mkt access×I{∆p < 0} -0.043∗ -0.046∗ -0.042∗ -0.045∗
(-35.75) (-37.50) (-33.53) (-21.21)
Former communist 0.154∗ 0.183∗ 0.160∗ 0.196∗ 0.154∗ 0.186∗ 0.120∗ 0.133∗ 0.135∗ 0.158∗
(15.19) (13.25) (15.17) (14.79) (15.40) (9.98) (13.10) (10.74) (14.24) (8.60)
Population Fundamentals
nat. pop ∆ 0.008∗ 0.008∗ 0.008∗ 0.009∗ 0.006∗ 0.009∗ 0.004∗ 0.004∗ 0.005∗ 0.005∗
(17.07) (16.68) (14.95) (15.60) (13.00) (4.91) (9.63) (8.89) (10.63) (11.73)
net migration 0.020∗ 0.019∗ 0.021∗ 0.020∗ 0.016∗ 0.020∗ 0.010∗ 0.009∗ 0.012∗ 0.010∗
(16.63) (14.62) (15.96) (13.93) (12.86) (6.18) (9.27) (7.56) (10.14) (8.07)
Urban Structure
center of agglom. -0.025∗ -0.020 -0.027∗ -0.021 -0.021 -0.020 -0.022∗ -0.018 -0.027∗ -0.021(-2.09) (-1.65) (-2.28) (-1.77) (-1.76) (-1.65) (-2.07) (-1.68) (-2.54) (-1.69)
periphery of agglom. 0.032∗ 0.035∗ 0.029∗ 0.033∗ 0.034∗ 0.035∗ 0.011 0.011 0.005 0.007(4.61) (5.01) (4.12) (4.61) (4.94) (4.94) (1.83) (1.82) (0.77) (0.68)
Secondary city 0.063∗ 0.070∗ 0.064∗ 0.072∗ 0.063∗ 0.070∗ 0.060∗ 0.063∗ 0.064∗ 0.069∗
(6.46) (7.00) (6.60) (7.30) (6.57) (6.84) (6.83) (7.09) (7.30) (2.80)
Capital city 0.096 0.164∗ 0.098 0.173∗ 0.082 0.172∗ 0.089 0.110 0.092 0.120(1.81) (2.18) (1.86) (2.30) (1.57) (2.09) (1.86) (1.62) (1.94) (1.43)
Location Fundamentals
pct forest -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 0.000 0.000 0.000 0.000(-0.17) (-0.84) (-0.26) (-1.05) (-0.42) (-0.84) (1.34) (0.95) (1.00) (0.45)
precipitation 0.000 0.000 0.000 0.000∗ -0.000 0.000 -0.000 -0.000 0.000 0.000(0.28) (1.44) (0.64) (2.11) (-0.38) (1.17) (-0.91) (-0.15) (0.04) (1.42)
temperature 0.006∗ 0.006∗ 0.006∗ 0.005∗ 0.006∗ 0.006∗ 0.003∗ 0.003∗ 0.002∗ 0.002∗
(10.42) (8.95) (9.44) (7.95) (9.56) (8.28) (6.21) (4.89) (4.16) (3.66)
land usability 0.000 0.000 0.000 0.000 0.000∗ 0.000 0.000 0.000 0.000 0.000(1.32) (1.11) (1.24) (0.94) (2.46) (0.58) (1.49) (1.29) (1.54) (1.33)
Constant -0.366∗ -0.453∗ -0.368∗ -0.481∗ -0.406∗ -0.456∗ -0.236∗ -0.269∗ -0.252∗ -0.328∗
(-7.45) (-7.61) (-7.49) (-8.36) (-8.35) (-7.51) (-5.32) (-5.02) (-5.68) (-2.85)Observations 5388 5343 5388 5343 5388 5343 5388 5343 5388 5343R2 0.147 0.143 0.148 0.141 0.169 0.132 0.311 0.307 0.316 0.309Adjusted R2 0.145 0.140 0.146 0.139 0.167 0.129 0.309 0.305 0.314 0.307
t statistics in parentheses∗ p < 0.05
42
Table 11: Russia OLS - IV (Population-Driving Distance)
(1) (2) (3) (4) (5) (6) (7) (8)OLS IV OLS IV OLS IV OLS IV
Market
local mkt -0.058∗ -0.057 -0.063∗ -0.062 -0.070∗ -0.069 -0.070∗ -0.069(-5.14) (-0.93) (-5.59) (-1.17) (-6.39) (-1.11) (-6.37) (-1.14)
mkt access 0.012 -0.007 0.021 0.009 0.040∗ 0.022 0.040∗ 0.022(0.91) (-0.21) (1.61) (0.14) (3.16) (0.81) (3.14) (0.69)
mkt access×I{∆regionp < 0} -0.014∗ -0.012 -0.000 -0.001(-3.47) (-0.36) (-0.11) (-0.18)
mkt access×I{∆p < 0} -0.040∗ -0.041∗ -0.040∗ -0.040∗
(-9.31) (-6.13) (-8.59) (-4.56)
Urban Structure
center of agglom. 0.045 0.044 0.054 0.053 0.048 0.047 0.048 0.048(1.15) (0.72) (1.40) (1.07) (1.28) (0.78) (1.28) (0.84)
periphery of agglom. 0.133∗ 0.134∗ 0.125∗ 0.126 0.090∗ 0.090 0.090∗ 0.090(5.13) (1.96) (4.82) (1.50) (3.55) (1.50) (3.55) (1.47)
Secondary city 0.145∗ 0.144 0.151∗ 0.150 0.137∗ 0.136 0.137∗ 0.137(4.59) (1.47) (4.80) (1.68) (4.51) (1.43) (4.51) (1.48)
Capital city 0.443 0.448 0.430 0.435 0.381 0.385 0.381 0.385(1.62) (1.45) (1.58) (1.30) (1.45) (1.31) (1.45) (1.30)
Location Fundamentals
pct forest -0.001∗ -0.001 -0.001∗ -0.001 -0.001∗ -0.001 -0.001∗ -0.001(-2.14) (-1.38) (-2.08) (-1.49) (-2.09) (-1.33) (-2.09) (-1.36)
precipitation 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000(0.71) (0.67) (0.79) (0.73) (0.68) (0.60) (0.69) (0.58)
temperature 0.003 0.004 0.002 0.003 0.002 0.003 0.002 0.003(1.45) (1.09) (1.27) (0.59) (0.87) (0.85) (0.86) (0.77)
land usability -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000 -0.000(-1.09) (-1.94) (-0.81) (-1.12) (-0.38) (-0.49) (-0.37) (-0.49)
Industrial Structure
Monotown -0.033 -0.031∗ -0.017 -0.018 0.003 0.006 0.003 0.007(-1.58) (-2.53) (-0.81) (-0.52) (0.15) (0.35) (0.17) (0.49)
founded 1922-1991 0.060∗ 0.057 0.050∗ 0.049 0.046∗ 0.042 0.046∗ 0.041(2.72) (1.43) (2.25) (0.86) (2.16) (1.16) (2.13) (1.07)
Constant 0.587∗ 0.677 0.605∗ 0.657 0.643∗ 0.739 0.644∗ 0.740(3.81) (0.86) (3.94) (0.80) (4.34) (0.95) (4.34) (0.93)
Observations 1071 1071 1071 1071 1071 1071 1071 1071R2 0.080 0.078 0.090 0.089 0.150 0.148 0.150 0.148Adjusted R2 0.069 0.068 0.079 0.078 0.139 0.137 0.138 0.136
t statistics in parentheses∗ p < 0.05
43